xref: /openbsd/gnu/usr.bin/gcc/bugs.html (revision a67f0032)
1<html>
2
3<head>
4<title>GCC Bugs</title>
5</head>
6
7<body>
8<h1>GCC Bugs</h1>
9
10<p>The latest version of this document is always available at
11<a href="http://gcc.gnu.org/bugs.html">http://gcc.gnu.org/bugs.html</a>.</p>
12
13<hr />
14
15<h2>Table of Contents</h2>
16<ul>
17<li><a href="#report">Reporting Bugs</a>
18  <ul>
19  <li><a href="#need">What we need</a></li>
20  <li><a href="#dontwant">What we DON'T want</a></li>
21  <li><a href="#where">Where to post it</a></li>
22  <li><a href="#detailed">Detailed bug reporting instructions</a></li>
23  <li><a href="#gnat">Detailed bug reporting instructions for GNAT</a></li>
24  <li><a href="#pch">Detailed bug reporting instructions when using a precompiled header</a></li>
25  </ul>
26</li>
27<li><a href="#known">Frequently Reported Bugs in GCC</a>
28  <ul>
29  <li><a href="#cxx">C++</a>
30    <ul>
31    <li><a href="#missing">Missing features</a></li>
32    <li><a href="#fixed34">Bugs fixed in the 3.4 series</a></li>
33    </ul>
34  </li>
35  <li><a href="#fortran">Fortran</a></li>
36  </ul>
37</li>
38<li><a href="#nonbugs">Non-bugs</a>
39  <ul>
40  <li><a href="#nonbugs_general">General</a></li>
41  <li><a href="#nonbugs_c">C</a></li>
42  <li><a href="#nonbugs_cxx">C++</a>
43    <ul>
44    <li><a href="#upgrading">Common problems when upgrading the compiler</a></li>
45    </ul>
46  </li>
47  </ul>
48</li>
49</ul>
50
51<hr />
52
53<h1><a name="report">Reporting Bugs</a></h1>
54
55<p>The main purpose of a bug report is to enable us to fix the bug. The
56most important prerequisite for this is that the report must be complete and
57self-contained, which we explain in detail below.</p>
58
59<p>Before you report a bug, please check the
60<a href="#known">list of well-known bugs</a> and, <strong>if possible
61in any way, try a current development snapshot</strong>.
62If you want to report a bug with versions of GCC before 3.1 we strongly
63recommend upgrading to the current release first.</p>
64
65<p>Before reporting that GCC compiles your code incorrectly, please
66compile it with <code>gcc -Wall</code> and see whether this shows
67anything wrong with your code that could be the cause instead of a bug
68in GCC.</p>
69
70<h2>Summarized bug reporting instructions</h2>
71
72<p>After this summary, you'll find detailed bug reporting
73instructions, that explain how to obtain some of the information
74requested in this summary.</p>
75
76<h3><a name="need">What we need</a></h3>
77
78<p>Please include in your bug report all of the following items, the first
79three of which can be obtained from the output of <code>gcc -v</code>:</p>
80
81<ul>
82  <li>the exact version of GCC;</li>
83  <li>the system type;</li>
84  <li>the options given when GCC was configured/built;</li>
85  <li>the complete command line that triggers the bug;</li>
86  <li>the compiler output (error messages, warnings, etc.); and</li>
87  <li>the <em>preprocessed</em> file (<code>*.i*</code>) that triggers the
88  bug, generated by adding <code>-save-temps</code> to the complete
89  compilation command, or, in the case of a bug report for the GNAT front end,
90  a complete set of source files (see below).</li>
91</ul>
92
93<h3><a name="dontwant">What we do <strong>not</strong> want</a></h3>
94
95<ul>
96  <li>A source file that <code>#include</code>s header files that are left
97  out of the bug report (see above)</li>
98
99  <li>That source file and a collection of header files.</li>
100
101  <li>An attached archive (tar, zip, shar, whatever) containing all
102  (or some :-) of the above.</li>
103
104  <li>A code snippet that won't cause the compiler to produce the
105  exact output mentioned in the bug report (e.g., a snippet with just
106  a few lines around the one that <b>apparently</b> triggers the bug,
107  with some pieces replaced with ellipses or comments for extra
108  obfuscation :-)</li>
109
110  <li>The location (URL) of the package that failed to build (we won't
111  download it, anyway, since you've already given us what we need to
112  duplicate the bug, haven't you? :-)</li>
113
114  <li>An error that occurs only some of the times a certain file is
115  compiled, such that retrying a sufficient number of times results in
116  a successful compilation; this is a symptom of a hardware problem,
117  not of a compiler bug (sorry)</li>
118
119  <li>E-mail messages that complement previous, incomplete bug
120  reports. Post a new, self-contained, full bug report instead, if
121  possible as a follow-up to the original bug report</li>
122
123  <li>Assembly files (<code>*.s</code>) produced by the compiler, or any
124  binary files, such as object files, executables, core files, or
125  precompiled header files</li>
126
127  <li>Duplicate bug reports, or reports of bugs already fixed in the
128  development tree, especially those that have already been reported
129  as fixed last week :-)</li>
130
131  <li>Bugs in the assembler, the linker or the C library.  These are
132  separate projects, with separate mailing lists and different bug
133  reporting procedures</li>
134
135  <li>Bugs in releases or snapshots of GCC not issued by the GNU
136  Project.  Report them to whoever provided you with the release</li>
137
138  <li>Questions about the correctness or the expected behavior of
139  certain constructs that are not GCC extensions.  Ask them in forums
140  dedicated to the discussion of the programming language</li>
141</ul>
142
143<h3><a name="where">Where to post it</a></h3>
144
145<p>Please submit your bug report directly to the
146<a href="http://gcc.gnu.org/bugzilla/">GCC bug database</a>.
147Alternatively, you can use the <code>gccbug</code> script that mails your bug
148report to the bug database.
149<br />
150Only if all this is absolutely impossible, mail all information to
151<a href="mailto:gcc-bugs@gcc.gnu.org">gcc-bugs@gcc.gnu.org</a>.</p>
152
153<h2><a name="detailed">Detailed bug reporting instructions</a></h2>
154
155<p>Please refer to the <a href="#gnat">next section</a> when reporting
156bugs in GNAT, the Ada compiler, or to the <a href="#pch">one after
157that</a> when reporting bugs that appear when using a precompiled header.</p>
158
159<p>In general, all the information we need can be obtained by
160collecting the command line below, as well as its output and the
161preprocessed file it generates.</p>
162
163<blockquote><p><code>gcc -v -save-temps <i>all-your-options
164source-file</i></code></p></blockquote>
165
166<p>Typically the preprocessed file (extension <code>.i</code> for C or
167<code>.ii</code> for C++, and <code>.f</code> if the preprocessor is used on
168Fortran files) will be large, so please compress the
169resulting file with one of the popular compression programs such as
170bzip2, gzip, zip or compress (in
171decreasing order of preference).  Use maximum compression
172(<code>-9</code>) if available.  Please include the compressed
173preprocessor output in your bug report, even if the source code is
174freely available elsewhere; it makes the job of our volunteer testers
175much easier.</p>
176
177<p>The <b>only</b> excuses to not send us the preprocessed sources are
178(i) if you've found a bug in the preprocessor, (ii) if you've reduced
179the testcase to a small file that doesn't include any other file or
180(iii) if the bug appears only when using precompiled headers.  If you
181can't post the preprocessed sources because they're proprietary code,
182then try to create a small file that triggers the same problem.</p>
183
184<p>Since we're supposed to be able to re-create the assembly output
185(extension <code>.s</code>), you usually should not include
186it in the bug report, although you may want to post parts of it to
187point out assembly code you consider to be wrong.</p>
188
189<p>Whether to use MIME attachments or <code>uuencode</code> is up to
190you.  In any case, make sure the compiler command line, version and
191error output are in plain text, so that we don't have to decode the
192bug report in order to tell who should take care of it.  A meaningful
193subject indicating language and platform also helps.</p>
194
195<p>Please avoid posting an archive (.tar, .shar or .zip); we generally
196need just a single file to reproduce the bug (the .i/.ii/.f preprocessed
197file), and, by storing it in an archive, you're just making our
198volunteers' jobs harder.  Only when your bug report requires multiple
199source files to be reproduced should you use an archive.  This is, for example,
200the case if you are using <code>INCLUDE</code> directives in Fortran code,
201which are not processed by the preprocessor, but the compiler.  In that case,
202we need the main file and all <code>INCLUDE</code>d files.  In any case,
203make sure the compiler version, error message, etc, are included in
204the body of your bug report as plain text, even if needlessly
205duplicated as part of an archive.</p>
206
207<p>If you fail to supply enough information for a bug report to be
208reproduced, someone will probably ask you to post additional
209information (or just ignore your bug report, if they're in a bad day,
210so try to get it right on the first posting :-).  In this case, please
211post the additional information to the bug reporting mailing list, not
212just to the person who requested it, unless explicitly told so.  If
213possible, please include in this follow-up all the information you had
214supplied in the incomplete bug report (including the preprocessor
215output), so that the new bug report is self-contained.</p>
216
217<h2><a name="gnat">Detailed bug reporting instructions for GNAT</a></h2>
218
219<p>See the <a href="#detailed">previous section</a> for bug reporting
220instructions for GCC language implementations other than Ada.</p>
221
222<p>Bug reports have to contain at least the following information in
223order to be useful:</p>
224
225<ul>
226<li>the exact version of GCC, as shown by "<code>gcc -v</code>";</li>
227<li>the system type;</li>
228<li>the options when GCC was configured/built;</li>
229<li>the exact command line passed to the <code>gcc</code> program
230triggering the bug
231(not just the flags passed to <code>gnatmake</code>, but
232<code>gnatmake</code> prints the parameters it passed to <code>gcc</code>)</li>
233<li>a collection of source files for reproducing the bug,
234preferably a minimal set (see below);</li>
235<li>a description of the expected behavior;</li>
236<li>a description of actual behavior.</li>
237</ul>
238
239<p>If your code depends on additional source files (usually package
240specifications), submit the source code for these compilation units in
241a single file that is acceptable input to <code>gnatchop</code>,
242i.e. contains no non-Ada text.  If the compilation terminated
243normally, you can usually obtain a list of dependencies using the
244"<code>gnatls -d <i>main_unit</i></code>" command, where
245<code><i>main_unit</i></code> is the file name of the main compilation
246unit (which is also passed to <code>gcc</code>).</p>
247
248<p>If you report a bug which causes the compiler to print a bug box,
249include that bug box in your report, and do not forget to send all the
250source files listed after the bug box along with your report.</p>
251
252<p>If you use <code>gnatprep</code>, be sure to send in preprocessed
253sources (unless you have to report a bug in <code>gnatprep</code>).</p>
254
255<p>When you have checked that your report meets these criteria, please
256submit it according to our <a href="#where">generic instructions</a>.
257(If you use a mailing list for reporting, please include an
258"<code>[Ada]</code>" tag in the subject.)</p>
259
260<h2><a name="pch">Detailed bug reporting instructions when using a
261precompiled header</a></h2>
262
263<p>If you're encountering a bug when using a precompiled header, the
264first thing to do is to delete the precompiled header, and try running
265the same GCC command again.  If the bug happens again, the bug doesn't
266really involve precompiled headers, please report it without using
267them by following the instructions <a href="#detailed">above</a>.</p>
268
269<p>If you've found a bug while <i>building</i> a precompiled header
270(for instance, the compiler crashes), follow the usual instructions
271<a href="#detailed">above</a>.</p>
272
273<p>If you've found a real precompiled header bug, what we'll need to
274reproduce it is the sources to build the precompiled header (as a
275single <code>.i</code> file), the source file that uses the
276precompiled header, any other headers that source file includes, and
277the command lines that you used to build the precompiled header and to
278use it.</p>
279
280<p>Please <strong>don't</strong> send us the actual precompiled
281header.  It is likely to be very large and we can't use it to
282reproduce the problem.</p>
283
284<hr />
285
286<h1><a name="known">Frequently Reported Bugs in GCC</a></h1>
287
288<p>This is a list of bugs in GCC that are reported very often, but not
289yet fixed. While it is certainly better to fix bugs instead of documenting
290them, this document might save people the effort of writing a bug report
291when the bug is already well-known.</p>
292
293<p>There are many reasons why a reported bug doesn't get fixed.
294It might be difficult to fix, or fixing it might break compatibility.
295Often, reports get a low priority when there is a simple work-around.
296In particular, bugs caused by invalid code have a simple work-around:
297<em>fix the code</em>.</p>
298
299<hr />
300
301<h2><a name="cxx">C++</a></h2>
302
303<h3><a name="missing">Missing features</a></h3>
304
305<dl>
306
307<dt>The <code>export</code> keyword is not implemented.</dt>
308<dd><p>Most C++ compilers (G++ included) do not yet implement
309<code>export</code>, which is necessary for separate compilation of
310template declarations and definitions. Without <code>export</code>, a
311template definition must be in scope to be used. The obvious
312workaround is simply to place all definitions in the header
313itself. Alternatively, the compilation unit containing template
314definitions may be included from the header.</p></dd>
315
316</dl>
317
318<h3><a name="fixed34">Bugs fixed in the 3.4 series</a></h3>
319
320<p>The following bugs are present up to (and including) GCC 3.3.x.
321They have been fixed in 3.4.0.</p>
322
323<dl>
324
325<dt>Two-stage name-lookup.</dt>
326
327<dd><p>GCC did not implement two-stage name-lookup (also see
328<a href="#new34">below</a>).</p></dd>
329
330<dt>Covariant return types.</dt>
331
332<dd><p>GCC did not implement non-trivial covariant returns.</p></dd>
333
334<dt>Parse errors for "simple" code.</dt>
335
336<dd><p>GCC gave parse errors for seemingly simple code, such as</p>
337
338<blockquote><pre>
339struct A
340{
341  A();
342  A(int);
343};
344
345struct B
346{
347  B(A);
348  B(A,A);
349  void foo();
350};
351
352A bar()
353{
354  B b(A(),A(1));  // Variable b, initialized with two temporaries
355  B(A(2)).foo();  // B temporary, initialized with A temporary
356  return (A());   // return A temporary
357}
358</pre></blockquote>
359
360<p>Although being valid code, each of the three lines with a comment was
361rejected by GCC.  The work-arounds for older compiler versions proposed
362below do not change the semantics of the programs at all.</p>
363
364<p>The problem in the first case was that GCC started to parse the
365declaration of <code>b</code> as a function called <code>b</code> returning
366<code>B</code>, taking a function returning <code>A</code> as an argument.
367When it encountered the <code>1</code>, it was too late.  To show the
368compiler that this should be really an expression, a comma operator with
369a dummy argument could be used:</p>
370
371<blockquote><pre>
372B b((0,A()),A(1));
373</pre></blockquote>
374
375<p>The work-around for simpler cases like the second one was to add
376additional parentheses around the expressions that were mistaken as
377declarations:</p>
378
379<blockquote><pre>
380(B(A(2))).foo();
381</pre></blockquote>
382
383<p>In the third case, however, additional parentheses were causing
384the problems: The compiler interpreted <code>A()</code> as a function
385(taking no arguments, returning <code>A</code>), and <code>(A())</code>
386as a cast lacking an expression to be casted, hence the parse error.
387The work-around was to omit the parentheses:</p>
388
389<blockquote><pre>
390return A();
391</pre></blockquote>
392
393<p>This problem occurred in a number of variants; in <code>throw</code>
394statements, people also frequently put the object in parentheses.</p></dd>
395
396</dl>
397
398<hr />
399
400<h2><a name="fortran">Fortran</a></h2>
401
402<p>Fortran bugs are documented in the G77 manual rather than
403explicitly listed here.  Please see
404<a href="http://gcc.gnu.org/onlinedocs/g77/Trouble.html">Known Causes of
405Trouble with GNU Fortran</a> in the G77 manual.</p>
406
407<hr />
408
409<h1><a name="nonbugs">Non-bugs</a></h1>
410
411<p>The following are not actually bugs, but are reported often
412enough to warrant a mention here.</p>
413
414<p>It is not always a bug in the compiler, if code which "worked" in a
415previous version, is now rejected.  Earlier versions of GCC sometimes were
416less picky about standard conformance and accepted invalid source code.
417In addition, programming languages themselves change, rendering code
418invalid that used to be conforming (this holds especially for C++).
419In either case, you should update your code to match recent language
420standards.</p>
421
422<hr />
423
424<h2><a name="nonbugs_general">General</a></h2>
425
426<dl>
427<dt>Problems with floating point numbers - the
428<a href="http://gcc.gnu.org/PR323">most often reported non-bug</a>.</dt>
429<dd><p>In a number of cases, GCC appears to perform floating point
430computations incorrectly. For example, the C++ program</p>
431<blockquote><pre>
432#include &lt;iostream&gt;
433
434int main()
435{
436  double a = 0.5;
437  double b = 0.01;
438  std::cout &lt;&lt; (int)(a / b) &lt;&lt; std::endl;
439  return 0;
440}
441</pre></blockquote>
442<p>might print 50 on some systems and optimization levels, and 49 on
443others.</p>
444
445<p>This is the result of <em>rounding</em>: The computer cannot
446represent all real numbers exactly, so it has to use
447approximations. When computing with approximation, the computer needs
448to round to the nearest representable number.</p>
449
450<p>This is not a bug in the compiler, but an inherent limitation of
451the floating point types. Please study
452<a href="http://www.validlab.com/goldberg/paper.ps">this paper</a>
453for more information.</p></dd>
454</dl>
455
456<hr />
457
458<h2><a name="nonbugs_c">C</a></h2>
459
460<dl>
461<dt>Increment/decrement operator (<code>++</code>/<code>--</code>) not
462working as expected - a <a href="http://gcc.gnu.org/PR11751">problem with
463many variations</a>.</dt>
464
465<dd><p>The following expressions have unpredictable results:</p>
466<blockquote><pre>
467x[i]=++i
468foo(i,++i)
469i*(++i)                 /* special case with foo=="operator*" */
470std::cout &lt;&lt; i &lt;&lt; ++i   /* foo(foo(std::cout,i),++i)          */
471</pre></blockquote>
472<p>since the <code>i</code> without increment can be evaluated before or
473after <code>++i</code>.</p>
474
475<p>The C and C++ standards have the notion of "sequence points". Everything
476that happens between two sequence points happens in an unspecified order,
477but it has to happen after the first and before the second sequence point.
478The end of a statement and a function call are examples for sequence points,
479whereas assignments and the comma between function arguments are not.</p>
480
481<p>Modifying a value twice between two sequence points as shown in the
482following examples is even worse:</p>
483<blockquote><pre>
484i=++i
485foo(++i,++i)
486(++i)*(++i)               /* special case with foo=="operator*" */
487std::cout &lt;&lt; ++i &lt;&lt; ++i   /* foo(foo(std::cout,++i),++i)        */
488</pre></blockquote>
489<p>This leads to undefined behavior (i.e. the compiler can do
490anything).</p></dd>
491
492
493<dt>Casting does not work as expected when optimization is turned on.</dt>
494
495<dd><p>This is often caused by a violation of aliasing rules, which are part
496of the ISO C standard.  These rules say that a program is invalid if you try
497to access a variable through a pointer of an incompatible type.  This is
498happening in the following example where a short is accessed through a
499pointer to integer (the code assumes 16-bit <code>short</code>s and 32-bit
500<code>int</code>s):</p>
501<blockquote><pre>
502#include &lt;stdio.h&gt;
503
504int main()
505{
506  short a[2];
507
508  a[0]=0x1111;
509  a[1]=0x1111;
510
511  *(int *)a = 0x22222222; /* violation of aliasing rules */
512
513  printf("%x %x\n", a[0], a[1]);
514  return 0;
515}
516</pre></blockquote>
517<p>The aliasing rules were designed to allow compilers more aggressive
518optimization.  Basically, a compiler can assume that all changes to variables
519happen through pointers or references to variables of a type compatible to
520the accessed variable.  Dereferencing a pointer that violates the aliasing
521rules results in undefined behavior.</p>
522
523<p>In the case above, the compiler may assume that no access through an
524integer pointer can change the array <code>a</code>, consisting of shorts.
525Thus, <code>printf</code> may be called with the original values of
526<code>a[0]</code> and <code>a[1]</code>.  What really happens is up to
527the compiler and may change with architecture and optimization level.</p>
528
529<p>Recent versions of GCC turn on the option <code>-fstrict-aliasing</code>
530(which allows alias-based optimizations) by default with <code>-O2</code>.
531And some architectures then really print "1111 1111" as result.  Without
532optimization the executable will generate the "expected" output
533"2222 2222".</p>
534
535<p>To disable optimizations based on alias-analysis for faulty legacy code,
536the option <code>-fno-strict-aliasing</code> can be used as a work-around.</p>
537
538<p>The option <code>-Wstrict-aliasing</code> (which is included in
539<code>-Wall</code>) warns about some - but not all - cases of violation
540of aliasing rules when <code>-fstrict-aliasing</code> is active.</p>
541
542<p>To fix the code above, you can use a <code>union</code> instead of a
543cast (note that this is a GCC extension which might not work with other
544compilers):</p>
545<blockquote><pre>
546#include &lt;stdio.h&gt;
547
548int main()
549{
550  union
551  {
552    short a[2];
553    int i;
554  } u;
555
556  u.a[0]=0x1111;
557  u.a[1]=0x1111;
558
559  u.i = 0x22222222;
560
561  printf("%x %x\n", u.a[0], u.a[1]);
562  return 0;
563}
564</pre></blockquote>
565<p>Now the result will always be "2222 2222".</p>
566
567<p>For some more insight into the subject, please have a look at
568<a href="http://mail-index.NetBSD.org/tech-kern/2003/08/11/0001.html">this
569article</a>.</p></dd>
570
571
572<dt>Cannot use preprocessor directive in macro arguments.</dt>
573<dd><p>Let me guess... you used an older version of GCC to compile code
574that looks something like this:</p>
575<blockquote><pre>
576  memcpy(dest, src,
577#ifdef PLATFORM1
578	 12
579#else
580	 24
581#endif
582	);
583</pre></blockquote>
584<p>and you got a whole pile of error messages:</p>
585<blockquote><pre>
586test.c:11: warning: preprocessing directive not recognized within macro arg
587test.c:11: warning: preprocessing directive not recognized within macro arg
588test.c:11: warning: preprocessing directive not recognized within macro arg
589test.c: In function `foo':
590test.c:6: undefined or invalid # directive
591test.c:8: undefined or invalid # directive
592test.c:9: parse error before `24'
593test.c:10: undefined or invalid # directive
594</pre></blockquote>
595
596<p>This is because your C library's <code>&lt;string.h&gt;</code> happens
597to define <code>memcpy</code> as a macro - which is perfectly legitimate.
598In recent versions of glibc, for example, <code>printf</code> is among those
599functions which are implemented as macros.</p>
600
601<p>Versions of GCC prior to 3.3 did not allow you to put <code>#ifdef</code>
602(or any other preprocessor directive) inside the arguments of a macro.  The
603code therefore would not compile.</p>
604
605<p>As of GCC 3.3 this kind of construct is always accepted and the
606preprocessor will probably do what you expect, but see the manual for
607detailed semantics.</p>
608
609<p>However, this kind of code is not portable.  It is "undefined behavior"
610according to the C standard; that means different compilers may do
611different things with it.  It is always possible to rewrite code which
612uses conditionals inside macros so that it doesn't.  You could write
613the above example</p>
614<blockquote><pre>
615#ifdef PLATFORM1
616   memcpy(dest, src, 12);
617#else
618   memcpy(dest, src, 24);
619#endif
620</pre></blockquote>
621<p>This is a bit more typing, but I personally think it's better style
622in addition to being more portable.</p></dd>
623
624
625<dt>Cannot initialize a static variable with <code>stdin</code>.</dt>
626<dd><p>This has nothing to do with GCC, but people ask us about it a
627lot.  Code like this:</p>
628
629<blockquote><pre>
630#include &lt;stdio.h&gt;
631
632FILE *yyin = stdin;
633</pre></blockquote>
634
635<p>will not compile with GNU libc, because <code>stdin</code> is not a
636constant.  This was done deliberately, to make it easier to maintain
637binary compatibility when the type <code>FILE</code> needs to be changed.
638It is surprising for people used to traditional Unix C libraries, but it
639is permitted by the C standard.</p>
640
641<p>This construct commonly occurs in code generated by old versions of
642lex or yacc.  We suggest you try regenerating the parser with a
643current version of flex or bison, respectively.  In your own code, the
644appropriate fix is to move the initialization to the beginning of
645main.</p>
646
647<p>There is a common misconception that the GCC developers are
648responsible for GNU libc.  These are in fact two entirely separate
649projects; please check the
650<a href="http://www.gnu.org/software/libc/">GNU libc web pages</a>
651for details.
652</p></dd>
653</dl>
654
655<hr />
656
657<h2><a name="nonbugs_cxx">C++</a></h2>
658
659<dl>
660<dt>Nested classes can access private members and types of the containing
661class.</dt>
662
663<dd><p>Defect report 45 clarifies that nested classes are members of the
664class they are nested in, and so are granted access to private members of
665that class.</p></dd>
666
667<dt>G++ emits two copies of constructors and destructors.</dt>
668
669<dd><p>In general there are <em>three</em> types of constructors (and
670destructors).</p>
671<ol>
672<li>The complete object constructor/destructor.</li>
673<li>The base object constructor/destructor.</li>
674<li>The allocating constructor/deallocating destructor.</li>
675</ol>
676<p>The first two are different, when virtual base classes are involved.
677</p></dd>
678
679<dt>Global destructors are not run in the correct order.</dt>
680
681<dd><p>Global destructors should be run in the reverse order of their
682constructors <em>completing</em>. In most cases this is the same as
683the reverse order of constructors <em>starting</em>, but sometimes it
684is different, and that is important. You need to compile and link your
685programs with <code>--use-cxa-atexit</code>. We have not turned this
686switch on by default, as it requires a <code>cxa</code> aware runtime
687library (<code>libc</code>, <code>glibc</code>, or equivalent).</p></dd>
688
689<dt>Classes in exception specifiers must be complete types.</dt>
690
691<dd><p>[15.4]/1 tells you that you cannot have an incomplete type, or
692pointer to incomplete (other than <code><i>cv</i> void *</code>) in
693an exception specification.</p></dd>
694
695<dt>Exceptions don't work in multithreaded applications.</dt>
696
697<dd><p>You need to rebuild g++ and libstdc++ with
698<code>--enable-threads</code>.  Remember, C++ exceptions are not like
699hardware interrupts. You cannot throw an exception in one thread and
700catch it in another. You cannot throw an exception from a signal
701handler and catch it in the main thread.</p></dd>
702
703<dt>Templates, scoping, and digraphs.</dt>
704
705<dd><p>If you have a class in the global namespace, say named <code>X</code>,
706and want to give it as a template argument to some other class, say
707<code>std::vector</code>, then <code>std::vector&lt;::X&gt;</code>
708fails with a parser error.</p>
709
710<p>The reason is that the standard mandates that the sequence
711<code>&lt;:</code> is treated as if it were the token <code>[</code>.
712(There are several such combinations of characters - they are called
713<em>digraphs</em>.) Depending on the version, the compiler then reports
714a parse error before the character <code>:</code> (the colon before
715<code>X</code>) or a missing closing bracket <code>]</code>.</p>
716
717<p>The simplest way to avoid this is to write <code>std::vector&lt;
718::X&gt;</code>, i.e. place a space between the opening angle bracket
719and the scope operator.</p></dd>
720
721
722<dt><a name="cxx_rvalbind">Copy constructor access check while
723initializing a reference.</a></dt>
724
725<dd><p>Consider this code:</p>
726
727<blockquote><pre>
728class A
729{
730public:
731  A();
732
733private:
734  A(const A&amp;);   // private copy ctor
735};
736
737A makeA(void);
738void foo(const A&amp;);
739
740void bar(void)
741{
742  foo(A());       // error, copy ctor is not accessible
743  foo(makeA());   // error, copy ctor is not accessible
744
745  A a1;
746  foo(a1);        // OK, a1 is a lvalue
747}</pre></blockquote>
748
749<p>Starting with GCC 3.4.0, binding an rvalue to a const reference requires
750an accessible copy constructor. This might be surprising at first sight,
751especially since most popular compilers do not correctly implement this
752rule.</p>
753
754<p>The C++ Standard says that a temporary object should be created in
755this context and its contents filled with a copy of the object we are
756trying to bind to the reference; it also says that the temporary copy
757can be elided, but the semantic constraints (eg. accessibility) of the
758copy constructor still have to be checked.</p>
759
760<p>For further information, you can consult the following paragraphs of
761the C++ standard: [dcl.init.ref]/5, bullet 2, sub-bullet 1, and
762[class.temporary]/2.</p></dd>
763</dl>
764
765<h3><a name="upgrading">Common problems when upgrading the compiler</a></h3>
766
767<h4>ABI changes</h4>
768
769<p>The C++ application binary interface (ABI) consists of two
770components: the first defines how the elements of classes are laid
771out, how functions are called, how function names are mangled, etc;
772the second part deals with the internals of the objects in libstdc++.
773Although we strive for a non-changing ABI, so far we have had to
774modify it with each major release.  If you change your compiler to a
775different major release <em>you must recompile all libraries that
776contain C++ code</em>.  If you fail to do so you risk getting linker
777errors or malfunctioning programs.  Some of our Java support libraries
778also contain C++ code, so you might want to recompile all libraries to
779be safe.  It should not be necessary to recompile if you have changed
780to a bug-fix release of the same version of the compiler; bug-fix
781releases are careful to avoid ABI changes. See also the
782<a href="http://gcc.gnu.org/onlinedocs/gcc/Compatibility.html">compatibility
783section</a> of the GCC manual.</p>
784
785<p>Remark: A major release is designated by a change to the first or second
786component of the two- or three-part version number.  A minor (bug-fix)
787release is designated by a change to the third component only.  Thus GCC
7883.2 and 3.3 are major releases, while 3.3.1 and 3.3.2 are bug-fix releases
789for GCC 3.3.  With the 3.4 series we are introducing a new naming scheme;
790the first release of this series is 3.4.0 instead of just 3.4.</p>
791
792<h4>Standard conformance</h4>
793
794<p>With each release, we try to make G++ conform closer to the ISO C++ standard
795(available at
796<a href="http://www.ncits.org/cplusplus.htm">http://www.ncits.org/cplusplus.htm</a>).
797We have also implemented some of the core and library defect reports
798(available at
799<a href="http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html">http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html</a>
800&amp;
801<a href="http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-defects.html">http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-defects.html</a>
802respectively).</p>
803
804<p>Non-conforming legacy code that worked with older versions of GCC may be
805rejected by more recent compilers.  There is no command-line switch to ensure
806compatibility in general, because trying to parse standard-conforming and
807old-style code at the same time would render the C++ frontend unmaintainable.
808However, some non-conforming constructs are allowed when the command-line
809option <code>-fpermissive</code> is used.</p>
810
811<p>Two milestones in standard conformance are GCC 3.0 (including a major
812overhaul of the standard library) and the 3.4.0 version (with its new C++
813parser).</p>
814
815<h4>New in GCC 3.0</h4>
816
817<ul>
818
819<li>The standard library is much more conformant, and uses the
820<code>std::</code> namespace (which is now a real namespace, not an
821alias for <code>::</code>).</li>
822
823<li>The standard header files for the c library don't end with
824<code>.h</code>, but begin with <code>c</code> (i.e.
825<code>&lt;cstdlib&gt;</code> rather than <code>&lt;stdlib.h&gt;</code>).
826The <code>.h</code> names are still available, but are deprecated.</li>
827
828<li><code>&lt;strstream&gt;</code> is deprecated, use
829<code>&lt;sstream&gt;</code> instead.</li>
830
831<li><code>streambuf::seekoff</code> &amp;
832<code>streambuf::seekpos</code> are private, instead use
833<code>streambuf::pubseekoff</code> &amp;
834<code>streambuf::pubseekpos</code> respectively.</li>
835
836<li>If <code>std::operator &lt;&lt; (std::ostream &amp;, long long)</code>
837doesn't exist, you need to recompile libstdc++ with
838<code>--enable-long-long</code>.</li>
839
840</ul>
841
842<p>If you get lots of errors about things like <code>cout</code> not being
843found, you've most likely forgotten to tell the compiler to look in the
844<code>std::</code> namespace.  There are several ways to do this:</p>
845
846<ul>
847
848<li>Say <code>std::cout</code> at the call.  This is the most explicit
849way of saying what you mean.</li>
850
851<li>Say <code>using std::cout;</code> somewhere before the call.  You
852will need to do this for each function or type you wish to use from the
853standard library.</li>
854
855<li>Say <code>using namespace std;</code> somewhere before the call.
856This is the quick-but-dirty fix. This brings the <em>whole</em> of the
857<code>std::</code> namespace into scope.  <em>Never</em> do this in a
858header file, as every user of your header file will be affected by this
859decision.</li>
860
861</ul>
862
863<h4><a name="new34">New in GCC 3.4.0</a></h4>
864
865<p>The new parser brings a lot of improvements, especially concerning
866name-lookup.</p>
867
868<ul>
869
870<li>The "implicit typename" extension got removed (it was already deprecated
871since GCC 3.1), so that the following code is now rejected, see [14.6]:
872<blockquote><pre>
873template &lt;typename&gt; struct A
874{
875    typedef int X;
876};
877
878template &lt;typename T&gt; struct B
879{
880    A&lt;T&gt;::X          x;  // error
881    typename A&lt;T&gt;::X y;  // OK
882};
883
884B&lt;void&gt; b;
885</pre></blockquote></li>
886
887<li>For similar reasons, the following code now requires the
888<code>template</code> keyword, see [14.2]:
889<blockquote><pre>
890template &lt;typename&gt; struct A
891{
892    template &lt;int&gt; struct X {};
893};
894
895template &lt;typename T&gt; struct B
896{
897    typename A&lt;T&gt;::X&lt;0&gt;          x;  // error
898    typename A&lt;T&gt;::template X&lt;0&gt; y;  // OK
899};
900
901B&lt;void&gt; b;
902</pre></blockquote></li>
903
904<li>We now have two-stage name-lookup, so that the following code is
905rejected, see [14.6]/9:
906<blockquote><pre>
907template &lt;typename T&gt; int foo()
908{
909    return i;  // error
910}
911</pre></blockquote></li>
912
913<li>This also affects members of base classes, see [14.6.2]:
914<blockquote><pre>
915template &lt;typename&gt; struct A
916{
917    int i, j;
918};
919
920template &lt;typename T&gt; struct B : A&lt;T&gt;
921{
922    int foo1() { return i; }       // error
923    int foo2() { return this-&gt;i; } // OK
924    int foo3() { return B&lt;T&gt;::i; } // OK
925    int foo4() { return A&lt;T&gt;::i; } // OK
926
927    using A&lt;T&gt;::j;
928    int foo5() { return j; }       // OK
929};
930</pre></blockquote></li>
931
932</ul>
933
934<p>In addition to the problems listed above, the manual contains a section on
935<a href="http://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Misunderstandings.html">
936Common Misunderstandings with GNU C++</a>.</p>
937
938</body>
939</html>
940