xref: /openbsd-src/gnu/lib/libstdc++/libstdc++/docs/html/27_io/howto.html (revision 03a78d155d6fff5698289342b62759a75b20d130)
1<?xml version="1.0" encoding="ISO-8859-1"?>
2<!DOCTYPE html
3          PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
4          "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
5
6<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
7<head>
8   <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1" />
9   <meta name="AUTHOR" content="pme@gcc.gnu.org (Phil Edwards)" />
10   <meta name="KEYWORDS" content="HOWTO, libstdc++, GCC, g++, libg++, STL" />
11   <meta name="DESCRIPTION" content="HOWTO for the libstdc++ chapter 27." />
12   <meta name="GENERATOR" content="vi and eight fingers" />
13   <title>libstdc++-v3 HOWTO:  Chapter 27</title>
14<link rel="StyleSheet" href="../lib3styles.css" />
15</head>
16<body>
17
18<h1 class="centered"><a name="top">Chapter 27:  Input/Output</a></h1>
19
20<p>Chapter 27 deals with iostreams and all their subcomponents
21   and extensions.  All <em>kinds</em> of fun stuff.
22</p>
23
24
25<!-- ####################################################### -->
26<hr />
27<h1>Contents</h1>
28<ul>
29   <li><a href="#1">Copying a file</a></li>
30   <li><a href="#2">The buffering is screwing up my program!</a></li>
31   <li><a href="#3">Binary I/O</a></li>
32   <li><a href="#5">What is this &lt;sstream&gt;/stringstreams thing?</a></li>
33   <li><a href="#6">Deriving a stream buffer</a></li>
34   <li><a href="#7">More on binary I/O</a></li>
35   <li><a href="#8">Pathetic performance?  Ditch C.</a></li>
36   <li><a href="#9">Threads and I/O</a></li>
37   <li><a href="#10">Which header?</a></li>
38   <li><a href="#11">Using FILE*s and file descriptors with IOStreams</a></li>
39</ul>
40
41<hr />
42
43<!-- ####################################################### -->
44
45<h2><a name="1">Copying a file</a></h2>
46   <p>So you want to copy a file quickly and easily, and most important,
47      completely portably.  And since this is C++, you have an open
48      ifstream (call it IN) and an open ofstream (call it OUT):
49   </p>
50   <pre>
51   #include &lt;fstream&gt;
52
53   std::ifstream  IN ("input_file");
54   std::ofstream  OUT ("output_file"); </pre>
55   <p>Here's the easiest way to get it completely wrong:
56   </p>
57   <pre>
58   OUT &lt;&lt; IN;</pre>
59   <p>For those of you who don't already know why this doesn't work
60      (probably from having done it before), I invite you to quickly
61      create a simple text file called &quot;input_file&quot; containing
62      the sentence
63   </p>
64      <pre>
65      The quick brown fox jumped over the lazy dog.</pre>
66   <p>surrounded by blank lines.  Code it up and try it.  The contents
67      of &quot;output_file&quot; may surprise you.
68   </p>
69   <p>Seriously, go do it.  Get surprised, then come back.  It's worth it.
70   </p>
71   <hr width="60%" />
72   <p>The thing to remember is that the <code>basic_[io]stream</code> classes
73      handle formatting, nothing else.  In particular, they break up on
74      whitespace.  The actual reading, writing, and storing of data is
75      handled by the <code>basic_streambuf</code> family.  Fortunately, the
76      <code>operator&lt;&lt;</code> is overloaded to take an ostream and
77      a pointer-to-streambuf, in order to help with just this kind of
78      &quot;dump the data verbatim&quot; situation.
79   </p>
80   <p>Why a <em>pointer</em> to streambuf and not just a streambuf?  Well,
81      the [io]streams hold pointers (or references, depending on the
82      implementation) to their buffers, not the actual
83      buffers.  This allows polymorphic behavior on the part of the buffers
84      as well as the streams themselves.  The pointer is easily retrieved
85      using the <code>rdbuf()</code> member function.  Therefore, the easiest
86      way to copy the file is:
87   </p>
88   <pre>
89   OUT &lt;&lt; IN.rdbuf();</pre>
90   <p>So what <em>was</em> happening with OUT&lt;&lt;IN?  Undefined
91      behavior, since that particular &lt;&lt; isn't defined by the Standard.
92      I have seen instances where it is implemented, but the character
93      extraction process removes all the whitespace, leaving you with no
94      blank lines and only &quot;Thequickbrownfox...&quot;.  With
95      libraries that do not define that operator, IN (or one of IN's
96      member pointers) sometimes gets converted to a void*, and the output
97      file then contains a perfect text representation of a hexidecimal
98      address (quite a big surprise).  Others don't compile at all.
99   </p>
100   <p>Also note that none of this is specific to o<b>*f*</b>streams.
101      The operators shown above are all defined in the parent
102      basic_ostream class and are therefore available with all possible
103      descendents.
104   </p>
105   <p>Return <a href="#top">to top of page</a> or
106      <a href="../faq/index.html">to the FAQ</a>.
107   </p>
108
109<hr />
110<h2><a name="2">The buffering is screwing up my program!</a></h2>
111<!--
112  This is not written very well.  I need to redo this section.
113-->
114   <p>First, are you sure that you understand buffering?  Particularly
115      the fact that C++ may not, in fact, have anything to do with it?
116   </p>
117   <p>The rules for buffering can be a little odd, but they aren't any
118      different from those of C.  (Maybe that's why they can be a bit
119      odd.)  Many people think that writing a newline to an output
120      stream automatically flushes the output buffer.  This is true only
121      when the output stream is, in fact, a terminal and not a file
122      or some other device -- and <em>that</em> may not even be true
123      since C++ says nothing about files nor terminals.  All of that is
124      system-dependent.  (The &quot;newline-buffer-flushing only occurring
125      on terminals&quot; thing is mostly true on Unix systems, though.)
126   </p>
127   <p>Some people also believe that sending <code>endl</code> down an
128      output stream only writes a newline.  This is incorrect; after a
129      newline is written, the buffer is also flushed.  Perhaps this
130      is the effect you want when writing to a screen -- get the text
131      out as soon as possible, etc -- but the buffering is largely
132      wasted when doing this to a file:
133   </p>
134   <pre>
135   output &lt;&lt; &quot;a line of text&quot; &lt;&lt; endl;
136   output &lt;&lt; some_data_variable &lt;&lt; endl;
137   output &lt;&lt; &quot;another line of text&quot; &lt;&lt; endl; </pre>
138   <p>The proper thing to do in this case to just write the data out
139      and let the libraries and the system worry about the buffering.
140      If you need a newline, just write a newline:
141   </p>
142   <pre>
143   output &lt;&lt; &quot;a line of text\n&quot;
144          &lt;&lt; some_data_variable &lt;&lt; '\n'
145          &lt;&lt; &quot;another line of text\n&quot;; </pre>
146   <p>I have also joined the output statements into a single statement.
147      You could make the code prettier by moving the single newline to
148      the start of the quoted text on the thing line, for example.
149   </p>
150   <p>If you do need to flush the buffer above, you can send an
151      <code>endl</code> if you also need a newline, or just flush the buffer
152      yourself:
153   </p>
154   <pre>
155   output &lt;&lt; ...... &lt;&lt; flush;    // can use std::flush manipulator
156   output.flush();               // or call a member fn </pre>
157   <p>On the other hand, there are times when writing to a file should
158      be like writing to standard error; no buffering should be done
159      because the data needs to appear quickly (a prime example is a
160      log file for security-related information).  The way to do this is
161      just to turn off the buffering <em>before any I/O operations at
162      all</em> have been done (note that opening counts as an I/O operation):
163   </p>
164   <pre>
165   std::ofstream    os;
166   std::ifstream    is;
167   int   i;
168
169   os.rdbuf()-&gt;pubsetbuf(0,0);
170   is.rdbuf()-&gt;pubsetbuf(0,0);
171
172   os.open(&quot;/foo/bar/baz&quot;);
173   is.open(&quot;/qux/quux/quuux&quot;);
174   ...
175   os &lt;&lt; &quot;this data is written immediately\n&quot;;
176   is &gt;&gt; i;   // and this will probably cause a disk read </pre>
177   <p>Since all aspects of buffering are handled by a streambuf-derived
178      member, it is necessary to get at that member with <code>rdbuf()</code>.
179      Then the public version of <code>setbuf</code> can be called.  The
180      arguments are the same as those for the Standard C I/O Library
181      function (a buffer area followed by its size).
182   </p>
183   <p>A great deal of this is implementation-dependent.  For example,
184      <code>streambuf</code> does not specify any actions for its own
185      <code>setbuf()</code>-ish functions; the classes derived from
186      <code>streambuf</code> each define behavior that &quot;makes
187      sense&quot; for that class:  an argument of (0,0) turns off buffering
188      for <code>filebuf</code> but has undefined behavior for its sibling
189      <code>stringbuf</code>, and specifying anything other than (0,0) has
190      varying effects.  Other user-defined class derived from streambuf can
191      do whatever they want.  (For <code>filebuf</code> and arguments for
192      <code>(p,s)</code> other than zeros, libstdc++ does what you'd expect:
193      the first <code>s</code> bytes of <code>p</code> are used as a buffer,
194      which you must allocate and deallocate.)
195   </p>
196   <p>A last reminder:  there are usually more buffers involved than
197      just those at the language/library level.  Kernel buffers, disk
198      buffers, and the like will also have an effect.  Inspecting and
199      changing those are system-dependent.
200   </p>
201   <p>Return <a href="#top">to top of page</a> or
202      <a href="../faq/index.html">to the FAQ</a>.
203   </p>
204
205<hr />
206<h2><a name="3">Binary I/O</a></h2>
207   <p>The first and most important thing to remember about binary I/O is
208      that opening a file with <code>ios::binary</code> is not, repeat
209      <em>not</em>, the only thing you have to do.  It is not a silver
210      bullet, and will not allow you to use the <code>&lt;&lt;/&gt;&gt;</code>
211      operators of the normal fstreams to do binary I/O.
212   </p>
213   <p>Sorry.  Them's the breaks.
214   </p>
215   <p>This isn't going to try and be a complete tutorial on reading and
216      writing binary files (because &quot;binary&quot;
217      <a href="#7">covers a lot of ground)</a>, but we will try and clear
218      up a couple of misconceptions and common errors.
219   </p>
220   <p>First, <code>ios::binary</code> has exactly one defined effect, no more
221      and no less.  Normal text mode has to be concerned with the newline
222      characters, and the runtime system will translate between (for
223      example) '\n' and the appropriate end-of-line sequence (LF on Unix,
224      CRLF on DOS, CR on Macintosh, etc).  (There are other things that
225      normal mode does, but that's the most obvious.)  Opening a file in
226      binary mode disables this conversion, so reading a CRLF sequence
227      under Windows won't accidentally get mapped to a '\n' character, etc.
228      Binary mode is not supposed to suddenly give you a bitstream, and
229      if it is doing so in your program then you've discovered a bug in
230      your vendor's compiler (or some other part of the C++ implementation,
231      possibly the runtime system).
232   </p>
233   <p>Second, using <code>&lt;&lt;</code> to write and <code>&gt;&gt;</code> to
234      read isn't going to work with the standard file stream classes, even
235      if you use <code>skipws</code> during reading.  Why not?  Because
236      ifstream and ofstream exist for the purpose of <em>formatting</em>,
237      not reading and writing.  Their job is to interpret the data into
238      text characters, and that's exactly what you don't want to happen
239      during binary I/O.
240   </p>
241   <p>Third, using the <code>get()</code> and <code>put()/write()</code> member
242      functions still aren't guaranteed to help you.  These are
243      &quot;unformatted&quot; I/O functions, but still character-based.
244      (This may or may not be what you want, see below.)
245   </p>
246   <p>Notice how all the problems here are due to the inappropriate use
247      of <em>formatting</em> functions and classes to perform something
248      which <em>requires</em> that formatting not be done?  There are a
249      seemingly infinite number of solutions, and a few are listed here:
250   </p>
251   <ul>
252      <li>&quot;Derive your own fstream-type classes and write your own
253          &lt;&lt;/&gt;&gt; operators to do binary I/O on whatever data
254          types you're using.&quot;  This is a Bad Thing, because while
255          the compiler would probably be just fine with it, other humans
256          are going to be confused.  The overloaded bitshift operators
257          have a well-defined meaning (formatting), and this breaks it.
258      </li>
259      <li>&quot;Build the file structure in memory, then <code>mmap()</code>
260          the file and copy the structure.&quot;  Well, this is easy to
261          make work, and easy to break, and is pretty equivalent to
262          using <code>::read()</code> and <code>::write()</code> directly, and
263          makes no use of the iostream library at all...
264      </li>
265      <li>&quot;Use streambufs, that's what they're there for.&quot;
266          While not trivial for the beginner, this is the best of all
267          solutions.  The streambuf/filebuf layer is the layer that is
268          responsible for actual I/O.  If you want to use the C++
269          library for binary I/O, this is where you start.
270      </li>
271   </ul>
272   <p>How to go about using streambufs is a bit beyond the scope of this
273      document (at least for now), but while streambufs go a long way,
274      they still leave a couple of things up to you, the programmer.
275      As an example, byte ordering is completely between you and the
276      operating system, and you have to handle it yourself.
277   </p>
278   <p>Deriving a streambuf or filebuf
279      class from the standard ones, one that is specific to your data
280      types (or an abstraction thereof) is probably a good idea, and
281      lots of examples exist in journals and on Usenet.  Using the
282      standard filebufs directly (either by declaring your own or by
283      using the pointer returned from an fstream's <code>rdbuf()</code>)
284      is certainly feasible as well.
285   </p>
286   <p>One area that causes problems is trying to do bit-by-bit operations
287      with filebufs.  C++ is no different from C in this respect:  I/O
288      must be done at the byte level.  If you're trying to read or write
289      a few bits at a time, you're going about it the wrong way.  You
290      must read/write an integral number of bytes and then process the
291      bytes.  (For example, the streambuf functions take and return
292      variables of type <code>int_type</code>.)
293   </p>
294   <p>Another area of problems is opening text files in binary mode.
295      Generally, binary mode is intended for binary files, and opening
296      text files in binary mode means that you now have to deal with all of
297      those end-of-line and end-of-file problems that we mentioned before.
298      An instructive thread from comp.lang.c++.moderated delved off into
299      this topic starting more or less at
300      <a href="http://www.deja.com/getdoc.xp?AN=436187505">this</a>
301      article and continuing to the end of the thread.  (You'll have to
302      sort through some flames every couple of paragraphs, but the points
303      made are good ones.)
304   </p>
305
306<hr />
307<h2><a name="5">What is this &lt;sstream&gt;/stringstreams thing?</a></h2>
308   <p>Stringstreams (defined in the header <code>&lt;sstream&gt;</code>)
309      are in this author's opinion one of the coolest things since
310      sliced time.  An example of their use is in the Received Wisdom
311      section for Chapter 21 (Strings),
312      <a href="../21_strings/howto.html#1.1internal"> describing how to
313      format strings</a>.
314   </p>
315   <p>The quick definition is:  they are siblings of ifstream and ofstream,
316      and they do for <code>std::string</code> what their siblings do for
317      files.  All that work you put into writing <code>&lt;&lt;</code> and
318      <code>&gt;&gt;</code> functions for your classes now pays off
319      <em>again!</em>  Need to format a string before passing the string
320      to a function?  Send your stuff via <code>&lt;&lt;</code> to an
321      ostringstream.  You've read a string as input and need to parse it?
322      Initialize an istringstream with that string, and then pull pieces
323      out of it with <code>&gt;&gt;</code>.  Have a stringstream and need to
324      get a copy of the string inside?  Just call the <code>str()</code>
325      member function.
326   </p>
327   <p>This only works if you've written your
328      <code>&lt;&lt;</code>/<code>&gt;&gt;</code> functions correctly, though,
329      and correctly means that they take istreams and ostreams as
330      parameters, not i<b>f</b>streams and o<b>f</b>streams.  If they
331      take the latter, then your I/O operators will work fine with
332      file streams, but with nothing else -- including stringstreams.
333   </p>
334   <p>If you are a user of the strstream classes, you need to update
335      your code.  You don't have to explicitly append <code>ends</code> to
336      terminate the C-style character array, you don't have to mess with
337      &quot;freezing&quot; functions, and you don't have to manage the
338      memory yourself.  The strstreams have been officially deprecated,
339      which means that 1) future revisions of the C++ Standard won't
340      support them, and 2) if you use them, people will laugh at you.
341   </p>
342
343<hr />
344<h2><a name="6">Deriving a stream buffer</a></h2>
345   <p>Creating your own stream buffers for I/O can be remarkably easy.
346      If you are interested in doing so, we highly recommend two very
347      excellent books:
348      <a href="http://home.camelot.de/langer/iostreams.htm">Standard C++
349      IOStreams and Locales</a> by Langer and Kreft, ISBN 0-201-18395-1, and
350      <a href="http://www.josuttis.com/libbook/">The C++ Standard Library</a>
351      by Nicolai Josuttis, ISBN 0-201-37926-0.  Both are published by
352      Addison-Wesley, who isn't paying us a cent for saying that, honest.
353   </p>
354   <p>Here is a simple example, io/outbuf1, from the Josuttis text.  It
355      transforms everything sent through it to uppercase.  This version
356      assumes many things about the nature of the character type being
357      used (for more information, read the books or the newsgroups):
358   </p>
359   <pre>
360    #include &lt;iostream&gt;
361    #include &lt;streambuf&gt;
362    #include &lt;locale&gt;
363    #include &lt;cstdio&gt;
364
365    class outbuf : public std::streambuf
366    {
367      protected:
368	/* central output function
369	 * - print characters in uppercase mode
370	 */
371	virtual int_type overflow (int_type c) {
372	    if (c != EOF) {
373		// convert lowercase to uppercase
374		c = std::toupper(static_cast&lt;char&gt;(c),getloc());
375
376		// and write the character to the standard output
377		if (putchar(c) == EOF) {
378		    return EOF;
379		}
380	    }
381	    return c;
382	}
383    };
384
385    int main()
386    {
387	// create special output buffer
388	outbuf ob;
389	// initialize output stream with that output buffer
390	std::ostream out(&amp;ob);
391
392	out &lt;&lt; "31 hexadecimal: "
393	    &lt;&lt; std::hex &lt;&lt; 31 &lt;&lt; std::endl;
394	return 0;
395    }
396   </pre>
397   <p>Try it yourself!  More examples can be found in 3.1.x code, in
398      <code>include/ext/*_filebuf.h</code>, and on
399      <a href="http://www.informatik.uni-konstanz.de/~kuehl/c++/iostream/">Dietmar
400      K&uuml;hl's IOStreams page</a>.
401   </p>
402
403<hr />
404<h2><a name="7">More on binary I/O</a></h2>
405   <p>Towards the beginning of February 2001, the subject of
406      &quot;binary&quot; I/O was brought up in a couple of places at the
407      same time.  One notable place was Usenet, where James Kanze and
408      Dietmar K&uuml;hl separately posted articles on why attempting
409      generic binary I/O was not a good idea.  (Here are copies of
410      <a href="binary_iostreams_kanze.txt">Kanze's article</a> and
411      <a href="binary_iostreams_kuehl.txt">K&uuml;hl's article</a>.)
412   </p>
413   <p>Briefly, the problems of byte ordering and type sizes mean that
414      the unformatted functions like <code>ostream::put()</code> and
415      <code>istream::get()</code> cannot safely be used to communicate
416      between arbitrary programs, or across a network, or from one
417      invocation of a program to another invocation of the same program
418      on a different platform, etc.
419   </p>
420   <p>The entire Usenet thread is instructive, and took place under the
421      subject heading &quot;binary iostreams&quot; on both comp.std.c++
422      and comp.lang.c++.moderated in parallel.  Also in that thread,
423      Dietmar K&uuml;hl mentioned that he had written a pair of stream
424      classes that would read and write XDR, which is a good step towards
425      a portable binary format.
426   </p>
427
428<hr />
429<h2><a name="8">Pathetic performance?  Ditch C.</a></h2>
430   <p>It sounds like a flame on C, but it isn't.  Really.  Calm down.
431      I'm just saying it to get your attention.
432   </p>
433   <p>Because the C++ library includes the C library, both C-style and
434      C++-style I/O have to work at the same time.  For example:
435   </p>
436   <pre>
437     #include &lt;iostream&gt;
438     #include &lt;cstdio&gt;
439
440     std::cout &lt;&lt; &quot;Hel&quot;;
441     std::printf (&quot;lo, worl&quot;);
442     std::cout &lt;&lt; &quot;d!\n&quot;;
443   </pre>
444   <p>This must do what you think it does.
445   </p>
446   <p>Alert members of the audience will immediately notice that buffering
447      is going to make a hash of the output unless special steps are taken.
448   </p>
449   <p>The special steps taken by libstdc++, at least for version 3.0,
450      involve doing very little buffering for the standard streams, leaving
451      most of the buffering to the underlying C library.  (This kind of
452      thing is <a href="../explanations.html#cstdio">tricky to get right</a>.)
453      The upside is that correctness is ensured.  The downside is that
454      writing through <code>cout</code> can quite easily lead to awful
455      performance when the C++ I/O library is layered on top of the C I/O
456      library (as it is for 3.0 by default).  Some patches have been applied
457      which improve the situation for 3.1.
458   </p>
459   <p>However, the C and C++ standard streams only need to be kept in sync
460      when both libraries' facilities are in use.  If your program only uses
461      C++ I/O, then there's no need to sync with the C streams.  The right
462      thing to do in this case is to call
463   </p>
464   <pre>
465     #include <em>any of the I/O headers such as ios, iostream, etc</em>
466
467     std::ios::sync_with_stdio(false);
468   </pre>
469   <p>You must do this before performing any I/O via the C++ stream objects.
470      Once you call this, the C++ streams will operate independently of the
471      (unused) C streams.  For GCC 3.x, this means that <code>cout</code> and
472      company will become fully buffered on their own.
473   </p>
474   <p>Note, by the way, that the synchronization requirement only applies to
475      the standard streams (<code>cin</code>, <code>cout</code>,
476      <code>cerr</code>,
477      <code>clog</code>, and their wide-character counterparts).  File stream
478      objects that you declare yourself have no such requirement and are fully
479      buffered.
480   </p>
481
482<hr />
483<h2><a name="9">Threads and I/O</a></h2>
484   <p>I'll assume that you have already read the
485      <a href="../17_intro/howto.html#3">general notes on library threads</a>,
486      and the
487      <a href="../23_containers/howto.html#3">notes on threaded container
488      access</a> (you might not think of an I/O stream as a container, but
489      the points made there also hold here).  If you have not read them,
490      please do so first.
491   </p>
492   <p>This gets a bit tricky.  Please read carefully, and bear with me.
493   </p>
494   <h3>Structure</h3>
495   <p>As described <a href="../explanations.html#cstdio">here</a>, a wrapper
496      type called <code>__basic_file</code> provides our abstraction layer
497      for the <code>std::filebuf</code> classes.  Nearly all decisions dealing
498      with actual input and output must be made in <code>__basic_file</code>.
499   </p>
500   <p>A generic locking mechanism is somewhat in place at the filebuf layer,
501      but is not used in the current code.  Providing locking at any higher
502      level is akin to providing locking within containers, and is not done
503      for the same reasons (see the links above).
504   </p>
505   <h3>The defaults for 3.0.x</h3>
506   <p>The __basic_file type is simply a collection of small wrappers around
507      the C stdio layer (again, see the link under Structure).  We do no
508      locking ourselves, but simply pass through to calls to <code>fopen</code>,
509      <code>fwrite</code>, and so forth.
510   </p>
511   <p>So, for 3.0, the question of &quot;is multithreading safe for I/O&quot;
512      must be answered with, &quot;is your platform's C library threadsafe
513      for I/O?&quot;  Some are by default, some are not; many offer multiple
514      implementations of the C library with varying tradeoffs of threadsafety
515      and efficiency.  You, the programmer, are always required to take care
516      with multiple threads.
517   </p>
518   <p>(As an example, the POSIX standard requires that C stdio FILE*
519       operations are atomic.  POSIX-conforming C libraries (e.g, on Solaris
520       and GNU/Linux) have an internal mutex to serialize operations on
521       FILE*s.  However, you still need to not do stupid things like calling
522       <code>fclose(fs)</code> in one thread followed by an access of
523       <code>fs</code> in another.)
524   </p>
525   <p>So, if your platform's C library is threadsafe, then your
526      <code>fstream</code> I/O operations will be threadsafe at the lowest
527      level.  For higher-level operations, such as manipulating the data
528      contained in the stream formatting classes (e.g., setting up callbacks
529      inside an <code>std::ofstream</code>), you need to guard such accesses
530      like any other critical shared resource.
531   </p>
532   <h3>The future</h3>
533   <p>As already mentioned <a href="../explanations.html#cstdio">here</a>, a
534      second choice is available for I/O implementations:  libio.  This is
535      disabled by default, and in fact will not currently work due to other
536      issues.  It will be revisited, however.
537   </p>
538   <p>The libio code is a subset of the guts of the GNU libc (glibc) I/O
539      implementation.  When libio is in use, the <code>__basic_file</code>
540      type is basically derived from FILE.  (The real situation is more
541      complex than that... it's derived from an internal type used to
542      implement FILE.  See libio/libioP.h to see scary things done with
543      vtbls.)  The result is that there is no &quot;layer&quot; of C stdio
544      to go through; the filebuf makes calls directly into the same
545      functions used to implement <code>fread</code>, <code>fwrite</code>,
546      and so forth, using internal data structures.  (And when I say
547      &quot;makes calls directly,&quot; I mean the function is literally
548      replaced by a jump into an internal function.  Fast but frightening.
549      *grin*)
550   </p>
551   <p>Also, the libio internal locks are used.  This requires pulling in
552      large chunks of glibc, such as a pthreads implementation, and is one
553      of the issues preventing widespread use of libio as the libstdc++
554      cstdio implementation.
555   </p>
556   <p>But we plan to make this work, at least as an option if not a future
557      default.  Platforms running a copy of glibc with a recent-enough
558      version will see calls from libstdc++ directly into the glibc already
559      installed.  For other platforms, a copy of the libio subsection will
560      be built and included in libstdc++.
561   </p>
562   <h3>Alternatives</h3>
563   <p>Don't forget that other cstdio implemenations are possible.  You could
564      easily write one to perform your own forms of locking, to solve your
565      &quot;interesting&quot; problems.
566   </p>
567
568<hr />
569<h2><a name="10">Which header?</a></h2>
570   <p>To minimize the time you have to wait on the compiler, it's good to
571      only include the headers you really need.  Many people simply include
572      &lt;iostream&gt; when they don't need to -- and that can <em>penalize
573      your runtime as well.</em>  Here are some tips on which header to use
574      for which situations, starting with the simplest.
575   </p>
576   <p><strong>&lt;iosfwd&gt;</strong> should be included whenever you simply
577      need the <em>name</em> of an I/O-related class, such as
578      &quot;ofstream&quot; or &quot;basic_streambuf&quot;.  Like the name
579      implies, these are forward declarations.  (A word to all you fellow
580      old school programmers:  trying to forward declare classes like
581      &quot;class istream;&quot; won't work.  Look in the iosfwd header if
582      you'd like to know why.)  For example,
583   </p>
584   <pre>
585    #include &lt;iosfwd&gt;
586
587    class MyClass
588    {
589        ....
590        std::ifstream   input_file;
591    };
592
593    extern std::ostream&amp; operator&lt;&lt; (std::ostream&amp;, MyClass&amp;);
594   </pre>
595   <p><strong>&lt;ios&gt;</strong> declares the base classes for the entire
596      I/O stream hierarchy, std::ios_base and std::basic_ios&lt;charT&gt;, the
597      counting types std::streamoff and std::streamsize, the file
598      positioning type std::fpos, and the various manipulators like
599      std::hex, std::fixed, std::noshowbase, and so forth.
600   </p>
601   <p>The ios_base class is what holds the format flags, the state flags,
602      and the functions which change them (setf(), width(), precision(),
603      etc).  You can also store extra data and register callback functions
604      through ios_base, but that has been historically underused.  Anything
605      which doesn't depend on the type of characters stored is consolidated
606      here.
607   </p>
608   <p>The template class basic_ios is the highest template class in the
609      hierarchy; it is the first one depending on the character type, and
610      holds all general state associated with that type:  the pointer to the
611      polymorphic stream buffer, the facet information, etc.
612   </p>
613   <p><strong>&lt;streambuf&gt;</strong> declares the template class
614      basic_streambuf, and two standard instantiations, streambuf and
615      wstreambuf.  If you need to work with the vastly useful and capable
616      stream buffer classes, e.g., to create a new form of storage
617      transport, this header is the one to include.
618   </p>
619   <p><strong>&lt;istream&gt;</strong>/<strong>&lt;ostream&gt;</strong> are
620      the headers to include when you are using the &gt;&gt;/&lt;&lt;
621      interface, or any of the other abstract stream formatting functions.
622      For example,
623   </p>
624   <pre>
625    #include &lt;istream&gt;
626
627    std::ostream&amp; operator&lt;&lt; (std::ostream&amp; os, MyClass&amp; c)
628    {
629       return os &lt;&lt; c.data1() &lt;&lt; c.data2();
630    }
631   </pre>
632   <p>The std::istream and std::ostream classes are the abstract parents of
633      the various concrete implementations.  If you are only using the
634      interfaces, then you only need to use the appropriate interface header.
635   </p>
636   <p><strong>&lt;iomanip&gt;</strong> provides &quot;extractors and inserters
637      that alter information maintained by class ios_base and its dervied
638      classes,&quot; such as std::setprecision and std::setw.  If you need
639      to write expressions like <code>os &lt;&lt; setw(3);</code> or
640      <code>is &gt;&gt; setbase(8);</code>, you must include &lt;iomanip&gt;.
641   </p>
642   <p><strong>&lt;sstream&gt;</strong>/<strong>&lt;fstream&gt;</strong>
643      declare the six stringstream and fstream classes.  As they are the
644      standard concrete descendants of istream and ostream, you will already
645      know about them.
646   </p>
647   <p>Finally, <strong>&lt;iostream&gt;</strong> provides the eight standard
648      global objects (cin, cout, etc).  To do this correctly, this header
649      also provides the contents of the &lt;istream&gt; and &lt;ostream&gt;
650      headers, but nothing else.  The contents of this header look like
651   </p>
652   <pre>
653    #include &lt;ostream&gt;
654    #include &lt;istream&gt;
655
656    namespace std
657    {
658        extern istream cin;
659        extern ostream cout;
660        ....
661
662        // this is explained below
663        <strong>static ios_base::Init __foo;</strong>    // not its real name
664    }
665   </pre>
666   <p>Now, the runtime penalty mentioned previously:  the global objects
667      must be initialized before any of your own code uses them; this is
668      guaranteed by the standard.  Like any other global object, they must
669      be initialized once and only once.  This is typically done with a
670      construct like the one above, and the nested class ios_base::Init is
671      specified in the standard for just this reason.
672   </p>
673   <p>How does it work?  Because the header is included before any of your
674      code, the <strong>__foo</strong> object is constructed before any of
675      your objects.  (Global objects are built in the order in which they
676      are declared, and destroyed in reverse order.)  The first time the
677      constructor runs, the eight stream objects are set up.
678   </p>
679   <p>The <code>static</code> keyword means that each object file compiled
680      from a source file containing &lt;iostream&gt; will have its own
681      private copy of <strong>__foo</strong>.  There is no specified order
682      of construction across object files (it's one of those pesky NP
683      problems that make life so interesting), so one copy in each object
684      file means that the stream objects are guaranteed to be set up before
685      any of your code which uses them could run, thereby meeting the
686      requirements of the standard.
687   </p>
688   <p>The penalty, of course, is that after the first copy of
689      <strong>__foo</strong> is constructed, all the others are just wasted
690      processor time.  The time spent is merely for an increment-and-test
691      inside a function call, but over several dozen or hundreds of object
692      files, that time can add up.  (It's not in a tight loop, either.)
693   </p>
694   <p>The lesson?  Only include &lt;iostream&gt; when you need to use one of
695      the standard objects in that source file; you'll pay less startup
696      time.  Only include the header files you need to in general; your
697      compile times will go down when there's less parsing work to do.
698   </p>
699
700
701<hr />
702<h2><a name="11">Using FILE*s and file descriptors with IOStreams</a></h2>
703   <!-- referenced by ext/howto.html#2, update link if numbering changes -->
704   <p>The v2 library included non-standard extensions to construct
705      <code>std::filebuf</code>s from C stdio types such as
706      <code>FILE*</code>s and POSIX file descriptors.
707      Today the recommended way to use stdio types with libstdc++-v3
708      IOStreams is via the <code>stdio_filebuf</code> class (see below),
709      but earlier releases provided slightly different mechanisms.
710   </p>
711   <ul>
712     <li>3.0.x <code>filebuf</code>s have another ctor with this signature:
713         <br />
714        <code>basic_filebuf(__c_file_type*, ios_base::openmode, int_type);</code>
715         <br />This comes in very handy in a number of places, such as
716         attaching Unix sockets, pipes, and anything else which uses file
717         descriptors, into the IOStream buffering classes.  The three
718         arguments are as follows:
719         <ul>
720          <li><code>__c_file_type*      F   </code>
721              // the __c_file_type typedef usually boils down to stdio's FILE
722          </li>
723          <li><code>ios_base::openmode  M   </code>
724              // same as all the other uses of openmode
725          </li>
726          <li><code>int_type            B   </code>
727              // buffer size, defaults to BUFSIZ if not specified
728          </li>
729         </ul>
730         For those wanting to use file descriptors instead of FILE*'s, I
731         invite you to contemplate the mysteries of C's <code>fdopen()</code>.
732     </li>
733     <li>In library snapshot 3.0.95 and later, <code>filebuf</code>s bring
734         back an old extension:  the <code>fd()</code> member function.  The
735         integer returned from this function can be used for whatever file
736         descriptors can be used for on your platform.  Naturally, the
737         library cannot track what you do on your own with a file descriptor,
738         so if you perform any I/O directly, don't expect the library to be
739         aware of it.
740     </li>
741     <li>Beginning with 3.1, the extra <code>filebuf</code> constructor and
742         the <code>fd()</code> function were removed from the standard
743         filebuf.  Instead, <code>&lt;ext/stdio_filebuf.h&gt;</code> contains
744         a derived class called
745         <a href="http://gcc.gnu.org/onlinedocs/libstdc++/latest-doxygen/class____gnu__cxx_1_1stdio__filebuf.html"><code>__gnu_cxx::stdio_filebuf</code></a>.
746         This class can be constructed from a C <code>FILE*</code> or a file
747         descriptor, and provides the <code>fd()</code> function.
748     </li>
749   </ul>
750   <p>If you want to access a <code>filebuf</code>s file descriptor to
751      implement file locking (e.g. using the <code>fcntl()</code> system
752      call) then you might be interested in Henry Suter's
753      <a href="http://suter.home.cern.ch/suter/RWLock.html">RWLock</a>
754      class.
755   </p>
756
757<!-- ####################################################### -->
758
759<hr />
760<p class="fineprint"><em>
761See <a href="../17_intro/license.html">license.html</a> for copying conditions.
762Comments and suggestions are welcome, and may be sent to
763<a href="mailto:libstdc++@gcc.gnu.org">the libstdc++ mailing list</a>.
764</em></p>
765
766
767</body>
768</html>
769
770
771