Re: [patches] Allow ports to define the version of libgcc_s

[Maxim Kuvyrkov <maxim@xxxxxxxxxxxxxxxx>]

Carlos O'Donell wrote:
> On Mon, Aug 17, 2009 at 7:17 AM, Maxim Kuvyrkov<maxim@xxxxxxxxxxxxxxxx> wrote:
> > The attached patch makes another attempt on GLIBC's PR4457
> > (http://sourceware.org/bugzilla/show_bug.cgi?id=4457).
> >
> > This patch is not exactly what Ulrich proposed to do about the issue; it
> > doesn't detects the version of libgcc_s.so installed, but rather relies on
> > shlib-versions files to specify it.  The entry in top-level shlib-versions
> > file specifies the version of libgcc_s to be '1' (what most architectures
> > have) and shlib-versions in sysdeps directories can override that.  This
> > patch sets libgcc_s' version to 2 for m68k and hppa architectures.
> >
> > FWIW, making GLIBC to autodetect the version of libgcc_s installed does not
> > seem trivial to me.
>
> There will always be multiple versions of libgcc_s.so installed, and
> the one to use for a new glibc is always the most recent version e.g.
> `ls -a /lib/libgcc_s.so* | tail -1`

This will not work for cases when libgcc_s.so installed in a different 
sysroot.  It is necessary to detect the very libgcc_s.so, which GCC 
links against; this can be done, e.g., by a configure script building an 
example program which calls, say, _Unwind_GetCFA() and then checking how 
GCC invokes the linker.  Maybe there's a simpler way, though... 
Suggestions?

...

> > +hppa-.*-.*              libgcc_s=2
>
> This shold be set to "4". Thanks.

HPPA is fast ;).  Thanks for checking this.

--
Maxim