4 The latest version of this document is always available at
5 [1]http://gcc.gnu.org/bugs.html.
6 _________________________________________________________________
12 + [4]What we DON'T want
14 + [6]Detailed bug reporting instructions
15 + [7]Detailed bug reporting instructions for GNAT
16 + [8]Detailed bug reporting instructions when using a
18 * [9]Frequently Reported Bugs in GCC
20 o [11]Missing features
21 o [12]Bugs fixed in the 3.4 series
27 o [18]Common problems when upgrading the compiler
28 _________________________________________________________________
32 The main purpose of a bug report is to enable us to fix the bug. The
33 most important prerequisite for this is that the report must be
34 complete and self-contained, which we explain in detail below.
36 Before you report a bug, please check the [19]list of well-known bugs
37 and, if possible in any way, try a current development snapshot. If
38 you want to report a bug with versions of GCC before 3.1 we strongly
39 recommend upgrading to the current release first.
41 Before reporting that GCC compiles your code incorrectly, please
42 compile it with gcc -Wall and see whether this shows anything wrong
43 with your code that could be the cause instead of a bug in GCC.
45 Summarized bug reporting instructions
47 After this summary, you'll find detailed bug reporting instructions,
48 that explain how to obtain some of the information requested in this
53 Please include in your bug report all of the following items, the
54 first three of which can be obtained from the output of gcc -v:
55 * the exact version of GCC;
57 * the options given when GCC was configured/built;
58 * the complete command line that triggers the bug;
59 * the compiler output (error messages, warnings, etc.); and
60 * the preprocessed file (*.i*) that triggers the bug, generated by
61 adding -save-temps to the complete compilation command, or, in the
62 case of a bug report for the GNAT front end, a complete set of
63 source files (see below).
67 * A source file that #includes header files that are left out of the
68 bug report (see above)
69 * That source file and a collection of header files.
70 * An attached archive (tar, zip, shar, whatever) containing all (or
71 some :-) of the above.
72 * A code snippet that won't cause the compiler to produce the exact
73 output mentioned in the bug report (e.g., a snippet with just a
74 few lines around the one that apparently triggers the bug, with
75 some pieces replaced with ellipses or comments for extra
77 * The location (URL) of the package that failed to build (we won't
78 download it, anyway, since you've already given us what we need to
79 duplicate the bug, haven't you? :-)
80 * An error that occurs only some of the times a certain file is
81 compiled, such that retrying a sufficient number of times results
82 in a successful compilation; this is a symptom of a hardware
83 problem, not of a compiler bug (sorry)
84 * E-mail messages that complement previous, incomplete bug reports.
85 Post a new, self-contained, full bug report instead, if possible
86 as a follow-up to the original bug report
87 * Assembly files (*.s) produced by the compiler, or any binary
88 files, such as object files, executables, core files, or
89 precompiled header files
90 * Duplicate bug reports, or reports of bugs already fixed in the
91 development tree, especially those that have already been reported
92 as fixed last week :-)
93 * Bugs in the assembler, the linker or the C library. These are
94 separate projects, with separate mailing lists and different bug
96 * Bugs in releases or snapshots of GCC not issued by the GNU
97 Project. Report them to whoever provided you with the release
98 * Questions about the correctness or the expected behavior of
99 certain constructs that are not GCC extensions. Ask them in forums
100 dedicated to the discussion of the programming language
104 Please submit your bug report directly to the [20]GCC bug database.
105 Alternatively, you can use the gccbug script that mails your bug
106 report to the bug database.
107 Only if all this is absolutely impossible, mail all information to
108 [21]gcc-bugs@gcc.gnu.org.
110 Detailed bug reporting instructions
112 Please refer to the [22]next section when reporting bugs in GNAT, the
113 Ada compiler, or to the [23]one after that when reporting bugs that
114 appear when using a precompiled header.
116 In general, all the information we need can be obtained by collecting
117 the command line below, as well as its output and the preprocessed
120 gcc -v -save-temps all-your-options source-file
122 Typically the preprocessed file (extension .i for C or .ii for C++,
123 and .f if the preprocessor is used on Fortran files) will be large, so
124 please compress the resulting file with one of the popular compression
125 programs such as bzip2, gzip, zip or compress (in decreasing order of
126 preference). Use maximum compression (-9) if available. Please include
127 the compressed preprocessor output in your bug report, even if the
128 source code is freely available elsewhere; it makes the job of our
129 volunteer testers much easier.
131 The only excuses to not send us the preprocessed sources are (i) if
132 you've found a bug in the preprocessor, (ii) if you've reduced the
133 testcase to a small file that doesn't include any other file or (iii)
134 if the bug appears only when using precompiled headers. If you can't
135 post the preprocessed sources because they're proprietary code, then
136 try to create a small file that triggers the same problem.
138 Since we're supposed to be able to re-create the assembly output
139 (extension .s), you usually should not include it in the bug report,
140 although you may want to post parts of it to point out assembly code
141 you consider to be wrong.
143 Whether to use MIME attachments or uuencode is up to you. In any case,
144 make sure the compiler command line, version and error output are in
145 plain text, so that we don't have to decode the bug report in order to
146 tell who should take care of it. A meaningful subject indicating
147 language and platform also helps.
149 Please avoid posting an archive (.tar, .shar or .zip); we generally
150 need just a single file to reproduce the bug (the .i/.ii/.f
151 preprocessed file), and, by storing it in an archive, you're just
152 making our volunteers' jobs harder. Only when your bug report requires
153 multiple source files to be reproduced should you use an archive. This
154 is, for example, the case if you are using INCLUDE directives in
155 Fortran code, which are not processed by the preprocessor, but the
156 compiler. In that case, we need the main file and all INCLUDEd files.
157 In any case, make sure the compiler version, error message, etc, are
158 included in the body of your bug report as plain text, even if
159 needlessly duplicated as part of an archive.
161 If you fail to supply enough information for a bug report to be
162 reproduced, someone will probably ask you to post additional
163 information (or just ignore your bug report, if they're in a bad day,
164 so try to get it right on the first posting :-). In this case, please
165 post the additional information to the bug reporting mailing list, not
166 just to the person who requested it, unless explicitly told so. If
167 possible, please include in this follow-up all the information you had
168 supplied in the incomplete bug report (including the preprocessor
169 output), so that the new bug report is self-contained.
171 Detailed bug reporting instructions for GNAT
173 See the [24]previous section for bug reporting instructions for GCC
174 language implementations other than Ada.
176 Bug reports have to contain at least the following information in
178 * the exact version of GCC, as shown by "gcc -v";
180 * the options when GCC was configured/built;
181 * the exact command line passed to the gcc program triggering the
182 bug (not just the flags passed to gnatmake, but gnatmake prints
183 the parameters it passed to gcc)
184 * a collection of source files for reproducing the bug, preferably a
185 minimal set (see below);
186 * a description of the expected behavior;
187 * a description of actual behavior.
189 If your code depends on additional source files (usually package
190 specifications), submit the source code for these compilation units in
191 a single file that is acceptable input to gnatchop, i.e. contains no
192 non-Ada text. If the compilation terminated normally, you can usually
193 obtain a list of dependencies using the "gnatls -d main_unit" command,
194 where main_unit is the file name of the main compilation unit (which
195 is also passed to gcc).
197 If you report a bug which causes the compiler to print a bug box,
198 include that bug box in your report, and do not forget to send all the
199 source files listed after the bug box along with your report.
201 If you use gnatprep, be sure to send in preprocessed sources (unless
202 you have to report a bug in gnatprep).
204 When you have checked that your report meets these criteria, please
205 submit it according to our [25]generic instructions. (If you use a
206 mailing list for reporting, please include an "[Ada]" tag in the
209 Detailed bug reporting instructions when using a precompiled header
211 If you're encountering a bug when using a precompiled header, the
212 first thing to do is to delete the precompiled header, and try running
213 the same GCC command again. If the bug happens again, the bug doesn't
214 really involve precompiled headers, please report it without using
215 them by following the instructions [26]above.
217 If you've found a bug while building a precompiled header (for
218 instance, the compiler crashes), follow the usual instructions
221 If you've found a real precompiled header bug, what we'll need to
222 reproduce it is the sources to build the precompiled header (as a
223 single .i file), the source file that uses the precompiled header, any
224 other headers that source file includes, and the command lines that
225 you used to build the precompiled header and to use it.
227 Please don't send us the actual precompiled header. It is likely to be
228 very large and we can't use it to reproduce the problem.
229 _________________________________________________________________
231 Frequently Reported Bugs in GCC
233 This is a list of bugs in GCC that are reported very often, but not
234 yet fixed. While it is certainly better to fix bugs instead of
235 documenting them, this document might save people the effort of
236 writing a bug report when the bug is already well-known.
238 There are many reasons why a reported bug doesn't get fixed. It might
239 be difficult to fix, or fixing it might break compatibility. Often,
240 reports get a low priority when there is a simple work-around. In
241 particular, bugs caused by invalid code have a simple work-around: fix
243 _________________________________________________________________
249 The export keyword is not implemented.
250 Most C++ compilers (G++ included) do not yet implement export,
251 which is necessary for separate compilation of template
252 declarations and definitions. Without export, a template
253 definition must be in scope to be used. The obvious workaround
254 is simply to place all definitions in the header itself.
255 Alternatively, the compilation unit containing template
256 definitions may be included from the header.
258 Bugs fixed in the 3.4 series
260 The following bugs are present up to (and including) GCC 3.3.x. They
261 have been fixed in 3.4.0.
263 Two-stage name-lookup.
264 GCC did not implement two-stage name-lookup (also see
267 Covariant return types.
268 GCC did not implement non-trivial covariant returns.
270 Parse errors for "simple" code.
271 GCC gave parse errors for seemingly simple code, such as
288 B b(A(),A(1)); // Variable b, initialized with two temporaries
289 B(A(2)).foo(); // B temporary, initialized with A temporary
290 return (A()); // return A temporary
293 Although being valid code, each of the three lines with a
294 comment was rejected by GCC. The work-arounds for older
295 compiler versions proposed below do not change the semantics of
298 The problem in the first case was that GCC started to parse the
299 declaration of b as a function called b returning B, taking a
300 function returning A as an argument. When it encountered the 1,
301 it was too late. To show the compiler that this should be
302 really an expression, a comma operator with a dummy argument
307 The work-around for simpler cases like the second one was to
308 add additional parentheses around the expressions that were
309 mistaken as declarations:
313 In the third case, however, additional parentheses were causing
314 the problems: The compiler interpreted A() as a function
315 (taking no arguments, returning A), and (A()) as a cast lacking
316 an expression to be casted, hence the parse error. The
317 work-around was to omit the parentheses:
321 This problem occured in a number of variants; in throw
322 statements, people also frequently put the object in
324 _________________________________________________________________
328 Fortran bugs are documented in the G77 manual rather than explicitly
329 listed here. Please see [29]Known Causes of Trouble with GNU Fortran
331 _________________________________________________________________
335 The following are not actually bugs, but are reported often enough to
336 warrant a mention here.
338 It is not always a bug in the compiler, if code which "worked" in a
339 previous version, is now rejected. Earlier versions of GCC sometimes
340 were less picky about standard conformance and accepted invalid source
341 code. In addition, programming languages themselves change, rendering
342 code invalid that used to be conforming (this holds especially for
343 C++). In either case, you should update your code to match recent
345 _________________________________________________________________
349 Problems with floating point numbers - the [30]most often reported
351 In a number of cases, GCC appears to perform floating point
352 computations incorrectly. For example, the C++ program
360 std::cout << (int)(a / b) << std::endl;
364 might print 50 on some systems and optimization levels, and 49
367 This is the result of rounding: The computer cannot represent
368 all real numbers exactly, so it has to use approximations. When
369 computing with approximation, the computer needs to round to
370 the nearest representable number.
372 This is not a bug in the compiler, but an inherent limitation
373 of the floating point types. Please study [31]this paper for
375 _________________________________________________________________
379 Casting does not work as expected when optimization is turned on.
380 This is often caused by a violation of aliasing rules, which
381 are part of the ISO C standard. These rules say that a program
382 is invalid if you try to access a variable through a pointer of
383 an incompatible type. This is happening in the following
384 example where a short is accessed through a pointer to integer
385 (the code assumes 16-bit shorts and 32-bit ints):
396 *(int *)a = 0x22222222; /* violation of aliasing rules */
398 printf("%x %x\n", a[0], a[1]);
402 The aliasing rules were designed to allow compilers more
403 aggressive optimization. Basically, a compiler can assume that
404 all changes to variables happen through pointers or references
405 to variables of a type compatible to the accessed variable.
406 Dereferencing a pointer that violates the aliasing rules
407 results in undefined behavior.
409 In the case above, the compiler may assume that no access
410 through an integer pointer can change the array a, consisting
411 of shorts. Thus, printf may be called with the original values
412 of a[0] and a[1]. What really happens is up to the compiler and
413 may change with architecture and optimization level.
415 Recent versions of GCC turn on the option -fstrict-aliasing
416 (which allows alias-based optimizations) by default with -O2.
417 And some architectures then really print "1111 1111" as result.
418 Without optimization the executable will generate the
419 "expected" output "2222 2222".
421 To disable optimizations based on alias-analysis for faulty
422 legacy code, the option -fno-strict-aliasing can be used as a
425 The option -Wstrict-aliasing (which is included in -Wall) warns
426 about some - but not all - cases of violation of aliasing rules
427 when -fstrict-aliasing is active.
429 To fix the code above, you can use a union instead of a cast
430 (note that this is a GCC extension which might not work with
448 printf("%x %x\n", u.a[0], u.a[1]);
452 Now the result will always be "2222 2222".
454 For some more insight into the subject, please have a look at
457 Cannot use preprocessor directive in macro arguments.
458 Let me guess... you used an older version of GCC to compile
459 code that looks something like this:
469 and you got a whole pile of error messages:
471 test.c:11: warning: preprocessing directive not recognized within macro arg
472 test.c:11: warning: preprocessing directive not recognized within macro arg
473 test.c:11: warning: preprocessing directive not recognized within macro arg
474 test.c: In function `foo':
475 test.c:6: undefined or invalid # directive
476 test.c:8: undefined or invalid # directive
477 test.c:9: parse error before `24'
478 test.c:10: undefined or invalid # directive
480 This is because your C library's <string.h> happens to define
481 memcpy as a macro - which is perfectly legitimate. In recent
482 versions of glibc, for example, printf is among those functions
483 which are implemented as macros.
485 Versions of GCC prior to 3.3 did not allow you to put #ifdef
486 (or any other preprocessor directive) inside the arguments of a
487 macro. The code therefore would not compile.
489 As of GCC 3.3 this kind of construct is always accepted and the
490 preprocessor will probably do what you expect, but see the
491 manual for detailed semantics.
493 However, this kind of code is not portable. It is "undefined
494 behavior" according to the C standard; that means different
495 compilers may do different things with it. It is always
496 possible to rewrite code which uses conditionals inside macros
497 so that it doesn't. You could write the above example
500 memcpy(dest, src, 12);
502 memcpy(dest, src, 24);
505 This is a bit more typing, but I personally think it's better
506 style in addition to being more portable.
508 Cannot initialize a static variable with stdin.
509 This has nothing to do with GCC, but people ask us about it a
516 will not compile with GNU libc, because stdin is not a
517 constant. This was done deliberately, to make it easier to
518 maintain binary compatibility when the type FILE needs to be
519 changed. It is surprising for people used to traditional Unix C
520 libraries, but it is permitted by the C standard.
522 This construct commonly occurs in code generated by old
523 versions of lex or yacc. We suggest you try regenerating the
524 parser with a current version of flex or bison, respectively.
525 In your own code, the appropriate fix is to move the
526 initialization to the beginning of main.
528 There is a common misconception that the GCC developers are
529 responsible for GNU libc. These are in fact two entirely
530 separate projects; please check the [33]GNU libc web pages for
532 _________________________________________________________________
536 Nested classes can access private members and types of the containing
538 Defect report 45 clarifies that nested classes are members of
539 the class they are nested in, and so are granted access to
540 private members of that class.
542 G++ emits two copies of constructors and destructors.
543 In general there are three types of constructors (and
546 1. The complete object constructor/destructor.
547 2. The base object constructor/destructor.
548 3. The allocating constructor/deallocating destructor.
550 The first two are different, when virtual base classes are
553 Global destructors are not run in the correct order.
554 Global destructors should be run in the reverse order of their
555 constructors completing. In most cases this is the same as the
556 reverse order of constructors starting, but sometimes it is
557 different, and that is important. You need to compile and link
558 your programs with --use-cxa-atexit. We have not turned this
559 switch on by default, as it requires a cxa aware runtime
560 library (libc, glibc, or equivalent).
562 Classes in exception specifiers must be complete types.
563 [15.4]/1 tells you that you cannot have an incomplete type, or
564 pointer to incomplete (other than cv void *) in an exception
567 Exceptions don't work in multithreaded applications.
568 You need to rebuild g++ and libstdc++ with --enable-threads.
569 Remember, C++ exceptions are not like hardware interrupts. You
570 cannot throw an exception in one thread and catch it in
571 another. You cannot throw an exception from a signal handler
572 and catch it in the main thread.
574 Templates, scoping, and digraphs.
575 If you have a class in the global namespace, say named X, and
576 want to give it as a template argument to some other class, say
577 std::vector, then std::vector<::X> fails with a parser error.
579 The reason is that the standard mandates that the sequence <:
580 is treated as if it were the token [. (There are several such
581 combinations of characters - they are called digraphs.)
582 Depending on the version, the compiler then reports a parse
583 error before the character : (the colon before X) or a missing
586 The simplest way to avoid this is to write std::vector< ::X>,
587 i.e. place a space between the opening angle bracket and the
590 Copy constructor access check while initializing a reference.
599 A(const A&); // private copy ctor
607 foo(A()); // error, copy ctor is not accessible
608 foo(makeA()); // error, copy ctor is not accessible
611 foo(a1); // OK, a1 is a lvalue
614 Starting with GCC 3.4.0, binding an rvalue to a const reference
615 requires an accessible copy constructor. This might be
616 surprising at first sight, especially since most popular
617 compilers do not correctly implement this rule.
619 The C++ Standard says that a temporary object should be created
620 in this context and its contents filled with a copy of the
621 object we are trying to bind to the reference; it also says
622 that the temporary copy can be elided, but the semantic
623 constraints (eg. accessibility) of the copy constructor still
626 For further information, you can consult the following
627 paragraphs of the C++ standard: [dcl.init.ref]/5, bullet 2,
628 sub-bullet 1, and [class.temporary]/2.
630 Common problems when upgrading the compiler
634 The C++ application binary interface (ABI) consists of two components:
635 the first defines how the elements of classes are laid out, how
636 functions are called, how function names are mangled, etc; the second
637 part deals with the internals of the objects in libstdc++. Although we
638 strive for a non-changing ABI, so far we have had to modify it with
639 each major release. If you change your compiler to a different major
640 release you must recompile all libraries that contain C++ code. If you
641 fail to do so you risk getting linker errors or malfunctioning
642 programs. Some of our Java support libraries also contain C++ code, so
643 you might want to recompile all libraries to be safe. It should not be
644 necessary to recompile if you have changed to a bug-fix release of the
645 same version of the compiler; bug-fix releases are careful to avoid
646 ABI changes. See also the [34]compatibility section of the GCC manual.
648 Remark: A major release is designated by a change to the first or
649 second component of the two- or three-part version number. A minor
650 (bug-fix) release is designated by a change to the third component
651 only. Thus GCC 3.2 and 3.3 are major releases, while 3.3.1 and 3.3.2
652 are bug-fix releases for GCC 3.3. With the 3.4 series we are
653 introducing a new naming scheme; the first release of this series is
654 3.4.0 instead of just 3.4.
658 With each release, we try to make G++ conform closer to the ISO C++
659 standard (available at [35]http://www.ncits.org/cplusplus.htm). We
660 have also implemented some of the core and library defect reports
662 [36]http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html &
663 [37]http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-defects.html
666 Non-conforming legacy code that worked with older versions of GCC may
667 be rejected by more recent compilers. There is no command-line switch
668 to ensure compatibility in general, because trying to parse
669 standard-conforming and old-style code at the same time would render
670 the C++ frontend unmaintainable. However, some non-conforming
671 constructs are allowed when the command-line option -fpermissive is
674 Two milestones in standard conformance are GCC 3.0 (including a major
675 overhaul of the standard library) and the 3.4.0 version (with its new
680 * The standard library is much more conformant, and uses the std::
681 namespace (which is now a real namespace, not an alias for ::).
682 * The standard header files for the c library don't end with .h, but
683 begin with c (i.e. <cstdlib> rather than <stdlib.h>). The .h names
684 are still available, but are deprecated.
685 * <strstream> is deprecated, use <sstream> instead.
686 * streambuf::seekoff & streambuf::seekpos are private, instead use
687 streambuf::pubseekoff & streambuf::pubseekpos respectively.
688 * If std::operator << (std::ostream &, long long) doesn't exist, you
689 need to recompile libstdc++ with --enable-long-long.
691 If you get lots of errors about things like cout not being found,
692 you've most likely forgotten to tell the compiler to look in the std::
693 namespace. There are several ways to do this:
694 * Say std::cout at the call. This is the most explicit way of saying
696 * Say using std::cout; somewhere before the call. You will need to
697 do this for each function or type you wish to use from the
699 * Say using namespace std; somewhere before the call. This is the
700 quick-but-dirty fix. This brings the whole of the std:: namespace
701 into scope. Never do this in a header file, as every user of your
702 header file will be affected by this decision.
706 The new parser brings a lot of improvements, especially concerning
708 * The "implicit typename" extension got removed (it was already
709 deprecated since GCC 3.1), so that the following code is now
710 rejected, see [14.6]:
712 template <typename> struct A
717 template <typename T> struct B
720 typename A<T>::X y; // OK
725 * For similar reasons, the following code now requires the template
728 template <typename> struct A
730 template <int> struct X {};
733 template <typename T> struct B
735 typename A<T>::X<0> x; // error
736 typename A<T>::template X<0> y; // OK
741 * We now have two-stage name-lookup, so that the following code is
742 rejected, see [14.6]/9:
744 template <typename T> int foo()
749 * This also affects members of base classes, see [14.6.2]:
751 template <typename> struct A
756 template <typename T> struct B : A<T>
758 int foo1() { return i; } // error
759 int foo2() { return this->i; } // OK
760 int foo3() { return B<T>::i; } // OK
761 int foo4() { return A<T>::i; } // OK
764 int foo5() { return j; } // OK
767 In addition to the problems listed above, the manual contains a
768 section on [38]Common Misunderstandings with GNU C++.
772 1. http://gcc.gnu.org/bugs.html
773 2. http://gcc.gnu.org/bugs.html#report
774 3. http://gcc.gnu.org/bugs.html#need
775 4. http://gcc.gnu.org/bugs.html#dontwant
776 5. http://gcc.gnu.org/bugs.html#where
777 6. http://gcc.gnu.org/bugs.html#detailed
778 7. http://gcc.gnu.org/bugs.html#gnat
779 8. http://gcc.gnu.org/bugs.html#pch
780 9. http://gcc.gnu.org/bugs.html#known
781 10. http://gcc.gnu.org/bugs.html#cxx
782 11. http://gcc.gnu.org/bugs.html#missing
783 12. http://gcc.gnu.org/bugs.html#fixed34
784 13. http://gcc.gnu.org/bugs.html#fortran
785 14. http://gcc.gnu.org/bugs.html#nonbugs
786 15. http://gcc.gnu.org/bugs.html#nonbugs_general
787 16. http://gcc.gnu.org/bugs.html#nonbugs_c
788 17. http://gcc.gnu.org/bugs.html#nonbugs_cxx
789 18. http://gcc.gnu.org/bugs.html#upgrading
790 19. http://gcc.gnu.org/bugs.html#known
791 20. http://gcc.gnu.org/bugzilla/
792 21. mailto:gcc-bugs@gcc.gnu.org
793 22. http://gcc.gnu.org/bugs.html#gnat
794 23. http://gcc.gnu.org/bugs.html#pch
795 24. http://gcc.gnu.org/bugs.html#detailed
796 25. http://gcc.gnu.org/bugs.html#where
797 26. http://gcc.gnu.org/bugs.html#detailed
798 27. http://gcc.gnu.org/bugs.html#detailed
799 28. http://gcc.gnu.org/bugs.html#new34
800 29. http://gcc.gnu.org/onlinedocs/g77/Trouble.html
801 30. http://gcc.gnu.org/PR323
802 31. http://www.validlab.com/goldberg/paper.ps
803 32. http://mail-index.NetBSD.org/tech-kern/2003/08/11/0001.html
804 33. http://www.gnu.org/software/libc/
805 34. http://gcc.gnu.org/onlinedocs/gcc/Compatibility.html
806 35. http://www.ncits.org/cplusplus.htm
807 36. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html
808 37. http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-defects.html
809 38. http://gcc.gnu.org/onlinedocs/gcc/C---Misunderstandings.html