[Bug libstdc++/54451] New: c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[Bug libstdc++/54451] New: c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[andris.pavenis at iki dot fi]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

             Bug #: 54451
           Summary: c++11/random.cc build failure when
                    _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: andris.pavenis@iki.fi


When _GLIBCXX_USE_C99_STDINT_TR1 defined in config.h neither of bits/random.h
and bits/random.tcc gets included and as result c++11/random.cc does not
compile:

random.cc:72:3: error: 'random_device' has not been declared

Noticed that when trying to build recent SVN version as Linux to DJGPP
cross-compiler. Tested versions 2012-08-30 and later.

Commenting out '#ifdef _GLIBCXX_USE_C99_STDINT_TR1' fixed build problem, but
I'm not sure that it is correct solution.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[rbmj at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

rbmj at verizon dot net changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rbmj at verizon dot net

--- Comment #1 from rbmj at verizon dot net 2012-09-09 02:36:26 UTC ---
I can confirm that the build fails as reported.

A bit of searching - turns out that acinclude.m4 has the following:

if test x"$glibcxx_cv_c99_stdint_tr1" = x"yes"; then
  AC_DEFINE(_GLIBCXX_USE_C99_STDINT_TR1, 1,
            [Define if C99 tyeps in <stdint.h> should be imported in
             <tr1/cstdint> in namespace std::tr1.])
fi

This is confirmed in include/tr1/cstdint:

#ifdef _GLIBCXX_USE_C99_STDINT_TR1

namespace std _GLIBCXX_VISIBILITY(default)
{
namespace tr1
{
  using ::int8_t;
  using ::int16_t;

etc.  However, this is also in include/c_global/cstdint:

#ifdef _GLIBCXX_USE_C99_STDINT_TR1

namespace std
{
  using ::int8_t;
  using ::int16_t;

etc.  Those are the only uses of it in code that I can find.  It seems like it
isn't exactly the best name for the define (it no longer just applies to TR1),
but it doesn't do too much.  I can't think of a case where this would not be
desired behavior (I don't remember, but I *think* that the C++ standard says
that those typenames should be in the standard namespace).

Anyway, it doesn't appear like removing that code will have any adverse
effects.

The relevant code is coming from revision 150312 (written by Paolo Carlini in
2009).  Since this just broke, I'm *guessing* that this is because of 4.8
moving more towards c++11, and because it was thought that this define was only
for tr1, it was removed from a default define set *somewhere* (I'm too scared
to venture too far into the build system...).  I can't seem to find any
evidence of this though... I know it worked ~3 months ago, but my git foo can't
seem to find the change.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |drepper.fsp at gmail dot
                   |                            |com

--- Comment #2 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-09-09 11:44:45 UTC ---
If it's not defined, it means that the configure time tests fails because the
target doesn't support the feature. In general - even for stdint which is
supported by most targets now - we should simply make sure that the build works
in both cases, simply the <random> features may end up not being available.

Adding Ulrich in CC, something needs tweaking in his recent work.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-09-09
     Ever Confirmed|0                           |1

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-09-09 11:50:18 UTC ---
(In reply to comment #0)
> Commenting out '#ifdef _GLIBCXX_USE_C99_STDINT_TR1' fixed build problem, but
> I'm not sure that it is correct solution.

It's not.

(In reply to comment #1)
> etc.  Those are the only uses of it in code that I can find.

It's used in many more headers.


>  It seems like it
> isn't exactly the best name for the define (it no longer just applies to TR1),
> but it doesn't do too much.  I can't think of a case where this would not be
> desired behavior (I don't remember, but I *think* that the C++ standard says
> that those typenames should be in the standard namespace).

But the C standard requires <stdint.h> and if the OS doesn't provide that we
can't provide a useful <cstdint>.

> Anyway, it doesn't appear like removing that code will have any adverse
> effects.

It would completely break libstdc++ on platforms without <stdint.h>, such as
djgpp

The fix is simply to check for the macro in random.cc

index 50b5359..1d0723d 100644
--- a/libstdc++-v3/src/c++11/random.cc
+++ b/libstdc++-v3/src/c++11/random.cc
@@ -24,6 +24,8 @@

 #include <random>

+#ifdef _GLIBCXX_USE_C99_STDINT_TR1
+
 #if defined __i386__ || defined __x86_64__
 # include <cpuid.h>
 #endif
@@ -142,5 +144,5 @@ namespace std _GLIBCXX_VISIBILITY(default)
     0xffffffffUL, 7,
     0x9d2c5680UL, 15,
     0xefc60000UL, 18, 1812433253UL>;
-
 }
+#endif

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[andris.pavenis at iki dot fi]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

--- Comment #4 from Andris Pavenis <andris.pavenis at iki dot fi> 2012-09-09 16:47:44 UTC ---
My test was done with DJGPP development version (2.04) only. It has stdint.h,
but it was recognized by configure as unusable due to bug unrelated to GCC
itself. After fixing this problem locally libstdc++-v3 builds OK for DJGPP
v2.04 (only tested cross-native build from Linux)

Stable version DJGPP v2.03 does not have stdint.h

As the result I agree with the suggestion in comment 3 to completely disable
random.cc when _GLIBCXX_USE_C99_STDINT_TR1 is not defined

Maybe it would be nice to use #error in header file to inform the user that
this feature is not supported for target system (nicer than to get linker error
later)

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-09-09 17:19:44 UTC ---
(In reply to comment #4)
> Maybe it would be nice to use #error in header file to inform the user that
> this feature is not supported for target system (nicer than to get linker error
> later)

You won't get a linker error, if the macro is not defined then the types in
<random> are not declared, so code using them won't compile.

Using #error would prevent #include <random> which I don't think is a good
idea. There's no harm including the header as long as you don't try to use
types which require a working <stdint.h>

The fix is to simply make random.cc check the macro. This is what we already do
in future.cc and thread.cc and mutex.cc and other files too.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[rbmj at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

--- Comment #6 from rbmj at verizon dot net 2012-09-09 18:08:11 UTC ---
Making local changes to bring stdint.h into compliance works for me as well.

(In reply to comment #5)
> (In reply to comment #4)
> > Maybe it would be nice to use #error in header file to inform the user that
> > this feature is not supported for target system (nicer than to get linker error
> > later)
> 
> You won't get a linker error, if the macro is not defined then the types in
> <random> are not declared, so code using them won't compile.

However, you will get weird "std::random_device" not declared errors, which
will seem strange to any end user.

> Using #error would prevent #include <random> which I don't think is a good
> idea. There's no harm including the header as long as you don't try to use
> types which require a working <stdint.h>

Agreed.  But could we use #warning to tell the user what's happening (if
portability is an issue we can check for __GNUC__) without going through the
source?

> The fix is to simply make random.cc check the macro. This is what we already do
> in future.cc and thread.cc and mutex.cc and other files too.

Also true.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-09-09 18:18:32 UTC ---
No, it should be consistent with how it's handled everywhere else.

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl.tools at gmail dot com

--- Comment #8 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-09-14 09:44:23 UTC ---
*** Bug 54576 has been marked as a duplicate of this bug. ***

[Bug libstdc++/54451] c++11/random.cc build failure when _GLIBCXX_USE_C99_STDINT_TR1 is not defined in config.h

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54451

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0

--- Comment #9 from H.J. Lu <hjl.tools at gmail dot com> 2012-09-14 11:44:20 UTC ---
Fixed by

http://gcc.gnu.org/ml/gcc-bugs/2012-09/msg01086.html