[Python-Dev] Python build trouble with the new gcc/binutils

Zack Weinberg zack@codesourcery.com
Sat, 17 Aug 2002 21:11:31 -0700

On Sat, Aug 17, 2002 at 02:43:33PM -0400, David Abrahams wrote:
> From: "Zack Weinberg" <zack@codesourcery.com>
> > On Fri, Aug 16, 2002 at 12:38:24PM -0400, Andrew Koenig wrote:
> > >
> > > #0  __register_frame_info_bases (begin=0xfed50000, ob=0xfed50000,
> tbase=0x0,
> > >     dbase=0x0) at /tmp/build1165/gcc-3.1.1/gcc/unwind-dw2-fde.c:83
> >
> > Er, is the directory name misleading, or have you picked up
> > libgcc_s.so from 3.1.1?  In theory that shouldn't be a problem; in
> > practice it could well be the problem.
> I'm still ploughing through several days of messages here (so this may have
> been discussed already) but I have recently learned that despite the
> existinence of a "-V" option, it has long been impossible to correctly
> install new versions of GCC on systems with existing versions without
> using --prefix= to select a unique location. Why GCC's configure doesn't
> issue a warning about this when you do it wrong, I don't know. The only
> clue that this is going to be a problem was buried in a FAQ somewhere as of
> three months ago.

I believe that this has been addressed in the current documentation --
the INSTALL file clearly states that multiple versions shouldn't be
installed in the same prefix, and the -V option (which never worked
properly) has been removed.  Would you mind reading the docs shipped
with 3.2, and reporting any remaining confusion as a bug?

(note that we're not going to add a configure check as you suggest,
because there are conditions where it's safe, and we don't want to
make life harder for people who really do want that.)