Re: libstdc++/3174: sparc-rtems (also seen on i386-rtems)

[Joel Sherrill <joel dot sherrill at OARcorp dot com>]

Benjamin Kosnik wrote:
> 
> Hmm. I still have the feeling that you and I are looking at different
> pieces of code.

I have switched to gcc-3.0-0614 and with a clean slate, this problem
has indeed gone away.  Thanks. PC3174 can be closed.

I now have the problem about selecting a threading model.  I have
posted this question separately to gcc-bugs.  The configury
changed a lot after 0527 and our old method of getting posix threads
is broken.
 
> Here's what I am looking at:
> 
> dnl
> dnl Check to see if g++ can compile this library, and if so, if any version-
> dnl specific precautions need to be taken.
> dnl
> dnl GLIBCPP_CHECK_COMPILER_VERSION
> AC_DEFUN(GLIBCPP_CHECK_COMPILER_VERSION, [
> if test ! -f stamp-sanity-compiler; then
>   AC_MSG_CHECKING([for g++ that will successfully compile libstdc++-v3])
>   AC_LANG_SAVE
>   AC_LANG_CPLUSPLUS
>   AC_TRY_COMPILE(, [
>   #if __GNUC__ < 3
>     not_ok
>   #endif
>   ], gpp_satisfactory=yes, AC_MSG_ERROR([please upgrade to GCC 3.0 or
> above]))
>   AC_LANG_RESTORE
>   AC_MSG_RESULT($gpp_satisfactory)
>   touch stamp-sanity-compiler
> fi
> ])
> 
> This is called after GLIBCPP_CONFIGURE(.) in configure.in.
> GLIBCPP_CONFIGURE correctly sets CXX, which AC_TRY_COMPILE then uses to
> try this test.
> 
> It's my understanding that CPP is not even an issue.
> 
> Again, this is working quite well for me for x86-x-powerpc-eabism crosses
> on a system with a /usr/bin/g++ that is not 3.0. I have no problems. I
> suspect that either the rtems bits are doing something funky, you are
> trying to build in previously-configured trees, or I there is something I
> really don't get and that you've failed to explain to me.
> 
> Please make sure we are talking about the same codebase. Also, can you
> try completely removing your build directory's (target)/libstdc++-v3
> directory and starting before starting the top-level make?
> 
> thanks,
> benjamin
> 
> > Remember the problem is not the version test itself -- it is the fact
> > that the
> > wrong CPP is being used -- the native instead of the cross one.  All other
> > tests use the cross compiler.

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985