précédentsommairesuivant

2. The Standard C++ Library

2-0. Introduction

Standard C++ not only incorporates all the Standard C libraries (with small additions and changes to support type safety), it also adds libraries of its own. These libraries are far more powerful than those in Standard C; the leverage you get from them is analogous to the leverage you get from changing from C to C++.

This section of the book gives you an in-depth introduction to key portions of the Standard C++ library.

The most complete and also the most obscure reference to the full libraries is the Standard itself. Bjarne Stroustrup's The C++ Programming Language, Third Edition (Addison Wesley, 2000) remains a reliable reference for both the language and the library. The most celebrated library-only reference is The C++ Standard Library: A Tutorial and Reference, by Nicolai Josuttis (Addison Wesley, 1999). The goal of the chapters in this part of the book is to provide you with an encyclopedia of descriptions and examples so that you'll have a good starting point for solving any problem that requires the use of the Standard libraries. However, some techniques and topics are rarely used and are not covered here. If you can't find it in these chapters, reach for the other two books; this book is not intended to replace those books but rather to complement them. In particular, we hope that after going through the material in the following chapters you'll have a much easier time understanding those books.

You will notice that these chapters do not contain exhaustive documentation describing every function and class in the Standard C++ library. We've left the full descriptions to others; in particular to P.J. Plauger's Dinkumware C/C++ Library Reference at http://www.dinkumware.com. This is an excellent online source of standard library documentation in HTML format that you can keep resident on your computer and view with a Web browser whenever you need to look something up. You can view this online or purchase it for local viewing. It contains complete reference pages for the both the C and C++ libraries (so it's good to use for all your Standard C/C++ programming questions). Electronic documentation is effective not only because you can always have it with you, but also because you can do an electronic search.

When you're actively programming, these resources should satisfy your reference needs (and you can use them to look up anything in this chapter that isn't clear to you). Appendix A lists additional references.

The first chapter in this section introduces the Standard C++ string class, which is a powerful tool that simplifies most of the text-processing chores you might have. Chances are, anything you've done to character strings with lines of code in C can be done with a member function call in the string class.

Chapter 4 covers the iostreams library, which contains classes for processing input and output with files, string targets, and the system console.

Although Chapter 5, “Templates in Depth,” is not explicitly a library chapter, it is necessary preparation for the two chapters that follow. In Chapter 6 we examine the generic algorithms offered by the Standard C++ library. Because they are implemented with templates, these algorithms can be applied to any sequence of objects. Chapter 7 covers the standard containers and their associated iterators. We cover algorithms first because they can be fully explored by using only arrays and the vector container (which we have been using since early in Volume 1). It is also natural to use the standard algorithms in connection with containers, so it's good to be familiar with the algorithms before studying the containers.

2-1. Strings in Depth

String processing with character arrays is one of the biggest time-wasters in C. Character arrays require the programmer to keep track of the difference between static quoted strings and arrays created on the stack and the heap, and the fact that sometimes you're passing around a char* and sometimes you must copy the whole array.

Especially because string manipulation is so common, character arrays are a great source of misunderstandings and bugs. Despite this, creating string classes remained a common exercise for beginning C++ programmers for many years. The Standard C++ library string class solves the problem of character array manipulation once and for all, keeping track of memory even during assignments and copy-constructions. You simply don't need to think about it.

This chapter(31) examines the Standard C++ string class, beginning with a look at what constitutes a C++ string and how the C++ version differs from a traditional C character array. You'll learn about operations and manipulations using string objects, and you'll see how C++ strings accommodate variation in character sets and string data conversion.

Handling text is one of the oldest programming applications, so it's not surprising that the C++ string draws heavily on the ideas and terminology that have long been used in C and other languages. As you begin to acquaint yourself with C++ strings, this fact should be reassuring. No matter which programming idiom you choose, there are three common things you want to do with a string:

  • Create or modify the sequence of characters stored in the string.
  • Detect the presence or absence of elements within the string.
  • Translate between various schemes for representing string characters.

You'll see how each of these jobs is accomplished using C++ string objects.

2-1-1. What's in a string?

In C, a string is simply an array of characters that always includes a binary zero (often called the null terminator) as its final array element. There are significant differences between C++ strings and their C progenitors. First, and most important, C++ strings hide the physical representation of the sequence of characters they contain. You don't need to be concerned about array dimensions or null terminators. A string also contains certain “housekeeping” information about the size and storage location of its data. Specifically, a C++ string object knows its starting location in memory, its content, its length in characters, and the length in characters to which it can grow before the string object must resize its internal data buffer. C++ strings thus greatly reduce the likelihood of making three of the most common and destructive C programming errors: overwriting array bounds, trying to access arrays through uninitialized or incorrectly valued pointers, and leaving pointers “dangling” after an array ceases to occupy the storage that was once allocated to it.

The exact implementation of memory layout for the string class is not defined by the C++ Standard. This architecture is intended to be flexible enough to allow differing implementations by compiler vendors, yet guarantee predictable behavior for users. In particular, the exact conditions under which storage is allocated to hold data for a string object are not defined. String allocation rules were formulated to allow but not require a reference-counted implementation, but whether or not the implementation uses reference counting, the semantics must be the same. To put this a bit differently, in C, every char array occupies a unique physical region of memory. In C++, individual string objects may or may not occupy unique physical regions of memory, but if reference counting avoids storing duplicate copies of data, the individual objects must look and act as though they exclusively own unique regions of storage. For example:

 
Sélectionnez
//: C03:StringStorage.h
#ifndef STRINGSTORAGE_H
#define STRINGSTORAGE_H
#include <iostream>
#include <string>
#include "../TestSuite/Test.h"
using std::cout;
using std::endl;
using std::string;
 
class StringStorageTest : public TestSuite::Test {
public:
  void run() {
    string s1("12345");
    // This may copy the first to the second or
    // use reference counting to simulate a copy:
    string s2 = s1;
    test_(s1 == s2);
    // Either way, this statement must ONLY modify s1:
    s1[0] = '6';
    cout << "s1 = " << s1
<< endl;  // 62345
    cout << "s2 = " << s2
<< endl;  // 12345
    test_(s1 != s2);
  }
};
#endif //
STRINGSTORAGE_H ///:~
 
Sélectionnez
//: C03:StringStorage.cpp
//{L} ../TestSuite/Test
#include "StringStorage.h"
 
int main() {
  StringStorageTest t;
  t.run();
  return t.report();
} ///:~

We say that an implementation that only makes unique copies when a string is modified uses a copy-on-write strategy. This approach saves time and space when strings are used only as value parameters or in other read-only situations.

Whether a library implementation uses reference counting or not should be transparent to users of the string class. Unfortunately, this is not always the case. In multithreaded programs, it is practically impossible to use a reference-counting implementation safely.(32)

2-1-2. Creating and initializing C++ strings

Creating and initializing strings is a straightforward proposition and fairly flexible. In the SmallString.cpp example below, the first string, imBlank, is declared but contains no initial value. Unlike a C char array, which would contain a random and meaningless bit pattern until initialization, imBlank does contain meaningful information. This string object is initialized to hold “no characters” and can properly report its zero length and absence of data elements using class member functions.

The next string, heyMom, is initialized by the literal argument “Where are my socks?” This form of initialization uses a quoted character array as a parameter to the string constructor. By contrast, standardReply is simply initialized with an assignment. The last string of the group, useThisOneAgain, is initialized using an existing C++ string object. Put another way, this example illustrates that string objects let you do the following:

  • Create an empty string and defer initializing it with character data.
  • Initialize a string by passing a literal, quoted character array as an argument to the constructor.
  • Initialize a string using the equal sign (=).
  • Use one string to initialize another.
 
Sélectionnez
//: C03:SmallString.cpp
#include <string>
using namespace std;
 
int main() {
  string imBlank;
  string heyMom("Where are my socks?");
  string standardReply = "Beamed into deep "
    "space on wide angle dispersion?";
  string useThisOneAgain(standardReply);
} ///:~

These are the simplest forms of string initialization, but variations offer more flexibility and control. You can do the following:

  • Use a portion of either a C char array or a C++ string.
  • Combine different sources of initialization data using operator+.
  • Use the string object's substr( ) member function to create a substring.

Here's a program that illustrates these features:

 
Sélectionnez
//: C03:SmallString2.cpp
#include <string>
#include <iostream>
using namespace std;
 
int main() {
  string s1("What is the sound of one clam
napping?");
  string s2("Anything worth doing is worth
overdoing.");
  string s3("I saw Elvis in a UFO");
  // Copy the first 8 chars:
  string s4(s1, 0, 8);
  cout << s4 << endl;
  // Copy 6 chars from the middle of the source:
  string s5(s2, 15, 6);
  cout << s5 << endl;
  // Copy from middle to end:
  string s6(s3, 6, 15);
  cout << s6 << endl;
  // Copy many different things:
  string quoteMe = s4 + "that" +
  // substr() copies 10 chars at element 20
  s1.substr(20, 10) + s5 +
  // substr() copies up to either 100 char
  // or eos starting at element 5
  "with" + s3.substr(5, 100) +
  // OK to copy a single char this way
  s1.substr(37, 1);
  cout << quoteMe << endl;
} ///:~

The string member function substr( ) takes a starting position as its first argument and the number of characters to select as the second argument. Both arguments have default values. If you say substr( ) with an empty argument list, you produce a copy of the entire string, so this is a convenient way to duplicate a string.

Here's the output from the program:

 
Sélectionnez
What is
doing
Elvis in a UFO
What is that one clam doing
with Elvis in a UFO?

Notice the final line of the example. C++ allows string initialization techniques to be mixed in a single statement, a flexible and convenient feature. Also notice that the last initializer copies just one character from the source string.

Another slightly more subtle initialization technique involves the use of the string iterators string::begin( ) and string::end( ). This technique treats a string like a container object (which you've seen primarily in the form of vector so far—you'll see many more containers in Chapter 7), which uses iterators to indicate the start and end of a sequence of characters. In this way you can hand a string constructor two iterators, and it copies from one to the other into the new string:

 
Sélectionnez
//: C03:StringIterators.cpp
#include <string>
#include <iostream>
#include <cassert>
using namespace std;
 
int main() {
  string source("xxx");
  string s(source.begin(), source.end());
  assert(s == source);
} ///:~

The iterators are not restricted to begin( ) and end( ); you can increment, decrement, and add integer offsets to them, allowing you to extract a subset of characters from the source string.

C++ strings may not be initialized with single characters or with ASCII or other integer values. You can initialize a string with a number of copies of a single character, however:

 
Sélectionnez
//: C03:UhOh.cpp
#include <string>
#include <cassert>
using namespace std;
 
int main() {
  // Error: no single char inits
  //! string nothingDoing1('a');
  // Error: no integer inits
  //! string nothingDoing2(0x37);
  // The following is legal:
  string okay(5, 'a');
  assert(okay == string("aaaaa"));
} ///:~

The first argument indicates the number of copies of the second argument to place in the string. The second argument can only be a single char, not a char array.

2-1-3. Operating on strings

If you've programmed in C, you are accustomed to the family of functions that write, search, modify, and copy char arrays. There are two unfortunate aspects of the Standard C library functions for handling char arrays. First, there are two loosely organized families of them: the “plain” group, and the ones that require you to supply a count of the number of characters to be considered in the operation at hand. The roster of functions in the C char array library shocks the unsuspecting user with a long list of cryptic, mostly unpronounceable names. Although the type and number of arguments to the functions are somewhat consistent, to use them properly you must be attentive to details of function naming and parameter passing.

The second inherent trap of the standard C char array tools is that they all rely explicitly on the assumption that the character array includes a null terminator. If by oversight or error the null is omitted or overwritten, there's little to keep the C char array functions from manipulating the memory beyond the limits of the allocated space, sometimes with disastrous results.

C++ provides a vast improvement in the convenience and safety of string objects. For purposes of actual string handling operations, there are about the same number of distinct member function names in the string class as there are functions in the C library, but because of overloading the functionality is much greater. Coupled with sensible naming practices and the judicious use of default arguments, these features combine to make the string class much easier to use than the C library char array functions.

2-1-3-1. Appending, inserting, and concatenating strings

One of the most valuable and convenient aspects of C++ strings is that they grow as needed, without intervention on the part of the programmer. Not only does this make string-handling code inherently more trustworthy, it also almost entirely eliminates a tedious “housekeeping” chore—keeping track of the bounds of the storage where your strings live. For example, if you create a string object and initialize it with a string of 50 copies of ‘X', and later store in it 50 copies of “Zowie”, the object itself will reallocate sufficient storage to accommodate the growth of the data. Perhaps nowhere is this property more appreciated than when the strings manipulated in your code change size and you don't know how big the change is. The string member functions append( ) and insert( ) transparently reallocate storage when a string grows:

 
Sélectionnez
//: C03:StrSize.cpp
#include <string>
#include <iostream>
using namespace std;
 
int main() {
  string bigNews("I saw Elvis in a UFO. ");
  cout << bigNews << endl;
  // How much data have we actually got?
  cout << "Size = " <<
bigNews.size() << endl;
  // How much can we store without reallocating?
  cout << "Capacity = " <<
bigNews.capacity() << endl;
  // Insert this string in bigNews immediately
  // before bigNews[1]:
  bigNews.insert(1, " thought I");
  cout << bigNews << endl;
  cout << "Size = " <<
bigNews.size() << endl;
  cout << "Capacity = " <<
bigNews.capacity() << endl;
  // Make sure that there will be this much space
  bigNews.reserve(500);
  // Add this to the end of the string:
  bigNews.append("I've been working too
hard.");
  cout << bigNews << endl;
  cout << "Size = " <<
bigNews.size() << endl;
  cout << "Capacity = " <<
bigNews.capacity() << endl;
} ///:~

Here is the output from one particular compiler:

 
Sélectionnez
I saw Elvis in a UFO.
Size = 22
Capacity = 31
I thought I saw Elvis in a UFO.
Size = 32
Capacity = 47
I thought I saw Elvis in a UFO. I've been
working too hard.
Size = 59
Capacity = 511

This example demonstrates that even though you can safely relinquish much of the responsibility for allocating and managing the memory your strings occupy, C++ strings provide you with several tools to monitor and manage their size. Notice the ease with which we changed the size of the storage allocated to the string. The size( ) function returns the number of characters currently stored in the string and is identical to the length( ) member function. The capacity( ) functionreturns the size of the current underlying allocation, meaning the number of characters the string can hold without requesting more storage. The reserve( ) function is an optimization mechanism that indicates your intention to specify a certain amount of storage for future use; capacity( ) always returns a value at least as large as the most recent call to reserve( ). A resize( ) function appends spaces if the new size is greater than the current string size or truncates the string otherwise. (An overload of resize( ) can specify a different character to append.)

The exact fashion that the string member functions allocate space for your data depends on the implementation of the library. When we tested one implementation with the previous example, it appeared that reallocations occurred on even word (that is, full-integer) boundaries, with one byte held back. The architects of the string class have endeavored to make it possible to mix the use of C char arrays and C++ string objects, so it is likely that figures reported by StrSize.cpp for capacity reflect that, in this particular implementation, a byte is set aside to easily accommodate the insertion of a null terminator.

2-1-3-2. Replacing string characters

The insert( ) functionis particularly nice because it absolves you from making sure the insertion of characters in a string won't overrun the storage space or overwrite the characters immediately following the insertion point. Space grows, and existing characters politely move over to accommodate the new elements. Sometimes this might not be what you want. If you want the size of the string to remain unchanged, use the replace( ) function to overwrite characters. There are a number of overloaded versions of replace( ), but the simplest one takes three arguments: an integer indicating where to start in the string, an integer indicating how many characters to eliminate from the original string, and the replacement string (which can be a different number of characters than the eliminated quantity). Here's a simple example:

 
Sélectionnez
//: C03:StringReplace.cpp
// Simple find-and-replace in strings.
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string s("A piece of text");
  string tag("$tag$");
  s.insert(8, tag + ' ');
  assert(s == "A piece $tag$
of text");
  int start = s.find(tag);
  assert(start == 8);
  assert(tag.size() == 5);
  s.replace(start, tag.size(), "hello
there");
  assert(s == "A piece hello there of text");
} ///:~

The tag is first inserted into s (notice that the insert happens before the value indicating the insert point and that an extra space was added after tag), and then it is found and replaced.

You should check to see if you've found anything before you perform a replace( ).The previous example replaces with a char*, but there's an overloaded version that replaces with a string.Here's a more complete demonstration replace( ):

 
Sélectionnez
//: C03:Replace.cpp
#include <cassert>
#include <cstddef>  // For size_t
#include <string>
using namespace std;
 
void replaceChars(string& modifyMe,
  const string& findMe, const string& newChars)
{
  // Look in modifyMe for the "find string"
  // starting at position 0:
  size_t i = modifyMe.find(findMe, 0);
  // Did we find the string to replace?
  if(i != string::npos)
    // Replace the find string with newChars:
    modifyMe.replace(i, findMe.size(), newChars);
}
 
int main() {
  string bigNews = "I thought I saw Elvis in a
UFO. "
                   "I have been working too
hard.";
  string replacement("wig");
  string findMe("UFO");
  // Find "UFO" in bigNews and overwrite it:
  replaceChars(bigNews, findMe, replacement);
  assert(bigNews == "I thought I saw Elvis in a
"
         "wig. I have been working too
hard.");
} ///:~

If replace doesn't find the search string, it returns string::npos. The npos data member is a static constant member of the string class that represents a nonexistent character position.(33)

Unlike insert( ), replace( ) won't grow the string's storage space if you copy new characters into the middle of an existing series of array elements. However, it will grow the storage space if needed, for example, when you make a “replacement” that would expand the original string beyond the end of the current allocation. Here's an example:

 
Sélectionnez
//: C03:ReplaceAndGrow.cpp
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string bigNews("I have been working the
grave.");
  string replacement("yard shift.");
  // The first argument says replace chars
  // beyond the end of the existing string:
  bigNews.replace(bigNews.size() - 1,
    replacement.size(), replacement);
  assert(bigNews == "I have been working the"
         "graveyard shift.");
} ///:~

The call to replace( ) begins “replacing” beyond the end of the existing array, which is equivalent to an append operation. Notice that in this example replace( ) expands the array accordingly.

You may have been hunting through this chapter trying to do something relatively simple such as replace all the instances of one character with a different character. Upon finding the previous material on replacing, you thought you found the answer, but then you started seeing groups of characters and counts and other things that looked a bit too complex. Doesn't string have a way to just replace one character with another everywhere?

You can easily write such a function using the find( ) and replace( ) member functions as follows:

 
Sélectionnez
//: C03:ReplaceAll.h
#ifndef REPLACEALL_H
#define REPLACEALL_H
#include <string>
 
std::string& replaceAll(std::string& context,
  const std::string& from, const std::string&
to);
#endif // REPLACEALL_H ///:~
 
Sélectionnez
//: C03:ReplaceAll.cpp {O}
#include <cstddef>
#include "ReplaceAll.h"
using namespace std;
 
string& replaceAll(string& context, const
string& from,
  const string& to) {
  size_t lookHere = 0;
  size_t foundHere;
  while((foundHere = context.find(from, lookHere))
    != string::npos) {
    context.replace(foundHere, from.size(), to);
    lookHere = foundHere + to.size();
  }
  return context;
} ///:~

The version of find( ) used here takes as a second argument the position to start looking in and returns string::npos if it doesn't find it. It is important to advance the position held in the variable lookHere past the replacement string, in case from is a substring of to. The following program tests the replaceAll function:

 
Sélectionnez
//: C03:ReplaceAllTest.cpp
//{L} ReplaceAll
#include <cassert>
#include <iostream>
#include <string>
#include "ReplaceAll.h"
using namespace std;
 
int main() {
  string text = "a man, a plan, a canal, Panama";
  replaceAll(text, "an", "XXX");
  assert(text == "a mXXX, a plXXX, a cXXXal, PXXXama");
} ///:~

As you can see, the string class by itself doesn't solve all possible problems. Many solutions have been left to the algorithms in the Standard library(34) because the string class can look just like an STL sequence (by virtue of the iterators discussed earlier). All the generic algorithms work on a “range” of elements within a container. Usually that range is just “from the beginning of the container to the end.” A string object looks like a container of characters: to get the beginning of the range you use string::begin( ), and to get the end of the range you use string::end( ). The following example shows the use of the replace( ) algorithm to replace all the instances of the single character ‘X' with ‘Y':

 
Sélectionnez
//: C03:StringCharReplace.cpp
#include <algorithm>
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string s("aaaXaaaXXaaXXXaXXXXaaa");
  replace(s.begin(), s.end(), 'X', 'Y');
  assert(s == "aaaYaaaYYaaYYYaYYYYaaa");
} ///:~

Notice that this replace( ) is not called as a member function of string. Also, unlike the string::replace( ) functions that only perform one replacement, the replace( ) algorithm replaces all instances of one character with another.

The replace( ) algorithm only works with single objects (in this case, char objects) and will not replace quoted char arrays or string objects. Since a string behaves like an STL sequence, a number of other algorithms can be applied to it, which might solve other problems that are not directly addressed by the string member functions.

2-1-3-3. Concatenation using nonmember overloaded operators

One of the most delightful discoveries awaiting a C programmer learning about C++ string handling is how simply strings can be combined and appended using operator+ and operator+=.These operators make combining strings syntactically similar to adding numeric data:

 
Sélectionnez
//: C03:AddStrings.cpp
#include <string>
#include <cassert>
using namespace std;
 
int main() {
  string s1("This ");
  string s2("That ");
  string s3("The other ");
  // operator+ concatenates strings
  s1 = s1 + s2;
  assert(s1 == "This That ");
  // Another way to concatenates strings
  s1 += s3;
  assert(s1 == "This That The other ");
  // You can index the string on the right
  s1 += s3 + s3[4] + "ooh lala";
  assert(s1 == "This That The other The other oooh
lala");
} ///:~

Using the operator+ and operator+= operatorsis a flexible andconvenient way to combine string data. On the right side of the statement, you can use almost any type that evaluates to a group of one or more characters.

2-1-4. Searching in strings

The find family of string member functions locates a character or group of characters within a given string. Here are the members of the find family and their general usage :

string find member function What/how it finds
find( ) Searches a string for a specified character or group of characters and returns the starting position of the first occurrence found or npos if no match is found.
find_first_of( ) Searches a target string and returns the position of the first match of any character in a specified group. If no match is found, it returns npos.
find_last_of( ) Searches a target string and returns the position of the last match of any character in a specified group. If no match is found, it returns npos.
find_first_not_of( ) Searches a target string and returns the position of the first element that doesn't match any character in a specified group. If no such element is found, it returns npos.
find_last_not_of( ) Searches a target string and returns the position of the element with the largest subscript that doesn't match any character in a specified group. If no such element is found, it returns npos.
rfind( ) Searches a string from end to beginning for a specified character or group of characters and returns the starting position of the match if one is found. If no match is found, it returns npos.

The simplest use of find( ) searches for one or more characters in a string. This overloaded version of find( ) takes a parameter that specifies the character(s) for which to search and optionally a parameter that tells it where in the string to begin searching for the occurrence of a substring. (The default position at which to begin searching is 0.) By setting the call to find inside a loop, you can easily move through a string, repeating a search to find all the occurrences of a given character or group of characters within the string.

The following program uses the method of The Sieve of Eratosthenes to find prime numbers less than 50. This method starts with the number 2, marks all subsequent multiples of 2 as not prime, and repeats the process for the next prime candidate. The SieveTest constructor initializes sieveChars by setting the initial size of the character array and writing the value ‘P' to each of its members.

 
Sélectionnez
//: C03:Sieve.h
#ifndef SIEVE_H
#define SIEVE_H
#include <cmath>
#include <cstddef>
#include <string>
#include "../TestSuite/Test.h"
using std::size_t;
using std::sqrt;
using std::string;
 
class SieveTest : public TestSuite::Test {
  string sieveChars;
public:
  // Create a 50 char string and set each
  // element to 'P' for Prime:
  SieveTest() : sieveChars(50, 'P') {}
  void run() {
    findPrimes();
    testPrimes();
  }
  bool isPrime(int p) {
    if(p == 0 || p == 1) return false;
    int root = int(sqrt(double(p)));
    for(int i = 2; i <= root; ++i)
      if(p % i == 0) return false;
    return true;
  }
  void findPrimes() {
    // By definition neither 0 nor 1 is prime.
    // Change these elements to "N" for Not
Prime:
    sieveChars.replace(0, 2, "NN");
    // Walk through the array:
    size_t sieveSize = sieveChars.size();
    int root = int(sqrt(double(sieveSize)));
    for(int i = 2; i <= root; ++i)
      // Find all the multiples:
      for(size_t factor = 2; factor * i < sieveSize;
           ++factor)
        sieveChars[factor * i] = 'N';
  }
  void testPrimes() {
    size_t i = sieveChars.find('P');
    while(i != string::npos) {
      test_(isPrime(i++));
      i = sieveChars.find('P', i);
    }
    i = sieveChars.find_first_not_of('P');
    while(i != string::npos) {
      test_(!isPrime(i++));
      i = sieveChars.find_first_not_of('P', i);
    }
  }
};
#endif // SIEVE_H ///:~
 
Sélectionnez
//: C03:Sieve.cpp
//{L} ../TestSuite/Test
#include "Sieve.h"
 
int main() {
  SieveTest t;
  t.run();
  return t.report();
} ///:~

The find( ) function can walk forward through a string, detecting multiple occurrences of a character or a group of characters, and find_first_not_of( ) finds other characters or substrings.

There are no functions in the string class to change the case of a string, but you can easily create these functions using the Standard C library functions toupper( ) and tolower( ), which change the case of one character at a time. The following example illustrates a case-insensitive search:

 
Sélectionnez
//: C03:Find.h
#ifndef FIND_H
#define FIND_H
#include <cctype>
#include <cstddef>
#include <string>
#include "../TestSuite/Test.h"
using std::size_t;
using std::string;
using std::tolower;
using std::toupper;
 
// Make an uppercase copy of s
inline string upperCase(const string& s) {
  string upper(s);
  for(size_t i = 0; i < s.length(); ++i)
    upper[i] = toupper(upper[i]);
  return upper;
}
 
// Make a lowercase copy of s
inline string lowerCase(const string& s) {
  string lower(s);
  for(size_t i = 0; i < s.length(); ++i)
    lower[i] = tolower(lower[i]);
  return lower;
}
 
class FindTest : public TestSuite::Test {
  string chooseOne;
public:
  FindTest() : chooseOne("Eenie, Meenie, Miney,
Mo") {}
  void testUpper() {
    string upper = upperCase(chooseOne);
    const string LOWER =
"abcdefghijklmnopqrstuvwxyz";
    test_(upper.find_first_of(LOWER) == string::npos);
  }
  void testLower() {
    string lower = lowerCase(chooseOne);
    const string UPPER =
"ABCDEFGHIJKLMNOPQRSTUVWXYZ";
    test_(lower.find_first_of(UPPER) == string::npos);
  }
  void testSearch() {
    // Case sensitive search
    size_t i = chooseOne.find("een");
    test_(i == 8);
    // Search lowercase:
    string test = lowerCase(chooseOne);
    i = test.find("een");
    test_(i == 0);
    i = test.find("een", ++i);
    test_(i == 8);
    i = test.find("een", ++i);
    test_(i == string::npos);
    // Search uppercase:
    test = upperCase(chooseOne);
    i = test.find("EEN");
    test_(i == 0);
    i = test.find("EEN", ++i);
    test_(i == 8);
    i = test.find("EEN", ++i);
    test_(i == string::npos);
  }
  void run() {
    testUpper();
    testLower();
    testSearch();
  }
};
#endif // FIND_H ///:~
 
Sélectionnez
//: C03:Find.cpp
//{L} ../TestSuite/Test
#include "Find.h"
#include "../TestSuite/Test.h"
 
int main() {
  FindTest t;
  t.run();
  return t.report();
} ///:~

Both the upperCase( ) and lowerCase( ) functions follow the same form: they make a copy of the argument string and change the case. The Find.cpp program isn't the best solution to the case-sensitivity problem, so we'll revisit it when we examine string comparisons.

2-1-4-1. Finding in reverse

If you need to search through a string from end to beginning (to find the data in “last in / first out” order), you can use the string member function rfind( ):

 
Sélectionnez
//: C03:Rparse.h
#ifndef RPARSE_H
#define RPARSE_H
#include <cstddef>
#include <string>
#include <vector>
#include "../TestSuite/Test.h"
using std::size_t;
using std::string;
using std::vector;
 
class RparseTest : public TestSuite::Test {
  // To store the words:
  vector<string> strings;
public:
  void parseForData() {
    // The ';' characters will be delimiters
    string
s("now.;sense;make;to;going;is;This");
    // The last element of the string:
    int last = s.size();
    // The beginning of the current word:
    size_t current = s.rfind(';');
    // Walk backward through the string:
    while(current != string::npos) {
      // Push each word into the vector.
      // Current is incremented before copying
      // to avoid copying the delimiter:
      ++current;
      strings.push_back(s.substr(current, last - current));
      // Back over the delimiter we just found,
      // and set last to the end of the next word:
      current -= 2;
      last = current + 1;
      // Find the next delimiter:
      current = s.rfind(';', current);
    }
    // Pick up the first word -- it's not
    // preceded by a delimiter:
    strings.push_back(s.substr(0, last));
  }
  void testData() {
    // Test them in the new order:
    test_(strings[0] == "This");
    test_(strings[1] == "is");
    test_(strings[2] == "going");
    test_(strings[3] == "to");
    test_(strings[4] == "make");
    test_(strings[5] == "sense");
    test_(strings[6] == "now.");
    string sentence;
    for(size_t i = 0; i < strings.size() - 1; i++)
      sentence += strings[i] += " ";
    // Manually put last word in to avoid an extra
space:
    sentence += strings[strings.size() - 1];
    test_(sentence == "This is going to make sense
now.");
  }
  void run() {
    parseForData();
    testData();
  }
};
#endif // RPARSE_H ///:~
 
Sélectionnez
//: C03:Rparse.cpp
//{L} ../TestSuite/Test
#include "Rparse.h"
 
int main() {
  RparseTest t;
  t.run();
  return t.report();
} ///:~

The string member function rfind( ) backs through the string looking for tokens and reports the array index of matching characters or string::npos if it is unsuccessful.

2-1-4-2. Finding first/last of a set of characters

The find_first_of( ) and find_last_of( ) member functions can be conveniently put to work to create a little utility that will strip whitespace characters from both ends of a string. Notice that it doesn't touch the original string, but instead returns a new string:

 
Sélectionnez
//: C03:Trim.h
// General tool to strip spaces from both ends.
#ifndef TRIM_H
#define TRIM_H
#include <string>
#include <cstddef>
 
inline std::string trim(const std::string& s) {
  if(s.length() == 0)
    return s;
  std::size_t beg = s.find_first_not_of("
\a\b\f\n\r\t\v");
  std::size_t end = s.find_last_not_of("
\a\b\f\n\r\t\v");
  if(beg == std::string::npos) // No non-spaces
    return "";
  return std::string(s, beg, end - beg + 1);
}
#endif // TRIM_H ///:~

The first test checks for an empty string; in that case, no tests are made, and a copy is returned. Notice that once the end points are found, the string constructor builds a new string from the old one, giving the starting count and the length.

Testing such a general-purpose tool needs to be thorough:

 
Sélectionnez
//: C03:TrimTest.h
#ifndef TRIMTEST_H
#define TRIMTEST_H
#include "Trim.h"
#include "../TestSuite/Test.h"
 
class TrimTest : public TestSuite::Test {
  enum {NTESTS = 11};
  static std::string s[NTESTS];
public:
  void testTrim() {
    test_(trim(s[0]) == "abcdefghijklmnop");
    test_(trim(s[1]) == "abcdefghijklmnop");
    test_(trim(s[2]) == "abcdefghijklmnop");
    test_(trim(s[3]) == "a");
    test_(trim(s[4]) == "ab");
    test_(trim(s[5]) == "abc");
    test_(trim(s[6]) == "a b c");
    test_(trim(s[7]) == "a b c");
    test_(trim(s[8]) == "a \t b \t c");
    test_(trim(s[9]) == "");
    test_(trim(s[10]) == "");
  }
  void run() {
    testTrim();
  }
};
#endif // TRIMTEST_H ///:~
 
Sélectionnez
//: C03:TrimTest.cpp {O}
#include "TrimTest.h"
 
// Initialize static data
std::string TrimTest::s[TrimTest::NTESTS] = {
  " \t abcdefghijklmnop \t ",
  "abcdefghijklmnop \t ",
  " \t abcdefghijklmnop",
  "a", "ab", "abc",
"a b c",
  " \t a b c \t ", " \t a \t b \t c \t
",
  "\t \n \r \v \f",
  "" // Must also test the empty string
}; ///:~
 
Sélectionnez
//: C03:TrimTestMain.cpp
//{L} ../TestSuite/Test TrimTest
#include "TrimTest.h"
 
int main() {
  TrimTest t;
  t.run();
  return t.report();
} ///:~

In the array of strings, you can see that the character arrays are automatically converted to string objects. This array provides cases to check the removal of spaces and tabs from both ends, as well as ensuring that spaces and tabs are not removed from the middle of a string.

2-1-4-3. Removing characters from strings

Removing characters is easy and efficient with the erase( ) member function, which takes two arguments: where to start removing characters (which defaults to 0), and how many to remove (which defaults to string::npos). If you specify more characters than remain in the string, the remaining characters are all erased anyway (so calling erase( ) without any arguments removes all characters from a string). Sometimes it's useful to take an HTML file and strip its tags and special characters so that you have something approximating the text that would be displayed in the Web browser, only as a plain text file. The following example uses erase( ) to do the job:

 
Sélectionnez
//: C03:HTMLStripper.cpp {RunByHand}
//{L} ReplaceAll
// Filter to remove html tags and markers.
#include <cassert>
#include <cmath>
#include <cstddef>
#include <fstream>
#include <iostream>
#include <string>
#include "ReplaceAll.h"
#include "../require.h"
using namespace std;
 
string& stripHTMLTags(string& s) {
  static bool inTag = false;
  bool done = false;
  while(!done) {
    if(inTag) {
      // The previous line started an HTML tag
      // but didn't finish. Must search for '>'.
      size_t rightPos = s.find('>');
      if(rightPos != string::npos) {
        inTag = false;
        s.erase(0, rightPos + 1);
      }
      else {
        done = true;
        s.erase();
      }
    }
    else {
      // Look for start of tag:
      size_t leftPos = s.find('<');
      if(leftPos != string::npos) {
        // See if tag close is in this line:
        size_t rightPos = s.find('>');
        if(rightPos == string::npos) {
          inTag = done = true;
          s.erase(leftPos);
        }
        else
          s.erase(leftPos, rightPos - leftPos + 1);
      }
      else
        done = true;
    }
  }
  // Remove all special HTML characters
  replaceAll(s, "&lt;",
"<");
  replaceAll(s, "&gt;",
">");
  replaceAll(s, "&amp;",
"&");
  replaceAll(s, "&nbsp;", " ");
  // Etc...
  return s;
}
 
int main(int argc, char* argv[]) {
  requireArgs(argc, 1,
    "usage: HTMLStripper InputFile");
  ifstream in(argv[1]);
  assure(in, argv[1]);
  string s;
  while(getline(in, s))
    if(!stripHTMLTags(s).empty())
      cout << s << endl;
} ///:~

This example will even strip HTML tags that span multiple lines.(35) This is accomplished with the static flag, inTag, which is true whenever the start of a tag is found, but the accompanying tag end is not found in the same line. All forms of erase( ) appear in the stripHTMLFlags( ) function.(36) The version of getline( ) we use here is a (global) function declared in the <string> header and is handy because it stores an arbitrarily long line in its string argument. You don't need to worry about the dimension of a character array as you do with istream::getline( ). Notice that this program uses the replaceAll( ) function from earlier in this chapter. In the next chapter, we'll use string streams to create a more elegant solution.

2-1-4-4. Comparing strings

Comparing strings is inherently different from comparing numbers. Numbers have constant, universally meaningful values. To evaluate the relationship between the magnitudes of two strings, you must make a lexical comparison. Lexical comparison means that when you test a character to see if it is “greater than” or “less than” another character, you are actually comparing the numeric representation of those characters as specified in the collating sequence of the character set being used. Most often this will be the ASCII collating sequence, which assigns the printable characters for the English language numbers in the range 32 through 127 decimal. In the ASCII collating sequence, the first “character” in the list is the space, followed by several common punctuation marks, and then uppercase and lowercase letters. With respect to the alphabet, this means that the letters nearer the front have lower ASCII values than those nearer the end. With these details in mind, it becomes easier to remember that when a lexical comparison that reports s1 is “greater than” s2, it simply means that when the two were compared, the first differing character in s1 came later in the alphabet than the character in that same position in s2.

C++ provides several ways to compare strings, and each has advantages. The simplest to use are the nonmember, overloaded operator functions: operator ==, operator != operator >, operator <, operator >=,and operator <=.

 
Sélectionnez
//: C03:CompStr.h
#ifndef COMPSTR_H
#define COMPSTR_H
#include <string>
#include "../TestSuite/Test.h"
using std::string;
 
class CompStrTest : public TestSuite::Test {
public:
  void run() {
    // Strings to compare
    string s1("This");
    string s2("That");
    test_(s1 == s1);
    test_(s1 != s2);
    test_(s1 > s2);
    test_(s1 >= s2);
    test_(s1 >= s1);
    test_(s2 < s1);
    test_(s2 <= s1);
    test_(s1 <= s1);
  }
};
#endif // COMPSTR_H ///:~
 
Sélectionnez
//: C03:CompStr.cpp
//{L} ../TestSuite/Test
#include "CompStr.h"
 
int main() {
  CompStrTest t;
  t.run();
  return t.report();
} ///:~

The overloaded comparison operators are useful for comparing both full strings and individual string character elements.

Notice in the following example the flexibility of argument types on both the left and right side of the comparison operators. For efficiency, the string class provides overloaded operators for the direct comparison of string objects, quoted literals, and pointers to C-style strings without having to create temporary string objects.

 
Sélectionnez
//: C03:Equivalence.cpp
#include <iostream>
#include <string>
using namespace std;
 
int main() {
  string s2("That"), s1("This");
  // The lvalue is a quoted literal
  // and the rvalue is a string:
  if("That" == s2)
    cout << "A match" << endl;
  // The left operand is a string and the right is
  // a pointer to a C-style null terminated string:
  if(s1 != s2.c_str())
    cout << "No match" << endl;
} ///:~

The c_str( ) function returns a const char* that points to a C-style, null-terminated string equivalent to the contents of the string object. This comes in handy when you want to pass a string to a standard C function, such as atoi( ) or any of the functions defined in the <cstring> header. It is an error to use the value returned by c_str( ) as non-const argument to any function.

You won't find the logical not (!) or the logical comparison operators (&& and ||) among operators for a string. (Neither will you find overloaded versions of the bitwise C operators &, |, ^, or ~.) The overloaded nonmember comparison operators for the string class are limited to the subset that has clear, unambiguous application to single characters or groups of characters.

The compare( ) member function offers you a great deal more sophisticated and precise comparison than the nonmember operator set. It provides overloaded versions to compare:

  • Two complete strings.
  • Part of either string to a complete string.
  • Subsets of two strings.

The following example compares complete strings:

 
Sélectionnez
//: C03:Compare.cpp
// Demonstrates compare() and swap().
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string first("This");
  string second("That");
  assert(first.compare(first) == 0);
  assert(second.compare(second) == 0);
  // Which is lexically greater?
  assert(first.compare(second) > 0);
  assert(second.compare(first) < 0);
  first.swap(second);
  assert(first.compare(second) < 0);
  assert(second.compare(first) > 0);
} ///:~

The swap( ) function in this example does what its name implies: it exchanges the contents of its object and argument. To compare a subset of the characters in one or both strings, you add arguments that define where to start the comparison and how many characters to consider. For example, we can use the following overloaded version of compare( ):

s1.compare(s1StartPos, s1NumberChars, s2, s2StartPos, s2NumberChars);

Here's an example:

 
Sélectionnez
//: C03:Compare2.cpp
// Illustrate overloaded compare().
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string first("This is a day that will live in
infamy");
  string second("I don't believe that this is what
"
                "I signed up for");
  // Compare "his is" in both strings:
  assert(first.compare(1, 7, second, 22, 7) == 0);
  // Compare "his is a" to "his is w":
  assert(first.compare(1, 9, second, 22, 9) < 0);
} ///:~

In the examples so far, we have used C-style array indexing syntax to refer to an individual character in a string. C++ strings provide an alternative to the s[n] notation: the at( ) member. These two indexing mechanisms produce the same result in C++ if all goes well:

 
Sélectionnez
//: C03:StringIndexing.cpp
#include <cassert>
#include <string>
using namespace std;
 
int main() {
  string s("1234");
  assert(s[1] == '2');
  assert(s.at(1) == '2');
} ///:~

There is one important difference, however, between [ ] and at( ). When you try to reference an array element that is out of bounds, at( ) will do you the kindness of throwing an exception, while ordinary [ ] subscripting syntax will leave you to your own devices:

 
Sélectionnez
//: C03:BadStringIndexing.cpp
#include <exception>
#include <iostream>
#include <string>
using namespace std;
 
int main() {
  string s("1234");
  // at() saves you by throwing an exception:
  try {
    s.at(5);
  } catch(exception& e) {
    cerr << e.what() << endl;
  }
} ///:~

Responsible programmers will not use errant indexes, but should you want to benefits of automatic index checking, using at( ) in place of [ ] will give you a chance to gracefully recover from references to array elements that don't exist. Execution of this program on one of our test compilers gave the following output:

 
Sélectionnez
invalid string position

The at( ) member throws an object of class out_of_range, which derives (ultimately) from std::exception. By catching this object in an exception handler, you can take appropriate remedial actions such as recalculating the offending subscript or growing the array. Using string::operator[ ]( ) gives no such protection and is as dangerous as char array processing in C.(37)

2-1-4-5. Strings and character traits

The program Find.cpp earlier in this chapter leads us to ask the obvious question: Why isn't case-insensitive comparison part of the standard string class? The answer provides interesting background on the true nature of C++ string objects.

Consider what it means for a character to have “case.” Written Hebrew, Farsi, and Kanji don't use the concept of upper- and lowercase, so for those languages this idea has no meaning. It would seem that if there were a way to designate some languages as “all uppercase” or “all lowercase,” we could design a generalized solution. However, some languages that employ the concept of “case” also change the meaning of particular characters with diacritical marks, for example: the cedilla in Spanish, the circumflex in French, and the umlaut in German. For this reason, any case-sensitive collating scheme that attempts to be comprehensive will be nightmarishly complex to use.

Although we usually treat the C++ string as a class, this is really not the case. The string type is a specialization of a more general constituent, the basic_string< > template. Observe how string is declared in the Standard C++ header file:(38)

 
Sélectionnez
typedef basic_string<char> string;

To understand the nature of the string class, look at the basic_string< > template:

 
Sélectionnez
template<class charT, class traits =
char_traits<charT>,
  class allocator =
allocator<charT> > class basic_string;

In Chapter 5, we examine templates in great detail (much more than in Chapter 16 of Volume 1). For now, just notice that the string type is created when the basic_string template is instantiated with char. Inside the basic_string< > template declaration, the line:

 
Sélectionnez
class traits = char_traits<charT>,

tells us that the behavior of the class made from the basic_string< > template is specified by a class based on the template char_traits< >. Thus, the basic_string< > template produces string-oriented classes that manipulate types other than char (wide characters, for example). To do this, the char_traits< > template controls the content and collating behaviors of a variety of character sets using the character comparison functions eq( ) (equal), ne( ) (not equal), and lt( ) (less than). The basic_string< > string comparison functions rely on these.

This is why the string class doesn't include case-insensitive member functions: that's not in its job description. To change the way the string class treats character comparison, you must supply a different char_traits< > template because that defines the behavior of the individual character comparison member functions.

You can use this information to make a new type of string class that ignores case. First, we'll define a new case-insensitive char_traits< > template that inherits from the existing template. Next, we'll override only the members we need to change to make character-by-character comparison case insensitive. (In addition to the three lexical character comparison members mentioned earlier, we'll also supply a new implementation for the char_traits functions find( ) and compare( )) . Finally, we'll typedef a new class based on basic_string, but using the case-insensitive ichar_traits template for its second argument:

 
Sélectionnez
//: C03:ichar_traits.h
// Creating your own character traits.
#ifndef ICHAR_TRAITS_H
#define ICHAR_TRAITS_H
#include <cassert>
#include <cctype>
#include <cmath>
#include <cstddef>
#include <ostream>
#include <string>
using std::allocator;
using std::basic_string;
using std::char_traits;
using std::ostream;
using std::size_t;
using std::string;
using std::toupper;
using std::tolower;
 
struct ichar_traits : char_traits<char> {
  // We'll only change character-by-
  // character comparison functions
  static bool eq(char c1st, char c2nd) {
    return toupper(c1st) == toupper(c2nd);
  }
  static bool ne(char c1st, char c2nd) {
    return !eq(c1st, c2nd);
  }
  static bool lt(char c1st, char c2nd) {
    return toupper(c1st) < toupper(c2nd);
  }
  static int
  compare(const char* str1, const char* str2, size_t n)
{
    for(size_t i = 0; i < n; ++i) {
      if(str1 == 0)
        return -1;
      else if(str2 == 0)
        return 1;
      else if(tolower(*str1) < tolower(*str2))
        return -1;
      else if(tolower(*str1) > tolower(*str2))
        return 1;
      assert(tolower(*str1) == tolower(*str2));
      ++str1; ++str2; // Compare the other chars
    }
    return 0;
  }
  static const char*
  find(const char* s1, size_t n, char c) {
    while(n-- > 0)
      if(toupper(*s1) == toupper(c))
        return s1;
      else
        ++s1;
    return 0;
  }
};
 
typedef basic_string<char, ichar_traits> istring;
 
inline ostream& operator<<(ostream& os,
const istring& s) {
  return os << string(s.c_str(), s.length());
}
#endif // ICHAR_TRAITS_H ///:~

We provide a typedef named istring so that our class will act like an ordinary string in every way, except that it will make all comparisons without respect to case. For convenience, we've also provided an overloaded operator<<( ) so that you can print istrings. Here's an example:

 
Sélectionnez
//: C03:ICompare.cpp
#include <cassert>
#include <iostream>
#include "ichar_traits.h"
using namespace std;
 
int main() {
  // The same letters except for case:
  istring first = "tHis";
  istring second = "ThIS";
  cout << first << endl;
  cout << second << endl;
  assert(first.compare(second) == 0);
  assert(first.find('h') == 1);
  assert(first.find('I') == 2);
  assert(first.find('x') == string::npos);
} ///:~

This is just a toy example. To make istring fully equivalent to string, we'd have to create the other functions necessary to support the new istring type.

The <string> header provides a wide string class via the following typedef:

 
Sélectionnez
typedef basic_string<wchar_t> wstring;

Wide string support also reveals itself in wide streams (wostream in place of ostream, also defined in <iostream>) and in the header <cwctype>, a wide-character version of <cctype>. This along with the wchar_t specialization of char_traits in the standard library allows us to do a wide-character version of ichar_traits:

 
Sélectionnez
//: C03:iwchar_traits.h {-g++}
// Creating your own wide-character traits.
#ifndef IWCHAR_TRAITS_H
#define IWCHAR_TRAITS_H
#include <cassert>
#include <cmath>
#include <cstddef>
#include <cwctype>
#include <ostream>
#include <string>
 
using std::allocator;
using std::basic_string;
using std::char_traits;
using std::size_t;
using std::towlower;
using std::towupper;
using std::wostream;
using std::wstring;
 
struct iwchar_traits : char_traits<wchar_t> {
  // We'll only change character-by-
  // character comparison functions
  static bool eq(wchar_t c1st, wchar_t c2nd) {
    return towupper(c1st) == towupper(c2nd);
  }
  static bool ne(wchar_t c1st, wchar_t c2nd) {
    return towupper(c1st) != towupper(c2nd);
  }
  static bool lt(wchar_t c1st, wchar_t c2nd) {
    return towupper(c1st) < towupper(c2nd);
  }
  static int compare(
    const wchar_t* str1, const wchar_t* str2, size_t n)
{
    for(size_t i = 0; i < n; i++) {
      if(str1 == 0)
        return -1;
      else if(str2 == 0)
        return 1;
      else if(towlower(*str1) < towlower(*str2))
        return -1;
      else if(towlower(*str1) > towlower(*str2))
        return 1;
      assert(towlower(*str1) == towlower(*str2));
      ++str1; ++str2; // Compare the other wchar_ts
    }
    return 0;
  }
  static const wchar_t*
  find(const wchar_t* s1, size_t n, wchar_t c) {
    while(n-- > 0)
      if(towupper(*s1) == towupper(c))
        return s1;
      else
        ++s1;
    return 0;
  }
};
 
typedef basic_string<wchar_t, iwchar_traits>
iwstring;
 
inline wostream& operator<<(wostream& os,
  const iwstring& s) {
  return os << wstring(s.c_str(), s.length());
}
#endif // IWCHAR_TRAITS_H  ///:~

As you can see, this is mostly an exercise in placing a ‘w' in the appropriate place in the source code. The test program looks like this:

 
Sélectionnez
//: C03:IWCompare.cpp {-g++}
#include <cassert>
#include <iostream>
#include "iwchar_traits.h"
using namespace std;
 
int main() {
  // The same letters except for case:
  iwstring wfirst = L"tHis";
  iwstring wsecond = L"ThIS";
  wcout << wfirst << endl;
  wcout << wsecond << endl;
  assert(wfirst.compare(wsecond) == 0);
  assert(wfirst.find('h') == 1);
  assert(wfirst.find('I') == 2);
  assert(wfirst.find('x') == wstring::npos);
} ///:~

Unfortunately, some compilers still do not provide robust support for wide characters.

2-1-5. A string application

If you've looked at the sample code in this book closely, you've noticed that certain tokens in the comments surround the code. These are used by a Python program that Bruce wrote to extract the code into files and set up makefiles for building the code. For example, a double-slash followed by a colon at the beginning of a line denotes the first line of a source file. The rest of the line contains information describing the file's name and location and whether it should be only compiled rather than fully built into an executable file. For example, the first line in the previous program above contains the string C03:IWCompare.cpp, indicating that the file IWCompare.cpp should be extracted into the directory C03.

The last line of a source file contains a triple-slash followed by a colon and a tilde. If the first line has an exclamation point immediately after the colon, the first and last lines of the source code are not to be output to the file (this is for data-only files). (If you're wondering why we're avoiding showing you these tokens, it's because we don't want to break the code extractor when applied to the text of the book!)

Bruce's Python program does a lot more than just extract code. If the token “{O}” follows the file name, its makefile entry will only be set up to compile the file and not to link it into an executable. (The Test Framework in Chapter 2 is built this way.) To link such a file with another source example, the target executable's source file will contain an “{L}” directive, as in:

 
Sélectionnez
//{L} ../TestSuite/Test

This section will present a program to just extract all the code so that you can compile and inspect it manually. You can use this program to extract all the code in this book by saving the document file as a text file(39) (let's call it TICV2.txt) and by executing something like the following on a shell command line:

 
Sélectionnez
C:> extractCode TICV2.txt /TheCode

This command reads the text file TICV2.txt and writes all the source code files in subdirectories under the top-level directory /TheCode. The directory tree will look like the following:

 
Sélectionnez
TheCode/
   C0B/
   C01/
   C02/
   C03/
   C04/
   C05/
   C06/
   C07/
   C08/
   C09/
   C10/
   C11/
   TestSuite/

The source files containing the examples from each chapter will be in the corresponding directory.

Here's the program:

 
Sélectionnez
//: C03:ExtractCode.cpp {-edg} {RunByHand}
// Extracts code from text.
#include <cassert>
#include <cstddef>
#include <cstdio>
#include <cstdlib>
#include <fstream>
#include <iostream>
#include <string>
using namespace std;
 
// Legacy non-standard C header for mkdir()
#if defined(__GNUC__) || defined(__MWERKS__)
#include <sys/stat.h>
#elif defined(__BORLANDC__) || defined(_MSC_VER) \
  || defined(__DMC__)
#include <direct.h>
#else
#error Compiler not supported
#endif
 
// Check to see if directory exists
// by attempting to open a new file
// for output within it.
bool exists(string fname) {
  size_t len = fname.length();
  if(fname[len-1] != '/' && fname[len-1] !=
'\\')
    fname.append("/");
  fname.append("000.tmp");
  ofstream outf(fname.c_str());
  bool existFlag = outf;
  if(outf) {
    outf.close();
    remove(fname.c_str());
  }
  return existFlag;
}
 
int main(int argc, char* argv[]) {
  // See if input file name provided
  if(argc == 1) {
    cerr << "usage: extractCode file
[dir]" << endl;
    exit(EXIT_FAILURE);
  }
  // See if input file exists
  ifstream inf(argv[1]);
  if(!inf) {
    cerr << "error opening file: "
<< argv[1] << endl;
    exit(EXIT_FAILURE);
  }
  // Check for optional output directory
  string root("./");  // current is default
  if(argc == 3) {
    // See if output directory exists
    root = argv[2];
    if(!exists(root)) {
      cerr << "no such directory: "
<< root << endl;
      exit(EXIT_FAILURE);
    }
    size_t rootLen = root.length();
    if(root[rootLen-1] != '/' &&
root[rootLen-1] != '\\')
      root.append("/");
  }
  // Read input file line by line
  // checking for code delimiters
  string line;
  bool inCode = false;
  bool printDelims = true;
  ofstream outf;
  while(getline(inf, line)) {
    size_t findDelim = line.find("//"
"/:~");
    if(findDelim != string::npos) {
      // Output last line and close file
      if(!inCode) {
        cerr << "Lines out of order"
<< endl;
        exit(EXIT_FAILURE);
      }
      assert(outf);
      if(printDelims)
        outf << line << endl;
      outf.close();
      inCode = false;
      printDelims = true;
    } else {
      findDelim = line.find("//"
":");
      if(findDelim == 0) {
        // Check for '!' directive
        if(line[3] == '!') {
          printDelims = false;
          ++findDelim;  // To skip '!' for next search
        }
        // Extract subdirectory name, if any
        size_t startOfSubdir =
          line.find_first_not_of(" \t",
findDelim+3);
        findDelim = line.find(':', startOfSubdir);
        if(findDelim == string::npos) {
          cerr << "missing filename
information\n" << endl;
          exit(EXIT_FAILURE);
        }
        string subdir;
        if(findDelim > startOfSubdir)
          subdir = line.substr(startOfSubdir,
                               findDelim -
startOfSubdir);
        // Extract file name (better be one!)
        size_t startOfFile = findDelim + 1;
        size_t endOfFile =
          line.find_first_of(" \t",
startOfFile);
        if(endOfFile == startOfFile) {
          cerr << "missing filename"
<< endl;
          exit(EXIT_FAILURE);
        }
        // We have all the pieces; build fullPath name
        string fullPath(root);
        if(subdir.length() > 0)
 
fullPath.append(subdir).append("/");
        assert(fullPath[fullPath.length()-1] == '/');
        if(!exists(fullPath))
#if defined(__GNUC__) || defined(__MWERKS__)
          mkdir(fullPath.c_str(), 0);  // Create subdir
#else
          mkdir(fullPath.c_str());  // Create subdir
#endif
        fullPath.append(line.substr(startOfFile,
                        endOfFile - startOfFile));
        outf.open(fullPath.c_str());
        if(!outf) {
          cerr << "error opening "
<< fullPath
               << " for output"
<< endl;
          exit(EXIT_FAILURE);
        }
        inCode = true;
        cout << "Processing " <<
fullPath << endl;
        if(printDelims)
          outf << line << endl;
      }
      else if(inCode) {
        assert(outf);
        outf << line << endl;  // Output middle
code line
      }
    }
  }
  exit(EXIT_SUCCESS);
} ///:~

First, you'll notice some conditional compilation directives. The mkdir( ) function, which creates a directory in the file system, is defined by the POSIX(40) standard in the header <sys/stat.h>. Unfortunately, many compilers still use a different header (<direct.h>). The respective signatures for mkdir( ) also differ: POSIX specifies two arguments, the older versions just one. For this reason, there is more conditional compilation later in the program to choose the right call to mkdir( ). We normally don't use conditional compilation in the examples in this book, but this particular program is too useful not to put a little extra work into, since you can use it to extract all the code with it.

The exists( ) function in ExtractCode.cpp tests whether a directory exists by opening a temporary file in it. If the open fails, the directory doesn't exist. You remove a file by sending its name as a char* to std::remove( ).

The main program validates the command-line arguments and then reads the input file a line at a time, looking for the special source code delimiters. The Boolean flag inCode indicates that the program is in the middle of a source file, so lines should be output. The printDelims flag will be true if the opening token is not followed by an exclamation point; otherwise the first and last lines are not written. It is important to check for the closing delimiter first, because the start token is a subset, and searching for the start token first would return a successful find for both cases. If we encounter the closing token, we verify that we are in the middle of processing a source file; otherwise, something is wrong with the way the delimiters are laid out in the text file. If inCode is true, all is well, and we (optionally) write the last line and close the file. When the opening token is found, we parse the directory and file name components and open the file. The following string-related functions were used in this example: length( ), append( ), getline( ), find( ) (two versions), find_first_not_of( ), substr( ), find_first_of( ), c_str( ), and, of course, operator<<( ).

2-1-6. Summary

C++ string objects provide developers with a number of great advantages over their C counterparts. For the most part, the string class makes referring to strings with character pointers unnecessary. This eliminates an entire class of software defects that arise from the use of uninitialized and incorrectly valued pointers.

C++ strings dynamically and transparently grow their internal data storage space to accommodate increases in the size of the string data. When the data in a string grows beyond the limits of the memory initially allocated to it, the string object will make the memory management calls that take space from and return space to the heap. Consistent allocation schemes prevent memory leaks and have the potential to be much more efficient than “roll your own” memory management.

The string class member functions provide a fairly comprehensive set of tools for creating, modifying, and searching in strings. String comparisons are always case sensitive, but you can work around this by copying string data to C-style null-terminated strings and using case-insensitive string comparison functions, temporarily converting the data held in string objects to a single case, or by creating a case-insensitive string class that overrides the character traits used to create the basic_string object.

2-1-7. Exercises

Solutions to selected exercises can be found in the electronic document The Thinking in C++ Volume 2 Annotated Solution Guide, available for a small fee from www.MindView.net.

  1. Write and test a function that reverses the order of the characters in a string.
  2. A palindrome is a word or group of words that read the same forward and backward. For example “madam” or “wow.” Write a program that takes a string argument from the command line and, using the function from the previous exercise, prints whether the string was a palindrome or not.
  3. Make your program from Exercise 2 return true even if symmetric letters differ in case. For example, “Civic” would still return true although the first letter is capitalized.
  4. Change your program from Exercise 3 to ignore punctuation and spaces as well. For example “Able was I, ere I saw Elba.” would report true.
  5. Using the following string declarations and only chars (no string literals or magic numbers):

    string one("I walked down the canyon with the moving mountain bikers.");
    string two("The bikers passed by me too close for comfort.");
    string three("I went hiking instead.");

    produce the following sentence:

    I moved down the canyon with the mountain bikers. The mountain bikers passed by me too close for comfort. So I went hiking instead.
  6. Write a program named replace that takes three command-line arguments representing an input text file, a string to replace (call it from), and a replacement string (call it to). The program should write a new file to standard output with all occurrences of from replaced by to.
  7. Repeat the previous exercise but replace all instances of from regardless of case.
  8. Make your program from Exercise 3 take a filename from the command-line, and then display all words that are palindromes (ignoring case) in the file. Do not display duplicates (even if their case differs). Do not try to look for palindromes that are larger than a word (unlike in Exercise 4).
  9. Modify HTMLStripper.cpp so that when it encounters a tag, it displays the tag's name, then displays the file's contents between the tag and the file's ending tag. Assume no nesting of tags, and that all tags have ending tags (denoted with </TAGNAME>).
  10. Write a program that takes three command-line arguments (a filename and two strings) and displays to the console all lines in the file that have both strings in the line, either string, only one string, or neither string, based on user input at the beginning of the program (the user will choose which matching mode to use). For all but the “neither string” option, highlight the input string(s) by placing an asterisk (*) at the beginning and end of each string's occurrence when it is displayed.
  11. Write a program that takes two command-line arguments (a filename and a string) and counts the number of times the string occurs in the file, even as a substring (but ignoring overlaps). For example, an input string of “ba” would match twice in the word “basketball,” but an input string of “ana” would match only once in the word “banana.” Display to the console the number of times the string is matched in the file, as well as the average length of the words where the string occurred. (If the string occurs more than once in a word, only count the word once in figuring the average.)
  12. Write a program that takes a filename from the command line and profiles the character usage, including punctuation and spaces (all character values of 0x21 [33] through 0x7E [126], as well as the space character). That is, count the number of occurrences of each character in the file, then display the results sorted either sequentially (space, then !, ", #, etc.) or by ascending or descending frequency based on user input at the beginning of the program. For space, display the word “Space” instead of the character ' '. A sample run might look something like this:
    Format sequentially, ascending, or descending (S/A/D): D
    t: 526
    r: 490
    etc.
  13. Using find( ) and rfind( ), write a program that takes two command-line arguments (a filename and a string) and displays the first and last words (and their indexes) not matching the string, as well as the indexes of the first and last instances of the string. Display “Not Found” if any of the searches fail.
  14. Using the find_first_of “family” of functions (but not exclusively), write a program that will remove all non-alphanumeric characters except spaces and periods from a file, then capitalize the first letter following a period.
  15. Again using the find_first_of “family” of functions, write a program that accepts a filename as a command-line argument and then formats all numbers in the file to currency. Ignore decimal points after the first until a non-numeric character is found, and round to the nearest hundredth. For example, the string 12.399abc29.00.6a would be formatted (in the USA) to $12.40abc$29.01a.
  16. Write a program that accepts two command-line arguments (a filename and a number) and scrambles each word in the file by randomly switching two of its letters the number of times specified in the second argument. (That is, if 0 is passed into your program from the command-line, the words should not be scrambled; if 1 is passed in, one pair of randomly-chosen letters should be swapped, for an input of 2, two random pairs should be swapped, etc.).
  17. Write a program that accepts a filename from the command line and displays the number of sentences (defined as the number of periods in the file), average number of characters per sentence, and the total number of characters in the file.
  18. Prove to yourself that the at( ) member function really will throw an exception if an attempt is made to go out of bounds, and that the indexing operator ([ ]) won't.

2-2. Iostreams

You can do much more with the general I/O problem than just take standard I/O and turn it into a class.

Wouldn't it be nice if you could make all the usual “receptacles”—standard I/O, files, and even blocks of memory—look the same so that you need to remember only one interface? That's the idea behind iostreams. They're much easier, safer, and sometimes even more efficient than the assorted functions from the Standard C stdio library.

The iostreams classes are usually the first part of the C++ library that new C++ programmers learn to use. This chapter discusses how iostreams are an improvement over C's stdio facilities and explores the behavior of file and string streams in addition to the standard console streams.

2-2-1. Why iostreams?

You might wonder what's wrong with the good old C library. Why not “wrap” the C library in a class and be done with it? Sometimes this is a fine solution. For example, suppose you want to make sure that the file represented by a stdio FILE pointer is always safely opened and properly closed without having to rely on the user to remember to call the close( ) function. The following program is such an attempt:

 
Sélectionnez
//: C04:FileClass.h
// stdio files wrapped.
#ifndef FILECLASS_H
#define FILECLASS_H
#include <cstdio>
#include <stdexcept>
 
class FileClass {
  std::FILE* f;
public:
  struct FileClassError : std::runtime_error {
    FileClassError(const char* msg)
    : std::runtime_error(msg) {}
  };
  FileClass(const char* fname, const char* mode =
"r");
  ~FileClass();
  std::FILE* fp();
};
#endif // FILECLASS_H ///:~

When you perform file I/O in C, you work with a naked pointer to a FILE struct, but this class wraps around the pointer and guarantees it is properly initialized and cleaned up using the constructor and destructor. The second constructor argument is the file mode, which defaults to “r” for “read.”

To fetch the value of the pointer to use in the file I/O functions, you use the fp( ) access function. Here are the member function definitions:

 
Sélectionnez
//: C04:FileClass.cpp {O}
// FileClass Implementation.
#include "FileClass.h"
#include <cstdlib>
#include <cstdio>
using namespace std;
 
FileClass::FileClass(const char* fname, const char*
mode) {
  if((f = fopen(fname, mode)) == 0)
    throw FileClassError("Error opening
file");
}
 
FileClass::~FileClass() { fclose(f); }
 
FILE* FileClass::fp() { return
f; } ///:~

The constructor calls fopen( ), as you would normally do, but it also ensures that the result isn't zero, which indicates a failure upon opening the file. If the file does not open as expected, an exception is thrown.

The destructor closes the file, and the access function fp( ) returns f. Here's a simple example using FileClass:

 
Sélectionnez
//: C04:FileClassTest.cpp
//{L} FileClass
#include <cstdlib>
#include <iostream>
#include "FileClass.h"
using namespace std;
 
int main() {
  try {
    FileClass f("FileClassTest.cpp");
    const int BSIZE = 100;
    char buf[BSIZE];
    while(fgets(buf, BSIZE, f.fp()))
      fputs(buf, stdout);
  } catch(FileClass::FileClassError& e) {
    cout << e.what() << endl;
    return EXIT_FAILURE;
  }
  return EXIT_SUCCESS;
} // File automatically closed by destructor
///:~

You create the FileClass object and use it in normal C file I/O function calls by calling fp( ). When you're done with it, just forget about it; the file is closed by the destructor at the end of its scope.

Even though the FILE pointer is private, it isn't particularly safe because fp( ) retrieves it. Since the only effect seems to be guaranteed initialization and cleanup, why not make it public or use a struct instead? Notice that while you can get a copy of f using fp( ), you cannot assign to f—that's completely under the control of the class. After capturing the pointer returned by fp( ), the client programmer can still assign to the structure elements or even close it, so the safety is in guaranteeing a valid FILE pointer rather than proper contents of the structure.

If you want complete safety, you must prevent the user from directly accessing the FILE pointer. Some version of all the normal file I/O functions must show up as class members so that everything you can do with the C approach is available in the C++ class:

 
Sélectionnez
//: C04:Fullwrap.h
// Completely hidden file IO.
#ifndef FULLWRAP_H
#define FULLWRAP_H
#include <cstddef>
#include <cstdio>
#undef getc
#undef putc
#undef ungetc
using std::size_t;
using std::fpos_t;
 
class File {
  std::FILE* f;
  std::FILE* F(); // Produces checked pointer to f
public:
  File(); // Create object but don't open file
  File(const char* path, const char* mode =
"r");
  ~File();
  int open(const char* path, const char* mode =
"r");
  int reopen(const char* path, const char* mode);
  int getc();
  int ungetc(int c);
  int putc(int c);
  int puts(const char* s);
  char* gets(char* s, int n);
  int printf(const char* format, ...);
  size_t read(void* ptr, size_t size, size_t n);
  size_t write(const void* ptr, size_t size, size_t n);
  int eof();
  int close();
  int flush();
  int seek(long offset, int whence);
  int getpos(fpos_t* pos);
  int setpos(const fpos_t* pos);
  long tell();
  void rewind();
  void setbuf(char* buf);
  int setvbuf(char* buf, int type, size_t sz);
  int error();
  void clearErr();
};
#endif // FULLWRAP_H ///:~

This class contains almost all the file I/O functions from <cstdio>. (vfprintf( ) is missing; it implements the printf( ) member function.)

File has the same constructor as in the previous example, and it also has a default constructor. The default constructor is important if you want to create an array of File objects or use a File object as a member of another class where the initialization doesn't happen in the constructor, but some time after the enclosing object is created.

The default constructor sets the private FILE pointer f to zero. But now, before any reference to f, its value must be checked to ensure it isn't zero. This is accomplished with F( ), which is private because it is intended to be used only by other member functions. (We don't want to give the user direct access to the underlying FILE structure in this class.)

This approach is not a terrible solution by any means. It's quite functional, and you could imagine making similar classes for standard (console) I/O and for in-core formatting (reading/writing a piece of memory rather than a file or the console).

The stumbling block is the runtime interpreter used for the variable argument list functions. This is the code that parses your format string at runtime and grabs and interprets arguments from the variable argument list. It's a problem for four reasons.

  1. Even if you use only a fraction of the functionality of the interpreter, the whole thing gets loaded into your executable. So if you say printf("%c", 'x');, you'll get the whole package, including the parts that print floating-point numbers and strings. There's no standard option for reducing the amount of space used by the program.
  2. Because the interpretation happens at runtime, you can't get rid of a performance overhead. It's frustrating because all the information is there in the format string at compile time, but it's not evaluated until runtime. However, if you could parse the arguments in the format string at compile time, you could make direct function calls that have the potential to be much faster than a runtime interpreter (although the printf( ) family of functions is usually quite well optimized).
  3. Because the format string is not evaluated until runtime, there can be no compile-time error checking. You're probably familiar with this problem if you've tried to find bugs that came from using the wrong number or type of arguments in a printf( ) statement. C++ makes a big deal out of compile-time error checking to find errors early and make your life easier. It seems a shame to throw type safety away for an I/O library, especially since I/O is used a lot.
  4. For C++, the most crucial problem is that the printf( ) family of functions is not particularly extensible. They're really designed to handle only the basic data types in C (char, int, float, double, wchar_t, char*, wchar_t*, and void*) and their variations. You might think that every time you add a new class, you could add overloaded printf( ) and scanf( ) functions (and their variants for files and strings), but remember, overloaded functions must have different types in their argument lists, and the printf( ) family hides its type information in the format string and in the variable argument list. For a language such as C++, whose goal is to be able to easily add new data types, this is an unacceptable restriction.

2-2-2. Iostreams to the rescue

These issues make it clear that I/O is one of the first priorities for the Standard C++ class libraries. Because “hello, world” is the first program just about everyone writes in a new language, and because I/O is part of virtually every program, the I/O library in C++ must be particularly easy to use. It also has the much greater challenge that it must accommodate any new class. Thus, its constraints require that this foundation class library be a truly inspired design. In addition to gaining a great deal of leverage and clarity in your dealings with I/O and formatting, you'll also see in this chapter how a really powerful C++ library can work.

2-2-2-1. Inserters and extractors

A stream is an object that transports and formats characters of a fixed width. You can have an input stream (via descendants of the istream class), an output stream (with ostream objects), or a stream that does both simultaneously (with objects derived from iostream). The iostreams library provides different types of such classes: ifstream,ofstream, and fstream for files, and istringstream, ostringstream, and stringstream for interfacing with the Standard C++ string class. All these stream classes have nearly identical interfaces, so you can use streams in a uniform manner, whether you're working with a file, standard I/O, a region of memory, or a string object. The single interface you learn also works for extensions added to support new classes. Some functions implement your formatting commands, and some functions read and write characters without formatting.

The stream classes mentioned earlier are actually template specializations,(41) much like the standard string class is a specialization of the basic_string template. The basic classes in the iostreams inheritance hierarchy are shown in the following figure:

Image non disponible

The ios_base class declares everything that is common to all streams, independent of the type of character the stream handles. These declarations are mostly constants and functions to manage them, some of which you'll see throughout this chapter. The rest of the classes are templates that have the underlying character type as a parameter. The istream class, for example, is defined as follows:

 
Sélectionnez
typedef basic_istream<char> istream;

All the classes mentioned earlier are defined via similar type definitions. There are also type definitions for all stream classes using wchar_t (the wide character type discussed in Chapter 3) instead of char. We'll look at these at the end of this chapter. The basic_ios template defines functions common to both input and output, but that depends on the underlying character type (we won't use these much). The template basic_istream defines generic functions for input, and basic_ostream does the same for output. The classes for file and string streams introduced later add functionality for their specific stream types.

In the iostreams library, two operators are overloaded to simplify the use of iostreams. The operator << is often referred to as an inserter for iostreams, and the operator >> is often referred to as an extractor.

Extractors parse the information that's expected by the destination object according to its type. To see an example of this, you can use the cin object, which is the iostream equivalent of stdin in C, that is, redirectable standard input. This object is predefined whenever you include the <iostream> header.

 
Sélectionnez
  int i;
  cin >> i;
 
  float f;
  cin >> f;
 
  char c;
  cin >> c;
 
  char buf[100];
  cin >> buf;

There's an overloaded operator >> for every built-in data type. You can also overload your own, as you'll see later.

To find out what you have in the various variables, you can use the cout object (corresponding to standard output; there's also a cerr object corresponding to standard error) with the inserter <<:

 
Sélectionnez
  cout << "i = ";
  cout << i;
  cout << "\n";
  cout << "f = ";
  cout << f;
  cout << "\n";
  cout << "c = ";
  cout << c;
  cout << "\n";
  cout << "buf = ";
  cout << buf;
  cout << "\n";

This is tedious and doesn't seem like much of an improvement over printf( ), despite improved type checking. Fortunately, the overloaded inserters and extractors are designed to be chained into a more complex expression that is much easier to write (and read):

 
Sélectionnez
  cout << "i = " << i <<
endl;
  cout << "f = " << f <<
endl;
  cout << "c = " << c <<
endl;
  cout << "buf = " << buf << endl;

Defining inserters and extractors for your own classes is just a matter of overloading the associated operators to do the right things, namely:

  • Make the first parameter a non-const reference to the stream (istream for input, ostream for output).
  • Perform the operation by inserting/extracting data to/from the stream (by processing the components of the object).
  • Return a reference to the stream.

The stream should be non-const because processing stream data changes the state of the stream. By returning the stream, you allow for chaining stream operations in a single statement, as shown earlier.

As an example, consider how to output the representation of a Date object in MM-DD-YYYY format. The following inserter does the job:

 
Sélectionnez
ostream& operator<<(ostream& os, const
Date& d) {
  char fillc = os.fill('0');
  os << setw(2) << d.getMonth() <<
'-'
     << setw(2) << d.getDay() << '-'
     << setw(4) << setfill(fillc) <<
d.getYear();
  return os;
}

This function cannot be a member of the Date class because the left operand of the << operator must be the output stream. The fill( ) member function of ostream changes the padding character used when the width of an output field, determined by the manipulatorsetw( ), is greater than needed for the data. We use a ‘0' character so that months preceding October will display a leading zero, such as “09” for September. The fill( ) function also returns the previous fill character (which defaults to a single space) so that we can restore it later with the manipulator setfill( ). We discuss manipulators in depth later in this chapter.

Extractors require a little more care because things can go wrong with input data. The way to signal a stream error is to set the stream's fail bit, as follows:

 
Sélectionnez
istream& operator>>(istream& is,
Date& d) {
  is >> d.month;
  char dash;
  is >> dash;
  if(dash != '-')
    is.setstate(ios::failbit);
  is >> d.day;
  is >> dash;
  if(dash != '-')
    is.setstate(ios::failbit);
  is >> d.year;
  return is;
}

When an error bit is set in a stream, all further streams operations are ignored until the stream is restored to a good state (explained shortly). That's why the code above continues extracting even if ios::failbit gets set. This implementation is somewhat forgiving in that it allows white space between the numbers and dashes in a date string (because the >> operator skips white space by default when reading built-in types). The following are valid date strings for this extractor:

 
Sélectionnez
"08-10-2003"
"8-10-2003"
"08 - 10 - 2003"

but these are not:

 
Sélectionnez
"A-10-2003" // No alpha characters allowed
"08/2003" // Only
dashes allowed as a delimiter

We'll discuss stream state in more depth in the section “Handling stream errors” later in this chapter.

2-2-2-2. Common usage

As the Date extractor illustrated, you must be on guard for erroneous input. If the input produces an unexpected value, the process is skewed, and it's difficult to recover. In addition, formatted input defaults to white space delimiters. Consider what happens when we collect the code fragments from earlier in this chapter into a single program:

 
Sélectionnez
//: C04:Iosexamp.cpp {RunByHand}
// Iostream examples.
#include <iostream>
using namespace std;
 
int main() {
  int i;
  cin >> i;
 
  float f;
  cin >> f;
 
  char c;
  cin >> c;
 
  char buf[100];
  cin >> buf;
 
  cout << "i = "
<< i << endl;
  cout << "f = " << f <<
endl;
  cout << "c = " << c <<
endl;
  cout << "buf = " << buf
<< endl;
 
  cout << flush;
  cout << hex << "0x" << i
<< endl;
} ///:~

and give it the following input:

 
Sélectionnez
12 1.4 c this is a test

We expect the same output as if we gave it

 
Sélectionnez
12
1.4
c
this is a test

but the output is, somewhat unexpectedly

 
Sélectionnez
i = 12
f = 1.4
c = c
buf = this
0xc

Notice that buf got only the first word because the input routine looked for a space to delimit the input, which it saw after “this.” In addition, if the continuous input string is longer than the storage allocated for buf, we overrun the buffer.

In practice, you'll usually want to get input from interactive programs a line at a time as a sequence of characters, scan them, and then perform conversions once they're safely in a buffer. This way you don't need to worry about the input routine choking on unexpected data.

Another consideration is the whole concept of a command-line interface. This made sense in the past when the console was little more than a glass typewriter, but the world is rapidly changing to one where the graphical user interface (GUI) dominates. What is the meaning of console I/O in such a world? It makes much more sense to ignore cin altogether, other than for simple examples or tests, and take the following approaches:

  1. If your program requires input, read that input from a file—you'll soon see that it's remarkably easy to use files with iostreams. Iostreams for files still works fine with a GUI.
  2. Read the input without attempting to convert it, as we just suggested. When the input is some place where it can't foul things up during conversion, you can safely scan it.
  3. Output is different. If you're using a GUI, cout doesn't necessarily work, and you must send it to a file (which is identical to sending it to cout) or use the GUI facilities for data display. Otherwise it often makes sense to send it to cout. In both cases, the output formatting functions of iostreams are highly useful.

Another common practice saves compile time on large projects. Consider, for example, how you would declare the Date stream operators introduced earlier in the chapter in a header file. You only need to include the prototypes for the functions, so it's not really necessary to include the entire <iostream> header in Date.h. The standard practice is to only declare classes, something like this:

 
Sélectionnez
class ostream;

This is an age-old technique for separating interface from implementation and is often called a forward declaration (and ostream at this point would be considered an incomplete type, since the class definition has not yet been seen by the compiler).

This will not work as is, however, for two reasons:

1.  The stream classes are defined in the std namespace.

2.  They are templates.

The proper declaration would be:

 
Sélectionnez
namespace std {
  template<class charT, class traits =
char_traits<charT> >
    class basic_ostream;
  typedef basic_ostream<char> ostream;
}

(As you can see, like the string class, the streams classes use the character traits classes mentioned in Chapter 3). Since it would be terribly tedious to type all that for every stream class you want to reference, the standard provides a header that does it for you: <iosfwd>. The Date header would then look something like this:

 
Sélectionnez
// Date.h
#include <iosfwd>
 
class Date {
  friend std::ostream&
operator<<(std::ostream&,
                                  const Date&);
  friend std::istream&
operator>>(std::istream&, Date&);
  // Etc.
2-2-2-3. Line-oriented input

To grab input a line at a time, you have three choices:

  • The member function get( )
  • The member function getline( )
  • The global function getline( ) defined in the <string> header

The first two functions take three arguments:

  1. A pointer to a character buffer in which to store the result.
  2. The size of that buffer (so it's not overrun).
  3. The terminating character, to know when to stop reading input.

The terminating character has a default value of '\n', which is what you'll usually use. Both functions store a zero in the result buffer when they encounter the terminating character in the input.

So what's the difference? Subtle, but important: get( ) stops when it sees the delimiter in the input stream, but it doesn't extract it from the input stream. Thus, if you did another get( ) using the same delimiter, it would immediately return with no fetched input. (Presumably, you either use a different delimiter in the next get( ) statement or a different input function.) The getline( ) function, on the other hand, extracts the delimiter from the input stream, but still doesn't store it in the result buffer.

The getline( ) function defined in <string> is convenient. It is not a member function, but rather a stand-alone function declared in the namespace std. It takes only two non-default arguments, the input stream and the string object to populate. Like its namesake, it reads characters until it encounters the first occurrence of the delimiter ('\n' by default) and consumes and discards the delimiter. The advantage of this function is that it reads into a string object, so you don't need to worry about buffer size.

Generally, when you're processing a text file that you read a line at a time, you'll want to use one of the getline( ) functions.

Overloaded versions of get( )

The get( ) function also comes in three other overloaded versions: one with no arguments that returns the next character using an int return value; one that stuffs a character into its char argument using a reference; and one that stores directly into the underlying buffer structure of another iostream object. The latter is explored later in the chapter.

Reading raw bytes

If you know exactly what you're dealing with and want to move the bytes directly into a variable, an array, or a structure in memory, you can use the unformatted I/O function read( ). The first argument for this function is a pointer to the destination memory, and the second is the number of bytes to read. This is especially useful if you've previously stored the information to a file, for example, in binary form using the complementary write( ) member function for an output stream (using the same compiler, of course). You'll see examples of all these functions later.

2-2-3. Handling stream errors

The Date extractor shown earlier sets a stream's fail bit under certain conditions. How does the user know when such a failure occurs? You can detect stream errors by either calling certain stream member functions to see if an error state has occurred, or if you don't care what the particular error was, you can just evaluate the stream in a Boolean context. Both techniques derive from the state of a stream's error bits.

Stream state

The ios_base class, from which ios derives,(42) defines four flags that you can use to test the state of a stream:

Flag Meaning
badbit Some fatal (perhaps physical) error occurred. The stream should be considered unusable.
eofbit End-of-input has occurred (either by encountering the physical end of a file stream or by the user terminating a console stream, such as with Ctrl-Z or Ctrl‑D).
failbit An I/O operation failed, most likely because of invalid data (e.g., letters were found when trying to read a number). The stream is still usable. The failbit flag is also set when end-of-input occurs.
goodbit All is well; no errors. End-of-input has not yet occurred.

You can test whether any of these conditions have occurred by calling corresponding member functions that return a Boolean value indicating whether any of these have been set. The good( ) stream member function returns true if none of the other three bits are set. The eof( ) function returns true if eofbit is set, which happens with an attempt to read from a stream that has no more data (usually a file). Because end-of-input happens in C++ when trying to read past the end of the physical medium, failbit is also set to indicate that the “expected” data was not successfully read. The fail( ) function returns true if either failbit or badbit is set, and bad( ) returns true only if the badbit is set.

Once any of the error bits in a stream's state are set, they remain set, which is not always what you want. When reading a file, you might want to reposition to an earlier place in the file before end-of-file occurred. Just moving the file pointer doesn't automatically reset eofbit or failbit; you must do it yourself with the clear( ) function, like this:

 
Sélectionnez
myStream.clear(); // Clears all error bits

After calling clear( ), good( ) will return true if called immediately. As you saw in the Date extractor earlier, the setstate( ) function sets the bits you pass it. It turns out that setstate( ) doesn't affect any other bits—if they're already set, they stay set. If you want to set certain bits but at the same time reset all the rest, you can call an overloaded version of clear( ), passing it a bitwise expression representing the bits you want to set, as in:

 
Sélectionnez
myStream.clear(ios::failbit | ios::eofbit);

Most of the time you won't be interested in checking the stream state bits individually. Usually you just want to know if everything is okay. This is the case when you read a file from beginning to end; you just want to know when the input data is exhausted. You can use a conversion function defined for void* that is automatically called when a stream occurs in a Boolean expression. Reading a stream until end-of-input using this idiom looks like the following:

 
Sélectionnez
int i;
while(myStream >> i)
  cout << i <<
endl;

Remember that operator>>( ) returns its stream argument, so the while statement above tests the stream as a Boolean expression. This particular example assumes that the input stream myStream contains integers separated by white space. The function ios_base::operator void*( ) simply calls good( ) on its stream and returns the result.(43) Because most stream operations return their stream, using this idiom is convenient.

Streams and exceptions

Iostreams existed as part of C++ long before there were exceptions, so checking stream state manually was just the way things were done. For backward compatibility, this is still the status quo, but modern iostreams can throw exceptions instead. The exceptions( ) stream member function takes a parameter representing the state bits for which you want exceptions to be thrown. Whenever the stream encounters such a state, it throws an exception of type std::ios_base::failure, which inherits from std::exception.

Although you can trigger a failure exception for any of the four stream states, it's not necessarily a good idea to enable exceptions for all of them. As Chapter 1 explains, use exceptions for truly exceptional conditions, but end-of-file is not only not exceptional—it's expected! For that reason, you might want to enable exceptions only for the errors represented by badbit, which you would do like this:

 
Sélectionnez
myStream.exceptions(ios::badbit);

You enable exceptions on a stream-by-stream basis, since exceptions( ) is a member function for streams. The exceptions( ) function returns a bitmask(44) (of type iostate, which is some compiler-dependent type convertible to int) indicating which stream states will cause exceptions. If those states have already been set, an exception is thrown immediately. Of course, if you use exceptions in connection with streams, you had better be ready to catch them, which means that you need to wrap all stream processing with a try block that has an ios::failure handler. Many programmers find this tedious and just check states manually where they expect errors to occur (since, for example, they don't expect bad( ) to return true most of the time anyway). This is another reason that having streams throw exceptions is optional and not the default. In any case, you can choose how you want to handle stream errors. For the same reasons that we recommend using exceptions for error handling in other contexts, we do so here.

2-2-4. File iostreams

Manipulating files with iostreams is much easier and safer than using stdio in C. All you do to open a file is create an object—the constructor does the work. You don't need to explicitly close a file (although you can, using the close( ) member function) because the destructor will close it when the object goes out of scope. To create a file that defaults to input, make an ifstream object. To create one that defaults to output, make an ofstream object. An fstream object can do both input and output.

The file stream classes fit into the iostreams classes as shown in the following figure:

Image non disponible

As before, the classes you actually use are template specializations defined by type definitions. For example, ifstream, which processes files of char, is defined as

 
Sélectionnez
typedef basic_ifstream<char> ifstream;
2-2-4-1. A File-Processing Example

Here's an example that shows many of the features discussed so far. Notice the inclusion of <fstream>to declare the file I/O classes. Although on many platforms this will also include <iostream> automatically, compilers are not required to do so. If you want portable code, always include both headers.

 
Sélectionnez
//: C04:Strfile.cpp
// Stream I/O with files;
// The difference between get() & getline().
#include <fstream>
#include <iostream>
#include "../require.h"
using namespace std;
 
int main() {
  const int SZ = 100; // Buffer size;
  char buf[SZ];
  {
    ifstream in("Strfile.cpp"); // Read
    assure(in, "Strfile.cpp"); // Verify open
    ofstream out("Strfile.out"); // Write
    assure(out, "Strfile.out");
    int i = 1; // Line counter
 
    // A less-convenient approach for line input:
    while(in.get(buf, SZ)) { // Leaves \n in input
      in.get(); // Throw away next character (\n)
      cout << buf << endl; // Must add \n
      // File output just like standard I/O:
      out << i++ << ": " <<
buf << endl;
    }
  } // Destructors close in & out
 
  ifstream in("Strfile.out");
  assure(in, "Strfile.out");
  // More convenient line input:
  while(in.getline(buf, SZ)) { // Removes \n
    char* cp = buf;
    while(*cp != ':')
      ++cp;
    cp += 2; // Past ": "
    cout << cp << endl; // Must still add
\n
  }
} ///:~

The creation of both the ifstream and ofstream are followed by an assure( ) to guarantee the file was successfully opened. Here again the object, used in a situation where the compiler expects a Boolean result, produces a value that indicates success or failure.

The first while loop demonstrates the use of two forms of the get( ) function. The first gets characters into a buffer and puts a zero terminator in the buffer when either SZ-1 characters have been read or the third argument (defaulted to '\n') is encountered. The get( ) function leaves the terminator character in the input stream, so this terminator must be thrown away via in.get( ) using the form of get( ) with no argument, which fetches a single byte and returns it as an int. You can also use the ignore( ) member function, which has two default arguments. The first argument is the number of characters to throw away and defaults to one. The second argument is the character at which the ignore( ) function quits (after extracting it) and defaults to EOF.

Next, you see two output statements that look similar: one to cout and one to the file out. Notice the convenience here—you don't need to worry about the object type because the formatting statements work the same with all ostream objects. The first one echoes the line to standard output, and the second writes the line out to the new file and includes a line number.

To demonstrate getline( ), open the file we just created and strip off the line numbers. To ensure the file is properly closed before opening it to read, you have two choices. You can surround the first part of the program with braces to force the out object out of scope, thus calling the destructor and closing the file, which is done here. You can also call close( ) for both files; if you do this, you can even reuse the in object by calling the open( ) member function.

The second while loop shows how getline( ) removes the terminator character (its third argument, which defaults to '\n') from the input stream when it's encountered. Although getline( ), like get( ), puts a zero in the buffer, it still doesn't insert the terminating character.

This example, as well as most of the examples in this chapter, assumes that each call to any overload of getline( ) will encounter a newline character. If this is not the case, the eofbit state of the stream will be set and the call to getline( ) will return false, causing the program to lose the last line of input.

2-2-4-2. Open modes

You can control the way a file is opened by overriding the constructor's default arguments. The following table shows the flags that control the mode of the file:

Flag Function
ios::in Opens an input file. Use this as an open mode for an ofstream to prevent truncating an existing file.
ios::out Opens an output file. When used for an ofstream without ios::app, ios::ate or ios::in, ios::trunc is implied.
ios::app Opens an output file for appending only.
ios::ate Opens an existing file (either input or output) and seeks to the end.
ios::trunc Truncates the old file if it already exists.
ios::binary Opens a file in binary mode. The default is text mode.

You can combine these flags using a bitwise or operation.

The binary flag, while portable, only has an effect on some non-UNIX systems, such as operating systems derived from MS-DOS, that have special conventions for storing end-of-line delimiters. For example, on MS-DOS systems in text mode (which is the default), every time you output a newline character ('\n'), the file system actually outputs two characters, a carriage-return/linefeed pair (CRLF), which is the pair of ASCII characters 0x0D and 0x0A. Conversely, when you read such a file back into memory in text mode, each occurrence of this pair of bytes causes a '\n' to be sent to the program in its place. If you want to bypass this special processing, you open files in binary mode. Binary mode has nothing whatsoever to do with whether you can write raw bytes to a file—you always can (by calling write( )) . You should, however, open a file in binary mode when you'll be using read( ) or write( ), because these functions take a byte count parameter. Having the extra '\r' characters will throw your byte count off in those instances. You should also open a file in binary mode if you're going to use the stream-positioning commands discussed later in this chapter.

You can open a file for both input and output by declaring an fstream object. When declaring an fstream object, you must use enough of the open mode flags mentioned earlier to let the file system know whether you want to input, output, or both. To switch from output to input, you need to either flush the stream or change the file position. To change from input to output, change the file position. To create a file via an fstream object, use the ios::trunc open mode flag in the constructor call to do both input and output.

2-2-5. Iostream buffering

Good design practice dictates that, whenever you create a new class, you should endeavor to hide the details of the underlying implementation as much as possible from the user of the class. You show them only what they need to know and make the rest private to avoid confusion. When using inserters and extractors, you normally don't know or care where the bytes are being produced or consumed, whether you're dealing with standard I/O, files, memory, or some newly created class or device.

A time comes, however, when it is important to communicate with the part of the iostream that produces and consumes bytes. To provide this part with a common interface and still hide its underlying implementation, the standard library abstracts it into its own class, called streambuf. Each iostream object contains a pointer to some kind of streambuf. (The type depends on whether it deals with standard I/O, files, memory, and so on.) You can access the streambuf directly; for example, you can move raw bytes into and out of the streambuf without formatting them through the enclosing iostream. This is accomplished by calling member functions for the streambuf object.

Currently, the most important thing for you to know is that every iostream object contains a pointer to a streambuf object, and the streambuf object has some member functions you can call if necessary. For file and string streams, there are specialized types of stream buffers, as the following figure illustrates:

Image non disponible

To allow you to access the streambuf, every iostream object has a member function called rdbuf( ) that returns the pointer to the object's streambuf. This way you can call any member function for the underlying streambuf. However, one of the most interesting things you can do with the streambuf pointer is to connect it to another iostream object using the << operator. This drains all the characters from your object into the one on the left side of the <<. If you want to move all the characters from one iostream to another, you don't need to go through the tedium (and potential coding errors) of reading them one character or one line at a time. This is a much more elegant approach.

Here's a simple program that opens a file and sends the contents to standard output (similar to the previous example):

 
Sélectionnez
//: C04:Stype.cpp
// Type a file to standard output.
#include <fstream>
#include <iostream>
#include "../require.h"
using namespace std;
 
int main() {
  ifstream in("Stype.cpp");
  assure(in, "Stype.cpp");
  cout << in.rdbuf(); // Outputs entire file
} ///:~

An ifstream is created using the source code file for this program as an argument. The assure( ) function reports a failure if the file cannot be opened. All the work really happens in the statement

 
Sélectionnez
cout << in.rdbuf();

which sends the entire contents of the file to cout. This is not only more succinct to code, it is often more efficient than moving the bytes one at a time.

A form of get( ) writes directly into the streambuf of another object. The first argument is a reference to the destination streambuf, and the second is the terminating character (‘\n' by default), which stops the get( ) function. So there is yet another way to print a file to standard output:

 
Sélectionnez
//: C04:Sbufget.cpp
// Copies a file to standard output.
#include <fstream>
#include <iostream>
#include "../require.h"
using namespace std;
 
int main() {
  ifstream in("Sbufget.cpp");
  assure(in);
  streambuf& sb = *cout.rdbuf();
  while(!in.get(sb).eof()) {
    if(in.fail())          // Found blank line
      in.clear();
    cout << char(in.get()); // Process '\n'
  }
} ///:~

The rdbuf( ) function returns a pointer, so it must be dereferenced to satisfy the function's need to see an object. Stream buffers are not meant to be copied (they have no copy constructor), so we define sb as a reference to cout's stream buffer. We need the calls to fail( ) and clear( ) in case the input file has a blank line (this one does). When this particular overloaded version of get( ) sees two newlines in a row (evidence of a blank line), it sets the input stream's fail bit, so we must call clear( ) to reset it so that the stream can continue to be read. The second call to get( ) extracts and echoes each newline delimiter. (Remember, the get( ) function doesn't extract its delimiter like getline( ) does.)

You probably won't need to use a technique like this often, but it's nice to know it exists.(45)

2-2-6. Seeking in iostreams

Each type of iostream has a concept of where its “next” character will come from (if it's an istream) or go (if it's an ostream). In some situations, you might want to move this stream position. You can do so using two models: one uses an absolute location in the stream called the streampos; the second works like the Standard C library functions fseek( ) for a file and moves a given number of bytes from the beginning, end, or current position in the file.

The streampos approach requires that you first call a “tell” function: tellp( ) for an ostream or tellg( ) for an istream. (The “p” refers to the “put pointer,” and the “g” refers to the “get pointer.”) This function returns a streampos you can later use in calls to seekp( ) for an ostream or seekg( ) for an istream when you want to return to that position in the stream.

The second approach is a relative seek and uses overloaded versions of seekp( ) and seekg( ). The first argument is the number of characters to move: it can be positive or negative. The second argument is the seek direction:

ios::beg From beginning of stream
ios::cur Current position in stream
ios::end From end of stream

Here's an example that shows the movement through a file, but remember, you're not limited to seeking within files as you are with C's stdio. With C++, you can seek in any type of iostream (although the standard stream objects, such as cin and cout,explicitly disallow it):

 
Sélectionnez
//: C04:Seeking.cpp
// Seeking in iostreams.
#include <cassert>
#include <cstddef>
#include <cstring>
#include <fstream>
#include "../require.h"
using namespace std;
 
int main() {
  const int STR_NUM = 5, STR_LEN = 30;
  char origData[STR_NUM][STR_LEN] = {
    "Hickory dickory dus. .
.",
    "Are you tired of C++?",
    "Well, if you have,",
    "That's just too bad,",
    "There's plenty more for us!"
  };
  char readData[STR_NUM][STR_LEN] = {{ 0
}};
  ofstream
out("Poem.bin", ios::out | ios::binary);
  assure(out, "Poem.bin");
  for(int i = 0; i < STR_NUM; i++)
    out.write(origData[i], STR_LEN);
  out.close();
  ifstream in("Poem.bin", ios::in |
ios::binary);
  assure(in, "Poem.bin");
  in.read(readData[0], STR_LEN);
  assert(strcmp(readData[0], "Hickory dickory dus.
. .")
    == 0);
  // Seek -STR_LEN bytes from the end of file
  in.seekg(-STR_LEN, ios::end);
  in.read(readData[1], STR_LEN);
  assert(strcmp(readData[1], "There's plenty more
for us!")
    == 0);
  // Absolute seek (like using operator[] with a file)
  in.seekg(3 * STR_LEN);
  in.read(readData[2], STR_LEN);
  assert(strcmp(readData[2], "That's just too
bad,") == 0);
  // Seek backwards from current position
  in.seekg(-STR_LEN * 2, ios::cur);
  in.read(readData[3], STR_LEN);
  assert(strcmp(readData[3], "Well, if you
have,") == 0);
  // Seek from the begining of the file
  in.seekg(1 * STR_LEN, ios::beg);
  in.read(readData[4], STR_LEN);
  assert(strcmp(readData[4], "Are you tired of
C++?")
    == 0);
} ///:~

This program writes a poem to a file using a binary output stream. Since we reopen it as an ifstream, we use seekg( ) to position the “get pointer.” As you can see, you can seek from the beginning or end of the file or from the current file position. Obviously, you must provide a positive number to move from the beginning of the file and a negative number to move back from the end.

Now that you know about the streambuf and how to seek, you can understand an alternative method (besides using an fstream object) for creating a stream object that will both read and write a file. The following code first creates an ifstream with flags that say it's both an input and an output file. You can't write to an ifstream, so you need to create an ostream with the underlying stream buffer:

 
Sélectionnez
ifstream in("filename", ios::in | ios::out);
ostream out(in.rdbuf());

You might wonder what happens when you write to one of these objects. Here's an example:

 
Sélectionnez
//: C04:Iofile.cpp
// Reading & writing one file.
#include <fstream>
#include <iostream>
#include "../require.h"
using namespace std;
 
int main() {
  ifstream in("Iofile.cpp");
  assure(in, "Iofile.cpp");
  ofstream out("Iofile.out");
  assure(out, "Iofile.out");
  out << in.rdbuf(); // Copy file
  in.close();
  out.close();
  // Open for reading and writing:
  ifstream in2("Iofile.out", ios::in |
ios::out);
  assure(in2, "Iofile.out");
  ostream out2(in2.rdbuf());
  cout << in2.rdbuf();  // Print whole file
  out2 << "Where does this end up?";
  out2.seekp(0, ios::beg);
  out2 << "And what about this?";
  in2.seekg(0, ios::beg);
  cout << in2.rdbuf();
} ///:~

The first five lines copy the source code for this program into a file called iofile.out and then close the files. This gives us a safe text file to play with. Then the aforementioned technique is used to create two objects that read and write to the same file. In cout << in2.rdbuf( ), you can see the “get” pointer is initialized to the beginning of the file. The “put” pointer, however, is set to the end of the file because “Where does this end up?” appears appended to the file. However, if the put pointer is moved to the beginning with a seekp( ), all the inserted text overwrites the existing text. Both writes are seen when the get pointer is moved back to the beginning with a seekg( ), and the file is displayed. The file is automatically saved and closed when out2 goes out of scope and its destructor is called.

2-2-7. String iostreams

A string stream works directly with memory instead of a file or standard output. It uses the same reading and formatting functions that you use with cin and cout to manipulate bytes in memory. On old computers, the memory was referred to as core, so this type of functionality is often called in-core formatting.

The class names for string streams echo those for file streams. If you want to create a string stream to extract characters from, you create an istringstream. If you want to put characters into a string stream, you create an ostringstream. All declarations for string streams are in the standard header <sstream>. As usual, there are class templates that fit into the iostreams hierarchy, as shown in the following figure:

Image non disponible
2-2-7-1. Input string streams

To read from a string using stream operations, you create an istringstream object initialized with the string. The following program shows how to use an istringstream object:

 
Sélectionnez
//: C04:Istring.cpp
// Input string streams.
#include <cassert>
#include <cmath>  // For fabs()
#include <iostream>
#include <limits> // For epsilon()
#include <sstream>
#include <string>
using namespace std;
 
int main() {
  istringstream s("47 1.414 This is a test");
  int i;
  double f;
  s >> i >> f; // Whitespace-delimited
input
  assert(i == 47);
  double relerr = (fabs(f) - 1.414) / 1.414;
  assert(relerr <=
numeric_limits<double>::epsilon());
  string buf2;
  s >> buf2;
  assert(buf2 == "This");
  cout << s.rdbuf(); // " is a test"
} ///:~

You can see that this is a more flexible and general approach to transforming character strings to typed values than the standard C library functions such as atof( )or atoi( ), even though the latter may be more efficient for single conversions.

In the expression s >> i >> f, the first number is extracted into i, and the second into f. This isn't “the first whitespace-delimited set of characters” because it depends on the data type it's being extracted into. For example, if the string were instead, “1.414 47 This is a test,” then i would get the value 1 because the input routine would stop at the decimal point. Then f would get 0.414. This could be useful if you want to break a floating-point number into a whole number and a fraction part. Otherwise it would seem to be an error. The second assert( ) calculates the relative error between what we read and what we expected; it's always better to do this than to compare floating-point numbers for equality. The constant returned by epsilon( ), defined in <limits>, represents the machine epsilon for double-precision numbers, which is the best tolerance you can expect comparisons of doubles to satisfy.(46)

As you may already have guessed, buf2 doesn't get the rest of the string, just the next white-space-delimited word. In general, it's best to use the extractor in iostreams when you know the exact sequence of data in the input stream and you're converting to some type other than a character string. However, if you want to extract the rest of the string all at once and send it to another iostream, you can use rdbuf( ) as shown.

To test the Date extractor at the beginning of this chapter, we used an input string stream with the following test program:

 
Sélectionnez
//: C04:DateIOTest.cpp
//{L} ../C02/Date
#include <iostream>
#include <sstream>
#include "../C02/Date.h"
using namespace std;
 
void testDate(const string& s) {
  istringstream os(s);
  Date d;
  os >> d;
  if(os)
    cout << d << endl;
  else
    cout << "input error with \""
<< s << "\"" << endl;
}
 
int main() {
  testDate("08-10-2003");
  testDate("8-10-2003");
  testDate("08 - 10 - 2003");
  testDate("A-10-2003");
  testDate("08/2003");
} ///:~

Each string literal in main( ) is passed by reference to testDate( ), which in turn wraps it in an istringstream so we can test thestream extractor we wrote for Date objects. The function testDate( ) also begins to test the inserter, operator<<( ).

2-2-7-2. Output string streams

To create an output string stream, you just create an ostringstream object, which manages a dynamically sized character buffer to hold whatever you insert. To get the formatted result as a string object, you call the str( ) member function. Here's an example:

 
Sélectionnez
//: C04:Ostring.cpp {RunByHand}
// Illustrates ostringstream.
#include <iostream>
#include <sstream>
#include <string>
using namespace std;
 
int main() {
  cout << "type an int, a float and a
string: ";
  int i;
  float f;
  cin >> i >> f;
  cin >> ws; // Throw away
white space
  string stuff;
  getline(cin, stuff); // Get rest of the line
  ostringstream os;
  os << "integer = "
<< i << endl;
  os << "float = " <<
f << endl;
  os << "string = "
<< stuff << endl;
  string result = os.str();
  cout << result << endl;
} ///:~

This is similar to the Istring.cpp example earlier that fetched an int and a float. A sample execution follows (the keyboard input is in bold type).

 
Sélectionnez
type an int, a float and a string: 10 20.5 the end
integer = 10
float = 20.5
string = the end

You can see that, like the other output streams, you can use the ordinary formatting tools, such as the << operator and endl, to send bytes to the ostringstream. The str( ) function returns a new string object every time you call it so the underlying stringbuf object owned by the string stream is left undisturbed.

In the previous chapter, we presented a program, HTMLStripper.cpp, that removed all HTML tags and special codes from a text file. As promised, here is a more elegant version using string streams.

 
Sélectionnez
//: C04:HTMLStripper2.cpp {RunByHand}
//{L} ../C03/ReplaceAll
// Filter to remove html tags and markers.
#include <cstddef>
#include <cstdlib>
#include <fstream>
#include <iostream>
#include <sstream>
#include <stdexcept>
#include <string>
#include "../C03/ReplaceAll.h"
#include "../require.h"
using namespace std;
 
string& stripHTMLTags(string& s)
throw(runtime_error) {
  size_t leftPos;
  while((leftPos = s.find('<')) != string::npos) {
    size_t rightPos = s.find('>', leftPos+1);
    if(rightPos == string::npos) {
      ostringstream msg;
      msg << "Incomplete HTML tag starting
in position "
          << leftPos;
      throw runtime_error(msg.str());
    }
    s.erase(leftPos, rightPos - leftPos + 1);
  }
  // Remove all special HTML characters
  replaceAll(s, "&lt;",
"<");
  replaceAll(s, "&gt;",
">");
  replaceAll(s, "&amp;",
"&");
  replaceAll(s, "&nbsp;", " ");
  // Etc...
  return s;
}
 
int main(int argc, char* argv[]) {
  requireArgs(argc, 1,
    "usage: HTMLStripper2 InputFile");
  ifstream in(argv[1]);
  assure(in, argv[1]);
  // Read entire file into string; then strip
  ostringstream ss;
  ss << in.rdbuf();
  try {
    string s = ss.str();
    cout << stripHTMLTags(s) << endl;
    return EXIT_SUCCESS;
  } catch(runtime_error& x) {
    cout << x.what() << endl;
    return EXIT_FAILURE;
  }
} ///:~

In this program we read the entire file into a string by inserting a rdbuf( ) call to the file stream into an ostringstream. Now it's an easy matter to search for HTML delimiter pairs and erase them without having to worry about crossing line boundaries like we had to with the previous version in Chapter 3.

The following example shows how to use a bidirectional (that is, read/write) string stream:

 
Sélectionnez
//: C04:StringSeeking.cpp {-bor}{-dmc}
// Reads and writes a string stream.
#include <cassert>
#include <sstream>
#include <string>
using namespace std;
 
int main() {
  string text = "We will hook no fish";
  stringstream ss(text);
  ss.seekp(0, ios::end);
  ss << " before its time.";
  assert(ss.str() ==
    "We will hook no fish before its time.");
  // Change "hook" to "ship"
  ss.seekg(8, ios::beg);
  string word;
  ss >> word;
  assert(word == "hook");
  ss.seekp(8, ios::beg);
  ss << "ship";
  // Change "fish" to "code"
  ss.seekg(16, ios::beg);
  ss >> word;
  assert(word == "fish");
  ss.seekp(16, ios::beg);
  ss << "code";
  assert(ss.str() ==
    "We will ship no code before its time.");
  ss.str("A horse of a different color.");
  assert(ss.str() == "A horse of a different
color.");
} ///:~

As always, to move the put pointer, you call seekp( ), and to reposition the get pointer, you call seekg( ). Even though we didn't show it with this example, string streams are a little more forgiving than file streams in that you can switch from reading to writing or vice-versa at any time. You don't need to reposition the get or put pointers or flush the stream. This program also illustrates the overload of str( ) that replaces the stream's underlying stringbuf with a new string.

2-2-8. Output stream formatting

The goal of the iostreams design is to allow you to easily move and/or format characters. It certainly wouldn't be useful if you couldn't do most of the formatting provided by C's printf( ) family of functions. In this section, you'll learn all the output formatting functions that are available for iostreams, so you can format your bytes the way you want them.

The formatting functions in iostreams can be somewhat confusing at first because there's often more than one way to control the formatting: through both member functions and manipulators. To further confuse things, a generic member function sets state flags to control formatting, such as left or right justification, to use uppercase letters for hex notation, to always use a decimal point for floating-point values, and so on. On the other hand, separate member functions set and read values for the fill character, the field width, and the precision.

In an attempt to clarify all this, we'll first examine the internal formatting data of an iostream, along with the member functions that can modify that data. (Everything can be controlled through the member functions, if desired.) We'll cover the manipulators separately.

2-2-8-1. Format flags

The class ios contains data members to store all the formatting information pertaining to a stream. Some of this data has a range of values and is stored in variables: the floating-point precision, the output field width, and the character used to pad the output (normally a space). The rest of the formatting is determined by flags, which are usually combined to save space and are referred to collectively as the format flags. You can find out the value of the format flags with the ios::flags( ) member function, which takes no arguments and returns an object of type fmtflags (usually a synonym for long) that contains the current format flags. All the rest of the functions make changes to the format flags and return the previous value of the format flags.

 
Sélectionnez
fmtflags ios::flags(fmtflags newflags);
fmtflags ios::setf(fmtflags ored_flag);
fmtflags ios::unsetf(fmtflags
clear_flag);
fmtflags ios::setf(fmtflags bits, fmtflags field);

The first function forces all the flags to change, which is sometimes what you want. More often, you change one flag at a time using the remaining three functions.

The use of setf( ) can seem somewhat confusing. To know which overloaded version to use, you must know what type of flag you're changing. There are two types of flags: those that are simply on or off, and those that work in a group with other flags. The on/off flags are the simplest to understand because you turn them on with setf(fmtflags) and off with unsetf(fmtflags). These flags are shown in the following table:

on/off flag Effect
ios::skipws Skip white space. (For input; this is the default.)
ios::showbase Indicate the numeric base (as set, for example, by dec, oct, or hex) when printing an integral value. Input streams also recognize the base prefix when showbase is on.
ios::showpoint Show decimal point and trailing zeros for floating-point values.
ios::uppercase Display uppercase A-F for hexadecimal values and E for scientific values.
ios::showpos Show plus sign (+) for positive values.
ios::unitbuf “Unit buffering.” The stream is flushed after each insertion.

For example, to show the plus sign for cout, you say cout.setf(ios::showpos). To stop showing the plus sign, you say cout.unsetf(ios::showpos).

The unitbuf flag controls unit buffering, which means that each insertion is flushed to its output stream immediately. This is handy for error tracing, so that in case of a program crash, your data is still written to the log file. The following program illustrates unit buffering:

 
Sélectionnez
//: C04:Unitbuf.cpp {RunByHand}
#include <cstdlib>  // For abort()
#include <fstream>
using namespace std;
 
int main() {
  ofstream out("log.txt");
  out.setf(ios::unitbuf);
  out << "one" << endl;
  out << "two" << endl;
  abort();
} ///:~

It is necessary to turn on unit buffering before any insertions are made to the stream. When we commented out the call to setf( ), one particular compiler had written only the letter ‘o' to the file log.txt. With unit buffering, no data was lost.

The standard error output stream cerr has unit buffering turned on by default. There is a cost for unit buffering, so if an output stream is heavily used, don't enable unit buffering unless efficiency is not a consideration.

2-2-8-2. Format fields

The second type of formatting flags work in a group. Only one of these flags can be set at a time, like the buttons on old car radios—you push one in, the rest pop out. Unfortunately this doesn't happen automatically, and you must pay attention to what flags you're setting so that you don't accidentally call the wrong setf( ) function. For example, there's a flag for each of the number bases: hexadecimal, decimal, and octal. Collectively, these flags are referred to as the ios::basefield. If the ios::dec flag is set and you call setf(ios::hex), you'll set the ios::hex flag, but you won't clear the ios::dec bit, resulting in undefined behavior. Instead, call the second form of setf( ) like this: setf(ios::hex, ios::basefield). This function first clears all the bits in the ios::basefield and then sets ios::hex. Thus, this form of setf( ) ensures that the other flags in the group “pop out” whenever you set one. The ios::hex manipulator does all this for you, automatically, so you don't need to concern yourself with the internal details of the implementation of this class or to even care that it's a set of binary flags. Later you'll see that there are manipulators to provide equivalent functionality in all the places you would use setf( ).

Here are the flag groups and their effects:

ios::basefield Effect
ios::dec Format integral values in base 10 (decimal) (the default radix—no prefix is visible).
ios::hex Format integral values in base 16 (hexadecimal).
ios::oct Format integral values in base 8 (octal).
ios::floatfield Effect
ios::scientific Display floating-point numbers in scientific format. Precision field indicates number of digits after the decimal point.
ios::fixed Display floating-point numbers in fixed format. Precision field indicates number of digits after the decimal point.
“automatic” (Neither bit is set.) Precision field indicates the total number of significant digits.
ios::adjustfield Effect
ios::left Left-align values; pad on the right with the fill character.
ios::right Right-align values. Pad on the left with the fill character. This is the default alignment.
ios::internal Add fill characters after any leading sign or base indicator, but before the value. (In other words, the sign, if printed, is left-justified while the number is right-justified.)
2-2-8-3. Width, fill, and precision

The internal variables that control the width of the output field, the fill character used to pad an output field, and the precision for printing floating-point numbers are read and written by member functions of the same name.

Function Effect
int ios::width( ) Returns the current width. Default is 0. Used for both insertion and extraction.
int ios::width(int n) Sets the width, returns the previous width.
int ios::fill( ) Returns the current fill character. Default is space.
int ios::fill(int n) Sets the fill character, returns the previous fill character.
int ios::precision( ) Returns current floating-point precision. Default is 6.
int ios::precision(int n) Sets floating-point precision, returns previous precision. See ios::floatfield table for the meaning of “precision.”

The fill and precision values are fairly straightforward, but width requires some explanation. When the width is zero, inserting a value produces the minimum number of characters necessary to represent that value. A positive width means that inserting a value will produce at least as many characters as the width; if the value has fewer than width characters, the fill character pad the field. However, the value will never be truncated, so if you try to print 123 with a width of two, you'll still get 123. The field width specifies a minimum number of characters; there's no way to specify a maximum number.

The width is also distinctly different because it's reset to zero by each inserter or extractor that could be influenced by its value. It's really not a state variable, but rather an implicit argument to the inserters and extractors. If you want a constant width, call width( ) after each insertion or extraction.

2-2-8-4. An exhaustive example

To make sure you know how to call all the functions previously discussed, here's an example that calls them all:

 
Sélectionnez
//: C04:Format.cpp
// Formatting Functions.
#include <fstream>
#include <iostream>
#include "../require.h"
using namespace std;
#define D(A) T << #A << endl; A
 
int main() {
  ofstream T("format.out");
  assure(T);
  D(int i = 47;)
  D(float f = 2300114.414159;)
  const char* s = "Is there any more?";
 
  D(T.setf(ios::unitbuf);)
  D(T.setf(ios::showbase);)
  D(T.setf(ios::uppercase | ios::showpos);)
  D(T << i << endl;) // Default is dec
  D(T.setf(ios::hex, ios::basefield);)
  D(T << i << endl;)
  D(T.setf(ios::oct, ios::basefield);)
  D(T << i << endl;)
  D(T.unsetf(ios::showbase);)
  D(T.setf(ios::dec, ios::basefield);)
  D(T.setf(ios::left, ios::adjustfield);)
  D(T.fill('0');)
  D(T << "fill char: " <<
T.fill() << endl;)
  D(T.width(10);)
  T << i << endl;
  D(T.setf(ios::right,
ios::adjustfield);)
  D(T.width(10);)
  T << i << endl;
  D(T.setf(ios::internal,
ios::adjustfield);)
  D(T.width(10);)
  T << i << endl;
  D(T << i << endl;)
// Without width(10)
 
  D(T.unsetf(ios::showpos);)
  D(T.setf(ios::showpoint);)
  D(T << "prec = " <<
T.precision() << endl;)
  D(T.setf(ios::scientific, ios::floatfield);)
  D(T << endl << f << endl;)
  D(T.unsetf(ios::uppercase);)
  D(T << endl << f << endl;)
  D(T.setf(ios::fixed, ios::floatfield);)
  D(T << f << endl;)
  D(T.precision(20);)
  D(T << "prec = " <<
T.precision() << endl;)
  D(T << endl << f << endl;)
  D(T.setf(ios::scientific, ios::floatfield);)
  D(T << endl << f << endl;)
  D(T.setf(ios::fixed, ios::floatfield);)
  D(T << f << endl;)
 
  D(T.width(10);)
  T << s << endl;
  D(T.width(40);)
  T << s << endl;
  D(T.setf(ios::left, ios::adjustfield);)
  D(T.width(40);)
  T << s << endl;
} ///:~

This example uses a trick to create a trace file so that you can monitor what's happening. The macro D(a) uses the preprocessor “stringizing” to turn a into a string to display. Then it reiterates a so the statement is executed. The macro sends all the information to a file called T, which is the trace file. The output is

 
Sélectionnez
int i = 47;
float f = 2300114.414159;
T.setf(ios::unitbuf);
T.setf(ios::showbase);
T.setf(ios::uppercase | ios::showpos);
T << i << endl;
+47
T.setf(ios::hex, ios::basefield);
T << i << endl;
0X2F
T.setf(ios::oct, ios::basefield);
T << i << endl;
057
T.unsetf(ios::showbase);
T.setf(ios::dec, ios::basefield);
T.setf(ios::left, ios::adjustfield);
T.fill('0');
T << "fill char: " << T.fill()
<< endl;
fill char: 0
T.width(10);
+470000000
T.setf(ios::right, ios::adjustfield);
T.width(10);
0000000+47
T.setf(ios::internal, ios::adjustfield);
T.width(10);
+000000047
T << i << endl;
+47
T.unsetf(ios::showpos);
T.setf(ios::showpoint);
T << "prec = " << T.precision()
<< endl;
prec = 6
T.setf(ios::scientific, ios::floatfield);
T << endl << f << endl;
 
2.300114E+06
T.unsetf(ios::uppercase);
T << endl << f << endl;
 
2.300114e+06
T.setf(ios::fixed, ios::floatfield);
T << f << endl;
2300114.500000
T.precision(20);
T << "prec = " << T.precision()
<< endl;
prec = 20
T << endl << f << endl;
 
2300114.50000000000000000000
T.setf(ios::scientific, ios::floatfield);
T << endl << f << endl;
 
2.30011450000000000000e+06
T.setf(ios::fixed, ios::floatfield);
T << f << endl;
2300114.50000000000000000000
T.width(10);
Is there any more?
T.width(40);
0000000000000000000000Is there any more?
T.setf(ios::left, ios::adjustfield);
T.width(40);
Is there any
more?0000000000000000000000

Studying this output should clarify your understanding of the iostream formatting member functions.

2-2-9. Manipulators

As you can see from the previous program, calling the member functions for stream formatting operations can get a bit tedious. To make things easier to read and write, a set of manipulators is supplied to duplicate the actions provided by the member functions. Manipulators are a convenience because you can insert them for their effect within a containing expression; you don't need to create a separate function-call statement.

Manipulators change the state of the stream instead of (or in addition to) processing data. When you insert endl in an output expression, for example, it not only inserts a newline character, but it also flushes the stream (that is, puts out all pending characters that have been stored in the internal stream buffer but not yet output). You can also just flush a stream like this:

 
Sélectionnez
cout << flush;

which causes a call to the flush( ) member function, as in:

 
Sélectionnez
cout.flush();

as a side effect (nothing is inserted into the stream). Additional basic manipulators will change the number base to oct (octal), dec (decimal) or hex (hexadecimal):

 
Sélectionnez
 cout << hex << "0x" << i << endl;

In this case, numeric output will continue in hexadecimal mode until you change it by inserting either dec or oct in the output stream.

There's also a manipulator for extraction that “eats” white space:

 
Sélectionnez
cin >> ws;

Manipulators with no arguments are provided in <iostream>. These include dec, oct, and hex, which perform the same action as, respectively, setf(ios::dec, ios::basefield), setf(ios::oct, ios::basefield), and setf(ios::hex, ios::basefield), albeit more succinctly. The <iostream> header also includes ws, endl, and flush and the additional set shown here:

Manipulator Effect
showbase noshowbase Indicate the numeric base (dec, oct, or hex) when printing an integral value.
showpos
noshowpos
Show plus sign (+) for positive values.
uppercase
nouppercase
Display uppercase A-F for hexadecimal values, and display E for scientific values.
showpoint
noshowpoint
Show decimal point and trailing zeros for floating-point values.
skipws
noskipws
Skip white space on input.
left
right

internal
Left-align, pad on right.
Right-align, pad on left.
Fill between leading sign or base indicator and value.
scientific
fixed
Indicates the display preference for floating-point output (scientific notation vs. fixed-point decimal).
2-2-9-1. Manipulators with arguments

There are six standard manipulators, such as setw( ), that take arguments. These are defined in the header file <iomanip>, and are summarized in the following table:

Manipulator effect
setiosflags (fmtflags n) Equivalent to a call to setf(n). The setting remains in effect until the next change, such as ios::setf( ).
resetiosflags (fmtflags n) Clears only the format flags specified by n. The setting remains in effect until the next change, such as ios::unsetf( ).
setbase (base n) Changes base to n, where n is 10, 8, or 16. (Anything else results in 0.) If n is zero, output is base 10, but input uses the C conventions: 10 is 10, 010 is 8, and 0xf is 15. You might as well use dec, oct, and hex for output.
setfill (char n) Changes the fill character to n, such as ios::fill( ).
setprecision (int n) Changes the precision to n, such as ios::precision( ).
setw(int n) Changes the field width to n, such as ios::width( ).

If you're doing a lot of formatting, you can see how using manipulators instead of calling stream member functions can clean up your code. As an example, here's the program from the previous section rewritten to use the manipulators. (The D( ) macro is removed to make it easier to read.)

 
Sélectionnez
//: C04:Manips.cpp
// Format.cpp using manipulators.
#include <fstream>
#include <iomanip>
#include <iostream>
using namespace std;
 
int main() {
  ofstream trc("trace.out");
  int i = 47;
  float f = 2300114.414159;
  char* s = "Is there any more?";
 
  trc << setiosflags(ios::unitbuf
           | ios::showbase | ios::uppercase
           | ios::showpos);
  trc << i << endl;
  trc << hex << i <<
endl
      << oct << i <<
endl;
  trc.setf(ios::left, ios::adjustfield);
  trc << resetiosflags(ios::showbase)
      << dec << setfill('0');
  trc << "fill char: " <<
trc.fill() << endl;
  trc << setw(10) << i << endl;
  trc.setf(ios::right, ios::adjustfield);
  trc << setw(10) << i << endl;
  trc.setf(ios::internal, ios::adjustfield);
  trc << setw(10) << i << endl;
  trc << i << endl; // Without setw(10)
 
  trc << resetiosflags(ios::showpos)
      << setiosflags(ios::showpoint)
      << "prec = " <<
trc.precision() << endl;
  trc.setf(ios::scientific, ios::floatfield);
  trc << f << resetiosflags(ios::uppercase)
<< endl;
  trc.setf(ios::fixed, ios::floatfield);
  trc << f << endl;
  trc << f << endl;
  trc << setprecision(20);
  trc << "prec = " <<
trc.precision() << endl;
  trc << f << endl;
  trc.setf(ios::scientific, ios::floatfield);
  trc << f << endl;
  trc.setf(ios::fixed, ios::floatfield);
  trc << f << endl;
  trc << f << endl;
 
  trc << setw(10) << s << endl;
  trc << setw(40) << s << endl;
  trc.setf(ios::left, ios::adjustfield);
  trc << setw(40) << s << endl;
} ///:~

You can see that a lot of the multiple statements have been condensed into a single chained insertion. Notice the call to setiosflags( ) in which the bitwise-OR of the flags is passed. This could also have been done with setf( ) and unsetf( ) as in the previous example.

When using setw( ) with an output stream, the output expression is formatted into a temporary string that is padded with the current fill character if needed, as determined by comparing the length of the formatted result to the argument of setw( ). In other words, setw( ) affects the result string of a formatted output operation. Likewise, using setw( ) with input streams only is meaningful when reading strings, as the following example makes clear:

 
Sélectionnez
//: C04:InputWidth.cpp
// Shows limitations of setw with input.
#include <cassert>
#include <cmath>
#include <iomanip>
#include <limits>
#include <sstream>
#include <string>
using namespace std;
 
int main() {
  istringstream is("one 2.34 five");
  string temp;
  is >> setw(2) >> temp;
  assert(temp == "on");
  is >> setw(2) >> temp;
  assert(temp == "e");
  double x;
  is >> setw(2) >> x;
  double relerr = fabs(x - 2.34) / x;
  assert(relerr <=
numeric_limits<double>::epsilon());
} ///:~

If you attempt to read a string, setw( ) will control the number of characters extracted quite nicely… up to a point. The first extraction gets two characters, but the second only gets one, even though we asked for two. That is because operator>>( ) uses white space as a delimiter (unless you turn off the skipws flag). When trying to read a number, however, such as x, you cannot use setw( ) to limit the characters read. With input streams, use only setw( ) for extracting strings.

2-2-9-2. Creating manipulators

Sometimes you'd like to create your own manipulators, and it turns out to be remarkably simple. A zero-argument manipulator such as endl is simply a function that takes as its argument an ostream reference and returns an ostream reference. The declaration for endl is

 
Sélectionnez
ostream& endl(ostream&);

Now, when you say:

 
Sélectionnez
cout << "howdy" << endl;

the endl produces the address of that function. So the compiler asks, “Is there a function that can be applied here that takes the address of a function as its argument?” Predefined functions in <iostream> do this; they're called applicators (because they apply a function to a stream). The applicator calls its function argument, passing it the ostream object as its argument. You don't need to know how applicators work to create your own manipulator; you only need to know that they exist. Here's the (simplified) code for an ostream applicator:

 
Sélectionnez
ostream& ostream::operator<<(ostream&
(*pf)(ostream&)) {
  return pf(*this);
}

The actual definition is a little more complicated since it involves templates, but this code illustrates the technique. When a function such as *pf (that takes a stream parameter and returns a stream reference) is inserted into a stream, this applicator function is called, which in turn executes the function to which pf points. Applicators for ios_base, basic_ios, basic_ostream, and basic_istream are predefined in the Standard C++ library.

To illustrate the process, here's a trivial example that creates a manipulator called nl that is equivalent to just inserting a newline into a stream (i.e., no flushing of the stream occurs, as with endl):

 
Sélectionnez
//: C04:nl.cpp
// Creating a manipulator.
#include <iostream>
using namespace std;
 
ostream& nl(ostream& os) {
  return os << '\n';
}
 
int main() {
  cout << "newlines" << nl
<< "between" << nl
       << "each" << nl <<
"word" << nl;
} ///:~

When you insert nl into an output stream, such as cout, the following sequence of calls ensues:

 
Sélectionnez
cout.operator<<(nl) è nl(cout)

The expression

 
Sélectionnez
os << '\n';

inside nl( ) calls ostream::operator(char), which returns the stream, which is what is ultimately returned from nl( ).(47)

2-2-9-3. Effectors

As you've seen, zero-argument manipulators are easy to create. But what if you want to create a manipulator that takes arguments? If you inspect the <iomanip> header, you'll see a type called smanip, which is what the manipulators with arguments return. You might be tempted to somehow use that type to define your own manipulators, but don't do it. The smanip type is implementation-dependent and thus not portable. Fortunately, you can define such manipulators in a straightforward way without any special machinery, based on a technique introduced by Jerry Schwarz, called an effector.(48) An effector is a simple class whose constructor formats a string representing the desired operation, along with an overloaded operator<< to insert that string into a stream. Here's an example with two effectors. The first outputs a truncated character string, and the second prints a number in binary.

 
Sélectionnez
//: C04:Effector.cpp
// Jerry Schwarz's "effectors."
#include <cassert>
#include <limits>  // For max()
#include <sstream>
#include <string>
using namespace std;
 
// Put out a prefix of a string:
class Fixw {
  string str;
public:
  Fixw(const string& s, int width) : str(s, 0,
width) {}
  friend ostream& operator<<(ostream& os,
const Fixw& fw) {
    return os << fw.str;
  }
};
 
// Print a number in binary:
typedef unsigned long ulong;
 
class Bin {
  ulong n;
public:
  Bin(ulong nn) { n = nn; }
  friend ostream& operator<<(ostream& os,
const Bin& b) {
    const ulong ULMAX =
numeric_limits<ulong>::max();
    ulong bit = ~(ULMAX >> 1); // Top bit set
    while(bit) {
      os << (b.n & bit ? '1' : '0');
      bit >>= 1;
    }
    return os;
  }
};
 
int main() {
  string words = "Things that make us happy, make
us wise";
  for(int i = words.size(); --i >= 0;) {
    ostringstream s;
    s << Fixw(words, i);
    assert(s.str() == words.substr(0, i));
  }
  ostringstream xs, ys;
  xs << Bin(0xCAFEBABEUL);
  assert(xs.str() ==
 
"1100""1010""1111""1110""1011""1010""1011""1110");
  ys << Bin(0x76543210UL);
  assert(ys.str() ==
 
"0111""0110""0101""0100""0011""0010""0001""0000");
} ///:~

The constructor for Fixw creates a shortened copy of its char* argument, and the destructor releases the memory created for this copy. The overloaded operator<< takes the contents of its second argument, the Fixw object, inserts it into the first argument, the ostream, and then returns the ostream so that it can be used in a chained expression. When you use Fixw in an expression like this:

 
Sélectionnez
cout << Fixw(string, i) << endl;

a temporary object is created by the call to the Fixw constructor, and that temporary object is passed to operator<<. The effect is that of a manipulator with arguments. The temporary Fixw object persists until the end of the statement.

The Bin effector relies on the fact that shifting an unsigned number to the right shifts zeros into the high bits. We use numeric_limits<unsigned long>::max( ) (the largest unsigned long value, from the standard header <limits>) to produce a value with the high bit set, and this value is moved across the number in question (by shifting it to the right), masking each bit in turn. We've juxtaposed string literals in the code for readability; the separate strings are concatenated into a single string by the compiler.

Historically, the problem with this technique was that once you created a class called Fixw for char* or Bin for unsigned long,no one else could create a different Fixw or Bin class for their type. However, with namespaces, this problem is eliminated. Effectors and manipulators aren't equivalent, although they can often be used to solve the same problem. If you find that an effector isn't enough, you will need to conquer the complexity of manipulators.

2-2-10. Iostream examples

In this section you'll see examples that use what you've learned in this chapter. Although many tools exist to manipulate bytes (stream editors such as sed and awk from UNIX are perhaps the most well known, but a text editor also fits this category), they generally have some limitations. Both sed and awk can be slow and can only handle lines in a forward sequence, and text editors usually require human interaction, or at least learning a proprietary macro language. The programs you write with iostreams have none of these limitations: they're fast, portable, and flexible.

2-2-10-1. Maintaining class library source code

Generally, when you create a class, you think in library terms: you make a header file Name.h for the class declaration, and then create a file called Name.cpp where the member functions are implemented. These files have certain requirements: a particular coding standard (the program shown here uses the coding format for this book), and preprocessor statements surrounding the code in the header file to prevent multiple declarations of classes. (Multiple declarations confuse the compiler—it doesn't know which one you want to use. They could be different, so it throws up its hands and gives an error message.)

This example creates a new header/implementation pair of files or modifies an existing pair. If the files already exist, it checks and potentially modifies the files, but if they don't exist, it creates them using the proper format.

 
Sélectionnez
//: C04:Cppcheck.cpp
// Configures .h & .cpp files to conform to style
// standard. Tests existing files for conformance.
#include <fstream>
#include <sstream>
#include <string>
#include <cstddef>
#include "../require.h"
using namespace std;
 
bool startsWith(const string& base, const
string& key) {
  return base.compare(0, key.size(), key) == 0;
}
 
void cppCheck(string fileName) {
  enum bufs { BASE, HEADER, IMPLEMENT, HLINE1, GUARD1,
    GUARD2, GUARD3, CPPLINE1, INCLUDE, BUFNUM };
  string part[BUFNUM];
  part[BASE] = fileName;
  // Find any '.' in the string:
  size_t loc = part[BASE].find('.');
  if(loc != string::npos)
    part[BASE].erase(loc); // Strip extension
  // Force to upper case:
  for(size_t i = 0; i < part[BASE].size(); i++)
    part[BASE][i] = toupper(part[BASE][i]);
  // Create file names and internal lines:
  part[HEADER] = part[BASE] + ".h";
  part[IMPLEMENT] = part[BASE] + ".cpp";
  part[HLINE1] = "/" ": " +
part[HEADER];
  part[GUARD1] = "ifndef " + part[BASE] +
"_H";
  part[GUARD2] = "define " + part[BASE] +
"_H";
  part[GUARD3] = "endif / " + part[BASE]
+"_H";
  part[CPPLINE1] = string("/") + ":
" + part[IMPLEMENT];
  part[INCLUDE] = "include \"" +
part[HEADER] + "\"";
  // First, try to open existing files:
  ifstream existh(part[HEADER].c_str()),
           existcpp(part[IMPLEMENT].c_str());
  if(!existh) { // Doesn't exist; create it
    ofstream newheader(part[HEADER].c_str());
    assure(newheader, part[HEADER].c_str());
    newheader << part[HLINE1] << endl
              << part[GUARD1] << endl
              << part[GUARD2] << endl
<< endl
              << part[GUARD3] << endl;
  } else { // Already exists; verify it
    stringstream hfile; // Write & read
    ostringstream newheader; // Write
    hfile << existh.rdbuf();
    // Check that first three lines conform:
    bool changed = false;
    string s;
    hfile.seekg(0);
    getline(hfile, s);
    bool lineUsed = false;
    // The call to good() is for Microsoft (later too):
    for(int line = HLINE1; hfile.good() && line
<= GUARD2;
         ++line) {
      if(startsWith(s, part[line])) {
        newheader << s << endl;
        lineUsed = true;
        if(getline(hfile, s))
          lineUsed = false;
      } else {
        newheader << part[line] << endl;
        changed = true;
        lineUsed = false;
      }
    }
    // Copy rest of file
    if(!lineUsed)
      newheader << s << endl;
    newheader << hfile.rdbuf();
    // Check for GUARD3
    string head = hfile.str();
    if(head.find(part[GUARD3]) == string::npos) {
      newheader << part[GUARD3] << endl;
      changed = true;
    }
    // If there were changes, overwrite file:
    if(changed) {
      existh.close();
      ofstream newH(part[HEADER].c_str());
      assure(newH, part[HEADER].c_str());
      newH << "//@//\n"  // Change
marker
           << newheader.str();
    }
  }
  if(!existcpp) { // Create cpp file
    ofstream newcpp(part[IMPLEMENT].c_str());
    assure(newcpp, part[IMPLEMENT].c_str());
    newcpp << part[CPPLINE1] << endl
           << part[INCLUDE] << endl;
  } else { // Already exists; verify it
    stringstream cppfile;
    ostringstream newcpp;
    cppfile << existcpp.rdbuf();
    // Check that first two lines conform:
    bool changed = false;
    string s;
    cppfile.seekg(0);
    getline(cppfile, s);
    bool lineUsed = false;
    for(int line = CPPLINE1;
        cppfile.good() && line <= INCLUDE;
++line) {
      if(startsWith(s, part[line])) {
        newcpp << s << endl;
        lineUsed = true;
        if(getline(cppfile, s))
          lineUsed = false;
      } else {
        newcpp << part[line] << endl;
        changed = true;
        lineUsed = false;
      }
    }
    // Copy rest of file
    if(!lineUsed)
      newcpp << s << endl;
    newcpp << cppfile.rdbuf();
    // If there were changes, overwrite file:
    if(changed) {
      existcpp.close();
      ofstream newCPP(part[IMPLEMENT].c_str());
      assure(newCPP, part[IMPLEMENT].c_str());
      newCPP << "//@//\n"  // Change
marker
             << newcpp.str();
    }
  }
}
 
int main(int argc, char* argv[]) {
  if(argc > 1)
    cppCheck(argv[1]);
  else
    cppCheck("cppCheckTest.h");
} ///:~

First notice the useful function startsWith( ), which does just what its name says—it returns true if the first string argument starts with the second argument. This is used when looking for the expected comments and include-related statements. Having the array of strings, part, allows for easy looping through the series of expected statements in source code. If the source file doesn't exist, we merely write the statements to a new file of the given name. If the file does exist, we search a line at a time, verifying that the expected lines occur. If they are not present, they are inserted. Special care must be taken to make sure we don't drop existing lines (see where we use the Boolean variable lineUsed). Notice that we use a stringstream for an existing file, so we can first write the contents of the file to it and then read from and search it.

The names in the enumeration are BASE, the capitalized base file name without extension; HEADER, the header file name; IMPLEMENT, the implementation file (cpp) name; HLINE1, the skeleton first line of the header file; GUARD1, GUARD2, and GUARD3, the “guard” lines in the header file (to prevent multiple inclusion); CPPLINE1, the skeleton first line of the cpp file; and INCLUDE, the line in the cpp file that includes the header file.

If you run this program without any arguments, the following two files are created:

 
Sélectionnez
// CPPCHECKTEST.h
#ifndef CPPCHECKTEST_H
#define CPPCHECKTEST_H
 
#endif // CPPCHECKTEST_H
 
Sélectionnez
// CPPCHECKTEST.cpp
#include
"CPPCHECKTEST.h"

(We removed the colon after the double-slash in the first comment lines so as not to confuse the book's code extractor. It will appear in the actual output produced by cppCheck.)

You can experiment by removing selected lines from these files and re-running the program. Each time you will see that the correct lines are added back in. When a file is modified, the string “//@//” is placed as the first line of the file to bring the change to your attention. You will need to remove this line before you process the file again (otherwise cppcheck will assume the initial comment line is missing).

2-2-10-2. Detecting compiler errors

All the code in this book is designed to compile as shown without errors. Lines of code that should generate a compile-time error may be commented out with the special comment sequence “//!”. The following program will remove these special comments and append a numbered comment to the line. When you run your compiler, it should generate error messages, and you will see all the numbers appear when you compile all the files. This program also appends the modified line to a special file so that you can easily locate any lines that don't generate errors.

 
Sélectionnez
//: C04:Showerr.cpp {RunByHand}
// Un-comment error generators.
#include <cstddef>
#include <cstdlib>
#include <cstdio>
#include <fstream>
#include <iostream>
#include <sstream>
#include <string>
#include "../require.h"
using namespace std;
 
const string USAGE =
  "usage: showerr filename chapnum\n"
  "where filename is a C++ source file\n"
  "and chapnum is the chapter name it's
in.\n"
  "Finds lines commented with //! and
removes\n"
  "the comment, appending //(#) where # is
unique\n"
  "across all files, so you can determine\n"
  "if your compiler finds the error.\n"
  "showerr /r\n"
  "resets the unique counter.";
 
class Showerr {
  const int CHAP;
  const string MARKER, FNAME;
  // File containing error number counter:
  const string ERRNUM;
  // File containing error lines:
  const string ERRFILE;
  stringstream edited; // Edited file
  int counter;
public:
  Showerr(const string& f, const string& en,
    const string& ef, int c)
  : CHAP(c), MARKER("//!"), FNAME(f),
ERRNUM(en),
    ERRFILE(ef), counter(0) {}
  void replaceErrors() {
    ifstream infile(FNAME.c_str());
    assure(infile, FNAME.c_str());
    ifstream count(ERRNUM.c_str());
    if(count) count >> counter;
    int linecount = 1;
    string buf;
    ofstream errlines(ERRFILE.c_str(), ios::app);
    assure(errlines, ERRFILE.c_str());
    while(getline(infile, buf)) {
      // Find marker at start of line:
      size_t pos = buf.find(MARKER);
      if(pos != string::npos) {
        // Erase marker:
        buf.erase(pos, MARKER.size() + 1);
        // Append counter & error info:
        ostringstream out;
        out << buf << " / ("
<< ++counter << ") "
            << "Chapter " << CHAP
            << " File: " << FNAME
            << " Line " <<
linecount << endl;
        edited << out.str();
        errlines << out.str(); // Append error
file
      }
      else
        edited << buf << "\n"; //
Just copy
      ++linecount;
    }
  }
  void saveFiles() {
    ofstream outfile(FNAME.c_str()); // Overwrites
    assure(outfile, FNAME.c_str());
    outfile << edited.rdbuf();
    ofstream count(ERRNUM.c_str()); // Overwrites
    assure(count, ERRNUM.c_str());
    count << counter; // Save new counter
  }
};
 
int main(int argc, char* argv[]) {
  const string ERRCOUNT("../errnum.txt"),
    ERRFILE("../errlines.txt");
  requireMinArgs(argc, 1, USAGE.c_str());
  if(argv[1][0] == '/' || argv[1][0] == '-') {
    // Allow for other switches:
    switch(argv[1][1]) {
      case 'r': case 'R':
        cout << "reset counter"
<< endl;
        remove(ERRCOUNT.c_str()); // Delete files
        remove(ERRFILE.c_str());
        return EXIT_SUCCESS;
      default:
        cerr << USAGE << endl;
        return EXIT_FAILURE;
    }
  }
  if(argc == 3) {
    Showerr s(argv[1], ERRCOUNT, ERRFILE, atoi(argv[2]));
    s.replaceErrors();
    s.saveFiles();
  }
} ///:~

You can replace the marker with one of your choice.

Each file is read a line at a time, and each line is searched for the marker appearing at the head of the line; the line is modified and put into the error line list and into the string stream, edited. When the whole file is processed, it is closed (by reaching the end of a scope), it is reopened as an output file, and edited is poured into the file. Also notice the counter is saved in an external file. The next time this program is invoked, it continues to increment the counter.

2-2-10-3. A simple data logger

This example shows an approach you might take to log data to disk and later retrieve it for processing. It is meant to produce a temperature-depth profile of the ocean at various points. The DataPoint class holds the data:

 
Sélectionnez
//: C04:DataLogger.h
// Datalogger record layout.
#ifndef DATALOG_H
#define DATALOG_H
#include <ctime>
#include <iosfwd>
#include <string>
using std::ostream;
 
struct Coord {
  int deg, min, sec;
  Coord(int d = 0, int m = 0, int s = 0)
  : deg(d), min(m), sec(s) {}
  std::string toString() const;
};
 
ostream& operator<<(ostream&, const
Coord&);
 
class DataPoint {
  std::time_t timestamp; // Time & day
  Coord latitude, longitude;
  double depth, temperature;
public:
  DataPoint(std::time_t ts, const Coord& lat,
            const Coord& lon, double dep, double
temp)
  : timestamp(ts), latitude(lat), longitude(lon),
      depth(dep), temperature(temp) {}
  DataPoint() : timestamp(0), depth(0), temperature(0)
{}
  friend ostream& operator<<(ostream&,
const DataPoint&);
};
#endif // DATALOG_H ///:~

A DataPoint consists of a time stamp, which is stored as a time_t value as defined in <ctime>, longitude and latitude coordinates, and values for depth and temperature. We use inserters for easy formatting. Here's the implementation file:

 
Sélectionnez
//: C04:DataLogger.cpp {O}
// Datapoint implementations.
#include "DataLogger.h"
#include <iomanip>
#include <iostream>
#include <sstream>
#include <string>
using namespace std;
 
ostream& operator<<(ostream& os, const
Coord& c) {
  return os << c.deg << '*'
<< c.min << '\''
            << c.sec
<< '"';
}
 
string Coord::toString() const {
  ostringstream os;
  os << *this;
  return os.str();
}
 
ostream& operator<<(ostream& os, const
DataPoint& d) {
  os.setf(ios::fixed, ios::floatfield);
  char fillc = os.fill('0'); // Pad on left with '0'
  tm* tdata =
localtime(&d.timestamp);
  os <<
setw(2) << tdata->tm_mon + 1 << '\\'
     << setw(2) <<
tdata->tm_mday << '\\'
     << setw(2) <<
tdata->tm_year+1900 << ' '
     << setw(2) << tdata->tm_hour
<< ':'
     << setw(2) <<
tdata->tm_min << ':'
     << setw(2) <<
tdata->tm_sec;
  os.fill(' '); // Pad on left with ' '
  streamsize prec = os.precision(4);
  os << " Lat:"    << setw(9)
<< d.latitude.toString()
     << ", Long:"  << setw(9)
<< d.longitude.toString()
     << ", depth:" << setw(9)
<< d.depth
     << ", temp:"  << setw(9)
<< d.temperature;
  os.fill(fillc);
  os.precision(prec);
  return os;
} ///:~

The Coord::toString( ) function is necessary because the DataPoint inserter calls setw( ) before it prints the latitude and longitude. If we used the stream inserter for Coord instead, the width would only apply to the first insertion (that is, to Coord::deg), since width changes are always reset immediately. The call to setf( ) causes the floating-point output to be fixed-precision, and precision( ) sets the number of decimal places to four. Notice how we restore the fill character and precision to whatever they were before the inserter was called.

To get the values from the time encoding stored in DataPoint::timestamp, we call the function std::localtime( ), which returns a static pointer to a tm object. The tm struct has the following layout:

 
Sélectionnez
struct tm {
  int tm_sec; // 0-59 seconds
  int tm_min; // 0-59 minutes
  int tm_hour; // 0-23 hours
  int tm_mday; // Day of month
  int tm_mon; // 0-11 months
  int tm_year; // Years since 1900
  int tm_wday; // Sunday == 0, etc.
  int tm_yday; // 0-365 day of year
  int tm_isdst; // Daylight savings?
};

Generating test data

Here's a program that creates a file of test data in binary form (using write( )) and a second file in ASCII form using the DataPoint inserter. You can also print it out to the screen, but it's easier to inspect in file form.

 
Sélectionnez
//: C04:Datagen.cpp
// Test data generator.
//{L} DataLogger
#include <cstdlib>
#include <ctime>
#include <cstring>
#include <fstream>
#include "DataLogger.h"
#include "../require.h"
using namespace std;
 
int main() {
  time_t timer;
  srand(time(&timer)); // Seed the random number
generator
  ofstream data("data.txt");
  assure(data, "data.txt");
  ofstream bindata("data.bin", ios::binary);
  assure(bindata, "data.bin");
  for(int i = 0; i < 100; i++, timer += 55) {
    // Zero to 199 meters:
    double newdepth  = rand() % 200;
    double fraction = rand() % 100 + 1;
    newdepth += 1.0 / fraction;
    double newtemp = 150 + rand() % 200; // Kelvin
    fraction = rand() % 100 + 1;
    newtemp += 1.0 / fraction;
    const DataPoint d(timer, Coord(45,20,31),
                      Coord(22,34,18), newdepth,
                      newtemp);
    data << d << endl;
    bindata.write(reinterpret_cast<const
char*>(&d),
                  sizeof(d));
  }
} ///:~

The file data.txt is created in the ordinary way as an ASCII file, but data.bin has the flag ios::binary to tell the constructor to set it up as a binary file. To illustrate the formatting used for the text file, here is the first line of data.txt (the line wraps because it's longer than this page will allow):

 
Sélectionnez
07\28\2003 12:54:40 Lat:45*20'31", Long:22*34'18", depth: 
16.0164, temp: 242.0122

The Standard C library function time( ) updates the time_t value its argument points to with an encoding of the current time, which on most platforms is the number of seconds elapsed since 00: 00: 00 GMT, January 1 1970 (the dawning of the age of Aquarius?). The current time is also a convenient way to seed the random number generator with the Standard C library function srand( ), as is done here.

After this, the timer is incremented by 55 seconds to give an interesting interval between readings in this simulation.

The latitude and longitude used are fixed values to indicate a set of readings at a single location. Both the depth and the temperature are generated with the Standard C library rand( ) function, which returns a pseudorandom number between zero and a platform-dependent constant, RAND_MAX, defined in <cstdlib> (usually the value of the platform's largest unsigned integer). To put this in a desired range, use the remainder operator % and the upper end of the range. These numbers are integral; to add a fractional part, a second call to rand( ) is made, and the value is inverted after adding one (to prevent divide-by-zero errors).

In effect, the data.bin file is being used as a container for the data in the program, even though the container exists on disk and not in RAM. write( ) sends the data out to the disk in binary form. The first argument is the starting address of the source block—notice it must be cast to a char* because that's what write( ) expects for narrow streams. The second argument is the number of characters to write, which in this case is the size of the DataPoint object (again, because we're using narrow streams). Because no pointers are contained in DataPoint, there is no problem in writing the object to disk. If the object is more sophisticated, you must implement a scheme for serialization, which writes the data referred to by pointers and defines new pointers when read back in later. (We don't talk about serialization in this volume—most vendor class libraries have some sort of serialization structure built into them.)

Verifying and viewing the data

To check the validity of the data stored in binary format, you can read it into memory with the read( ) member function for input streams, and compare it to the text file created earlier by Datagen.cpp. The following example just writes the formatted results to cout, but you can redirect this to a file and then use a file comparison utility to verify that it is identical to the original:

 
Sélectionnez
//: C04:Datascan.cpp
//{L} DataLogger
#include <fstream>
#include <iostream>
#include "DataLogger.h"
#include "../require.h"
using namespace std;
 
int main() {
  ifstream bindata("data.bin", ios::binary);
  assure(bindata, "data.bin");
  DataPoint d;
  while(bindata.read(reinterpret_cast<char*>(&d),
         sizeof d))
    cout << d << endl;
} ///:~

2-2-11. Internationalization

The software industry is now a healthy, worldwide economic market, with demand for applications that can run in various languages and cultures. As early as the late 1980s, the C Standards Committee added support for non-U.S. formatting conventions with their locale mechanism. A locale is a set of preferences for displaying certain entities such as dates and monetary quantities. In the 1990s, the C Standards Committee approved an addendum to Standard C that specified functions to handle wide characters (denoted by the type wchar_t), which allow support for character sets other than ASCII and its commonly used Western European extensions. Although the size of a wide character is not specified, some platforms implement them as 32-bit quantities, so they can hold the encodings specified by the Unicode Consortium, as well as mappings to multi-byte characters sets defined by Asian standards bodies. C++ has integrated support for both wide characters and locales into the iostreams library.

2-2-11-1. Wide Streams

A wide stream is a stream class that handles wide characters. All the examples so far (except for the last traits example in Chapter 3) have used narrow streams that hold instances of char. Since stream operations are essentially the same no matter the underlying character type, they are encapsulated generically as templates. So all input streams, for example, are connected to the basic_istream class template:

 
Sélectionnez
template<class charT, class traits =
char_traits<charT> >
class basic_istream {...};

In fact, all input stream types are specializations of this template, according to the following type definitions:

 
Sélectionnez
typedef basic_istream<char> istream;
typedef basic_istream<wchar_t> wistream;
typedef basic_ifstream<char> ifstream;
typedef basic_ifstream<wchar_t> wifstream;
typedef basic_istringstream<char> istringstream;
typedef
basic_istringstream<wchar_t> wistringstream;

All other stream types are defined in similar fashion.

In a perfect world, this is all you'd need to create streams of different character types. But things aren't that simple. The reason is that the character-processing functions provided for char and wchar_t don't have the same names. To compare two narrow strings, for example, you use the strcmp( ) function. For wide characters, that function is named wcscmp( ). (Remember these originated in C, which does not have function overloading, hence unique names are required.) For this reason, a generic stream can't just call strcmp( ) in response to a comparison operator. There needs to be a way for the correct low-level functions to be called automatically.

The solution is to factor out the differences into a new abstraction. The operations you can perform on characters have been abstracted into the char_traits template, which has predefined specializations for char and wchar_t, as we discussed at the end of the previous chapter. To compare two strings, then, basic_string just calls traits::compare( ) (remember that traits is the second template parameter), which in turn calls either strcmp( ) or wcscmp( ), depending on which specialization is being used (transparent to basic_string).

You only need to be concerned about char_traits if you access the low-level character processing functions; most of the time you don't care. Consider, however, making your inserters and extractors more robust by defining them as templates, just in case someone wants to use them on a wide stream.

To illustrate, recall again the Date class inserter from the beginning of this chapter. We originally declared it as:

 
Sélectionnez
ostream& operator<<(ostream&, const Date&);

This accommodates only narrow streams. To make it generic, we simply make it a template based on basic_ostream:

 
Sélectionnez
template<class charT, class traits>
std::basic_ostream<charT, traits>&
operator<<(std::basic_ostream<charT,
traits>& os,
           const Date& d) {
  charT fillc = os.fill(os.widen('0'));
  charT dash = os.widen('-');
  os << setw(2) << d.month << dash
     << setw(2) << d.day << dash
     << setw(4) << d.year;
  os.fill(fillc);
  return os;
}

Notice that we also have to replace char with the template parameter charT in the declaration of fillc, since it could be either char or wchar_t, depending on the template instantiation being used.

Since you don't know when you're writing the template which type of stream you have, you need a way to automatically convert character literals to the correct size for the stream. This is the job of the widen( ) member function. The expression widen('-'), for example, converts its argument to L'-' (the literal syntax equivalent to the conversion wchar_t(‘-')) if the stream is a wide stream and leaves it alone otherwise. There is also a narrow( ) function that converts to a char if needed.

We can use widen( ) to write a generic version of the nl manipulator we presented earlier in the chapter.

 
Sélectionnez
template<class charT, class traits>
basic_ostream<charT,traits>&
nl(basic_ostream<charT,traits>& os) {
  return os << charT(os.widen('\n'));
}
2-2-11-2. Locales

Perhaps the most notable difference in typical numeric computer output from country to country is the punctuator used to separate the integer and fractional parts of a real number. In the United States, a period denotes a decimal point, but in much of the world, a comma is expected instead. It would be quite inconvenient to do all your own formatting for locale-dependent displays. Once again, creating an abstraction that handles these differences solves the problem.

That abstraction is the locale. All streams have an associated locale object that they use for guidance on how to display certain quantities for different cultural environments. A locale manages the categories of culture-dependent display rules, which are defined as follows:

Category Effect
collate Allows comparing strings according to different, supported collating sequences.
ctype Abstracts the character classification and conversion facilities found in <cctype>.
monetary Supports different displays of monetary quantities.
numeric Supports different display formats of real numbers, including radix (decimal point) and grouping (thousands) separators.
time Supports various international formats for display of date and time.
messages Scaffolding to implement context-dependent message catalogs (such as for error messages in different languages).

The following program illustrates basic locale behavior:

 
Sélectionnez
//: C04:Locale.cpp {-g++}{-bor}{-edg} {RunByHand}
// Illustrates effects of locales.
#include <iostream>
#include <locale>
using namespace std;
 
int main() {
  locale def;
  cout << def.name() << endl;
  locale current = cout.getloc();
  cout << current.name() << endl;
  float val = 1234.56;
  cout << val << endl;
  // Change to French/France
  cout.imbue(locale("french"));
  current = cout.getloc();
  cout << current.name() << endl;
  cout << val << endl;
 
  cout << "Enter the literal 7890,12:
";
  cin.imbue(cout.getloc());
  cin >> val;
  cout << val << endl;
  cout.imbue(def);
  cout << val << endl;
} ///:~

Here's the output:

 
Sélectionnez
C
C
1234.56
French_France.1252
1234,56
Enter the literal 7890,12: 7890,12
7890,12
7890.12

The default locale is the “C” locale, which is what C and C++ programmers have been used to all these years (basically, English language and American culture). All streams are initially “imbued” with the “C” locale. The imbue( ) member function changes the locale that a stream uses. Notice that the full ISO name for the “French” locale is displayed (that is, French used in France vs. French used in another country). This example shows that this locale uses a comma for a radix point in numeric display. We have to change cin to the same locale if we want to do input according to the rules of this locale.

Each locale category is divided into number of facets, which are classes encapsulating the functionality that pertains to that category. For example, the time category has the facets time_put and time_get, which contain functions for doing time and date input and output respectively. The monetary category has facets money_get, money_put, and moneypunct. (The latter facet determines the currency symbol.) The following program illustrates the moneypunct facet. (The time facet requires a sophisticated use of iterators which is beyond the scope of this chapter.)

 
Sélectionnez
//: C04:Facets.cpp {-bor}{-g++}{-mwcc}{-edg}
#include <iostream>
#include <locale>
#include <string>
using namespace std;
 
int main() {
  // Change to French/France
  locale loc("french");
  cout.imbue(loc);
  string currency =
    use_facet<moneypunct<char>
>(loc).curr_symbol();
  char point =
    use_facet<moneypunct<char>
>(loc).decimal_point();
  cout << "I made " << currency
<< 12.34 << " today!"
       << endl;
} ///:~

The output shows the French currency symbol and decimal separator:

 
Sélectionnez
I made Ç12,34 today!

You can also define your own facets to construct customized locales.(49) Be aware that the overhead for locales is considerable. In fact, some library vendors provide different “flavors” of the Standard C++ library to accommodate environments that have limited space.(50)

2-2-12. Summary

This chapter has given you a fairly thorough introduction to the iostream class library. What you've seen here is likely to be all you need to create programs using iostreams. However, be aware that some additional features in iostreams are not used often, but you can discover them by looking at the iostream header files and by reading your compiler's documentation on iostreams or the references mentioned in this chapter and in the appendices.

2-2-13. Exercises

Solutions to selected exercises can be found in the electronic document The Thinking in C++ Volume 2 Annotated Solution Guide, available for a small fee from www.MindView.net.

  1. Open a file by creating an ifstream object. Make an ostringstream object and read the entire contents into the ostringstream using the rdbuf( ) member function. Extract a string copy of the underlying buffer and capitalize every character in the file using the Standard C toupper( ) macro defined in <cctype>. Write the result out to a new file.
  2. Create a program that opens a file (the first argument on the command line) and searches it for any one of a set of words (the remaining arguments on the command line). Read the input a line at a time, and write out the lines (with line numbers) that match to the new file.
  3. Write a program that adds a copyright notice to the beginning of all source-code files indicated by the program's command-line arguments.
  4. Use your favorite text-searching program (grep, for example) to output the names (only) of all the files that contain a particular pattern. Redirect the output into a file. Write a program that uses the contents of that file to generate a batch file that invokes your editor on each of the files found by the search program.
  5. We know that setw( ) allows for a minimum of characters read in, but what if you wanted to read a maximum? Write an effector that allows the user to specify a maximum number of characters to extract. Have your effector also work for output, in such a way that output fields are truncated, if necessary, to stay within width limits.
  6. Demonstrate to yourself that if the fail or bad bit is set, and you subsequently turn on stream exceptions, that the stream will immediately throw an exception.
  7. String streams accommodate easy conversions, but they come with a price. Write a program that races atoi( ) against the stringstream conversion system to see the effect of the overhead involved with stringstream.
  8. Make a Person struct with fields such as name, age, address, etc. Make the string fields fixed-size arrays. The social security number will be the key for each record. Implement the following Database class:
     
    Sélectionnez
    class DataBase {
    public:
      // Find where a record is on disk
      size_t query(size_t ssn);
      // Return the person at rn (record number)
      Person retrieve(size_t rn);
      // Record a record on disk
      void add(const Person& p);
    };
  9. Write some Person records to disk (do not keep them all in memory). When the user requests a record, read it off the disk and return it. The I/O operations in the DataBase class use read( ) and write( ) to process all Person records.
  10. Write an operator<< inserter for the Person struct that can be used to display records in a format for easy reading. Demonstrate it by writing data out to a file.
  11. Suppose the database for your Person structs was lost but that you have the file you wrote from the previous exercise. Recreate your database using this file. Be sure to use error checking.
  12. Write size_t(-1) (the largest unsigned int on your platform) to a text file 1,000,000 times. Repeat, but write to a binary file. Compare the size of the two files, and see how much room is saved using the binary format. (You may first want to calculate how much will be saved on your platform.)
  13. Discover the maximum number of digits of precision your implementation of iostreams will print by repeatedly increasing the value of the argument to precision( ) when printing a transcendental number such as sqrt(2.0).
  14. Write a program that reads real numbers from a file and prints their sum, average, minimum, and maximum.
  15. Determine the output of the following program before it is executed:
     
    Sélectionnez
    //: C04:Exercise14.cpp
    #include <fstream>
    #include <iostream>
    #include <sstream>
    #include "../require.h"
    using namespace std;
     
    #define d(a) cout << #a " ==\t"
    << a << endl;
     
    void tellPointers(fstream& s) {
      d(s.tellp());
      d(s.tellg());
      cout << endl;
    }
    void tellPointers(stringstream& s) {
      d(s.tellp());
      d(s.tellg());
      cout << endl;
    }
    int main() {
      fstream in("Exercise14.cpp");
      assure(in, "Exercise14.cpp");
      in.seekg(10);
      tellPointers(in);
      in.seekp(20);
      tellPointers(in);
      stringstream memStream("Here is a
    sentence.");
      memStream.seekg(10);
      tellPointers(memStream);
      memStream.seekp(5);
      tellPointers(memStream);
    } ///:~
  16. Suppose you are given line-oriented data in a file formatted as follows:
     
    Sélectionnez
    //: C04:Exercise15.txt
    Australia
    5E56,7667230284,Langler,Tyson,31.2147,0.00042117361
    2B97,7586701,Oneill,Zeke,553.429,0.0074673053156065
    4D75,7907252710,Nickerson,Kelly,761.612,0.010276276
    9F2,6882945012,Hartenbach,Neil,47.9637,0.0006471644
    Austria
    480F,7187262472,Oneill,Dee,264.012,0.00356226040013
    1B65,4754732628,Haney,Kim,7.33843,0.000099015948475
    DA1,1954960784,Pascente,Lester,56.5452,0.0007629529
    3F18,1839715659,Elsea,Chelsy,801.901,0.010819887645
    Belgium
    BDF,5993489554,Oneill,Meredith,283.404,0.0038239127
    5AC6,6612945602,Parisienne,Biff,557.74,0.0075254727
    6AD,6477082,Pennington,Lizanne,31.0807,0.0004193544
    4D0E,7861652688,Sisca,Francis,704.751,0.00950906238
    Bahamas
    37D8,6837424208,Parisienne,Samson,396.104,0.0053445
    5E98,6384069,Willis,Pam,90.4257,0.00122009564059246
    1462,1288616408,Stover,Hazal,583.939,0.007878970561
    5FF3,8028775718,Stromstedt,Bunk,39.8712,0.000537974
    1095,3737212,Stover,Denny,3.05387,0.000041205248883
    7428,2019381883,Parisienne,Shane,363.272,0.00490155
    ///:~

    The heading of each section is a region, and every line under that heading is a seller in that region. Each comma-separated field represents the data about each seller. The first field in a line is the SELLER_ID which unfortunately was written out in hexadecimal format. The second is the PHONE_NUMBER (notice that some are missing area codes). LAST_NAME and FIRST_NAME then follow. TOTAL_SALES is the second to the last column. The last column is the decimal amount of the total sales that the seller represents for the company. You are to format the data on the terminal window so that an executive can easily interpret the trends. Sample output is given below.

    Australia

    ---------------------------------

    *Last Name* *First Name* *ID* *Phone* *Sales* *Percent*

    Langler Tyson 24150 766-723-0284 31.24 4.21E-02

    Oneill Zeke 11159 XXX-758-6701 553.43 7.47E-01

    (etc.)

2-3. Templates in Depth

The C++ template facility goes far beyond simple “containers of T.” Although the original motivation was to enable type-safe, generic containers, in modern C++, templates are also used to generate custom code and to optimize program execution through compile-time programming constructs.

In this chapter we offer a practical look at the power (and pitfalls) of programming with templates in modern C++. For a more complete analysis of template-related language issues and pitfalls, we recommend the superb book by Daveed Vandevoorde and Nico Josuttis.(51)

2-3-1. Template parameters

As illustrated in Volume 1, templates come in two flavors: function templates and class templates. Both are wholly characterized by their parameters. Each template parameter can represent one of the following:

1.  Types (either built-in or user-defined).

2.  Compile-time constant values (for example, integers, and pointers and references to static entities; often referred to as non-type parameters).

3.  Other templates.

The examples in Volume 1 all fall into the first category and are the most common. The canonical example for simple container-like templates nowadays seems to be a Stack class. Being a container, a Stack object is not concerned with the type of object it stores; the logic of holding objects is independent of the type of objects being held. For this reason you can use a type parameter to represent the contained type:

 
Sélectionnez
template<class T> class Stack {
  T* data;
  size_t count;
public:
  void push(const T& t);
  // Etc.
};

The actual type to be used for a particular Stack instance is determined by the argument for the parameter T:

 
Sélectionnez
Stack<int> myStack;  // A Stack of ints

The compiler then provides an int-version of Stack by substituting int for T and generating the corresponding code. The name of the class instance generated from the template in this case is Stack<int>.

2-3-1-1. Non-type template parameters

A non-type template parameter must be an integral value that is known at compile time. You can make a fixed-size Stack, for instance, by specifying a non-type parameter to be used as the dimension for the underlying array, as follows.

 
Sélectionnez
template<class T, size_t N> class Stack {
  T data[N];  // Fixed capacity is N
  size_t count;
public:
  void push(const T& t);
  // Etc.
};

You must provide a compile-time constant value for the parameter N when you request an instance of this template, such as

 
Sélectionnez
Stack<int, 100> myFixedStack;

Because the value of N is known at compile time, the underlying array (data) can be placed on the runtime stack instead of on the free store. This can improve runtime performance by avoiding the overhead associated with dynamic memory allocation. Following the pattern mentioned earlier, the name of the class above is Stack<int, 100>. This means that each distinct value of N results in a unique class type. For example, Stack<int, 99> is a distinct class from Stack<int, 100>.

The bitset class template, discussed in detail in Chapter 7, is the only class in the Standard C++ library that uses a non-type template parameter (which specifies the number of bits the bitset object can hold). The following random number generator example uses a bitset to track numbers so all the numbers in its range are returned in random order without repetition before starting over. This example also overloads operator( ) to produce a familiar function-call syntax.

 
Sélectionnez
//: C05:Urand.h {-bor}
// Unique randomizer.
#ifndef URAND_H
#define URAND_H
#include <bitset>
#include <cstddef>
#include <cstdlib>
#include <ctime>
using std::size_t;
using std::bitset;
 
template<size_t UpperBound> class Urand {
  bitset<UpperBound> used;
public:
  Urand() { srand(time(0)); } // Randomize
  size_t operator()(); // The "generator"
function
};
 
template<size_t UpperBound>
inline size_t Urand<UpperBound>::operator()() {
  if(used.count() == UpperBound)
    used.reset();  // Start over (clear bitset)
  size_t newval;
  while(used[newval = rand() % UpperBound])
    ; // Until unique value is found
  used[newval] = true;
  return newval;
}
#endif // URAND_H ///:~

The numbers generated by Urand are unique because the bitset used tracks all the possible numbers in the random space (the upper bound is set with the template argument) and records each used number by setting the corresponding position bit. When the numbers are all used up, the bitset is cleared to start over. Here's a simple driver that illustrates how to use a Urand object:

 
Sélectionnez
//: C05:UrandTest.cpp {-bor}
#include <iostream>
#include "Urand.h"
using namespace std;
 
int main() {
  Urand<10> u;
  for(int i = 0; i < 20; ++i)
    cout << u() << ' ';
} ///:~

As we explain later in this chapter, non-type template arguments are also important in the optimization of numeric computations.

2-3-1-2. Default template arguments

You can provide default arguments for template parameters in class templates, but not function templates. As with default function arguments, they should only be defined once, the first time a template declaration or definition is seen by the compiler. Once you introduce a default argument, all the subsequent template parameters must also have defaults. To make the fixed-size Stack template shown earlier a little friendlier, for example, you can add a default argument like this:

 
Sélectionnez
template<class T, size_t N = 100> class Stack {
  T data[N];  // Fixed capacity is N
  size_t count;
public:
  void push(const T& t);
  // Etc.
};

Now, if you omit the second template argument when declaring a Stack object, the value for N will default to 100.

You can choose to provide defaults for all arguments, but you must use an empty set of brackets when declaring an instance so that the compiler knows that a class template is involved:

 
Sélectionnez
template<class T = int, size_t N = 100>  // Both
defaulted
class Stack {
  T data[N];  // Fixed capacity is N
  size_t count;
public:
  void push(const T& t);
  // Etc.
};
 
Stack<> myStack;  // Same as Stack<int, 100>

Default arguments are used heavily in the Standard C++ library. The vector class template, for instance, is declared as follows:

 
Sélectionnez
template<class T, class Allocator =
allocator<T> >
class vector;

Note the space between the last two right angle bracket characters. This prevents the compiler from interpreting those two characters (>>) as the right-shift operator.

This declaration reveals that vector takes two arguments: the type of the contained objects it holds, and a type that represents the allocator used by the vector. Whenever you omit the second argument, the standard allocator template is used, parameterized by the first template parameter. This declaration also shows that you can use template parameters in subsequent template parameters, as T is used here.

Although you cannot use default template arguments in function templates, you can use template parameters as default arguments to normal functions. The following function template adds the elements in a sequence:

 
Sélectionnez
//: C05:FuncDef.cpp
#include <iostream>
using namespace std;
 
template<class T> T sum(T* b, T* e, T init = T())
{
  while(b != e)
    init += *b++;
  return init;
}
 
int main() {
  int a[] = { 1, 2, 3 };
  cout << sum(a, a + sizeof a / sizeof a[0])
<< endl; // 6
} ///:~

The third argument to sum( ) is the initial value for the accumulation of the elements. Since we omitted it, this argument defaults to T( ), which in the case of int and other built-in types invokes a pseudo-constructor that performs zero-initialization.

2-3-1-3. Template template parameters

The third type of parameter a template can accept is another class template. If you are going to use a template type parameter itself as a template in your code, the compiler needs to know that the parameter is a template in the first place. The following example illustrates a template template parameter:

 
Sélectionnez
//: C05:TempTemp.cpp
// Illustrates a template template parameter.
#include <cstddef>
#include <iostream>
using namespace std;
 
template<class T>
class Array { // A simple, expandable sequence
  enum { INIT = 10 };
  T* data;
  size_t capacity;
  size_t count;
public:
  Array() {
    count = 0;
    data = new T[capacity = INIT];
  }
  ~Array() { delete [] data; }
  void push_back(const T& t) {
    if(count == capacity) {
      // Grow underlying array
      size_t newCap = 2 * capacity;
      T* newData = new T[newCap];
      for(size_t i = 0; i < count; ++i)
        newData[i] = data[i];
      delete [] data;
      data = newData;
      capacity = newCap;
    }
    data[count++] = t;
  }
  void pop_back() {
    if(count > 0)
      --count;
  }
  T* begin() { return data; }
  T* end() { return data + count; }
};
 
template<class T,
template<class> class Seq>
class Container {
  Seq<T> seq;
public:
  void append(const T& t) { seq.push_back(t); }
  T* begin() { return seq.begin(); }
  T* end() { return seq.end(); }
};
 
int main() {
  Container<int, Array> container;
  container.append(1);
  container.append(2);
  int* p = container.begin();
  while(p != container.end())
    cout << *p++ << endl;
} ///:~

The Array class template is a trivial sequence container. The Container template takes two parameters: the type that it holds, and a sequence data structure to do the holding. The following line in the implementation of the Container class requires that we inform the compiler that Seq is a template:

 
Sélectionnez
  Seq<T> seq;

If we hadn't declared Seq to be a template template parameter, the compiler would complain here that Seq is not a template, since we're using it as such. In main( ) a Container is instantiated to use an Array to hold integers, so Seq stands for Array in this example.

Note that it is not necessary in this case to name the parameter for Seq inside Container's declaration. The line in question is:

 
Sélectionnez
template<class T, template<class> class Seq>

Although we could have written

 
Sélectionnez
template<class T, template<class U> class Seq>

the parameter U is not needed anywhere. All that matters is that Seq is a class template that takes a single type parameter. This is analogous to omitting the names of function parameters when they're not needed, such as when you overload the post-increment operator:

 
Sélectionnez
T operator++(int);

The int here is merely a placeholder and so needs no name.

The following program uses a fixed-size array, which has an extra template parameter representing the array length:

 
Sélectionnez
//: C05:TempTemp2.cpp
// A multi-variate template template
parameter.
#include <cstddef>
#include <iostream>
using namespace std;
 
template<class T, size_t N> class Array {
  T data[N];
  size_t count;
public:
  Array() { count = 0; }
  void push_back(const T& t) {
    if(count < N)
      data[count++] = t;
  }
  void pop_back() {
    if(count > 0)
      --count;
  }
  T* begin() { return data; }
  T* end() { return data + count; }
};
 
template<class T,size_t
N,template<class,size_t> class Seq>
class Container {
  Seq<T,N> seq;
public:
  void append(const T& t) { seq.push_back(t); }
  T* begin() { return seq.begin(); }
  T* end() { return seq.end(); }
};
 
int main() {
  const size_t N = 10;
  Container<int, N, Array> container;
  container.append(1);
  container.append(2);
  int* p = container.begin();
  while(p != container.end())
    cout << *p++ << endl;
} ///:~

Once again, parameter names are not needed in the declaration of Seq inside Container's declaration, but we need two parameters to declare the data member seq, hence the appearance of the non-type parameter N at the top level.

Combining default arguments with template template parameters is slightly more problematic. When the compiler looks at the inner parameters of a template template parameter, default arguments are not considered, so you have to repeat the defaults in order to get an exact match. The following example uses a default argument for the fixed-size Array template and shows how to accommodate this quirk in the language:

 
Sélectionnez
//: C05:TempTemp3.cpp {-bor}{-msc}
// Template template parameters and default arguments.
#include <cstddef>
#include <iostream>
using namespace std;
 
template<class T, size_t N = 10>  // A default
argument
class Array {
  T data[N];
  size_t count;
public:
  Array() { count = 0; }
  void push_back(const T& t) {
    if(count < N)
      data[count++] = t;
  }
  void pop_back() {
    if(count > 0)
      --count;
  }
  T* begin() { return data; }
  T* end() { return data + count; }
};
 
template<class T, template<class, size_t = 10>
class Seq>
class Container {
  Seq<T> seq;  // Default used
public:
  void append(const T& t) { seq.push_back(t); }
  T* begin() { return seq.begin(); }
  T* end() { return seq.end(); }
};
 
int main() {
  Container<int, Array> container;
  container.append(1);
  container.append(2);
  int* p = container.begin();
  while(p != container.end())
    cout << *p++ << endl;
} ///:~

The default dimension of 10 is required in the line:

 
Sélectionnez
template<class T, template<class, size_t = 10> class Seq>

Both the definition of seq in Container and container in main( ) use the default. The only way to use something other than the default value was shown in TempTemp2.cpp. This is the only exception to the rule stated earlier that default arguments should appear only once in a compilation unit.

Since the standard sequence containers (vector, list, and deque, discussed in depth in Chapter 7) have a default allocator argument, the technique shown above is helpful should you ever want to pass one of these sequences as a template parameter. The following program passes a vector and then a list to two instances of Container:

 
Sélectionnez
//: C05:TempTemp4.cpp {-bor}{-msc}
// Passes standard sequences as template arguments.
#include <iostream>
#include <list>
#include <memory>  // Declares allocator<T>
#include <vector>
using namespace std;
 
template<class T,
template<class U, class = allocator<U> >
         class Seq>
class Container {
  Seq<T> seq; // Default of allocator<T>
applied implicitly
public:
  void push_back(const T& t) { seq.push_back(t); }
  typename Seq<T>::iterator begin() { return
seq.begin(); }
  typename Seq<T>::iterator end() { return
seq.end(); }
};
 
int main() {
  // Use a vector
  Container<int, vector> vContainer;
  vContainer.push_back(1);
  vContainer.push_back(2);
  for(vector<int>::iterator p = vContainer.begin();
      p != vContainer.end();
++p) {
    cout << *p <<
endl;
  }
  // Use a list
  Container<int, list> lContainer;
  lContainer.push_back(3);
  lContainer.push_back(4);
  for(list<int>::iterator p2 = lContainer.begin();
      p2 != lContainer.end(); ++p2) {
    cout << *p2 << endl;
  }
} ///:~

Here we name the first parameter of the inner template Seq (with the name U) because the allocators in the standard sequences must themselves be parameterized with the same type as the contained objects in the sequence. Also, since the default allocator parameter is known, we can omit it in the subsequent references to Seq<T>, as we did in the previous program. To fully explain this example, however, we have to discuss the semantics of the typename keyword.

2-3-1-4. The typename keyword

Consider the following:

 
Sélectionnez
//: C05:TypenamedID.cpp {-bor}
// Uses 'typename' as a prefix for nested types.
 
template<class T> class X {
  // Without typename, you should get an error:
  typename T::id i;
public:
  void f() { i.g(); }
};
 
class Y {
public:
  class id {
  public:
    void g() {}
  };
};
 
int main() {
  X<Y> xy;
  xy.f();
} ///:~

The template definition assumes that the class T that you hand it must have a nested identifier of some kind called id. Yet id could also be a static data member of T, in which case you can perform operations on id directly, but you can't “create an object” of “the type id.” In this example, the identifier id is being treated as if it were a nested type inside T. In the case of class Y, id is in fact a nested type, but (without the typename keyword) the compiler can't know that when it's compiling X.

If the compiler has the option of treating an identifier as a type or as something other than a type when it sees an identifier in a template, it will assume that the identifier refers to something other than a type. That is, it will assume that the identifier refers to an object (including variables of primitive types), an enumeration, or something similar. However, it will not—cannot—just assume that it is a type.

Because the default behavior of the compiler is to assume that a name that fits the above two points is not a type, you must use typename for nested names (except in constructor initializer lists, where it is neither needed nor allowed). In the above example, when the compiler sees typenameT::id, it knows (because of the typename keyword) that id refers to a nested type and thus it can create an object of that type.

The short version of the rule is: if a type referred to inside template code is qualified by a template type parameter, you must use the typename keyword as a prefix, unless it appears in a base class specification or initializer list in the same scope (in which case you must not).

The above explains the use of the typename keyword in the program TempTemp4.cpp. Without it, the compiler would assume that the expression Seq<T>::iterator is not a type, but we were using it to define the return type of the begin( ) and end( ) member functions.

The following example, which defines a function template that can print any Standard C++ sequence, shows a similar use of typename:

 
Sélectionnez
//: C05:PrintSeq.cpp {-msc}{-mwcc}
// A print function for Standard C++ sequences.
#include <iostream>
#include <list>
#include <memory>
#include <vector>
using namespace std;
 
template<class T, template<class U, class =
allocator<U> >
         class Seq>
void printSeq(Seq<T>& seq) {
  for(typename Seq<T>::iterator b = seq.begin();
       b != seq.end();)
    cout << *b++ << endl;
}
 
int main() {
  // Process a vector
  vector<int> v;
  v.push_back(1);
  v.push_back(2);
  printSeq(v);
  // Process a list
  list<int> lst;
  lst.push_back(3);
  lst.push_back(4);
  printSeq(lst);
} ///:~

Once again, without the typename keyword the compiler will interpret iterator as a static data member of Seq<T>, which is a syntax error, since a type is required.

Typedefing a typename

It's important not to assume that the typename keyword creates a new type name. It doesn't. Its purpose is to inform the compiler that the qualified identifier is to be interpreted as a type. A line that reads:

 
Sélectionnez
typename Seq<T>::iterator It;

causes a variable named It to be declared of type Seq<T>::iterator. If you mean to create a new type name, you should use typedef, as usual, as in:

 
Sélectionnez
typedef typename Seq<It>::iterator It;

Using typename instead of class

Another role of the typename keyword is to provide you the option of using typename instead of class in the template argument list of a template definition:

 
Sélectionnez
//: C05:UsingTypename.cpp
// Using 'typename' in the template argument list.
 
template<typename T> class X {};
 
int main() {
  X<int> x;
} ///:~

To some, this produces clearer code.

2-3-1-5. Using the template keyword as a hint

Just as the typename keyword helps the compiler when a type identifier is not expected, there is also a potential difficulty with tokens that are not identifiers, such as the < and > characters. Sometimes these represent the less-than or greater-than symbols, and sometimes they delimit template parameter lists. As an example, we'll once more use the bitset class:

 
Sélectionnez
//: C05:DotTemplate.cpp
// Illustrate the .template construct.
#include <bitset>
#include <cstddef>
#include <iostream>
#include <string>
using namespace std;
 
template<class charT, size_t N>
basic_string<charT> bitsetToString(const
bitset<N>& bs) {
  return bs. template to_string<charT,
char_traits<charT>,
                                allocator<charT>
>();
}
 
int main() {
  bitset<10> bs;
  bs.set(1);
  bs.set(5);
  cout << bs << endl; // 0000100010
  string s = bitsetToString<char>(bs);
  cout << s << endl;  // 0000100010
} ///:~

The bitset class supports conversion to string object via its to_string member function. To support multiple string classes, to_string is itself a template, following the pattern established by the basic_string template discussed in Chapter 3. The declaration of to_string inside of bitset looks like this:

 
Sélectionnez
template<class charT, class traits, class
Allocator>
basic_string<charT, traits, Allocator> to_string() const;

Our bitsetToString( ) function template above requests different types of string representations of a bitset. To get a wide string, for instance, you change the call to the following:

 
Sélectionnez
  wstring s = bitsetToString<wchar_t>(bs);

Note that basic_string uses default template arguments, so we don't need to repeat the char_traits and allocator arguments in the return value. Unfortunately, bitset::to_string does not use default arguments. Using bitsetToString<char>( bs) is more convenient than typing a fully-qualified call to bs.template to_string<char, char_traits, allocator<char> >( ) every time.

The return statement in bitsetToString( ) contains the template keyword in an odd place—right after the dot operator applied to the bitset object bs. This is because when the template is parsed, the < character after the to_string token would be interpreted as a less-than operation instead of the beginning of a template argument list. The template keyword used in this context tells the compiler that what follows is the name of a template, causing the < character to be interpreted correctly. The same reasoning applies to the -> and ::operators when applied to templates. As with the typename keyword, this template disambiguation technique can only be used within template code.(52)

2-3-1-6. Member Templates

The bitset::to_string( ) function template is an example of a member template: a template declared within another class or class template. This allows many combinations of independent template arguments to be combined. A useful example is found in the complex class template in the Standard C++ library. The complex template has a type parameter meant to represent an underlying floating-point type to hold the real and imaginary parts of a complex number. The following code snippet from the standard library shows a member-template constructor in the complex class template:

 
Sélectionnez
template<typename T> class
complex {
public:
  template<class X> complex(const complex<X>&);

The standard complex template comes ready-made with specializations that use float, double, and long double for the parameter T. The member-template constructor above creates a new complex number that uses a different floating-point type as its base type, as seen in the code below:

 
Sélectionnez
  complex<float> z(1, 2);
  complex<double> w(z);

In the declaration of w, the complex template parameter T is double and X is float. Member templates make this kind of flexible conversion easy.

Defining a template within a template is a nesting operation, so the prefixes that introduce the templates must reflect this nesting if you define the member template outside the outer class definition. For example, if you implement the complex class template, and if you define the member-template constructor outside the complex template class definition, you do it like this:

 
Sélectionnez
template<typename T>
template<typename X>
complex<T>::complex(const
complex<X>& c) {/* Body here… */}

Another use of member function templates in the standard library is in the initialization of containers. Suppose we have a vector of ints and we want to initialize a new vector of doubles with it, like this:

 
Sélectionnez
  int data[5] = { 1, 2, 3, 4, 5 };
  vector<int> v1(data, data+5);
  vector<double> v2(v1.begin(), v1.end());

As long as the elements in v1 are assignment-compatible with the elements in v2 (as double and int are here), all is well. The vector class template has the following member template constructor:

 
Sélectionnez
template<class InputIterator>
vector(InputIterator first, InputIterator last,
       const Allocator& = Allocator());

This constructor is used twice in the vector declarations above. When v1 is initialized from the array of ints, the type InputIterator is int*. When v2 is initialized from v1, an instance of the member template constructor is used with InputIterator representing vector<int>::iterator.

Member templates can also be classes. (They don't need to be functions.) The following example shows a member class template inside an outer class template:

 
Sélectionnez
//: C05:MemberClass.cpp
// A member class template.
#include <iostream>
#include <typeinfo>
using namespace std;
 
template<class T> class Outer {
public:
  template<class R> class Inner {
  public:
    void f();
  };
};
 
template<class T> template<class R>
void Outer<T>::Inner<R>::f() {
  cout << "Outer == " <<
typeid(T).name() << endl;
  cout << "Inner == " <<
typeid(R).name() << endl;
  cout << "Full Inner == " <<
typeid(*this).name() << endl;
}
 
int main() {
  Outer<int>::Inner<bool> inner;
  inner.f();
} ///:~

The typeid operator, covered in Chapter 8, takes a single argument and returns a type_info object whose name( ) function yields a string representing the argument type. For example, typeid(int).name( ) might return the string “int.” (The actual string is platform-dependent.) The typeid operator can also take an expression and return a type_info object representing the type of that expression. For example, with int i, typeid(i).name( ) returns something like “int,” and typeid(&i).name( ) returns something like “int *.”

The output of the program above should be something like this:

 
Sélectionnez
Outer == int
Inner == bool
Full Inner == Outer<int>::Inner<bool>

The declaration of the variable inner in the main program instantiates both Inner<bool> and Outer<int>.

Member template functions cannot be declared virtual. Current compiler technology expects to be able to determine the size of a class's virtual function table when the class is parsed. Allowing virtual member template functions would require knowing all calls to such member functions everywhere in the program ahead of time. This is not feasible, especially for multi-file projects.

2-3-2. Function template issues

Just as a class template describes a family of classes, a function template describes a family of functions. The syntax for creating either type of template is virtually identical, but they differ somewhat in how they are used. You must always use angle brackets when instantiating class templates and you must supply all non-default template arguments. However, with function templates you can often omit the template arguments, and default template arguments are not even allowed. Consider a typical implementation of the min( ) function template declared in the <algorithm> header, which looks something like this:

 
Sélectionnez
template<typename T> const T& min(const
T& a, const T& b) {
  return (a < b) ? a : b;
}

You could invoke this template by providing the type of the arguments in angle brackets, just like you do with class templates, as in:

 
Sélectionnez
int z = min<int>(i, j);

This syntax tells the compiler that a specialization of the min( ) template is needed with int used in place of the parameter T, whereupon the compiler generates the corresponding code. Following the pattern of naming the classes generated from class templates, you can think of the name of the instantiated function as min<int>( ).

2-3-2-1. Type deduction of function template arguments

You can always use such explicit function template specification as in the example above, but it is often convenient to leave off the template arguments and let the compiler deduce them from the function arguments, like this:

 
Sélectionnez
int z = min(i, j);

If both i and j are ints, the compiler knows that you need min<int>( ), which it then instantiates automatically. The types must be identical because the template was originally specified with only one template type argument used for both function parameters. No standard conversions are applied for function arguments whose type is specified by a template parameter. For example, if you wanted to find the minimum of an int and a double, the following attempt at a call to min( ) would fail:

 
Sélectionnez
int z = min(x, j); // x is a double

Since x and j are distinct types, no single parameter matches the template parameter T in the definition of min( ), so the call does not match the template declaration. You can work around this difficulty by casting one argument to the other's type or by reverting to the fully-specified call syntax, as in:

 
Sélectionnez
int z = min<double>(x, j);

This tells the compiler to generate the double version of min( ), after which j can be promoted to a double by normal standard conversion rules (because the function min<double>(const double&, const double&) would then exist).

You might be tempted to require two parameters for min( ), allowing the types of the arguments to be independent, like this:

 
Sélectionnez
template<typename T, typename
U>
const T& min(const T& a,
const U& b) {
  return (a < b) ? a : b;
}

This is often a good strategy, but here it is problematic because min( ) must return a value, and there is no satisfactory way to determine which type it should be (T or U?).

If the return type of a function template is an independent template parameter, you must always specify its type explicitly when you call it, since there is no argument from which to deduce it. Such is the case with the fromString template below.

 
Sélectionnez
//: C05:StringConv.h
// Function templates to convert to and from strings.
#ifndef STRINGCONV_H
#define STRINGCONV_H
#include <string>
#include <sstream>
 
template<typename T> T fromString(const
std::string& s) {
  std::istringstream is(s);
  T t;
  is >> t;
  return t;
}
 
template<typename T> std::string toString(const
T& t) {
  std::ostringstream s;
  s << t;
  return s.str();
}
#endif // STRINGCONV_H ///:~

These function templates provide conversions to and from std::string for any types that provide a stream inserter or extractor, respectively. Here's a test program that includes the use of the standard library complex number type:

 
Sélectionnez
//: C05:StringConvTest.cpp
#include <complex>
#include <iostream>
#include "StringConv.h"
using namespace std;
 
int main() {
  int i = 1234;
  cout << "i == \""
<< toString(i) << "\"" << endl;
  float x = 567.89;
  cout << "x == \"" <<
toString(x) << "\"" << endl;
  complex<float> c(1.0, 2.0);
  cout << "c == \"" <<
toString(c) << "\"" << endl;
  cout << endl;
 
  i =
fromString<int>(string("1234"));
  cout << "i == "
<< i << endl;
  x = fromString<float>(string("567.89"));
  cout << "x == " << x <<
endl;
  c = fromString<complex<float>
>(string("(1.0,2.0)"));
  cout << "c == " << c <<
endl;
} ///:~

The output is what you'd expect:

 
Sélectionnez
i == "1234"
x == "567.89"
c == "(1,2)"
 
i == 1234
x == 567.89
c == (1,2)

Notice that in each of the instantiations of fromString( ), the template parameter is specified in the call. If you have a function template with template parameters for the parameter types as well as the return types, it is important to declare the return type parameter first, otherwise you won't be able to omit the type parameters for the function parameters. As an illustration, consider the following well-known function template:(53)

 
Sélectionnez
//: C05:ImplicitCast.cpp
 
template<typename R, typename
P>
R implicit_cast(const P& p) {
  return p;
}
 
int main() {
  int i = 1;
  float x = implicit_cast<float>(i);
  int j = implicit_cast<int>(x);
  //! char* p = implicit_cast<char*>(i);
} ///:~

If you interchange R and P in the template parameter list near the top of the file, it will be impossible to compile this program because the return type will remain unspecified (the first template parameter would be the function's parameter type). The last line (which is commented out) is illegal because there is no standard conversion from int to char*. implicit_cast is for revealing conversions in your code that are allowed naturally.

With a little care you can even deduce array dimensions. This example has an array-initialization function template (init2) that performs such a deduction:

 
Sélectionnez
//: C05:ArraySize.cpp
#include <cstddef>
using std::size_t;
 
template<size_t R, size_t C, typename T>
void init1(T a[R][C]) {
  for(size_t i = 0; i < R; ++i)
    for(size_t j = 0; j < C; ++j)
      a[i][j] = T();
}
 
template<size_t R, size_t C, class T>
void init2(T (&a)[R][C]) {  // Reference parameter
  for(size_t i = 0; i < R; ++i)
    for(size_t j = 0; j < C; ++j)
      a[i][j] = T();
}
 
int main() {
  int a[10][20];
  init1<10,20>(a);  // Must specify
  init2(a);         // Sizes deduced
} ///:~

Array dimensions are not passed as part of a function parameter's type unless that parameter is passed by pointer or reference. The function template init2 declares a to be a reference to a two-dimensional array, so its dimensions R and C are deduced by the template facility, making init2 a handy way to initialize a two-dimensional array of any size. The template init1 does not pass the array by reference, so the sizes must be explicitly specified, although the type parameter can still be deduced.

2-3-2-2. Function template overloading

As with ordinary functions, you can overload function templates that have the same name. When the compiler processes a function call in a program, it has to decide which template or ordinary function is the “best” fit for the call. Along with the min( ) function template introduced earlier, let's add some ordinary functions to the mix:

 
Sélectionnez
//: C05:MinTest.cpp
#include <cstring>
#include <iostream>
using std::strcmp;
using std::cout;
using std::endl;
 
template<typename T> const T& min(const
T& a, const T& b) {
  return (a < b) ? a : b;
}
 
const char* min(const char* a, const char* b) {
  return (strcmp(a, b) < 0) ? a : b;
}
 
double min(double x, double y) {
  return (x < y) ? x : y;
}
 
int main() {
  const char *s2 = "say \"Ni-!\"",
*s1 = "knights who";
  cout << min(1, 2) << endl;      // 1: 1
(template)
  cout << min(1.0, 2.0) << endl;  // 2: 1
(double)
  cout << min(1, 2.0) << endl;    // 3: 1
(double)
  cout << min(s1, s2) << endl;    // 4:
knights who (const
                                  //                
char*)
  cout << min<>(s1, s2) << endl;  //
5: say "Ni-!"
                                  //    (template)
} ///:~

In addition to the function template, this program defines two non-template functions: a C-style string version of min( ) and a double version. If the template doesn't exist, the call in line 1 above invokes the double version of min( ) because of the standard conversion from int to double. The template can generate an int version which is considered a better match, so that's what happens. The call in line 2 is an exact match for the double version, and the call in line 3 also invokes the same function, implicitly converting 1 to 1.0. In line 4 the const char* version of min( ) is called directly. In line 5 we force the compiler to use the template facility by appending empty angle brackets to the function name, whereupon it generates a const char* version from the template and uses it (which is verified by the wrong answer—it's just comparing addresses!(54)). If you're wondering why we have using declarations in lieu of the using namespace std directive, it's because some compilers include headers behind the scenes that bring in std::min( ), which would conflict with our declarations of the name min( ).

As stated above, you can overload templates of the same name, as long as they can be distinguished by the compiler. You could, for example, declare a min( ) function template that processes three arguments:

 
Sélectionnez
template<typename T>
const T& min(const T& a, const T& b, const T& c);

Versions of this template will be generated only for calls to min( ) that have three arguments of the same type.

2-3-2-3. Taking the address of a generated function template

In some situations you need to take the address of a function. For example, you may have a function that takes an argument of a pointer to another function. It's possible that this other function might be generated from a template function, so you need some way to take that kind of address:(55)

 
Sélectionnez
//: C05:TemplateFunctionAddress.cpp {-mwcc}
// Taking the address of a function generated
// from a template.
 
template<typename T> void f(T*) {}
 
void h(void (*pf)(int*)) {}
 
template<typename T> void g(void (*pf)(T*)) {}
 
int main() {
  h(&f<int>); // Full type specification
  h(&f); // Type deduction
  g<int>(&f<int>); // Full type
specification
  g(&f<int>); // Type deduction
  g<int>(&f); // Partial (but sufficient)
specification
} ///:~

This example demonstrates a number of issues. First, even though you're using templates, the signatures must match. The function h( ) takes a pointer to a function that takes an int* and returns void, and that's what the template f( ) produces. Second, the function that wants the function pointer as an argument can itself be a template, as in the case of the template g( ).

In main( ) you can see that type deduction works here, too. The first call to h( ) explicitly gives the template argument for f( ), but since h( ) says that it will only take the address of a function that takes an int*, that part can be deduced by the compiler. With g( ) the situation is even more interesting because two templates are involved. The compiler cannot deduce the type with nothing to go on, but if either f( ) or g( ) is given int, the rest can be deduced.

An obscure issue arises when trying to pass the functions tolower or toupper, declared in <cctype>, as parameters. It is possible to use these, for example, with the transform algorithm (which is covered in detail in the next chapter) to convert a string to lower or upper case. You must be careful because there are multiple declarations for these functions. A naive approach would be something like this:

 
Sélectionnez
// The variable s is a std::string
transform(s.begin(), s.end(), s.begin(), tolower);

The transform algorithm applies its fourth parameter (tolower( ) in this case) to each character in the string s and places the result in s itself, thus overwriting each character in s with its lower-case equivalent. As it is written, this statement may or may not work! It fails in the following context:

 
Sélectionnez
//: C05:FailedTransform.cpp {-xo}
#include <algorithm>
#include <cctype>
#include <iostream>
#include <string>
using namespace std;
 
int main() {
  string s("LOWER");
  transform(s.begin(), s.end(), s.begin(), tolower);
  cout << s << endl;
} ///:~

Even if your compiler lets you get away with this, it is illegal. The reason is that the <iostream> header also makes available a two-argument version of tolower( ) and toupper( ):

 
Sélectionnez
template<class charT> charT toupper(charT c,
                                    const locale&
loc);
template<class charT> charT tolower(charT c,
                                    const locale& loc);

These function templates take a second argument of type locale. The compiler has no way of knowing whether it should use the one-argument version of tolower( ) defined in <cctype> or the one mentioned above. You can solve this problem (almost!) with a cast in the call to transform, as follows:

 
Sélectionnez
  transform(s.begin(),s.end(),s.begin()
            static_cast<int (*)(int)>(tolower));

(Recall that tolower( ) and toupper( ) work with int instead of char.) The cast above makes clear that the single-argument version of tolower( ) is desired. This works with some compilers, but it is not required to. The reason, albeit obscure, is that a library implementation is allowed to give “C linkage” (meaning that the function name does not contain all the auxiliary information(56) that normal C++ functions do) to functions inherited from the C language. If this is the case, the cast fails because transform is a C++ function template and expects its fourth argument to have C++ linkage—and a cast is not allowed to change the linkage. What a predicament!

The solution is to place tolower( ) calls in an unambiguous context. For example, you could write a function named strTolower( ) and place it in its own file without including <iostream>, like this:

 
Sélectionnez
//: C05:StrTolower.cpp {O} {-mwcc}
#include <algorithm>
#include <cctype>
#include <string>
using namespace std;
 
string strTolower(string s) {
  transform(s.begin(), s.end(), s.begin(), tolower);
  return s;
} ///:~

The header <iostream> is not involved here, and the compilers we use do not introduce the two-argument version of tolower( ) in this context,(57) so there's no problem. You can then use this function normally:

 
Sélectionnez
//: C05:Tolower.cpp {-mwcc}
//{L} StrTolower
#include <algorithm>
#include <cctype>
#include <iostream>
#include <string>
using namespace std;
string strTolower(string);
 
int main() {
  string s("LOWER");
  cout << strTolower(s) << endl;
} ///:~

Another solution is to write a wrapper function template that calls the correct version of tolower( ) explicitly:

 
Sélectionnez
//: C05:ToLower2.cpp {-mwcc}
#include <algorithm>
#include <cctype>
#include <iostream>
#include <string>
using namespace std;
 
template<class charT> charT strTolower(charT c) {
  return tolower(c);  // One-arg version called
}
 
int main() {
  string s("LOWER");
 
transform(s.begin(),s.end(),s.begin(),&strTolower<char>);
  cout << s << endl;
} ///:~

This version has the advantage that it can process both wide and narrow strings since the underlying character type is a template parameter. The C++ Standards Committee is working on modifying the language so that the first example (without the cast) will work, and some day these workarounds can be ignored.(58)

2-3-2-4. Applying a function to an STL sequence

Suppose you want to take an STL sequence container (which you'll learn more about in subsequent chapters; for now we can just use the familiar vector) and apply a member function to all the objects it contains. Because a vector can contain any type of object, you need a function that works with any type of vector:

 
Sélectionnez
//: C05:ApplySequence.h
// Apply a function to an STL sequence container.
 
// const, 0 arguments, any type of return value:
template<class Seq, class T, class R>
void apply(Seq& sq, R (T::*f)() const) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)();
}
 
// const, 1 argument, any type of return value:
template<class Seq, class T, class R, class A>
void apply(Seq& sq, R(T::*f)(A) const, A a) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)(a);
}
 
// const, 2 arguments, any type of return value:
template<class Seq, class T, class R,
         class A1, class A2>
void apply(Seq& sq, R(T::*f)(A1, A2) const,
    A1 a1, A2 a2) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)(a1, a2);
}
 
// Non-const, 0 arguments, any type of return value:
template<class Seq, class T, class R>
void apply(Seq& sq, R (T::*f)()) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)();
}
 
// Non-const, 1 argument, any type of return value:
template<class Seq, class T, class R, class A>
void apply(Seq& sq, R(T::*f)(A), A a) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)(a);
}
 
// Non-const, 2 arguments, any type of return value:
template<class Seq, class T, class R,
         class A1, class A2>
void apply(Seq& sq, R(T::*f)(A1, A2),
    A1 a1, A2 a2) {
  typename Seq::iterator it = sq.begin();
  while(it != sq.end())
    ((*it++)->*f)(a1, a2);
}
// Etc., to handle maximum
likely arguments ///:~

The apply( ) function template above takes a reference to the container class and a pointer-to-member for a member function of the objects contained in the class. It uses an iterator to move through the sequence and apply the function to every object. We have overloaded on the const-ness of the function so you can use it with both const and non-const functions.

Notice that there are no STL header files (or any header files, for that matter) included in applySequence.h, so it is not limited to use with an STL container. However, it does make assumptions (primarily, the name and behavior of the iterator) that apply to STL sequences, and it also assumes that the elements of the container be of pointer type.

You can see there is more than one version of apply( ), further illustrating overloading of function templates. Although these templates allow any type of return value (which is ignored, but the type information is required to match the pointer-to-member), each version takes a different number of arguments, and because it's a template, those arguments can be of any type. The only limitation here is that there's no “super template” to create templates for you; you must decide how many arguments will ever be required and make the appropriate definitions.

To test the various overloaded versions of apply( ), the class Gromit(59) is created containing functions with different numbers of arguments, and both const and non-const member functions:

 
Sélectionnez
//: C05:Gromit.h
// The techno-dog. Has member functions
// with various numbers of arguments.
#include <iostream>
 
class Gromit {
  int arf;
  int totalBarks;
public:
  Gromit(int arf = 1) : arf(arf + 1), totalBarks(0) {}
  void speak(int) {
    for(int i = 0; i < arf; i++) {
      std::cout << "arf! ";
      ++totalBarks;
    }
    std::cout << std::endl;
  }
  char eat(float) const {
    std::cout << "chomp!" <<
std::endl;
    return 'z';
  }
  int sleep(char, double) const {
    std::cout << "zzz..." <<
std::endl;
    return 0;
  }
  void sit() const {
    std::cout << "Sitting..." <<
std::endl;
  }
}; ///:~

Now you can use the apply( ) template functions to apply the Gromit member functions to a vector<Gromit*>, like this:

 
Sélectionnez
//: C05:ApplyGromit.cpp
// Test ApplySequence.h.
#include <cstddef>
#include <iostream>
#include <vector>
#include "ApplySequence.h"
#include "Gromit.h"
#include "../purge.h"
using namespace std;
 
int main() {
  vector<Gromit*> dogs;
  for(size_t i = 0; i < 5; i++)
    dogs.push_back(new Gromit(i));
  apply(dogs, &Gromit::speak, 1);
  apply(dogs, &Gromit::eat, 2.0f);
  apply(dogs, &Gromit::sleep, 'z', 3.0);
  apply(dogs, &Gromit::sit);
  purge(dogs);
} ///:~

The purge( ) function is a small utility that calls delete on every element of sequence. You'll find it defined in Chapter 7, and used various places in this book.

Although the definition of apply( ) is somewhat complex and not something you'd ever expect a novice to understand, its use is remarkably clean and simple, and a novice could use it knowing only what it is intended to accomplish, not how. This is the type of division you should strive for in all your program components. The tough details are all isolated on the designer's side of the wall. Users are concerned only with accomplishing their goals and don't see, know about, or depend on details of the underlying implementation. We'll explore even more flexible ways to apply functions to sequences in the next chapter.

2-3-2-5. Partial ordering of function templates

We mentioned earlier that an ordinary function overload of min( ) is preferable to using the template. If a function already exists to match a function call, why generate another? In the absence of ordinary functions, however, it is possible that overloaded function templates can lead to ambiguities. To minimize the chances of this, an ordering is defined for function templates, which chooses the most specialized template, if such exists. A function template is considered more specialized than another if every possible list of arguments that matches it also matches the other, but not the other way around. Consider the following function template declarations, taken from an example in the C++ Standard document:

 
Sélectionnez
template<class T> void f(T);
template<class T> void f(T*);
template<class T> void f(const T*);

The first template can be matched with any type. The second template is more specialized than the first because only pointer types match it. In other words, you can look upon the set of possible calls that match the second template as a subset of the first. A similar relationship exists between the second and third template declarations above: the third can only be called for pointers to const, but the second accommodates any pointer type. The following program illustrates these rules:

 
Sélectionnez
//: C05:PartialOrder.cpp
// Reveals ordering of function templates.
#include <iostream>
using namespace std;
 
template<class T> void f(T) {
  cout << "T" << endl;
}
 
template<class T> void f(T*) {
  cout << "T*" << endl;
}
 
template<class T> void f(const T*) {
  cout << "const T*" << endl;
}
 
int main() {
  f(0);            // T
  int i = 0;
  f(&i);           // T*
  const int j = 0;
  f(&j);           // const T*
} ///:~

The call f(&i) certainly matches the first template, but since the second is more specialized, it is called. The third can't be called here since the pointer is not a pointer to const. The call f(&j) matches all three templates (for example, T would be const int in the second template), but again, the third template is more specialized, so it is used instead.

If there is no “most specialized” template among a set of overloaded function templates, an ambiguity remains, and the compiler will report an error. That is why this feature is called a “partial ordering”—it may not be able to resolve all possibilities. Similar rules exist for class templates (see the section “Partial specialization” below).

2-3-3. Template specialization

The term specialization has a specific, template-related meaning in C++. A template definition is, by its very nature, a generalization, because it describes a family of functions or classes in general terms. When template arguments are supplied, the result is a specialization of the template because it determines a unique instance out of the many possible instances of the family of functions or classes. The min( ) function template seen at the beginning of this chapter is a generalization of a minimum-finding function because the type of its parameters is not specified. When you supply the type for the template parameter, whether explicitly or implicitly via argument deduction, the resultant code generated by the compiler (for example, min<int>( )) is a specialization of the template. The code generated is also considered an instantiation of the template, as are all code bodies generated by the template facility.

2-3-3-1. Explicit specialization

You can also provide the code yourself for a given template specialization, should the need arise. Providing your own template specializations is often needed with class templates, but we will begin with the min( ) function template to introduce the syntax.

Recall that in MinTest.cpp earlier in this chapter we introduced the following ordinary function:

 
Sélectionnez
const char* min(const char* a, const char* b) {
  return (strcmp(a, b) < 0) ? a : b;
}

This was so that a call to min( ) would compare strings and not addresses. Although it would provide no advantage here, we could instead define a const char* specialization for min( ), as in the following program:

 
Sélectionnez
//: C05:MinTest2.cpp
#include <cstring>
#include <iostream>
using std::strcmp;
using std::cout;
using std::endl;
 
template<class T> const T& min(const T&
a, const T& b) {
  return (a < b) ? a : b;
}
 
// An explicit specialization of the min template
template<>
const char* const& min<const char*>(const
char* const& a,
                                    const char*
const& b) {
  return (strcmp(a, b) < 0) ? a : b;
}
 
int main() {
  const char *s2 = "say \"Ni-!\"",
*s1 = "knights who";
  cout << min(s1, s2) << endl;
  cout << min<>(s1, s2) << endl;
} ///:~

The “template<>” prefix tells the compiler that what follows is a specialization of a template. The type for the specialization must appear in angle brackets immediately following the function name, as it normally would in an explicitly specified call. Note that we carefully substitute const char* for T in the explicit specialization. Whenever the original template specifies const T, that const modifies the whole type T. It is the pointer to a const char* that is const. So we must write const char* const in place of const T in the specialization. When the compiler sees a call to min( ) with const char* arguments in the program, it will instantiate our const char* version of min( ) so it can be called. The two calls to min( ) in this program call the same specialization of min( ).

Explicit specializations tend to be more useful for class templates than for function templates. When you provide a full specialization for a class template, though, you may need to implement all the member functions. This is because you are providing a separate class, and client code may expect the complete interface to be implemented.

The standard library has an explicit specialization for vector when it holds objects of type bool. The purpose for vector<bool> is to allow library implementations to save space by packing bits into integers.(60)

As you saw earlier in this chapter, the declaration for the primary vector class template is:

 
Sélectionnez
template<class T, class Allocator =
allocator<T> >
class vector {...};

To specialize for objects of type bool, you could declare an explicit specialization as follows:

 
Sélectionnez
template<> class
vector<bool, allocator<bool> > {...};

Again, this is quickly recognized as a full, explicit specialization because of the template<> prefix and because all the primary template's parameters are satisfied by the argument list appended to the class name.

It turns out that vector<bool> is a little more flexible than we have described, as seen in the next section.

2-3-3-2. Partial Specialization

Class templates can also be partially specialized, meaning that at least one of the template parameters is left “open” in some way in the specialization. vector<bool> specifies the object type (bool), but leaves the allocator type unspecified. Here is the actual declaration of vector<bool>:

 
Sélectionnez
template<class Allocator> class vector<bool,
Allocator>;

You can recognize a partial specialization because non-empty parameter lists appear in angle brackets both after the template keyword (the unspecified parameters) and after the class (the specified arguments). Because of the way vector<bool> is defined, a user can provide a custom allocator type, even though the contained type of bool is fixed. In other words, specialization, and partial specialization in particular, constitute a sort of “overloading” for class templates.

Partial ordering of class templates

The rules that determine which template is selected for instantiation are similar to the partial ordering for function templates—the “most specialized” template is selected. The string in each f( ) member function in the illustration below explains the role of each template definition:

 
Sélectionnez
//: C05:PartialOrder2.cpp
// Reveals partial ordering of class templates.
#include <iostream>
using namespace std;
 
template<class T, class U> class C {
public:
  void f() { cout << "Primary Template\n";
}
};
 
template<class U> class C<int, U> {
public:
  void f() { cout << "T == int\n"; }
};
 
template<class T> class C<T, double> {
public:
  void f() { cout << "U == double\n”; }
};
 
template<class T, class U> class C<T*, U> {
public:
  void f() { cout << "T* used\n”; }
};
 
template<class T, class U> class C<T, U*> {
public:
  void f() { cout << "U* used\n”; }
};
 
template<class T, class U> class C<T*, U*>
{
public:
  void f() { cout << "T* and U* used\n”; }
};
 
template<class T> class C<T, T> {
public:
  void f() { cout << "T == U\n”; }
};
 
int main() {
  C<float, int>().f();    // 1: Primary template
  C<int, float>().f();    // 2: T == int
  C<float, double>().f(); // 3: U == double
  C<float, float>().f();  // 4: T == U
  C<float*, float>().f(); // 5: T* used [T is
float]
  C<float, float*>().f(); // 6: U* used [U is
float]
  C<float*, int*>().f();  // 7: T* and U* used
[float,int]
  // The following are ambiguous:
//   8: C<int, int>().f();
//   9: C<double, double>().f();
//  10: C<float*, float*>().f();
//  11: C<int, int*>().f();
//  12: C<int*, int*>().f();
} ///:~

As you can see, you can partially specify template parameters according to whether they are pointer types, or whether they are equal. When the T* specialization is used, such as is the case in line 5, T itself is not the top-level pointer type that was passed—it is the type that the pointer refers to (float, in this case). The T* specification is a pattern to allow matching against pointer types. If you use int** as the first template argument, T becomes int*. Line 8 is ambiguous because having the first parameter as an int versus having the two parameters equal are independent issues—one is not more specialized than the other. Similar logic applies to lines 9 through 12.

2-3-3-3. A practical example

You can easily derive from a class template, and you can create a new template that instantiates and inherits from an existing template. If the vector template does most everything you want, for example, but in a certain application you'd also like a version that can sort itself, you can easily reuse the vector code. The following example derives from vector<T> and adds sorting. Note that deriving from vector, which doesn't have a virtual destructor, would be dangerous if we needed to perform cleanup in our destructor.

 
Sélectionnez
//: C05:Sortable.h
// Template specialization.
#ifndef SORTABLE_H
#define SORTABLE_H
#include <cstring>
#include <cstddef>
#include <string>
#include <vector>
using std::size_t;
 
template<class T>
class Sortable : public std::vector<T> {
public:
  void sort();
};
 
template<class T>
void Sortable<T>::sort() { // A simple sort
  for(size_t i = this->size(); i > 0; --i)
    for(size_t j = 1; j < i; ++j)
      if(this->at(j-1) > this->at(j)) {
        T t = this->at(j-1);
        this->at(j-1) = this->at(j);
        this->at(j) = t;
      }
}
 
// Partial specialization for
pointers:
template<class T>
class Sortable<T*> :
public std::vector<T*> {
public:
  void sort();
};
 
template<class T>
void Sortable<T*>::sort() {
  for(size_t i = this->size(); i > 0; --i)
    for(size_t j = 1; j < i; ++j)
      if(*this->at(j-1) > *this->at(j)) {
        T* t = this->at(j-1);
        this->at(j-1) = this->at(j);
        this->at(j) = t;
      }
}
 
// Full specialization for char*
// (Made inline here for
convenience -- normally you would
// place the function body in a separate
file and only
// leave the declaration here).
template<> inline void Sortable<char*>::sort()
{
  for(size_t i = this->size(); i > 0; --i)
    for(size_t j = 1; j < i; ++j)
      if(std::strcmp(this->at(j-1), this->at(j))
> 0) {
        char* t = this->at(j-1);
        this->at(j-1) = this->at(j);
        this->at(j) = t;
      }
}
#endif // SORTABLE_H ///:~

The Sortable template imposes a restriction on all but one of the classes for which it is instantiated: they must contain a > operator. It works correctly only with non-pointer objects (including objects of built-in types). The full specialization compares the elements using strcmp( ) to sort vectors of char* according to the null-terminated strings to which they refer. The use of “this->” above is mandatory(61) and is explained in the section entitled “Name lookup issues” later in this chapter.(62)

Here's a driver for Sortable.h that uses the randomizer introduced earlier in the chapter:

 
Sélectionnez
//: C05:Sortable.cpp
//{-bor} (Because of bitset in Urand.h)
// Testing template specialization.
#include <cstddef>
#include <iostream>
#include "Sortable.h"
#include "Urand.h"
using namespace std;
 
#define asz(a) (sizeof a / sizeof a[0])
 
char* words[] = { "is", "running",
"big", "dog", "a", };
char* words2[] = { "this", "that",
"theother", };
 
int main() {
  Sortable<int> is;
  Urand<47> rnd;
  for(size_t i = 0; i < 15; ++i)
    is.push_back(rnd());
  for(size_t i = 0; i < is.size(); ++i)
    cout << is[i] << ' ';
  cout << endl;
  is.sort();
  for(size_t i = 0; i < is.size(); ++i)
    cout << is[i] << ' ';
  cout << endl;
 
  // Uses the template partial specialization:
  Sortable<string*> ss;
  for(size_t i = 0; i < asz(words); ++i)
    ss.push_back(new string(words[i]));
  for(size_t i = 0; i < ss.size(); ++i)
    cout << *ss[i] << ' ';
  cout << endl;
  ss.sort();
  for(size_t i = 0; i < ss.size(); ++i) {
    cout << *ss[i] << ' ';
    delete ss[i];
  }
  cout << endl;
 
  // Uses the full char* specialization:
  Sortable<char*> scp;
  for(size_t i = 0; i < asz(words2); ++i)
    scp.push_back(words2[i]);
  for(size_t i = 0; i < scp.size(); ++i)
    cout << scp[i] << ' ';
  cout << endl;
  scp.sort();
  for(size_t i = 0; i < scp.size(); ++i)
    cout << scp[i] << ' ';
  cout << endl;
} ///:~

Each of the template instantiations above uses a different version of the template. Sortable<int> uses the primary template. Sortable<string*> uses the partial specialization for pointers. Last, Sortable<char*> uses the full specialization for char*. Without this full specialization, you could be fooled into thinking that things were working correctly because the words array would still sort out to “a big dog is running” since the partial specialization would end up comparing the first character of each array. However, words2 would not sort correctly.

2-3-3-4. Preventing template code bloat

Whenever a class template is instantiated, the code from the class definition for the particular specialization is generated, along with all the member functions that are called in the program. Only the member functions that are called are generated. This is good, as you can see in the following program:

 
Sélectionnez
//: C05:DelayedInstantiation.cpp
// Member functions of class templates are not
// instantiated until they're needed.
 
class X {
public:
  void f() {}
};
 
class Y {
public:
  void g() {}
};
 
template<typename T> class Z {
  T t;
public:
  void a() { t.f(); }
  void b() { t.g(); }
};
 
int main() {
  Z<X> zx;
  zx.a(); // Doesn't create Z<X>::b()
  Z<Y> zy;
  zy.b(); // Doesn't create Z<Y>::a()
} ///:~

Here, even though the template Z purports to use both f( ) and g( ) member functions of T, the fact that the program compiles shows you that it only generates Z<X>::a( ) when it is explicitly called for zx. (If Z<X>::b( ) were also generated at the same time, a compile-time error message would be generated because it would attempt to call X::g( ), which doesn't exist.) Similarly, the call to zy.b( ) doesn't generate Z<Y>::a( ). As a result, the Z template can be used with X and Y, whereas if all the member functions were generated when the class was first instantiated the use of many templates would become significantly limited.

Suppose you have a templatized Stack container and you use specializations for int, int*, and char*. Three versions of Stack code will be generated and linked as part of your program. One of the reasons for using a template in the first place is so you don't need to replicate code by hand; but code still gets replicated—it's just the compiler that does it instead of you. You can factor the bulk of the implementation for storing pointer types into a single class by using a combination of full and partial specialization. The key is to fully specialize for void* and then derive all other pointer types from the void* implementation so the common code can be shared. The program below illustrates this technique:

 
Sélectionnez
//: C05:Nobloat.h
// Shares code for storing pointers in a Stack.
#ifndef NOBLOAT_H
#define NOBLOAT_H
#include <cassert>
#include <cstddef>
#include <cstring>
 
// The primary template
template<class T> class Stack {
  T* data;
  std::size_t count;
  std::size_t capacity;
  enum { INIT = 5 };
public:
  Stack() {
    count = 0;
    capacity = INIT;
    data = new T[INIT];
  }
  void push(const T& t) {
    if(count == capacity) {
      // Grow array store
      std::size_t newCapacity = 2 * capacity;
      T* newData = new T[newCapacity];
      for(size_t i = 0; i < count; ++i)
        newData[i] = data[i];
      delete [] data;
      data = newData;
      capacity = newCapacity;
    }
    assert(count < capacity);
    data[count++] = t;
  }
  void pop() {
    assert(count > 0);
    --count;
  }
  T top() const {
    assert(count > 0);
    return data[count-1];
  }
  std::size_t size() const { return count; }
};
 
// Full specialization for void*
template<> class Stack<void *> {
  void** data;
  std::size_t count;
  std::size_t capacity;
  enum { INIT = 5 };
public:
  Stack() {
    count = 0;
    capacity = INIT;
    data = new void*[INIT];
  }
  void push(void* const & t) {
    if(count == capacity) {
      std::size_t newCapacity = 2*capacity;
      void** newData = new void*[newCapacity];
      std::memcpy(newData, data, count*sizeof(void*));
      delete [] data;
      data = newData;
      capacity = newCapacity;
    }
    assert(count < capacity);
    data[count++] = t;
  }
  void pop() {
    assert(count > 0);
    --count;
  }
  void* top() const {
    assert(count > 0);
    return data[count-1];
  }
  std::size_t size() const { return count; }
};
 
// Partial specialization for
other pointer types
template<class T> class
Stack<T*> : private Stack<void *> {
  typedef Stack<void *>
Base;
public:
  void push(T* const & t) { Base::push(t); }
  void pop() {Base::pop();}
  T* top() const { return
static_cast<T*>(Base::top()); }
  std::size_t size() { return Base::size(); }
};
#endif // NOBLOAT_H ///:~

This simple stack expands as it fills its capacity. The void* specialization stands out as a full specialization by virtue of the template<> prefix (that is, the template parameter list is empty). As mentioned earlier, it is necessary to implement all member functions in a class template specialization. The savings occurs with all other pointer types. The partial specialization for other pointer types derives from Stack<void*> privately, since we are merely using Stack<void*> for implementation purposes, and do not wish to expose any of its interface directly to the user. The member functions for each pointer instantiation are small forwarding functions to the corresponding functions in Stack<void*>. Hence, whenever a pointer type other than void* is instantiated, it is a fraction of the size it would have been had the primary template alone been used.(63) Here is a driver program:

 
Sélectionnez
//: C05:NobloatTest.cpp
#include <iostream>
#include <string>
#include "Nobloat.h"
using namespace std;
 
template<class StackType>
void emptyTheStack(StackType& stk) {
  while(stk.size() > 0) {
    cout << stk.top() << endl;
    stk.pop();
  }
}
 
// An overload for emptyTheStack (not a
specialization!)
template<class T>
void emptyTheStack(Stack<T*>& stk) {
  while(stk.size() > 0) {
    cout << *stk.top() << endl;
    stk.pop();
  }
}
 
int main() {
  Stack<int> s1;
  s1.push(1);
  s1.push(2);
  emptyTheStack(s1);
  Stack<int *> s2;
  int i = 3;
  int j = 4;
  s2.push(&i);
  s2.push(&j);
  emptyTheStack(s2);
} ///:~

For convenience we include two emptyStack function templates. Since function templates don't support partial specialization, we provide overloaded templates. The second version of emptyStack is more specialized than the first, so it is chosen whenever pointer types are used. Three class templates are instantiated in this program: Stack<int>, Stack<void*>, and Stack<int*>. Stack<void*> is implicitly instantiated because Stack<int*> derives from it. A program using instantiations for many pointer types can produce substantial savings in code size over just using a single Stack template.

2-3-4. Name lookup issues

When the compiler encounters an identifier it must determine the type and scope (and in the case of variables, the lifetime) of the entity the identifier represents. Templates add complexity to the situation. Because the compiler doesn't know everything about a template when it first sees the definition, it can't tell whether the template is being used properly until it sees the template instantiation. This predicament leads to a two-phase process for template compilation.

2-3-4-1. Names in templates

In the first phase, the compiler parses the template definition looking for obvious syntax errors and resolving all the names it can. It can resolve names that do not depend on template parameters using normal name lookup, and if necessary through argument-dependent lookup (discussed below). The names it can't resolve are the so-called dependent names, which depend on template parameters in some way. These can't be resolved until the template is instantiated with its actual arguments. So instantiation is the second phase of template compilation. Here, the compiler determines whether to use an explicit specialization of the template instead of the primary template.

Before you see an example, you must understand two more terms. A qualified name is a name with a class-name prefix, a name with an object name and a dot operator, or a name with a pointer to an object and an arrow operator. Examples of qualified names are:

 
Sélectionnez
MyClass::f();
x.f();
p->f();

We use qualified names many times in this book, and most recently in connection with the typename keyword. These are called qualified names because the target names (like f above) are explicitly associated with a class or namespace, which tells the compiler where to look for the declarations of those names.

The other term is argument-dependent lookup(64) (ADL), a mechanism originally designed to simplify non-member function calls (including operators) declared in namespaces. Consider the following:

 
Sélectionnez
#include <iostream>
#include <string>
// ...
  std::string s("hello");
  std::cout << s << std::endl;

Note that, following the typical practice in header files, there is no using namespace std directive. Without such a directive, you must use the “std::”qualifier on the items that are in the std namespace. We have, however, not qualified everything from std that we are using. Can you see what is unqualified?

We have not specified which operator functions to use. We want the following to happen, but we don't want to have to type it!

 
Sélectionnez
std::operator<<(std::operator<<(std::cout,s),std::endl);

To make the original output statement work as desired, ADL specifies that when an unqualified function call appears and its declaration is not in (normal) scope, the namespaces of each of its arguments are searched for a matching function declaration. In the original statement, the first function call is:

 
Sélectionnez
operator<<(std::cout, s);

Since there is no such function in scope in our original excerpt, the compiler notes that this function's first argument (std::cout) is in the namespace std; so it adds that namespace to the list of scopes to search for a unique function that best matches the signature operator<<(std::ostream&, std::string). It finds this function declared in the std namespace via the <string> header.

Namespaces would be very inconvenient without ADL. Note that ADL generally brings in all declarations of the name in question from all eligible namespaces—if there is no single best match, an ambiguity will result.

To turn off ADL, you can enclose the function name in parentheses:

 
Sélectionnez
(f)(x, y);  // ADL suppressed

Now consider the following program:(65)

 
Sélectionnez
//: C05:Lookup.cpp
// Only produces correct behavior with EDG,
// and Metrowerks using a special option.
#include <iostream>
using std::cout;
using std::endl;
 
void f(double) { cout << "f(double)"
<< endl; }
 
template<class T> class X {
public:
  void g() { f(1); }
};
 
void f(int) { cout << "f(int)" <<
endl; }
 
int main() {
  X<int>().g();
} ///:~

The only compiler we have that produces correct behavior without modification is the Edison Design Group front end,(66) although some compilers, such as Metrowerks, have an option to enable the correct lookup behavior. The output should be:

 
Sélectionnez
f(double)

because f is a non-dependent name that can be resolved early by looking in the context where the template is defined, when only f(double) is in scope. Unfortunately, there is a lot of existing code in the industry that depends on the non-standard behavior of binding the call to f(1) inside g( ) to the latter f(int), so compiler writers have been reluctant to make the change.

Here is a more detailed example:(67)

 
Sélectionnez
//: C05:Lookup2.cpp {-bor}{-g++}{-dmc}
// Microsoft: use option -Za (ANSI mode)
#include <algorithm>
#include <iostream>
#include <typeinfo>
using std::cout;
using std::endl;
 
void g() { cout << "global g()” <<
endl; }
 
template<class T> class Y {
public:
  void g() {
    cout << "Y<" <<
typeid(T).name() << ">::g()” << endl;
  }
  void h() {
    cout << "Y<" <<
typeid(T).name() << ">::h()” << endl;
  }
  typedef int E;
};
 
typedef double E;
 
template<class T> void swap(T& t1, T& t2)
{
  cout << "global swap” << endl;
  T temp = t1;
  t1 = t2;
  t2 = temp;
}
 
template<class T> class X : public Y<T> {
public:
  E f() {
    g();
    this->h();
    T t1 = T(), t2 = T(1);
    cout << t1 << endl;
    swap(t1, t2);
    std::swap(t1, t2);
    cout << typeid(E).name() << endl;
    return E(t2);
  }
};
 
int main() {
  X<int> x;
  cout << x.f() << endl;
} ///:~

The output from this program should be:

 
Sélectionnez
global g()
Y<int>::h()
0
global swap
double
1

Looking at the declarations inside of X::f( ):

  • E, the return type of X::f( ), is not a dependent name, so it is looked up when the template is parsed, and the typedef naming E as a double is found. This may seem strange, since with non-template classes the declaration of E in the base class would be found first, but those are the rules. (The base class, Y, is a dependent base class, so it can't be searched at template definition time).
  • The call to g( ) is also non-dependent, since there is no mention of T. If g had parameters that were of class type of defined in another namespace, ADL would take over, since there is no g with parameters in scope. As it is, this call matches the global declaration of g( ).
  • The call this->h( ) is a qualified name, and the object that qualifies it (this) refers to the current object, which is of type X, which in turn depends on the name Y<T> by inheritance. There is no function h( ) inside of X, so the lookup will search the scope of X's base class, Y<T>. Since this is a dependent name, it is looked up at instantiation time, when Y<T> are reliably known (including any potential specializations that might have been written after the definition of X), so it calls Y<int>::h( ).
  • The declarations of t1 and t2 are dependent.
  • The call to operator<<(cout, t1) is dependent, since t1 is of type T. This is looked up later when T is int, and the inserter for int is found in std.
  • The unqualified call to swap( ) is dependent because its arguments are of type T. This ultimately causes a global swap(int&, int&) to be instantiated.
  • The qualified call to std::swap( ) is not dependent, because std is a fixed namespace. The compiler knows to look in std for the proper declaration. (The qualifier on the left of the “::” must mention a template parameter for a qualified name to be considered dependent.) The std::swap( ) function template later generates std::swap(int&, int&), at instantiation time. No more dependent names remain in X<T>::f( ).

To clarify and summarize: name lookup is done at the point of instantiation if the name is dependent, except that for unqualified dependent names the normal name lookup is also attempted early, at the point of definition. All non-dependent names in templates are looked up early, at the time the template definition is parsed. (If necessary, another lookup occurs at instantiation time, when the type of the actual argument is known.)

If you have studied this example to the point that you understand it, prepare yourself for yet another surprise in the following section on friend declarations.

2-3-4-2. Templates and friends

A friend function declaration inside a class allows a non-member function to access non-public members of that class. If the friend function name is qualified, it will be found in the namespace or class that qualifies it. If it is unqualified, however, the compiler must make an assumption about where the definition of the friend function will be, since all identifiers must have a unique scope. The expectation is that the function will be defined in the nearest enclosing namespace (non-class) scope that contains the class granting friendship. Often this is just the global scope. The following non-template example clarifies this issue:

 
Sélectionnez
//: C05:FriendScope.cpp
#include <iostream>
using namespace std;
 
class Friendly {
  int i;
public:
  Friendly(int theInt) { i = theInt; }
  friend void f(const Friendly&); // Needs global
def.
  void g() { f(*this); }
};
 
void h() {
  f(Friendly(1));  // Uses ADL
}
 
void f(const Friendly& fo) {  // Definition of
friend
  cout << fo.i << endl;
}
 
int main() {
  h(); // Prints 1
  Friendly(2).g(); // Prints 2
} ///:~

The declaration of f( ) inside the Friendly class is unqualified, so the compiler will expect to be able to eventually link that declaration to a definition at file scope (the namespace scope that contains Friendly in this case). That definition appears after the definition of the function h( ). The linking of the call to f( ) inside h( ) to the same function is a separate matter, however. This is resolved by ADL. Since the argument of f( ) inside h( ) is a Friendly object, the Friendly class is searched for a declaration of f( ), which succeeds. If the call were f(1) instead (which makes some sense since 1 can be implicitly converted to Friendly(1)), the call should fail, since there is no hint of where the compiler should look for the declaration of f( ). The EDG compiler correctly complains that f is undefined in that case.

Now suppose that Friendly and f are both templates, as in the following program:

 
Sélectionnez
//: C05:FriendScope2.cpp
#include <iostream>
using namespace std;
 
// Necessary forward declarations:
template<class T> class Friendly;
template<class T> void f(const
Friendly<T>&);
 
template<class T> class Friendly {
  T t;
public:
  Friendly(const T& theT) : t(theT) {}
  friend void f<>(const Friendly<T>&);
  void g() { f(*this); }
};
 
void h() {
  f(Friendly<int>(1));
}
 
template<class T> void f(const
Friendly<T>& fo) {
  cout << fo.t << endl;
}
 
int main() {
  h();
  Friendly<int>(2).g();
} ///:~

First notice that angle brackets in the declaration of f inside Friendly. This is necessary to tell the compiler that f is a template. Otherwise, the compiler will look for an ordinary function named f and not find it. We could have inserted the template parameter (<T>) in the brackets, but it is easily deduced from the declaration.

The forward declaration of the function template f before the class definition is necessary, even though it wasn't in the previous example when f was a not a template; the language specifies that friend function templates must be previously declared. To properly declare f, Friendly must also have been declared, since f takes a Friendly argument, hence the forward declaration of Friendly in the beginning. We could have placed the full definition of f right after the initial declaration of Friendly instead of separating its definition and declaration, but we chose instead to leave it in a form that more closely resembles the previous example.

One last option remains for using friends inside templates: fully define them inside the host class template definition itself. Here is how the previous example would appear with that change:

 
Sélectionnez
//: C05:FriendScope3.cpp {-bor}
// Microsoft: use the -Za (ANSI-compliant) option
#include <iostream>
using namespace std;
 
template<class T> class Friendly {
  T t;
public:
  Friendly(const T& theT) : t(theT) {}
  friend void f(const Friendly<T>& fo) {
    cout << fo.t << endl;
  }
  void g() { f(*this); }
};
 
void h() {
  f(Friendly<int>(1));
}
 
int main() {
  h();
  Friendly<int>(2).g();
} ///:~

There is an important difference between this and the previous example: f is not a template here, but is an ordinary function. (Remember that angle brackets were necessary before to imply that f( ) was a template.) Every time the Friendly class template is instantiated, a new, ordinary function overload is created that takes an argument of the current Friendly specialization. This is what Dan Saks has called “making new friends.”(68) This is the most convenient way to define friend functions for templates.

To clarify, suppose you want to add non-member friend operators to a class template. Here is a class template that simply holds a generic value:

 
Sélectionnez
template<class T> class Box {
  T t;
public:
  Box(const T& theT) : t(theT) {}
};

Without understanding the previous examples in this section, novices find themselves frustrated because they can't get a simple stream output inserter to work. If you don't define your operators inside the definition of Box, you must provide the forward declarations we showed earlier:

 
Sélectionnez
//: C05:Box1.cpp
// Defines template operators.
#include <iostream>
using namespace std;
 
// Forward declarations
template<class T> class Box;
 
template<class T>
Box<T> operator+(const Box<T>&, const
Box<T>&);
 
template<class T>
ostream& operator<<(ostream&, const
Box<T>&);
 
template<class T> class Box {
  T t;
public:
  Box(const T& theT) : t(theT) {}
  friend Box operator+<>(const Box<T>&,
const Box<T>&);
  friend ostream& operator<< <>(ostream&,
const Box<T>&);
};
 
template<class T>
Box<T> operator+(const Box<T>& b1,
const Box<T>& b2) {
  return Box<T>(b1.t + b2.t);
}
 
template<class T>
ostream& operator<<(ostream& os, const
Box<T>& b) {
  return os << '[' << b.t << ']';
}
 
int main() {
  Box<int> b1(1), b2(2);
  cout << b1 + b2 << endl;  // [3]
//  cout << b1 + 2 << endl; // No implicit
conversions!
} ///:~

Here we are defining both an addition operator and an output stream operator. The main program reveals a disadvantage of this approach: you can't depend on implicit conversions (the expression b1 + 2) because templates do not provide them. Using the in-class, non-template approach is shorter and more robust:

 
Sélectionnez
//: C05:Box2.cpp
// Defines non-template operators.
#include <iostream>
using namespace std;
 
template<class T> class Box {
  T t;
public:
  Box(const T& theT) : t(theT) {}
  friend Box<T> operator+(const Box<T>&
b1,
                          const Box<T>& b2) {
    return Box<T>(b1.t + b2.t);
  }
  friend ostream&
  operator<<(ostream& os, const
Box<T>& b) {
    return os << '[' << b.t << ']';
  }
};
 
int main() {
  Box<int> b1(1), b2(2);
  cout << b1 + b2 << endl; // [3]
  cout << b1 + 2 << endl; // [3]
} ///:~

Because the operators are normal functions (overloaded for each specialization of Box—just int in this case), implicit conversions are applied as normal; so the expression b1 + 2 is valid.

Note that there's one type in particular that cannot be made a friend of Box, or any other class template for that matter, and that type is T—or rather, the type that the class template is parameterized upon. To the best of our knowledge, there are really no good reasons why this shouldn't be allowed, but as is, the declaration friend class T is illegal, and should not compile.

Friend templates

You can be precise as to which specializations of a template are friends of a class. In the examples in the previous section, only the specialization of the function template f with the same type that specialized Friendly was a friend. For example, only the specialization f<int>(const Friendly<int>&) is a friend of the class Friendly<int>. This was accomplished by using the template parameter for Friendly to specialize f in its friend declaration. If we had wanted to, we could have made a particular, fixed specialization of f a friend to all instances of Friendly, like this:

 
Sélectionnez
  // Inside Friendly:
  friend void f<>(const Friendly<double>&);

By using double instead of T, the double specialization of f has access to the non-public members of any Friendly specialization. The specialization f<double>( ) still isn't instantiated unless it is explicitly called.

Likewise, if you declare a non-template function with no parameters dependent on T, that single function is a friend to all instances of Friendly:

 
Sélectionnez
  // Inside Friendly:
  friend void g(int);  // g(int) befriends all Friendlys

As always, since g(int) is unqualified, it must be defined at file scope (the namespace scope containing Friendly).

It is also possible to arrange for all specializations of f to be friends for all specializations of Friendly, with a so-called friend template, as follows:

 
Sélectionnez
template<class T> class Friendly {
  template<class U> friend void f<>(const Friendly<U>&);

Since the template argument for the friend declaration is independent of T, any combination of T and U is allowed, achieving the friendship objective. Like member templates, friend templates can appear within non-template classes as well.

2-3-5. Template programming idioms

Since language is a tool of thought, new language features tend to spawn new programming techniques. In this section we cover some commonly used template programming idioms that have emerged in the years since templates were added to the C++ language.(69)

2-3-5-1. Traits

The traits template technique, pioneered by Nathan Myers, is a means of bundling type-dependent declarations together. In essence, using traits you can “mix and match” certain types and values with contexts that use them in a flexible manner, while keeping your code readable and maintainable.

The simplest example of a traits template is the numeric_limits class template defined in <limits>. The primary template is defined as follows:

 
Sélectionnez
template<class T> class numeric_limits {
public:
  static const bool is_specialized = false;
  static T min() throw();
  static T max() throw();
  static const int digits = 0;
  static const int digits10 = 0;
  static const bool is_signed = false;
  static const bool is_integer = false;
  static const bool is_exact = false;
  static const int radix = 0;
  static T epsilon() throw();
  static T round_error() throw();
  static const int min_exponent = 0;
  static const int min_exponent10 = 0;
  static const int max_exponent = 0;
  static const int max_exponent10 = 0;
  static const bool has_infinity = false;
  static const bool has_quiet_NaN = false;
  static const bool has_signaling_NaN = false;
  static const float_denorm_style has_denorm =
                                  denorm_absent;
  static const bool has_denorm_loss = false;
  static T infinity() throw();
  static T quiet_NaN() throw();
  static T signaling_NaN() throw();
  static T denorm_min() throw();
  static const bool is_iec559 = false;
  static const bool is_bounded = false;
  static const bool is_modulo = false;
  static const bool traps = false;
  static const bool tinyness_before = false;
  static const float_round_style round_style =
                                 round_toward_zero;
};

The <limits> header defines specializations for all fundamental, numeric types (when the member is_specialized is set to true). To obtain the base for the double version of your floating-point number system, for example, you can use the expression numeric_limits<double>::radix. To find the smallest integer value available, you can use numeric_limits<int>::min( ). Not all members of numeric_limits apply to all fundamental types. (For example, epsilon( ) is only meaningful for floating-point types.)

The values that will always be integral are static data members of numeric_limits. Those that may not be integral, such as the minimum value for float, are implemented as static inline member functions. This is because C++ allows only integral static data member constants to be initialized inside a class definition.

In Chapter 3 you saw how traits are used to control the character-processing functionality used by the string classes. The classes std::string and std::wstring are specializations of the std::basic_string template, which is defined as follows:

 
Sélectionnez
template<class charT,
  class traits = char_traits<charT>,
  class allocator = allocator<charT> >
  class
basic_string;

The template parameter charT represents the underlying character type, which is usually either char or wchar_t. The primary char_traits template is typically empty, and specializations for char and wchar_t are provided by the standard library. Here is the specification of the specialization char_traits<char> according to the C++ Standard:

 
Sélectionnez
template<> struct char_traits<char> {
  typedef char char_type;
  typedef int int_type;
  typedef streamoff off_type;
  typedef streampos pos_type;
  typedef mbstate_t state_type;
  static void assign(char_type& c1, const
char_type& c2);
  static bool eq(const char_type& c1, const
char_type& c2);
  static bool lt(const char_type& c1, const
char_type& c2);
  static int compare(const char_type* s1,
                     const char_type* s2, size_t n);
  static size_t length(const char_type* s);
  static const char_type* find(const char_type* s,
                               size_t n,
                               const char_type& a);
  static char_type* move(char_type* s1,
                         const char_type* s2, size_t
n);
  static char_type* copy(char_type* s1,
                         const char_type* s2, size_t
n);
  static char_type* assign(char_type* s, size_t n,
                           char_type a);
  static int_type not_eof(const int_type& c);
  static char_type to_char_type(const int_type& c);
  static int_type to_int_type(const char_type& c);
  static bool eq_int_type(const int_type& c1,
                          const int_type& c2);
  static int_type eof();
};

These functions are used by the basic_string class template for character-based operations common to string processing. When you declare a string variable, such as:

 
Sélectionnez
std::string
s;

you are actually declaring s as follows (because of the default template arguments in the specification of basic_string):

 
Sélectionnez
std::basic_string<char,
std::char_traits<char>,
                  std::allocator<char> > s;

Because the character traits have been separated from the basic_string class template, you can supply a custom traits class to replace std::char_traits. The following example illustrates this flexibility:

 
Sélectionnez
//: C05:BearCorner.h
#ifndef BEARCORNER_H
#define BEARCORNER_H
#include <iostream>
using std::ostream;
 
// Item classes (traits of guests):
class Milk {
public:
  friend ostream& operator<<(ostream& os,
const Milk&) {
    return os << "Milk";
  }
};
 
class CondensedMilk {
public:
  friend ostream&
  operator<<(ostream& os, const CondensedMilk
&) {
    return os << "Condensed Milk";
  }
};
 
class Honey {
public:
  friend ostream& operator<<(ostream& os,
const Honey&) {
    return os << "Honey";
  }
};
 
class Cookies {
public:
  friend ostream& operator<<(ostream& os,
const Cookies&) {
    return os << "Cookies";
  }
};
 
// Guest classes:
class Bear {
public:
  friend ostream& operator<<(ostream& os,
const Bear&) {
    return os << "Theodore";
  }
};
 
class Boy {
public:
  friend ostream& operator<<(ostream& os,
const Boy&) {
    return os << "Patrick";
  }
};
 
// Primary traits template (empty-could hold common
types)
template<class Guest> class GuestTraits;
 
// Traits specializations for Guest types
template<> class GuestTraits<Bear> {
public:
  typedef CondensedMilk beverage_type;
  typedef Honey snack_type;
};
 
template<> class GuestTraits<Boy> {
public:
  typedef Milk beverage_type;
  typedef Cookies snack_type;
};
#endif // BEARCORNER_H ///:~
 
Sélectionnez
//: C05:BearCorner.cpp
// Illustrates traits classes.
#include <iostream>
#include "BearCorner.h"
using namespace std;
 
// A custom traits class
class MixedUpTraits {
public:
  typedef Milk beverage_type;
  typedef Honey snack_type;
};
 
// The Guest template (uses a traits class)
template<class Guest, class traits =
GuestTraits<Guest> >
class BearCorner {
  Guest theGuest;
  typedef typename traits::beverage_type beverage_type;
  typedef typename traits::snack_type snack_type;
  beverage_type bev;
  snack_type snack;
public:
  BearCorner(const Guest& g) : theGuest(g) {}
  void entertain() {
    cout << "Entertaining " <<
theGuest
         << " serving " << bev
         << " and " << snack
<< endl;
  }
};
 
int main() {
  Boy cr;
  BearCorner<Boy> pc1(cr);
  pc1.entertain();
  Bear pb;
  BearCorner<Bear> pc2(pb);
  pc2.entertain();
  BearCorner<Bear, MixedUpTraits> pc3(pb);
  pc3.entertain();
} ///:~

In this program, instances of the guest classes Boy and Bear are served items appropriate to their tastes. Boys like milk and cookies, and Bears like condensed milk and honey. This association of guests to items is done via specializations of a primary (empty) traits class template. The default arguments to BearCorner ensure that guests get their proper items, but you can override this by simply providing a class that meets the requirements of the traits class, as we do with the MixedUpTraits class above. The output of this program is:

 
Sélectionnez
Entertaining Patrick serving Milk and Cookies
Entertaining Theodore serving Condensed Milk and Honey
Entertaining Theodore serving
Milk and Honey

Using traits provides two key advantages: (1) it allows flexibility and extensibility in pairing objects with associated attributes or functionality, and (2) it keeps template parameter lists small and readable. If 30 types were associated with a guest, it would be inconvenient to have to specify all 30 arguments directly in each BearCorner declaration. Factoring the types into a separate traits class simplifies things considerably.

The traits technique is also used in implementing streams and locales, as we showed in Chapter 4. An example of iterator traits is found in the header file PrintSequence.h in Chapter 6.

2-3-5-2. Policies

If you inspect the char_traits specialization for wchar_t, you'll see that it is practically identical to its char counterpart:

 
Sélectionnez
template<> struct char_traits<wchar_t> {
  typedef wchar_t char_type;
  typedef wint_t int_type;
  typedef streamoff off_type;
  typedef wstreampos pos_type;
  typedef mbstate_t state_type;
  static void assign(char_type& c1, const
char_type& c2);
  static bool eq(const char_type& c1, const
char_type& c2);
  static bool lt(const char_type& c1, const
char_type& c2);
  static int compare(const char_type* s1,
                     const char_type*  s2, size_t n);
  static size_t length(const char_type* s);
  static const char_type* find(const char_type* s,
                               size_t n,
                               const char_type& a);
  static char_type* move(char_type* s1,
                         const char_type* s2, size_t
n);
  static char_type* copy(char_type* s1,
                         const char_type* s2, size_t
n);
  static char_type* assign(char_type* s, size_t n,
                           char_type a);
  static int_type not_eof(const int_type& c);
  static char_type to_char_type(const int_type& c);
  static int_type to_int_type(const char_type& c);
  static bool eq_int_type(const int_type& c1,
                          const int_type& c2);
  static int_type eof();
};

The only real difference between the two versions is the set of types involved (char and int vs. wchar_t and wint_t). The functionality provided is the same.(70) This highlights the fact that traits classes are indeed for traits, and the things that change between related traits classes are usually types and constant values, or fixed algorithms that use type-related template parameters. Traits classes tend to be templates themselves, since the types and constants they contain are seen as characteristics of the primary template parameter(s) (for example, char and wchar_t).

It is also useful to be able to associate functionality with template arguments, so that client programmers can easily customize behavior when they code. The following version of the BearCorner program, for instance, supports different types of entertainment:

 
Sélectionnez
//: C05:BearCorner2.cpp
// Illustrates policy classes.
#include <iostream>
#include "BearCorner.h"
using namespace std;
 
// Policy classes (require a static doAction()
function):
class Feed {
public:
  static const char* doAction() { return
"Feeding"; }
};
 
class Stuff {
public:
  static const char* doAction() { return
"Stuffing"; }
};
 
// The Guest template (uses a policy and a traits
class)
template<class Guest, class Action,
         class traits = GuestTraits<Guest> >
class BearCorner {
  Guest theGuest;
  typedef typename traits::beverage_type beverage_type;
  typedef typename traits::snack_type snack_type;
  beverage_type bev;
  snack_type snack;
public:
  BearCorner(const Guest& g) : theGuest(g) {}
  void entertain() {
    cout << Action::doAction() << "
" << theGuest
         << " with " << bev
         << " and " << snack
<< endl;
  }
};
 
int main() {
  Boy cr;
  BearCorner<Boy, Feed> pc1(cr);
  pc1.entertain();
  Bear pb;
  BearCorner<Bear, Stuff> pc2(pb);
  pc2.entertain();
} ///:~

The Action template parameter in the BearCorner class expects to have a static member function named doAction( ), which is used in BearCorner<>::entertain( ). Users can choose Feed or Stuff at will, both of which provide the required function. Classes that encapsulate functionality in this way are referred to as policy classes. The entertainment “policies” are provided above through Feed::doAction( ) and Stuff::doAction( ). These policy classes happen to be ordinary classes, but they can be templates, and can be combined with inheritance to great advantage. For more in-depth information on policy-based design, see Andrei Alexandrescu's book,(71) the definitive source on the subject.

2-3-5-3. The curiously recurring template pattern

Any novice C++ programmer can figure out how to modify a class to keep track of the number of objects of that class that currently exist. All you have to do is to add static members, and modify constructor and destructor logic, as follows:

 
Sélectionnez
//: C05:CountedClass.cpp
// Object counting via static members.
#include <iostream>
using namespace std;
 
class CountedClass {
  static int count;
public:
  CountedClass() { ++count; }
  CountedClass(const CountedClass&) { ++count; }
  ~CountedClass() { --count; }
  static int getCount() { return count; }
};
 
int CountedClass::count = 0;
 
int main() {
  CountedClass a;
  cout << CountedClass::getCount() <<
endl;   // 1
  CountedClass b;
  cout << CountedClass::getCount() <<
endl;   // 2
  { // An arbitrary scope:
    CountedClass c(b);
    cout << CountedClass::getCount() <<
endl; // 3
    a = c;
    cout << CountedClass::getCount() <<
endl; // 3
  }
  cout << CountedClass::getCount() <<
endl;   // 2
} ///:~

All constructors of CountedClass increment the static data member count, and the destructor decrements it. The static member function getCount( ) yields the current number of objects.

It would be tedious to manually add these members every time you wanted to add object counting to a class. The usual object-oriented device used to repeat or share code is inheritance, which is only half a solution in this case. Observe what happens when we collect the counting logic into a base class.

 
Sélectionnez
//: C05:CountedClass2.cpp
// Erroneous attempt to count objects.
#include <iostream>
using namespace std;
 
class Counted {
  static int count;
public:
  Counted() { ++count; }
  Counted(const Counted&) { ++count; }
  ~Counted() { --count; }
  static int getCount() { return count; }
};
 
int Counted::count = 0;
 
class CountedClass : public Counted {};
class CountedClass2 : public Counted {};
 
int main() {
  CountedClass a;
  cout << CountedClass::getCount() <<
endl;    // 1
  CountedClass b;
  cout << CountedClass::getCount() <<
endl;    // 2
  CountedClass2 c;
  cout << CountedClass2::getCount() <<
endl;   // 3 (Error)
} ///:~

All classes that derive from Counted share the same, single static data member, so the number of objects is tracked collectively across all classes in the Counted hierarchy. What is needed is a way to automatically generate a different base class for each derived class. This is accomplished by the curious template construct illustrated below:

 
Sélectionnez
//: C05:CountedClass3.cpp
#include <iostream>
using namespace std;
 
template<class T> class Counted {
  static int count;
public:
  Counted() { ++count; }
  Counted(const Counted<T>&) { ++count; }
  ~Counted() { --count; }
  static int getCount() { return count; }
};
 
template<class T> int Counted<T>::count =
0;
 
// Curious class definitions
class CountedClass : public Counted<CountedClass>
{};
class CountedClass2 : public
Counted<CountedClass2> {};
 
int main() {
  CountedClass a;
  cout << CountedClass::getCount() <<
endl;    // 1
  CountedClass b;
  cout << CountedClass::getCount() <<
endl;    // 2
  CountedClass2 c;
  cout << CountedClass2::getCount() <<
endl;   // 1 (!)
} ///:~

Each derived class derives from a unique base class that is determined by using itself (the derived class) as a template parameter! This may seem like a circular definition, and it would be, had any base class members used the template argument in a computation. Since no data members of Counted are dependent on T, the size of Counted (which is zero!) is known when the template is parsed. So it doesn't matter which argument is used to instantiate Counted because the size is always the same. Any derivation from an instance of Counted can be completed when it is parsed, and there is no recursion. Since each base class is unique, it has its own static data, thus constituting a handy technique for adding counting to any class whatsoever. Jim Coplien was the first to mention this interesting derivation idiom in print, which he cited in an article, entitled “Curiously Recurring Template Patterns.” (72)

2-3-6. Template metaprogramming

In 1993 compilers were beginning to support simple template constructs so that users could define generic containers and functions. About the same time that the STL was being considered for adoption into Standard C++, clever and surprising examples such as the following were passed around among members of the C++ Standards Committee:(73)

 
Sélectionnez
//: C05:Factorial.cpp
// Compile-time computation using templates.
#include <iostream>
using namespace std;
 
template<int n> struct Factorial {
  enum { val = Factorial<n-1>::val * n };
};
 
template<> struct Factorial<0> {
  enum { val = 1 };
};
 
int main() {
  cout << Factorial<12>::val << endl;
// 479001600
} ///:~

That this program prints the correct value of 12! is not alarming. What is alarming is that the computation is complete before the program even runs!

When the compiler attempts to instantiate Factorial<12>, it finds it must also instantiate Factorial<11>, which requires Factorial<10>, and so on. Eventually the recursion ends with the specialization Factorial<1>, and the computation unwinds. Eventually, Factorial<12>::val is replaced by the integral constant 479001600, and compilation ends. Since all the computation is done by the compiler, the values involved must be compile-time constants, hence the use of enum. When the program runs, the only work left to do is print that constant followed by a newline. To convince yourself that a specialization of Factorial results in the correct compile-time value, you could use it as an array dimension, such as:

 
Sélectionnez
double nums[Factorial<5>::val];
assert(sizeof
nums == sizeof(double)*120);
2-3-6-1. Compile-time programming

So what was meant to be a convenient way to perform type parameter substitution turned out to be a mechanism to support compile-time programming. Such a program is called a template metaprogram (since you're in effect “programming a program”), and it turns out that you can do quite a lot with it. In fact, template metaprogramming is Turing complete because it supports selection (if-else) and looping (through recursion). Theoretically, then, you can perform any computation with it.(74) The factorial example above shows how to implement repetition: write a recursive template and provide a stopping criterion via a specialization. The following example shows how to compute Fibonacci numbers at compile time by the same technique:

 
Sélectionnez
//: C05:Fibonacci.cpp
#include <iostream>
using namespace std;
 
template<int n> struct Fib {
  enum { val = Fib<n-1>::val +
Fib<n-2>::val };
};
 
template<> struct Fib<1> { enum { val = 1 };
};
 
template<> struct
Fib<0> { enum { val = 0 }; };
 
int main() {
  cout << Fib<5>::val << endl;   // 6
  cout << Fib<20>::val << endl;  //
6765
} ///:~

Fibonacci numbers are defined mathematically as:

Image non disponible

The first two cases lead to the template specializations above, and the rule in the third line becomes the primary template.

Compile-time looping

To compute any loop in a template metaprogram, it must first be reformulated recursively. For example, to raise the integer n to the power p, instead of using a loop such as in the following lines:

 
Sélectionnez
int val = 1;
while(p--)
  val *=
n;

you cast it as a recursive procedure:

 
Sélectionnez
int power(int n, int p) {
  return (p == 0) ? 1 : n*power(n, p - 1);
}

This can now be easily rendered as a template metaprogram:

 
Sélectionnez
//: C05:Power.cpp
#include <iostream>
using namespace std;
 
template<int N, int P> struct Power {
  enum { val = N * Power<N, P-1>::val };
};
 
template<int N> struct Power<N, 0> {
  enum { val = 1 };
};
 
int main() {
  cout << Power<2, 5>::val << endl; 
// 32
} ///:~

We need to use a partial specialization for the stopping condition, since the value N is still a free template parameter. Note that this program only works for non-negative powers.

The following metaprogram adapted from Czarnecki and Eisenecker(75) is interesting in that it uses a template template parameter, and simulates passing a function as a parameter to another function, which “loops through” the numbers 0..n:

 
Sélectionnez
//: C05:Accumulate.cpp
// Passes a "function" as a parameter at
compile time.
#include <iostream>
using namespace std;
 
// Accumulates the results of F(0)..F(n)
template<int n, template<int> class F> struct
Accumulate {
  enum { val = Accumulate<n-1, F>::val +
F<n>::val };
};
 
// The stopping criterion (returns the value F(0))
template<template<int> class F> struct
Accumulate<0, F> {
  enum { val = F<0>::val };
};
 
// Various "functions":
template<int n> struct
Identity {
  enum { val = n };
};
 
template<int n> struct Square {
  enum { val = n*n };
};
 
template<int n> struct Cube {
  enum { val = n*n*n };
};
 
int main() {
  cout << Accumulate<4, Identity>::val
<< endl; // 10
  cout << Accumulate<4, Square>::val
<< endl;   // 30
  cout << Accumulate<4, Cube>::val <<
endl;     // 100
} ///:~

The primary Accumulate template attempts to compute the sum F(n)+F(n‑1)…F(0). The stopping criterion is obtained by a partial specialization, which “returns” F(0). The parameter F is itself a template, and acts like a function as in the previous examples in this section. The templates Identity, Square, and Cube compute the corresponding functions of their template parameter that their names suggest. The first instantiation of Accumulate in main( ) computes the sum 4+3+2+1+0, because the Identity function simply “returns” its template parameter. The second line in main( ) adds the squares of those numbers (16+9+4+1+0), and the last computes the sum of the cubes (64+27+8+1+0).

Loop unrolling

Algorithm designers have always endeavored to optimize their programs. One time-honored optimization, especially for numeric programming, is loop unrolling, a technique that minimizes loop overhead. The quintessential loop-unrolling example is matrix multiplication. The following function multiplies a matrix and a vector. (Assume that the constants ROWS and COLS have been previously defined.):

 
Sélectionnez
void mult(int a[ROWS][COLS], int x[COLS], int y[COLS])
{
  for(int i = 0; i < ROWS; ++i) {
      y[i] = 0;
      for(int j = 0; j < COLS; ++j)
        y[i] += a[i][j]*x[j];
  }
}

If COLS is an even number, the overhead of incrementing and comparing the loop control variable j can be cut in half by “unrolling” the computation into pairs in the inner loop:

 
Sélectionnez
void mult(int a[ROWS][COLS], int x[COLS], int y[COLS])
{
  for(int i = 0; i < ROWS; ++i) {
      y[i] = 0;
      for(int j = 0; j < COLS; j += 2)
        y[i] += a[i][j]*x[j] + a[i][j+1]*x[j+1];
  }
}

In general, if COLS is a factor of k, k operations can be performed each time the inner loop iterates, greatly reducing the overhead. The savings is only noticeable on large arrays, but that is precisely the case with industrial-strength mathematical computations.

Function inlining also constitutes a form of loop unrolling. Consider the following approach to computing powers of integers:

 
Sélectionnez
//: C05:Unroll.cpp
// Unrolls an implicit loop via inlining.
#include <iostream>
using namespace std;
 
template<int n> inline int power(int m) {
  return power<n-1>(m) * m;
}
 
template<> inline int power<1>(int m) {
  return m;
}
 
template<> inline int power<0>(int m) {
  return 1;
}
 
int main() {
  int m = 4;
  cout << power<3>(m) << endl;
} ///:~

Conceptually, the compiler must generate three specializations of power<>, one each for the template parameters 3, 2, and 1. Because the code for each of these functions can be inlined, the actual code that is inserted into main( ) is the single expression m*m*m. Thus, a simple template specialization coupled with inlining provides a way to totally avoid loop control overhead.(76) This approach to loop unrolling is limited by your compiler's inlining depth.

Compile-time selection

To simulate conditionals at compile time, you can use the conditional ternary operator in an enum declaration. The following program uses this technique to calculate the maximum of two integers at compile time:

 
Sélectionnez
//: C05:Max.cpp
#include <iostream>
using namespace std;
 
template<int n1, int n2> struct Max {
  enum { val = n1 > n2 ? n1 : n2 };
};
 
int main() {
  cout << Max<10, 20>::val << endl; 
// 20
} ///:~

If you want to use compile-time conditions to govern custom code generation, you can use specializations of the values true and false:

 
Sélectionnez
//: C05:Conditionals.cpp
// Uses compile-time conditions to choose code.
#include <iostream>
using namespace std;
 
template<bool cond> struct Select {};
 
template<> class Select<true> {
  static void statement1() {
    cout << "This is statement1 executing\n";
  }
public:
  static void f() { statement1(); }
};
 
template<> class Select<false> {
  static void statement2() {
    cout << "This is statement2
executing\n";
  }
public:
  static void f() { statement2(); }
};
 
template<bool cond> void execute() {
  Select<cond>::f();
}
 
int main() {
  execute<sizeof(int) == 4>();
} ///:~

This program is equivalent to the expression:

 
Sélectionnez
if(cond)
  statement1();
else
 
statement2();

except that the condition cond is evaluated at compile time, and the appropriate versions of execute<>( ) and Select<> are instantiatedby the compiler. The function Select<>::f( ) executes at runtime. A switch statement can be emulated in similar fashion, but specializing on each case value instead of the values true and false.

Compile-time assertions

In Chapter 2 we touted the virtues of using assertions as part of an overall defensive programming strategy. An assertion is basically an evaluation of a Boolean expression followed by a suitable action: do nothing if the condition is true, or halt with a diagnostic message otherwise. It's best to discover assertion failures as soon as possible. If you can evaluate an expression at compile time, use a compile-time assertion. The following example uses a technique that maps a Boolean expression to an array declaration:

 
Sélectionnez
//: C05:StaticAssert1.cpp {-xo}
// A simple, compile-time assertion facility
 
#define STATIC_ASSERT(x) \
  do { typedef int a[(x) ? 1 : -1]; } while(0)
 
int main() {
  STATIC_ASSERT(sizeof(int) <= sizeof(long)); //
Passes
  STATIC_ASSERT(sizeof(double) <= sizeof(int)); //
Fails
} ///:~

The do loop creates a temporary scope for the definition of an array, a, whose size is determined by the condition in question. It is illegal to define an array of size -1, so when the condition is false the statement should fail.

The previous section showed how to evaluate compile-time Boolean expressions. The remaining challenge in emulating assertions at compile time is to print a meaningful error message and halt. All that is required to halt the compiler is a compile error; the trick is to insert helpful text in the error message. The following example from Alexandrescu(77) uses template specialization, a local class, and a little macro magic to do the job:

 
Sélectionnez
//: C05:StaticAssert2.cpp {-g++}
#include <iostream>
using namespace std;
 
// A template and a specialization
template<bool> struct StaticCheck {
  StaticCheck(...);
};
 
template<> struct StaticCheck<false> {};
 
// The macro (generates a local class)
#define STATIC_CHECK(expr, msg) {             \
  class Error_##msg {};                       \
  sizeof((StaticCheck<expr>(Error_##msg()))); \
}
 
// Detects narrowing conversions
template<class To, class From> To safe_cast(From
from) {
  STATIC_CHECK(sizeof(From) <= sizeof(To),
               NarrowingConversion);
  return reinterpret_cast<To>(from);
}
 
int main() {
  void* p = 0;
  int i = safe_cast<int>(p);
  cout << "int cast okay” << endl;
  //! char c = safe_cast<char>(p);
} ///:~

This example defines a function template, safe_cast<>( ), that checks to see if the object it is casting from is no larger than the type of object it casts to. If the size of the target object type is smaller, then the user will be notified at compile time that a narrowing conversion was attempted. Notice that the StaticCheck class template has the curious feature that anything can be converted to an instance of StaticCheck<true> (because of the ellipsis in its constructor(78)), and nothing can be converted to a StaticCheck<false> because no conversions are supplied for that specialization. The idea is to attempt to create an instance of a new class and attempt to convert it to StaticCheck<true>at compile time whenever the condition of interest is true, or to a StaticCheck<false> object when the condition being tested is false. Since the sizeof operator does its work at compile time, it is used to attempt the conversion. If the condition is false, the compiler will complain that it doesn't know how to convert from the new class type to StaticCheck<false>. (The extra parentheses inside the sizeof invocation in STATIC_CHECK( ) are to prevent the compiler from thinking that we're trying to invoke sizeof on a function, which is illegal.) To get some meaningful information inserted into the error message, the new class name carries key text in its name.

The best way to understand this technique is to walk through a specific case. Consider the line in main( ) above which reads:

 
Sélectionnez
   int i =
safe_cast<int>(p);

The call to safe_cast<int>(p) contains the following macro expansion replacing its first line of code:

 
Sélectionnez
{                                                   \
   class Error_NarrowingConversion {};              \
   sizeof(StaticCheck<sizeof(void*) <= sizeof(int)>
\
           (Error_NarrowingConversion()));          \
}

(Recall that the token-pasting preprocessing operator, ##, concatenates its operand into a single token, so Error_##NarrowingConversion becomes the token Error_NarrowingConversion after preprocessing). The class Error_NarrowingConversion is a local class (meaning that it is declared inside a non-namespace scope) because it is not needed elsewhere in the program. The application of the sizeof operator here attempts to determine the size of an instance of StaticCheck<true> (because sizeof(void*) <= sizeof(int) is true on our platforms), created implicitly from the temporary object returned by the call Error_NarrowingConversion( ). The compiler knows the size of the new class Error_NarrowingConversion (it's empty), and so the compile-time use of sizeof at the outer level in STATIC_CHECK( ) is valid. Since the conversion from the Error_NarrowingConversion temporary to StaticCheck<true> succeeds, so does this outer application of sizeof, and execution continues.

Now consider what would happen if the comment were removed from the last line of main( ):

 
Sélectionnez
   char c
= safe_cast<char>(p);

Here the STATIC_CHECK( ) macro inside safe_cast<char>(p) expands to:

 
Sélectionnez
{                                                    \
   class Error_NarrowingConversion {};               \
   sizeof(StaticCheck<sizeof(void*) <=
sizeof(char)> \
           (Error_NarrowingConversion()));           \
}

Since the expression sizeof(void*) <= sizeof(char) is false, a conversion from an Error_NarrowingConversion temporary to StaticCheck<false> is attempted, as follows:

 
Sélectionnez
sizeof(StaticCheck<false>(Error_NarrowingConversion()));

which fails, so the compiler halts with a message something like the following:

 
Sélectionnez
Cannot cast from
'Error_NarrowingConversion' to 'StaticCheck<0>' in function
char safe_cast<char,void *>(void *)

The class name Error_NarrowingConversion is the meaningful message judiciously arranged by the coder. In general, to perform a static assertion with this technique, you just invoke the STATIC_CHECK macro with the compile-time condition to check and with a meaningful name to describe the error.

2-3-6-2. Expression templates

Perhaps the most powerful application of templates is a technique discovered independently in 1994 by Todd Veldhuizen(79) and Daveed Vandevoorde:(80) expression templates. Expression templates enable extensive compile-time optimization of certain computations that result in code that is at least as fast as hand-optimized Fortran, and yet preserves the natural notation of mathematics via operator overloading. Although you wouldn't be likely to use this technique in everyday programming, it is the basis for a number of sophisticated, high-performance mathematical libraries written in C++.(81)

To motivate the need for expression templates, consider typical numerical linear algebra operations, such as adding together two matrices or vectors,(82) such as in the following:

 
Sélectionnez
D = A + B
+ C;

In naive implementations, this expression would result in a number of temporaries—one for A+B, and one for (A+B)+C. When these variables represent immense matrices or vectors, the coincident drain on resources is unacceptable. Expression templates allow you to use the same expression without temporaries.

The following program defines a MyVector class to simulate mathematical vectors of any size. We use a non-type template argument for the length of the vector. We also define a MyVectorSum class to act as a proxy class for a sum of MyVector objects. This allows us to use lazy evaluation, so the addition of vector components is performed on demand without the need for temporaries.

 
Sélectionnez
//: C05:MyVector.cpp
// Optimizes away temporaries via templates.
#include <cstddef>
#include <cstdlib>
#include <ctime>
#include <iostream>
using namespace std;
 
// A proxy class for sums of vectors
template<class, size_t> class MyVectorSum;
 
template<class T, size_t N> class MyVector {
  T data[N];
public:
  MyVector<T,N>& operator=(const
MyVector<T,N>& right) {
    for(size_t i = 0; i < N; ++i)
      data[i] = right.data[i];
    return *this;
  }
  MyVector<T,N>& operator=(const
MyVectorSum<T,N>& right);
  const T& operator[](size_t i) const { return
data[i]; }
  T& operator[](size_t i) { return data[i]; }
};
 
// Proxy class hold references; uses lazy addition
template<class T, size_t N> class MyVectorSum {
  const MyVector<T,N>& left;
  const MyVector<T,N>& right;
public:
  MyVectorSum(const MyVector<T,N>& lhs,
              const MyVector<T,N>& rhs)
  : left(lhs), right(rhs) {}
  T operator[](size_t i) const {
    return left[i] + right[i];
  }
};
 
// Operator to support v3 = v1 + v2
template<class T, size_t N> MyVector<T,N>&
MyVector<T,N>::operator=(const
MyVectorSum<T,N>& right) {
  for(size_t i = 0; i < N; ++i)
    data[i] = right[i];
  return *this;
}
 
// operator+ just stores references
template<class T, size_t N> inline
MyVectorSum<T,N>
operator+(const MyVector<T,N>& left,
          const MyVector<T,N>& right) {
  return MyVectorSum<T,N>(left, right);
}
 
// Convenience functions for the test program below
template<class T, size_t N> void
init(MyVector<T,N>& v) {
  for(size_t i = 0; i < N; ++i)
    v[i] = rand() % 100;
}
 
template<class T, size_t N> void
print(MyVector<T,N>& v) {
  for(size_t i = 0; i < N; ++i)
    cout << v[i] << ' ';
  cout << endl;
}
 
int main() {
  srand(time(0));
  MyVector<int, 5> v1;
  init(v1);
  print(v1);
  MyVector<int, 5> v2;
  init(v2);
  print(v2);
  MyVector<int, 5> v3;
  v3 = v1 + v2;
  print(v3);
  MyVector<int, 5> v4;
  // Not yet supported:
//!  v4 = v1 + v2 + v3;
} ///:~

The MyVectorSum class does no computation when it is created; it merely holds references to the two vectors to be added. Calculations happen only when you access a component of a vector sum (see its operator[ ]( )). The overload of the assignment operator for MyVector that takes a MyVectorSum argument is for an expression such as:

 
Sélectionnez
v1 = v2 + v3; 
// Add two vectors

When the expression v1+v2 is evaluated, a MyVectorSum object is returned (or actually, inserted inline, since that operator+( ) is declared inline). This is a small, fixed-size object (it holds only two references). Then the assignment operator mentioned above is invoked:

 
Sélectionnez
v3.operator=<int,5>(MyVectorSum<int,5>(v2,
v3));

This assigns to each element of v3 the sum of the corresponding elements of v1 and v2, computed in real time. No temporary MyVector objects are created.

This program does not support an expression that has more than two operands, however, such as

 
Sélectionnez
v4 = v1 +
v2 + v3;

The reason is that, after the first addition, a second addition is attempted:

 
Sélectionnez
(v1 + v2) +
v3;

which would require an operator+( ) with a first argument of MyVectorSum and a second argument of type MyVector. We could attempt to provide a number of overloads to meet all situations, but it is better to let templates do the work, as in the following version of the program:

 
Sélectionnez
//: C05:MyVector2.cpp
// Handles sums of any length with expression templates.
#include <cstddef>
#include <cstdlib>
#include <ctime>
#include <iostream>
using namespace std;
 
// A proxy class for sums of vectors
template<class, size_t, class, class> class
MyVectorSum;
 
template<class T, size_t N> class MyVector {
  T data[N];
public:
  MyVector<T,N>& operator=(const
MyVector<T,N>& right) {
    for(size_t i = 0; i < N; ++i)
      data[i] = right.data[i];
    return *this;
  }
  template<class Left, class
Right> MyVector<T,N>&
  operator=(const
MyVectorSum<T,N,Left,Right>& right);
  const T& operator[](size_t i) const {
    return data[i];
  }
  T& operator[](size_t i) {
    return data[i];
  }
};
 
// Allows mixing MyVector and MyVectorSum
template<class T, size_t N, class Left, class
Right>
class MyVectorSum {
  const Left& left;
  const Right& right;
public:
  MyVectorSum(const Left& lhs, const Right&
rhs)
  : left(lhs), right(rhs) {}
  T operator[](size_t i) const {
    return left[i] + right[i];
  }
};
 
template<class T, size_t N>
template<class Left, class Right>
MyVector<T,N>&
MyVector<T,N>::
operator=(const MyVectorSum<T,N,Left,Right>&
right) {
  for(size_t i = 0; i < N; ++i)
    data[i] = right[i];
  return *this;
}
// operator+ just stores references
template<class T, size_t N>
inline
MyVectorSum<T,N,MyVector<T,N>,MyVector<T,N> >
operator+(const MyVector<T,N>& left,
          const MyVector<T,N>& right) {
  return
MyVectorSum<T,N,MyVector<T,N>,MyVector<T,N> >
      (left,right);
}
 
template<class T, size_t N, class Left, class
Right>
inline MyVectorSum<T, N,
MyVectorSum<T,N,Left,Right>,
            MyVector<T,N> >
operator+(const MyVectorSum<T,N,Left,Right>&
left,
          const MyVector<T,N>& right) {
  return
MyVectorSum<T,N,MyVectorSum<T,N,Left,Right>,
                         MyVector<T,N> >
    (left, right);
}
// Convenience functions for the test program below
template<class T, size_t N> void
init(MyVector<T,N>& v) {
  for(size_t i = 0; i < N; ++i)
    v[i] = rand() % 100;
}
 
template<class T, size_t N> void
print(MyVector<T,N>& v) {
  for(size_t i = 0; i < N; ++i)
    cout << v[i] << ' ';
  cout << endl;
}
 
int main() {
  srand(time(0));
  MyVector<int, 5> v1;
  init(v1);
  print(v1);
  MyVector<int, 5> v2;
  init(v2);
  print(v2);
  MyVector<int, 5> v3;
  v3 = v1 + v2;
  print(v3);
  // Now supported:
  MyVector<int, 5> v4;
  v4 = v1 + v2 + v3;
  print(v4);
  MyVector<int, 5> v5;
  v5 = v1 + v2 + v3 + v4;
  print(v5);
} ///:~

The template facility deduces the argument types of a sum using the template arguments, Left and Right, instead of committing to those types ahead of time. The MyVectorSum template takes these extra two parameters so it can represent a sum of any combination of pairs of MyVector and MyVectorSum.

The assignment operator is now a member function template. This allows any <T, N> pair to be coupled with any <Left,Right> pair, so a MyVector object can be assigned from a MyVectorSum holding references to any possible pair of the types MyVector and MyVectorSum.

As we did before, let's trace through a sample assignment to understand exactly what takes place, beginning with the expression

 
Sélectionnez
v4 = v1 +
v2 + v3;

Since the resulting expressions become quite unwieldy, in the explanation that follows, we will use MVS as shorthand for MyVectorSum, and will omit the template arguments.

The first operation is v1+v2, which invokes the inline operator+( ), which in turn inserts MVS(v1, v2) into the compilation stream. This is then added to v3, which results in a temporary object according to the expression MVS(MVS(v1, v2), v3). The final representation of the entire statement is

 
Sélectionnez
v4.operator+(MVS(MVS(v1,
v2), v3));

This transformation is all arranged by the compiler and explains why this technique carries the moniker “expression templates.” The template MyVectorSum represents an expression (an addition, in this case), and the nested calls above are reminiscent of the parse tree of the left-associative expression v1+v2+v3.

An excellent article by Angelika Langer and Klaus Kreft explains how this technique can be extended to more complex computations.(83)

2-3-7. Template compilation models

You may have noticed that all our template examples place fully-defined templates within each compilation unit. (For example, we place them completely within single-file programs, or in header files for multi-file programs.) This runs counter to the conventional practice of separating ordinary function definitions from their declarations by placing the latter in header files and the function implementations in separate (that is, .cpp) files.

The reasons for this traditional separation are:

  • Non-inline function bodies in header files lead to multiple function definitions, resulting in linker errors.
  • Hiding the implementation from clients helps reduce compile-time coupling.
  • Vendors can distribute pre-compiled code (for a particular compiler) along with headers so that users cannot see the function implementations.
  • Compile times are shorter since header files are smaller.
2-3-7-1. The inclusion model

Templates, on the other hand, are not code per se, but instructions for code generation. Only template instantiations are real code. When a compiler has seen a complete template definition during a compilation and then encounters a point of instantiation for that template in the same translation unit, it must deal with the fact that an equivalent point of instantiation may be present in another translation unit. The most common approach consists of generating the code for the instantiation in every translation unit and letting the linker weed out duplicates. That particular approach also works well with inline functions that cannot be inlined and with virtual function tables, which is one of the reasons for its popularity. Nonetheless, several compilers prefer instead to rely on more complex schemes to avoid generating a particular instantiation more than once. Either way, it is the responsibility of the C++ translation system to avoid errors due to multiple equivalent points of instantiation.

A drawback of this approach is that all template source code is visible to the client, so there is little opportunity for library vendors to hide their implementation strategies. Another disadvantage of the inclusion model is that header files are much larger than they would be if function bodies were compiled separately. This can increase compile times dramatically over traditional compilation models.

To help reduce the large headers required by the inclusion model, C++ offers two (non-exclusive) alternative code organization mechanisms: you can manually instantiate each specialization using explicit instantiation or you can use exported templates, which support a large degree of separate compilation.

2-3-7-2. Explicit instantiation

You can manually direct the compiler to instantiate any template specializations of your choice. When you use this technique, there must be one and only one such directive for each such specialization; otherwise you mightget multiple definition errors, just as you would with ordinary, non-inline functions with identical signatures. To illustrate, we first (erroneously) separate the declaration of the min( ) template from earlier in this chapter from its definition, following the normal pattern for ordinary, non-inline functions. The following example consists of five files:

  • OurMin.h: contains the declaration of the min( ) function template.
  • OurMin.cpp: contains the definition of the min( ) function template.
  • UseMin1.cpp: attempts to use an int-instantiation of min( ).
  • UseMin2.cpp: attempts to use a double-instantiation of min( ).
  • MinMain.cpp: calls usemin1( ) and usemin2( ).
 
Sélectionnez
//: C05:OurMin.h
#ifndef OURMIN_H
#define OURMIN_H
// The declaration of min()
template<typename T> const T& min(const
T&, const T&);
#endif //
OURMIN_H ///:~
 
Sélectionnez
// OurMin.cpp
#include "OurMin.h"
// The definition of min()
template<typename T> const T& min(const
T& a, const T& b) {
  return (a < b) ? a : b;
}
 
Sélectionnez
//: C05:UseMin1.cpp {O}
#include <iostream>
#include "OurMin.h"
void usemin1() {
  std::cout << min(1,2) <<
std::endl;
} ///:~
 
Sélectionnez
//: C05:UseMin2.cpp {O}
#include <iostream>
#include "OurMin.h"
void usemin2() {
  std::cout << min(3.1,4.2)
<< std::endl;
} ///:~
 
Sélectionnez
//: C05:MinMain.cpp
//{L} UseMin1 UseMin2 MinInstances
void usemin1();
void usemin2();
 
int main() {
  usemin1();
  usemin2();
} ///:~

When we attempt to build this program, the linker reports unresolved external references for min<int>( ) and min<double>( ). The reason is that when the compiler encounters the calls to specializations of min( ) in UseMin1 and UseMin2, only the declaration of min( ) is visible. Since the definition is not available, the compiler assumes it will come from some other translation unit, and the needed specializations are thus not instantiated at that point, leaving the linker to eventually complain that it cannot find them.

To solve this problem, we will introduce a new file, MinInstances.cpp, that explicitly instantiates the needed specializations of min( ):

 
Sélectionnez
//: C05:MinInstances.cpp {O}
#include "OurMin.cpp"
 
// Explicit Instantiations for int and double
template const int& min<int>(const int&,
const int&);
template const double& min<double>(const
double&,
 
             const double&);
///:~

To manually instantiate a particular template specialization, you precede the specialization's declaration with the template keyword. Note that we must include OurMin.cpp, not OurMin.h, here, because the compiler needs the template definition to perform the instantiation. This is the only place where we have to do this in this program,(84) however, since it gives us the unique instantiations of min( ) that we need—the declarations alone suffice for the other files. Since we are including OurMin.cpp with the macro preprocessor, we add include guards:

 
Sélectionnez
//: C05:OurMin.cpp {O}
#ifndef OURMIN_CPP
#define OURMIN_CPP
#include "OurMin.h"
 
template<typename T> const T& min(const
T& a, const T& b) {
  return (a < b) ? a : b;
}
#endif //
OURMIN_CPP ///:~

Now when we compile all the files together into a complete program, the unique instances of min( ) are found, and the program executes correctly, giving the output:

 
Sélectionnez
1
3.1

You can also manually instantiate classes and static data members. When explicitly instantiating a class, all member functions for the requested specialization are instantiated, except any that may have been explicitly instantiated previously. This is important, as it will render many templates useless when using this mechanism—specifically, templates that implement different functionality depending on their parameterization type. Implicit instantiation has the advantage here: only member functions that get called are instantiated.

Explicit instantiation is intended for large projects where a hefty chunk of compilation time can be avoided. Whether you use implicit or explicit instantiation is independent of which template compilation you use. You can use manual instantiation with either the inclusion model or the separation model (discussed in the next section).

2-3-7-3. The separation model

The separation model of template compilation separates function template definitions or static data member definitions from their declarations across translation units, just like you do with ordinary functions and data, by exporting templates. After reading the preceding two sections, this must sound strange. Why bother to have the inclusion model in the first place if you can just adhere to the status quo? The reasons are both historical and technical.

Historically, the inclusion model was the first to experience widespread commercial use—all C++ compilers support the inclusion model. Part of the reason for that was that the separation model was not well specified until late in the standardization process, but also that the inclusion model is easier to implement. A lot of working code was in existence long before the semantics of the separation model were finalized.

The separation model is so difficult to implement that, as of Summer 2003, only one compiler front end (EDG) supports the separation model, and at the moment it still requires that template source code be available at compile time to perform instantiation on demand. Plans are in place to use some form of intermediate code instead of requiring that the original source be at hand, at which point you will be able to ship “pre-compiled” templates without shipping source code. Because of the lookup complexities explained earlier in this chapter (about dependent names being looked up in the template definition context), a full template definition still has to be available in some form when you compile a program that instantiates it.

The syntax to separate the source code of a template definition from its declaration is easy enough. You use the export keyword:

 
Sélectionnez
// C05:OurMin2.h
// Declares min as an exported template
// (Only works with EDG-based compilers)
#ifndef OURMIN2_H
#define OURMIN2_H
export template<typename T> const T&
min(const T&,
                                         const T&);
#endif //
OURMIN2_H ///:~

Similar to inline or virtual, the export keyword need only be mentioned once in a compilation stream, where an exported template is introduced. For this reason, we need not repeat it in the implementation file, but it is considered good practice to do so:

 
Sélectionnez
// C05:OurMin2.cpp
// The definition of the exported min template
// (Only works with EDG-based compilers)
#include "OurMin2.h"
export
template<typename T> const T& min(const
T& a, const T& b) {
  return (a < b) ? a : b;
} ///:~

The UseMin files used previously only need to include the correct header file (OurMin2.h), and the main program doesn't change. Although this appears to give true separation, the file with the template definition (OurMin2.cpp) must still be shipped to users (because it must be processed for each instantiation of min( )) until such time as some form of intermediate code representation of template definitions is supported. So while the standard does provide for a true separation model, not all of its benefits can be reaped today. Only one family of compilers currently supports export (those based on the EDG front end), and these compilers currently do not exploit the potential ability to distribute template definitions in compiled form.

2-3-8. Summary

Templates go far beyond simple type parameterization. When you combine argument type deduction, custom specialization, and template metaprogramming, C++ templates emerge as a powerful code generation mechanism.

One of the weaknesses of C++ templates that we did not mention is the difficulty in interpreting compile-time error messages. The quantity of inscrutable text spewed out by the compiler can be quite overwhelming. C++ compilers have improved their template error messages, and Leor Zolman has written a tool called STLFilt that renders these error messages much more readable by extracting the useful information and throwing away the rest.(85)

Another important idea to take away from this chapter is that a template implies an interface. That is, even though the template keyword says “I'll take any type,” the code in a template definition requires that certain operators and member functions be supported—that's the interface. So in reality, a template definition is saying, “I'll take any type that supports this interface.” Things would be much nicer if the compiler could simply say, “Hey, this type that you're trying to instantiate the template with doesn't support that interface—can't do it.” Using templates constitutes a sort of “latent type checking” that is more flexible than the pure object-oriented practice of requiring all types to derive from certain base classes.

In Chapters 6 and 7 we explore in depth the most famous application of templates, the subset of the Standard C++ library commonly known as the Standard Template Library (STL). Chapters 9 and 10 also use template techniques not found in this chapter.

2-3-9. Exercises

Solutions to selected exercises can be found in the electronic document The Thinking in C++ Volume 2 Annotated Solution Guide, available for a small fee from www.MindView.net.

  1. Write a unary function template that takes a single type template parameter. Create a full specialization for the type int. Also create a non-template overload for this function that takes a single int parameter. Have your main program invoke three function variations.
  2. Write a class template that uses a vector to implement a stack data structure.
  3. Modify your solution to the previous exercise so that the type of the container used to implement the stack is a template template parameter.
  4. In the following code, the class NonComparable does not have an operator=( ). Why would the presence of the class HardLogic cause a compile error, but SoftLogic would not?
     
    Sélectionnez
    //: C05:Exercise4.cpp {-xo}
    class Noncomparable {};
     
    struct HardLogic {
      Noncomparable nc1, nc2;
      void compare() {
        return nc1 == nc2; // Compiler error
      }
    };
     
    template<class T> struct SoftLogic {
      Noncomparable nc1, nc2;
      void noOp() {}
      void compare() {
        nc1 == nc2;
      }
    };
     
    int main() {
      SoftLogic<Noncomparable> l;
      l.noOp();
    } ///:~
  5. Write a function template that takes a single type parameter (T) and accepts four function arguments: an array of T, a start index, a stop index (inclusive), and an optional initial value. The function returns the sum of all the array elements in the specified range and the initial value. Use the default constructor of T for the default initial value.
  6. Repeat the previous exercise but use explicit instantiation to manually create specializations for int and double, following the technique explained in this chapter.
  7. Why does the following code not compile? (Hint: what do class member functions have access to?)
     
    Sélectionnez
    //: C05:Exercise7.cpp {-xo}
    class Buddy {};
     
    template<class T> class My {
      int i;
    public:
      void play(My<Buddy>& s) {
        s.i = 3;
      }
    };
     
    int main() {
      My<int> h;
      My<Buddy> me, bud;
      h.play(bud);
      me.play(bud);
    } ///:~
  8. Why does the following code not compile?
     
    Sélectionnez
    //: C05:Exercise8.cpp {-xo}
    template<class T> double pythag(T a, T b, T c) {
      return (-b + sqrt(double(b*b - 4*a*c))) / 2*a;
    }
     
    int main() {
      pythag(1, 2, 3);
      pythag(1.0, 2.0, 3.0);
      pythag(1, 2.0, 3.0);
      pythag<double>(1, 2.0, 3.0);
    } ///:~
  9. Write templates that take non-type parameters of the following variety: an int, a pointer to an int, a pointer to a static class member of type int, and a pointer to a static member function.
  10. Write a class template that takes two type parameters. Define a partial specialization for the first parameter, and another partial specialization that specifies the second parameter. In each specialization, introduce members that are not in the primary template.
  11. Define a class template named Bob that takes a single type parameter. Make Bob a friend of all instances of a template class named Friendly, and a friend of a class template named Picky only when the type parameter of Bob and Picky are identical. Give Bob member functions that demonstrate its friendship.

2-4. Generic Algorithms

Algorithms are at the core of computing. To be able to write an algorithm that works with any type of sequence makes your programs both simpler and safer. The ability to customize algorithms at runtime has revolutionized software development.

The subset of the Standard C++ library known as the Standard Template Library (STL) was originally designed around generic algorithms—code that processes sequences of any type of values in a type-safe manner. The goal was to use predefined algorithms for almost every task, instead of hand-coding loops every time you need to process a collection of data. This power comes with a bit of a learning curve, however. By the time you get to the end of this chapter, you should be able to decide for yourself whether you find the algorithms addictive or too confusing to remember. If you're like most people, you'll resist them at first but then tend to use them more and more as time goes on.

2-4-1. A first look

Among other things, the generic algorithms in the standard library provide a vocabulary with which to describe solutions. Once you become familiar with the algorithms, you'll have a new set of words with which to discuss what you're doing, and these words are at a higher level than what you had before. You don't need to say, “This loop moves through and assigns from here to there … oh, I see, it's copying!” Instead, you just say copy( ). This is what we've been doing in computer programming from the beginning—creating high-level abstractions to express what you're doing and spending less time saying how you're doing it. The how has been solved once and for all and is hidden in the algorithm's code, ready to be reused on demand.

Here's an example of how to use the copy algorithm:

 
Sélectionnez
//: C06:CopyInts.cpp
// Copies ints without an explicit loop.
#include <algorithm>
#include <cassert>
#include <cstddef>  // For size_t
using namespace std;
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  int b[SIZE];
  copy(a, a + SIZE, b);
  for(size_t i = 0; i < SIZE; ++i)
    assert(a[i] == b[i]);
} ///:~

The copy( ) algorithm's first two parameters represent the range of the input sequence—in this case the array a. Ranges are denoted by a pair of pointers. The first points to the first element of the sequence, and the second points one position past the end of the array (right after the last element). This may seem strange at first, but it is an old C idiom that comes in quite handy. For example, the difference of these two pointers yields the number of elements in the sequence. More important, in implementing copy, the second pointer can act as a sentinel to stop the iteration through the sequence. The third argument refers to the beginning of the output sequence, which is the array b in this example. It is assumed that the array that b represents has enough space to receive the copied elements.

The copy( ) algorithm wouldn't be very exciting if it could only process integers. It can copy any kind of sequence. The following example copies string objects:

 
Sélectionnez
//: C06:CopyStrings.cpp
// Copies strings.
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <string>
using namespace std;
 
int main() {
  string a[] = {"read", "my",
"lips"};
  const size_t SIZE = sizeof a / sizeof a[0];
  string b[SIZE];
  copy(a, a + SIZE, b);
  assert(equal(a, a + SIZE, b));
} ///:~

This example introduces another algorithm, equal( ), which returns true only if each element in the first sequence is equal (using its operator==( )) to the corresponding element in the second sequence. This example traverses each sequence twice, once for the copy, and once for the comparison, without a single explicit loop!

Generic algorithms achieve this flexibility because they are function templates. If you think that the implementation of copy( ) looks like the following, you're almost right:

 
Sélectionnez
template<typename T> void copy(T* begin, T* end,
T* dest) {
  while(begin != end)
    *dest++ = *begin++;
}

We say “almost” because copy( ) can process sequences delimited by anything that acts like a pointer, such as an iterator. In this way, copy( ) can be used to duplicate a vector:

 
Sélectionnez
//: C06:CopyVector.cpp
// Copies the contents of a vector.
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <vector>
using namespace std;
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  vector<int> v1(a, a + SIZE);
  vector<int> v2(SIZE);
  copy(v1.begin(), v1.end(), v2.begin());
  assert(equal(v1.begin(), v1.end(), v2.begin()));
} ///:~

The first vector, v1, is initialized from the sequence of integers in the array a. The definition of the vectorv2 uses a different vector constructor that makes room for SIZE elements, initialized to zero (the default value for integers).

As with the array example earlier, it's important that v2 have enough space to receive a copy of the contents of v1. For convenience, a special library function, back_inserter( ), returns a special type of iterator that inserts elements instead of overwriting them, so memory is expanded automatically by the container as needed. The following example uses back_inserter( ), so it doesn't have to establish the size of the output vector, v2, ahead of time:

 
Sélectionnez
//: C06:InsertVector.cpp
// Appends the contents of a vector to another.
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <iterator>
#include <vector>
using namespace std;
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  vector<int> v1(a, a + SIZE);
  vector<int> v2;  // v2 is empty here
  copy(v1.begin(), v1.end(), back_inserter(v2));
  assert(equal(v1.begin(), v1.end(), v2.begin()));
} ///:~

The back_inserter( ) function is defined in the <iterator> header. We'll explain how insert iterators work in depth in the next chapter.

Since iterators are identical to pointers in all essential ways, you can write the algorithms in the standard library in such a way as to allow both pointer and iterator arguments. For this reason, the implementation of copy( ) looks more like the following code:

 
Sélectionnez
template<typename Iterator>
void copy(Iterator begin, Iterator end, Iterator dest)
{
  while(begin != end)
    *begin++ = *dest++;
}

Whichever argument type you use in the call, copy( ) assumes it properly implements the indirection and increment operators. If it doesn't, you'll get a compile-time error.

2-4-1-1. Predicates

At times, you might want to copy only a well-defined subset of one sequence to another, such as only those elements that satisfy a particular condition. To achieve this flexibility, many algorithms have alternate calling sequences that allow you to supply a predicate, which is simply a function that returns a Boolean value based on some criterion. Suppose, for example, that you only want to extract from a sequence of integers those numbers that are less than or equal to 15. A version of copy( ) called remove_copy_if( ) can do the job, like this:

 
Sélectionnez
//: C06:CopyInts2.cpp
// Ignores ints that satisfy a predicate.
#include <algorithm>
#include <cstddef>
#include <iostream>
using namespace std;
 
// You supply this predicate
bool gt15(int x) { return 15 < x; }
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  int b[SIZE];
  int* endb = remove_copy_if(a, a+SIZE, b, gt15);
  int* beginb = b;
  while(beginb != endb)
    cout << *beginb++ << endl; // Prints 10
only
} ///:~

The remove_copy_if( ) function template takes the usual range-delimiting pointers, followed by a predicate of your choosing. The predicate must be a pointer to a function(86) that takes a single argument of the same type as the elements in the sequence, and it must return a bool. Here, the function gt15 returns true if its argument is greater than 15. The remove_copy_if( ) algorithm applies gt15( ) to each element in the input sequence and ignores those elements where the predicate yields true when writing to the output sequence.

The following program illustrates yet another variation of the copy algorithm:

 
Sélectionnez
//: C06:CopyStrings2.cpp
// Replaces strings that satisfy a predicate.
#include <algorithm>
#include <cstddef>
#include <iostream>
#include <string>
using namespace std;
 
// The predicate
bool contains_e(const string& s) {
  return s.find('e') != string::npos;
}
 
int main() {
  string a[] = {"read", "my",
"lips"};
  const size_t SIZE = sizeof a / sizeof a[0];
  string b[SIZE];
  string* endb = replace_copy_if(a, a + SIZE, b,
    contains_e, string("kiss"));
  string* beginb = b;
  while(beginb != endb)
    cout << *beginb++ << endl;
} ///:~

Instead of just ignoring elements that don't satisfy the predicate, replace_copy_if( ) substitutes a fixed value for such elements when populating the output sequence. The output is:

 
Sélectionnez
kiss
my
lips

because the original occurrence of “read,” the only input string containing the letter e, is replaced by the word “kiss,” as specified in the last argument in the call to replace_copy_if( ).

The replace_if( ) algorithm changes the original sequence in place, instead of writing to a separate output sequence, as the following program shows:

 
Sélectionnez
//: C06:ReplaceStrings.cpp
// Replaces strings in-place.
#include <algorithm>
#include <cstddef>
#include <iostream>
#include <string>
using namespace std;
 
bool contains_e(const string& s) {
  return s.find('e') != string::npos;
}
 
int main() {
  string a[] = {"read", "my",
"lips"};
  const size_t SIZE = sizeof a / sizeof a[0];
  replace_if(a, a + SIZE, contains_e,
string("kiss"));
  string* p = a;
  while(p != a + SIZE)
    cout << *p++ << endl;
} ///:~
2-4-1-2. Stream iterators

Like any good software library, the Standard C++ Library attempts to provide convenient ways to automate common tasks. We mentioned in the beginning of this chapter that you can use generic algorithms in place of looping constructs. So far, however, our examples have still used an explicit loop to print their output. Since printing output is one of the most common tasks, you would hope for a way to automate that too.

That's where stream iterators come in. A stream iterator uses a stream as either an input or an output sequence. To eliminate the output loop in the CopyInts2.cpp program, for instance, you can do something like the following:

 
Sélectionnez
//: C06:CopyInts3.cpp
// Uses an output stream iterator.
#include <algorithm>
#include <cstddef>
#include <iostream>
#include <iterator>
using namespace std;
 
bool gt15(int x) { return 15 < x; }
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  remove_copy_if(a, a + SIZE,
                 ostream_iterator<int>(cout,
"\n"), gt15);
} ///:~

In this example we've replaced the output sequence b in the third argument of remove_copy_if( ) with an output stream iterator, which is an instance of the ostream_iterator class template declared in the <iterator> header. Output stream iterators overload their copy-assignment operators to write to their stream. This particular instance of ostream_iterator is attached to the output stream cout. Every time remove_copy_if( ) assigns an integer from the sequence a to cout through this iterator, the iterator writes the integer to cout and also automatically writes an instance of the separator string found in its second argument, which in this case contains the newline character.

It is just as easy to write to a file by providing an output file stream, instead of cout:

 
Sélectionnez
//: C06:CopyIntsToFile.cpp
// Uses an output file stream iterator.
#include <algorithm>
#include <cstddef>
#include <fstream>
#include <iterator>
using namespace std;
 
bool gt15(int x) { return 15 < x; }
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  ofstream outf("ints.out");
  remove_copy_if(a, a + SIZE,
                 ostream_iterator<int>(outf,
"\n"), gt15);
} ///:~

An input stream iterator allows an algorithm to get its input sequence from an input stream. This is accomplished by having both the constructor and operator++( ) read the next element from the underlying stream and by overloading operator*( ) to yield the value previously read. Since algorithms require two pointers to delimit an input sequence, you can construct an istream_iteratorin two ways, as you can see in the program that follows.

 
Sélectionnez
//: C06:CopyIntsFromFile.cpp
// Uses an input stream iterator.
#include <algorithm>
#include <fstream>
#include <iostream>
#include <iterator>
#include "../require.h"
using namespace std;
 
bool gt15(int x) { return 15 < x; }
 
int main() {
  ofstream ints("someInts.dat");
  ints << "1 3 47 5 84 9";
  ints.close();
  ifstream inf("someInts.dat");
  assure(inf, "someInts.dat");
  remove_copy_if(istream_iterator<int>(inf),
                 istream_iterator<int>(),
                 ostream_iterator<int>(cout,
"\n"), gt15);
} ///:~

The first argument to replace_copy_if( ) in this program attaches an istream_iterator object to the input file stream containing ints. The second argument uses the default constructor of the istream_iterator class. This call constructs a special value of istream_iterator that indicates end-of-file, so that when the first iterator finally encounters the end of the physical file, it compares equal to the value istream_iterator<int>( ), allowing the algorithm to terminate correctly. Note that this example avoids using an explicit array altogether.

2-4-1-3. Algorithm complexity

Using a software library is a matter of trust. You trust the implementers to not only provide correct functionality, but you also hope that the functions execute as efficiently as possible. It's better to write your own loops than to use algorithms that degrade performance.

To guarantee quality library implementations, the C++ Standard not only specifies what an algorithm should do, but how fast it should do it and sometimes how much space it should use. Any algorithm that does not meet the performance requirements does not conform to the standard. The measure of an algorithm's operational efficiency is called its complexity.

When possible, the standard specifies the exact number of operation counts an algorithm should use. The count_if( ) algorithm, for example, returns the number of elements in a sequence satisfying a given predicate. The following call to count_if( ), if applied to a sequence of integers similar to the examples earlier in this chapter, yields the number of integer elements that are greater than 15:

 
Sélectionnez
size_t n = count_if(a, a + SIZE, gt15);

Since count_if( ) must look at every element exactly once, it is specified to make a number of comparisons exactly equal to the number of elements in the sequence. The copy( ) algorithm has the same specification.

Other algorithms can be specified to take at most a certain number of operations. The find( ) algorithm searches through a sequence in order until it encounters an element equal to its third argument:

 
Sélectionnez
int* p = find(a, a + SIZE, 20);

It stops as soon as the element is found and returns a pointer to that first occurrence. If it doesn't find one, it returns a pointer one position past the end of the sequence (a+SIZE in this example). So find() makes at most a number of comparisons equal to the number of elements in the sequence.

Sometimes the number of operations an algorithm takes cannot be measured with such precision. In such cases, the standard specifies the algorithm's asymptotic complexity, which is a measure of how the algorithm behaves with large sequences compared to well-known formulas. A good example is the sort( ) algorithm, which the standard says takes “approximately n log n comparisons on average” (n is the number of elements in the sequence).(87) Such complexity measures give a “feel” for the cost of an algorithm and at least give a meaningful basis for comparing algorithms. As you'll see in the next chapter, the find( ) member function for the set container has logarithmic complexity, which means that the cost of searching for an element in a set will, for large sets, be proportional to the logarithm of the number of elements. This is much smaller than the number of elements for large n, so it is always better to search a set by using its find( ) member function rather than by using the generic find( ) algorithm.

2-4-2. Function objects

As you study some of the examples earlier in this chapter, you will probably notice the limited utility of the function gt15( ). What if you want to use a number other than 15 as a comparison threshold? You may need a gt20( ) or gt25( ) or others as well. Having to write a separate function is time consuming, but also unreasonable because you must know all required values when you write your application code.

The latter limitation means that you can't use runtime values(88) to govern your searches, which is unacceptable. Overcoming this difficulty requires a way to pass information to predicates at runtime. For example, you would need a greater-than function that you can initialize with an arbitrary comparison value. Unfortunately, you can't pass that value as a function parameter because unary predicates, such as our gt15( ), are applied to each value in a sequence individually and must therefore take only one parameter.

The way out of this dilemma is, as always, to create an abstraction. Here, we need an abstraction that can act like a function as well as store state, without disturbing the number of function parameters it accepts when used. This abstraction is called a function object.(89)

A function object is an instance of a class that overloads operator( ), the function call operator. This operator allows an object to be used with function call syntax. As with any other object, you can initialize it via its constructors. Here is a function object that can be used in place of gt15( ):

 
Sélectionnez
//: C06:GreaterThanN.cpp
#include <iostream>
using namespace std;
 
class gt_n {
  int value;
public:
  gt_n(int val) : value(val) {}
  bool operator()(int n) { return n > value; }
};
 
int main() {
  gt_n f(4);
  cout << f(3) << endl;  // Prints 0 (for
false)
  cout << f(5) << endl;  // Prints 1 (for
true)
} ///:~

The fixed value to compare against (4) is passed when the function object f is created. The expression f(3) is then evaluated by the compiler as the following function call:

 
Sélectionnez
f.operator()(3);

which returns the value of the expression 3 > value, which is false when value is 4, as it is in this example.

Since such comparisons apply to types other than int, it would make sense to define gt_n( ) as a class template. It turns out you don't need to do it yourself, though—the standard library has already done it for you. The following descriptions of function objects should not only make that topic clear, but also give you a better understanding of how the generic algorithms work.

2-4-2-1. Classification of function objects

The Standard C++ library classifies a function object based on the number of arguments its operator( ) takes and the kind of value it returns. This classification is based on whether a function object's operator( ) takes zero, one, or two arguments:

Generator: A type of function object that takes no arguments and returns a value of an arbitrary type. A random number generator is an example of a generator. The standard library provides one generator, the function rand( ) declared in <cstdlib>, and has some algorithms, such as generate_n( ), which apply generators to a sequence.

Unary Function: A type of function object that takes a single argument of any type and returns a value that may be of a different type (which may be void).

Binary Function: A type of function object that takes two arguments of any two (possibly distinct) types and returns a value of any type (including void).

Unary Predicate: A Unary Function that returns a bool.

Binary Predicate: A Binary Function that returns a bool.

Strict Weak Ordering: A binary predicate that allows for a more general interpretation of “equality.” Some of the standard containers consider two elements equivalent if neither is less than the other (using operator<( )). This is important when comparing floating-point values, and objects of other types where operator==( ) is unreliable or unavailable. This notion also applies if you want to sort a sequence of data records (structs) on a subset of the struct's fields. That comparison scheme is considered a strict weak ordering because two records with equal keys are not really “equal” as total objects, but they are equal as far as the comparison you're using is concerned. The importance of this concept will become clearer in the next chapter.

In addition, certain algorithms make assumptions about the operations available for the types of objects they process. We will use the following terms to indicate these assumptions:

LessThanComparable: A class that has a less-than operator<.

Assignable: A class that has a copy-assignment operator= for its own type.

EqualityComparable: A class that has an equivalence operator== for its own type.

We will use these terms later in this chapter to describe the generic algorithms in the standard library.

2-4-2-2. Automatic creation of function objects

The <functional> header defines a number of useful generic function objects. They are admittedly simple, but you can use them to compose more complicated function objects. Consequently, in many instances, you can construct complicated predicates without writing a single function. You do so by using function object adaptors(90) to take the simple function objects and adapt them for use with other function objects in a chain of operations.

To illustrate, let's use only standard function objects to accomplish what gt15( ) did earlier. The standard function object, greater, is a binary function object that returns true if its first argument is greater than its second argument. We cannot apply this directly to a sequence of integers through an algorithm such as remove_copy_if( ) because remove_copy_if( ) expects a unary predicate. We can construct a unary predicate on the fly that uses greater to compare its first argument to a fixed value. We fix the value of the second parameter at 15 using the function object adaptor bind2nd, like this:

 
Sélectionnez
//: C06:CopyInts4.cpp
// Uses a standard function object and adaptor.
#include <algorithm>
#include <cstddef>
#include <functional>
#include <iostream>
#include <iterator>
using namespace std;
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  remove_copy_if(a, a + SIZE,
                 ostream_iterator<int>(cout,
"\n"),
                 bind2nd(greater<int>(), 15));
} ///:~

This program produces the same result as CopyInts3.cpp, but without writing our own predicate function gt15( ). The function object adaptor bind2nd( ) is a template function that creates a function object of type binder2nd, which simply stores the two arguments passed to bind2nd( ), the first of which must be a binary function or function object (that is, anything that can be called with two arguments). The operator( ) function in binder2nd, which is itself a unary function, calls the binary function it stored, passing it its incoming parameter and the fixed value it stored.

To make the explanation concrete for this example, let's call the instance of binder2nd created by bind2nd( ) by the name b. When b is created, it receives two parameters (greater<int>( ) and 15) and stores them. Let's call the instance of greater<int> by the name g, and call the instance of the output stream iterator by the name o. Then the call to remove_copy_if( ) earlier conceptually becomes the following:

 
Sélectionnez
remove_copy_if(a, a + SIZE, o, b(g, 15).operator());

As remove_copy_if( ) iterates through the sequence, it calls b on each element, to determine whether to ignore the element when copying to the destination. If we denote the current element by the name e, that call inside remove_copy_if( ) is equivalent to

 
Sélectionnez
if(b(e))

but binder2nd's function call operator just turns around and calls g(e,15), so the earlier call is the same as

 
Sélectionnez
if(greater<int>(e, 15))

which is the comparison we were seeking. There is also a bind1st( ) adaptor that creates a binder1st object, which fixes the first argument of the associated input binary function.

As another example, let's count the number of elements in the sequence not equal to 20. This time we'll use the algorithm count_if( ), introduced earlier. There is a standard binary function object, equal_to, and also a function object adaptor, not1( ), that takes a unary function object as a parameter and invert its truth value. The following program will do the job:

 
Sélectionnez
//: C06:CountNotEqual.cpp
// Count elements not equal to 20.
#include <algorithm>
#include <cstddef>
#include <functional>
#include <iostream>
using namespace std;
 
int main() {
  int a[] = { 10, 20, 30 };
  const size_t SIZE = sizeof a / sizeof a[0];
  cout << count_if(a, a + SIZE,
                   not1(bind1st(equal_to<int>(),
20)));// 2
} ///:~

As remove_copy_if( ) did in the previous example, count_if( ) calls the predicate in its third argument (let's call it n) for each element of its sequence and increments its internal counter each time true is returned. If, as before, we call the current element of the sequence by the name e, the statement

 
Sélectionnez
if(n(e))

in the implementation of count_if is interpreted as

 
Sélectionnez
if(!bind1st(equal_to<int>, 20)(e))

which ends up as

 
Sélectionnez
if(!equal_to<int>(20, e))

because not1( ) returns the logical negation of the result of calling its unary function argument. The first argument to equal_to is 20 because we used bind1st( ) instead of bind2nd( ). Since testing for equality is symmetric in its arguments, we could have used either bind1st( ) or bind2nd( ) in this example.

The following table shows the templates that generate the standard function objects, along with the kinds of expressions to which they apply:

Name Type Result produced
plus BinaryFunction arg1 + arg2
minus BinaryFunction arg1 - arg2
multiplies BinaryFunction arg1 * arg2
divides BinaryFunction arg1 / arg2
modulus BinaryFunction arg1 % arg2
negate UnaryFunction - arg1
equal_to BinaryPredicate arg1 == arg2
not_equal_to BinaryPredicate arg1 != arg2
greater BinaryPredicate arg1 > arg2
less BinaryPredicate arg1 < arg2
greater_equal BinaryPredicate arg1 >= arg2
less_equal BinaryPredicate arg1 <= arg2
logical_and BinaryPredicate arg1 && arg2
Logical_or BinaryPredicate arg1 || arg2
logical_not UnaryPredicate !arg1
unary_negate Unary Logical !(UnaryPredicate(arg1))
binary_negate Binary Logical !(BinaryPredicate(arg1, arg2))
2-4-2-3. Adaptable function objects

Standard function adaptors such as bind1st( ) and bind2nd( ) make some assumptions about the function objects they process. Consider the following expression from the last line of the earlier CountNotEqual.cpp program:

 
Sélectionnez
not1(bind1st(equal_to<int>(), 20))

The bind1st( ) adaptor creates a unary function object of type binder1st, which simply stores an instance of equal_to<int> and the value 20. The binder1st::operator( ) function needs to know its argument type and its return type; otherwise, it will not have a valid declaration. The convention to solve this problem is to expect all function objects to provide nested type definitions for these types. For unary functions, the type names are argument_type and result_type; for binary function objects they are first_argument_type, second_argument_type, and result_type. Looking at the implementation of bind1st( ) and binder1st in the <functional> header reveals these expectations. First inspect bind1st( ), as it might appear in a typical library implementation:

 
Sélectionnez
template<class Op, class T>
binder1st<Op> bind1st(const Op& f, const
T& val) {
  typedef typename Op::first_argument_type Arg1_t;
  return binder1st<Op>(f, Arg1_t(val));
}

Note that the template parameter, Op, which represents the type of the binary function being adapted by bind1st( ), must have a nested type named first_argument_type. (Note also the use of typename to inform the compiler that it is a member type name, as explained in Chapter 5.) Now see how binder1st uses the type names in Op in its declaration of its function call operator:

 
Sélectionnez
// Inside the implementation for binder1st<Op>
typename Op::result_type
operator()(const typename Op::second_argument_type&
x)
  const;

Function objects whose classes provide these type names are called adaptable function objects.

Since these names are expected of all standard function objects as well as of any function objects you create to use with function object adaptors, the <functional> header provides two templates that define these types for you: unary_function and binary_function. You simply derive from these classes while filling in the argument types as template parameters. Suppose, for example, that we want to make the function object gt_n, defined earlier in this chapter, adaptable. All we need to do is the following:

 
Sélectionnez
class gt_n : public unary_function<int, bool> {
  int value;
public:
  gt_n(int val) : value(val) {}
  bool operator()(int n) {
    return n > value;
  }
};

All unary_function does is to provide the appropriate type definitions, which it infers from its template parameters as you can see in its definition:

 
Sélectionnez
template<class Arg, class Result> struct
unary_function {
  typedef Arg argument_type;
  typedef Result result_type;
};

These types become accessible through gt_n because it derives publicly from unary_function. The binary_function template behaves in a similar manner.

2-4-2-4. More function object examples

The following FunctionObjects.cpp example provides simple tests for most of the built-in basic function object templates. This way, you can see how to use each template, along with the resulting behavior. This example uses one of the following generators for convenience:

 
Sélectionnez
//: C06:Generators.h
// Different ways to fill sequences.
#ifndef GENERATORS_H
#define GENERATORS_H
#include <cstring>
#include <set>
#include <cstdlib>
 
// A generator that can skip over numbers:
class SkipGen {
  int i;
  int skp;
public:
  SkipGen(int start = 0, int skip = 1)
  : i(start), skp(skip) {}
  int operator()() {
    int r = i;
    i += skp;
    return r;
  }
};
 
// Generate unique random numbers from 0 to mod:
class URandGen {
  std::set<int> used;
  int limit;
public:
  URandGen(int lim) : limit(lim) {}
  int operator()() {
    while(true) {
      int i = int(std::rand()) % limit;
      if(used.find(i) == used.end()) {
        used.insert(i);
        return i;
      }
    }
  }
};
 
// Produces random characters:
class CharGen {
  static const char* source;
  static const int len;
public:
  char operator()() {
    return source[std::rand() % len];
  }
};
#endif // GENERATORS_H ///:~
 
Sélectionnez
//: C06:Generators.cpp {O}
#include "Generators.h"
const char* CharGen::source = "ABCDEFGHIJK"
 
"LMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
const int CharGen::len = std::strlen(source);
///:~

We'll be using these generating functions in various examples throughout this chapter. The SkipGen function object returns the next number of an arithmetic sequence whose common difference is held in its skp data member. A URandGen object generates a unique random number in a specified range. (It uses a set container, which we'll discuss in the next chapter.) A CharGen object returns a random alphabetic character. Here is a sample program using UrandGen:

 
Sélectionnez
//: C06:FunctionObjects.cpp {-bor}
// Illustrates selected predefined function object
// templates from the Standard C++ library.
//{L} Generators
#include <algorithm>
#include <cstdlib>
#include <ctime>
#include <functional>
#include <iostream>
#include <iterator>
#include <vector>
#include "Generators.h"
#include "PrintSequence.h"
using namespace std;
 
template<typename Contain, typename UnaryFunc>
void testUnary(Contain& source, Contain& dest,
  UnaryFunc f) {
  transform(source.begin(), source.end(), dest.begin(),
f);
}
 
template<typename Contain1, typename Contain2,
  typename BinaryFunc>
void testBinary(Contain1& src1, Contain1& src2,
  Contain2& dest, BinaryFunc f) {
  transform(src1.begin(), src1.end(),
    src2.begin(), dest.begin(), f);
}
 
// Executes the expression, then stringizes the
// expression into the print statement:
#define T(EXPR) EXPR; print(r.begin(), r.end(), \
  "After " #EXPR);
// For Boolean tests:
#define B(EXPR) EXPR; print(br.begin(), br.end(), \
  "After " #EXPR);
 
// Boolean random generator:
struct BRand {
  bool operator()() { return rand() % 2 == 0; }
};
 
int main() {
  const int SZ = 10;
  const int MAX = 50;
  vector<int> x(SZ), y(SZ), r(SZ);
  // An integer random number generator:
  URandGen urg(MAX);
  srand(time(0));  // Randomize
  generate_n(x.begin(), SZ, urg);
  generate_n(y.begin(), SZ, urg);
  // Add one to each to guarantee nonzero divide:
  transform(y.begin(), y.end(), y.begin(),
    bind2nd(plus<int>(), 1));
  // Guarantee one pair of elements is ==:
  x[0] = y[0];
  print(x.begin(), x.end(), "x");
  print(y.begin(), y.end(), "y");
  // Operate on each element pair of x & y,
  // putting the result into r:
  T(testBinary(x, y, r, plus<int>()));
  T(testBinary(x, y, r, minus<int>()));
  T(testBinary(x, y, r, multiplies<int>()));
  T(testBinary(x, y, r, divides<int>()));
  T(testBinary(x, y, r, modulus<int>()));
  T(testUnary(x, r, negate<int>()));
  vector<bool> br(SZ); // For Boolean results
  B(testBinary(x, y, br, equal_to<int>()));
  B(testBinary(x, y, br, not_equal_to<int>()));
  B(testBinary(x, y, br, greater<int>()));
  B(testBinary(x, y, br, less<int>()));
  B(testBinary(x, y, br, greater_equal<int>()));
  B(testBinary(x, y, br, less_equal<int>()));
  B(testBinary(x, y, br,
not2(greater_equal<int>())));
  B(testBinary(x,y,br,not2(less_equal<int>())));
  vector<bool> b1(SZ), b2(SZ);
  generate_n(b1.begin(), SZ, BRand());
  generate_n(b2.begin(), SZ, BRand());
  print(b1.begin(), b1.end(), "b1");
  print(b2.begin(), b2.end(), "b2");
  B(testBinary(b1, b2, br, logical_and<int>()));
  B(testBinary(b1, b2, br, logical_or<int>()));
  B(testUnary(b1, br, logical_not<int>()));
  B(testUnary(b1, br, not1(logical_not<int>())));
} ///:~

This example uses a handy function template, print( ), which is capable of printing a sequence of any type along with an optional message. This template appears in the header file PrintSequence.h, and is explained later in this chapter.

The two template functions automate the process of testing the various function object templates. There are two because the function objects are either unary or binary. The testUnary( ) function takes a source vector, a destination vector, and a unary function object to apply to the source vector to produce the destination vector. In testBinary( ), two source vectors are fed to a binary function to produce the destination vector. In both cases, the template functions simply turn around and call the transform( ) algorithm, which applies the unary function or function object found in its fourth parameter to each sequence element, writing the result to the sequence indicated by its third parameter, which in this case is the same as the input sequence.

For each test, you want to see a string describing the test, followed by the results of the test. To automate this, the preprocessor comes in handy; the T( ) and B( ) macros each take the expression you want to execute. After evaluating the expression, they pass the appropriate range to print( ). To produce the message the expression is “stringized” using the preprocessor. That way you see the code of the expression that is executed followed by the result vector.

The last little tool, BRand, is a generator object that creates random bool values. To do this, it gets a random number from rand( ) and tests to see if it's greater than (RAND_MAX+1)/2. If the random numbers are evenly distributed, this should happen half the time.

In main( ), three vectors of int are created: x and y for source values, and r for results. To initialize x and y with random values no greater than 50, a generator of type URandGen from Generators.h is used. The standard generate_n( ) algorithm populates the sequence specified in its first argument by invoking its third argument (which must be a generator) a given number of times (specified in its second argument). Since there is one operation where elements of x are divided by elements of y, we must ensure that there are no zero values of y. This is accomplished by once again using the transform( ) algorithm, taking the source values from y and putting the results back into y. The function object for this is created with the expression:

 
Sélectionnez
bind2nd(plus<int>(), 1)

This expression uses the plus function object to add 1 to its first argument. As we did earlier in this chapter, we use a binder adaptor to make this a unary function so it can applied to the sequence by a single call to transform( ).

Another test in the program compares the elements in the two vectors for equality, so it is interesting to guarantee that at least one pair of elements is equivalent; here element zero is chosen.

Once the two vectors are printed, T( ) tests each of the function objects that produces a numeric value, and then B( ) tests each function object that produces a Boolean result. The result is placed into a vector<bool>, and when this vector is printed, it produces a ‘1' for a true value and a ‘0' for a false value. Here is the output from an execution of FunctionObjects.cpp:

 
Sélectionnez
x:
4 8 18 36 22 6 29 19 25 47
y:
4 14 23 9 11 32 13 15 44 30
After testBinary(x, y, r, plus<int>()):
8 22 41 45 33 38 42 34 69 77
After testBinary(x, y, r, minus<int>()):
0 -6 -5 27 11 -26 16 4 -19 17
After testBinary(x, y, r, multiplies<int>()):
16 112 414 324 242 192 377 285 1100 1410
After testBinary(x, y, r, divides<int>()):
1 0 0 4 2 0 2 1 0 1
After testBinary(x, y, r, limit<int>()):
0 8 18 0 0 6 3 4 25 17
After testUnary(x, r, negate<int>()):
-4 -8 -18 -36 -22 -6 -29 -19 -25 -47
After testBinary(x, y, br, equal_to<int>()):
1 0 0 0 0 0 0 0 0 0
After testBinary(x, y, br, not_equal_to<int>()):
0 1 1 1 1 1 1 1 1 1
After testBinary(x, y, br, greater<int>()):
0 0 0 1 1 0 1 1 0 1
After testBinary(x, y, br, less<int>()):
0 1 1 0 0 1 0 0 1 0
After testBinary(x, y, br, greater_equal<int>()):
1 0 0 1 1 0 1 1 0 1
After testBinary(x, y, br, less_equal<int>()):
1 1 1 0 0 1 0 0 1 0
After testBinary(x, y, br,
not2(greater_equal<int>())):
0 1 1 0 0 1 0 0 1 0
After testBinary(x,y,br,not2(less_equal<int>())):
0 0 0 1 1 0 1 1 0 1
b1:
0 1 1 0 0 0 1 0 1 1
b2:
0 1 1 0 0 0 1 0 1 1
After testBinary(b1, b2, br, logical_and<int>()):
0 1 1 0 0 0 1 0 1 1
After testBinary(b1, b2, br, logical_or<int>()):
0 1 1 0 0 0 1 0 1 1
After testUnary(b1, br, logical_not<int>()):
1 0 0 1 1 1 0 1 0 0
After testUnary(b1, br,
not1(logical_not<int>())):
0 1 1 0 0 0 1 0 1 1

If you want the Boolean values to display as “true” and “false” instead of 1 and 0, call cout.setf(ios::boolalpha).

A binder doesn't have to produce a unary predicate; it can also create any unary function (that is, a function that returns something other than bool). For example, you can to multiply every element in a vector by 10 using a binder with the transform( ) algorithm:

 
Sélectionnez
//: C06:FBinder.cpp
// Binders aren't limited to producing predicates.
//{L} Generators
#include <algorithm>
#include <cstdlib>
#include <ctime>
#include <functional>
#include <iostream>
#include <iterator>
#include <vector>
#include "Generators.h"
using namespace std;
 
int main() {
  ostream_iterator<int> out(cout," ");
  vector<int> v(15);
  srand(time(0));  // Randomize
  generate(v.begin(), v.end(), URandGen(20));
  copy(v.begin(), v.end(), out);
  transform(v.begin(), v.end(),
v.begin(),
 
bind2nd(multiplies<int>(), 10));
  copy(v.begin(), v.end(), out);
} ///:~

Since the third argument to transform( ) is the same as the first, the resulting elements are copied back into the source vector. The function object created by bind2nd( ) in this case produces an int result.

The “bound” argument to a binder cannot be a function object, but it does not have to be a compile-time constant. For example:

 
Sélectionnez
//: C06:BinderValue.cpp
// The bound argument can vary.
#include <algorithm>
#include <functional>
#include <iostream>
#include <iterator>
#include <cstdlib>
using namespace std;
 
int boundedRand() { return rand() % 100; }
 
int main() {
  const int SZ = 20;
  int a[SZ], b[SZ] = {0};
  generate(a, a + SZ, boundedRand);
  int val = boundedRand();
  int* end = remove_copy_if(a, a + SZ, b,
 
bind2nd(greater<int>(), val));
  // Sort for easier viewing:
  sort(a, a + SZ);
  sort(b, end);
  ostream_iterator<int> out(cout, " ");
  cout << "Original Sequence:” << endl;
  copy(a, a + SZ, out); cout << endl;
  cout << "Values <= " << val
<< endl;
  copy(b, end, out); cout << endl;
} ///:~

Here, an array is filled with 20 random numbers between 0 and 100, and the user provides a value on the command line. In the remove_copy_if( ) call, you can see that the bound argument to bind2nd( ) is random number in the same range as the sequence. Here is the output from one run:

 
Sélectionnez
Original Sequence:
4 12 15 17 19 21 26 30 47 48 56 58 60 63 71 79 82 90 92
95
Values <= 41
4 12 15 17 19 21 26 30
2-4-2-5. Function pointer adaptors

Wherever a function-like entity is expected by an algorithm, you can supply either a pointer to an ordinary function or a function object. When the algorithm issues a call, if it is through a function pointer, than the native function-call mechanism is used. If it is through a function object, then that object's operator( ) member executes. In CopyInts2.cpp, we passed the raw function gt15( ) as a predicate to remove_copy_if( ). We also passed pointers to functions returning random numbers to generate( ) and generate_n( ).

You cannot use raw functions with function object adaptors such as bind2nd( ) because they assume the existence of type definitions for the argument and result types. Instead of manually converting your native functions into function objects yourself, the standard library provides a family of adaptors to do the work for you. The ptr_fun( ) adaptors take a pointer to a function and turn it into a function object. They are not designed for a function that takes no arguments—they must only be used with unary functions or binary functions.

The following program uses ptr_fun( ) to wrap a unary function:

 
Sélectionnez
//: C06:PtrFun1.cpp
// Using ptr_fun() with a unary function.
#include <algorithm>
#include <cmath>
#include <functional>
#include <iostream>
#include <iterator>
#include <vector>
using namespace std;
 
int d[] = { 123, 94, 10, 314, 315 };
const int DSZ = sizeof d / sizeof *d;
 
bool isEven(int x) { return x % 2 == 0; }
 
int main() {
  vector<bool> vb;
  transform(d, d + DSZ, back_inserter(vb),
    not1(ptr_fun(isEven)));
  copy(vb.begin(), vb.end(),
    ostream_iterator<bool>(cout, " "));
  cout << endl;
  // Output: 1 0 0 0 1
} ///:~

We can't simply pass isEven to not1, because not1 needs to know the actual argument type and return type its argument uses. The ptr_fun( ) adaptor deduces those types through template argument deduction. The definition of the unary version of ptr_fun( ) looks something like this:

 
Sélectionnez
template<class Arg, class Result>
pointer_to_unary_function<Arg, Result>
ptr_fun(Result (*fptr)(Arg)) {
  return
pointer_to_unary_function<Arg, Result>(fptr);
}

As you can see, this version of ptr_fun( ) deduces the argument and result types from fptr and uses them to initialize a pointer_to_unary_function object that stores fptr. The function call operator for pointer_to_unary_function just calls fptr, as you can see by the last line of its code:

 
Sélectionnez
template<class Arg, class Result>
class pointer_to_unary_function
: public unary_function<Arg, Result> {
  Result (*fptr)(Arg); // Stores the f-ptr
public:
  pointer_to_unary_function(Result (*x)(Arg)) : fptr(x)
{}
  Result operator()(Arg x) const { return fptr(x); }
};

Since pointer_to_unary_function derives from unary_function, the appropriate type definitions come along for the ride and are available to not1.

There is also a binary version of ptr_fun( ), which returns a pointer_to_binary_function object (which derives from binary_function) that behaves analogously to the unary case. The following program uses the binary version of ptr_fun( ) to raise numbers in a sequence to a power. It also reveals a pitfall when passing overloaded functions to ptr_fun( ).

 
Sélectionnez
//: C06:PtrFun2.cpp {-edg}
// Using ptr_fun() for a binary function.
#include <algorithm>
#include <cmath>
#include <functional>
#include <iostream>
#include <iterator>
#include <vector>
using namespace std;
 
double d[] = { 01.23, 91.370, 56.661,
  023.230, 19.959, 1.0, 3.14159 };
const int DSZ = sizeof d / sizeof *d;
 
int main() {
  vector<double> vd;
  transform(d, d + DSZ, back_inserter(vd),
    bind2nd(ptr_fun<double, double, double>(pow),
2.0));
  copy(vd.begin(), vd.end(),
    ostream_iterator<double>(cout, "
"));
  cout << endl;
} ///:~

The pow( ) function is overloaded in the Standard C++ header <cmath> for each of the floating-point data types, as follows:

 
Sélectionnez
float pow(float, int);  // Efficient int power versions
...
double pow(double, int);
long double pow(long double, int);
float pow(float, float);
double pow(double, double);
long double pow(long double, long double);

Since there are multiple versions of pow( ), the compiler has no way of knowing which to choose. Here, we have to help the compiler by using explicit function template specialization, as explained in the previous chapter.(91)

It's even trickier to convert a member function into a function object suitable for using with the generic algorithms. As a simple example, suppose we have the classical “shape” problem and want to apply the draw( ) member function to each pointer in a container of Shape:

 
Sélectionnez
//: C06:MemFun1.cpp
// Applying pointers to member functions.
#include <algorithm>
#include <functional>
#include <iostream>
#include <vector>
#include "../purge.h"
using namespace std;
 
class Shape {
public:
  virtual void draw() = 0;
  virtual ~Shape() {}
};
 
class Circle : public Shape {
public:
  virtual void draw() { cout <<
"Circle::Draw()" << endl; }
  ~Circle() { cout <<
"Circle::~Circle()" << endl; }
};
 
class Square : public Shape {
public:
  virtual void draw() { cout << "Square::Draw()"
<< endl; }
  ~Square() { cout <<
"Square::~Square()" << endl; }
};
 
int main() {
  vector<Shape*> vs;
  vs.push_back(new Circle);
  vs.push_back(new Square);
  for_each(vs.begin(), vs.end(),
mem_fun(&Shape::draw));
  purge(vs);
} ///:~

The for_each( ) algorithm passes each element in a sequence to the function object denoted by its third argument. Here, we want the function object to wrap one of the member functions of the class itself, and so the function object's “argument” becomes the pointer to the object for the member function call. To produce such a function object, the mem_fun( ) template takes a pointer to a member as its argument.

The mem_fun( ) functions are for producing function objects that are called using a pointer to the object that the member function is called for, while mem_fun_ref( ) calls the member function directly for an object. One set of overloads of both mem_fun( ) and mem_fun_ref( ) is for member functions that take zero arguments and one argument, and this is multiplied by two to handle const vs. non-const member functions. However, templates and overloading take care of sorting all that out—all you need to remember is when to use mem_fun( ) vs. mem_fun_ref( ).

Suppose you have a container of objects (not pointers), and you want to call a member function that takes an argument. The argument you pass should come from a second container of objects. To accomplish this, use the second overloaded form of the transform( ) algorithm:

 
Sélectionnez
//: C06:MemFun2.cpp
// Calling member functions through an object reference.
#include <algorithm>
#include <functional>
#include <iostream>
#include <iterator>
#include <vector>
using namespace std;
 
class Angle {
  int degrees;
public:
  Angle(int deg) : degrees(deg) {}
  int mul(int times) { return degrees *= times; }
};
 
int main() {
  vector<Angle> va;
  for(int i = 0; i < 50; i += 10)
    va.push_back(Angle(i));
  int x[] = { 1, 2, 3, 4, 5 };
  transform(va.begin(), va.end(), x,
    ostream_iterator<int>(cout, " "),
    mem_fun_ref(&Angle::mul));
  cout << endl;
  // Output: 0 20 60 120 200
} ///:~

Because the container is holding objects, mem_fun_ref( ) must be used with the pointer-to-member function. This version of transform( ) takes the start and end point of the first range (where the objects live); the starting point of the second range, which holds the arguments to the member function; the destination iterator, which in this case is standard output; and the function object to call for each object. This function object is created with mem_fun_ref( ) and the desired pointer to member. Notice that the transform( ) and for_each( ) template functions are incomplete; transform( ) requires that the function it calls return a value, and there is no for_each( ) that passes two arguments to the function it calls. Thus, you cannot call a member function that returns void and takes an argument using transform( ) or for_each( ).

Most any member function works with mem_fun_ref( ). You can also use standard library member functions, if your compiler doesn't add any default arguments beyond the normal arguments specified in the standard.(92) For example, suppose you'd like to read a file and search for blank lines. Your compiler may allow you to use the string::empty( ) member function like this:

 
Sélectionnez
//: C06:FindBlanks.cpp
// Demonstrates mem_fun_ref() with string::empty().
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <fstream>
#include <functional>
#include <string>
#include <vector>
#include