This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: libstdc++-v3 and LIBCXXFLAGS
- To: Carlo Wood <carlo at alinoe dot com>
- Subject: Re: libstdc++-v3 and LIBCXXFLAGS
- From: Benjamin Kosnik <bkoz at redhat dot com>
- Date: Thu, 24 May 2001 12:33:15 -0700 (PDT)
- cc: Alexandre Oliva <aoliva at redhat dot com>, libstdc++ at gcc dot gnu dot org
Carlo a lot of engineers are busy with various bits right now: you'll
have to give us a bit of slack, and have some patience.
> 1) libstdc++ cores when creating std::ios_base::Init when
> locale.cc is compiled with -fdefer-pop -fomit-frame-pointer.
> This is probably a bug in gcc itself.
> The reason I ran into that is:
> 2) It's very unclear how to dictate the C[XX]FLAGS that need to be
> used for compiling the compiler and the library seperately.
> The way that I used to do this doesn't work anymore.
> Using --enable-debug doesn't work because I want to use
> -fomit-frame-pointer in the CFLAGS that is being used for
> the compiler files (not that I succeeded in that *frustrated*).
> 3) When using --enable-debug --enable-cxx-flags='-fno-omit-frame-pointer'
> then libstdc++ ends up with references to undefined functions
> (which turn out to be defined as inline in headerfiles).
Part of the problem is that this is likely not a library bug, but an
optimization issue related to either the c++ front end (say, file under
g++) or a backend issue (say file under optmization). The rule is if you
can compile the library with compiler sources and have it work, and then
change your compiler sources or compiler flags and the library no longer
works, then it's likely not a library issue per se.
Because of this, your time would be best spent trying to reduce your
testcases so that no standard library files are included, and it becomes
a pure compiler issue. Compiler hackers and optimization people will be
more willing to look at this without wading through 10k of includes.
thanks again for your patience,
benjamin