This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: RFC: Generate lib-names.h and stubs.h from ABI-LIST
On Thu, 24 May 2012, H.J. Lu wrote:
> On Thu, May 24, 2012 at 3:53 PM, Joseph S. Myers
> <joseph@codesourcery.com> wrote:
> > On Thu, 24 May 2012, H.J. Lu wrote:
> >
> >> > See what I suggested in
> >> > <http://sourceware.org/ml/libc-alpha/2012-05/msg01022.html>: keying
> >> > shlib-versions contents off the name of a variant rather than a triplet,
> >> > since once they are in sysdeps directories they'll no longer need the
> >> > triplet name as it stands at present.
> >>
> >> How do you pick which variant to use? Where will the
> >> selection be made?
> >
> > You can still use sysdeps makefiles to determine the default variant and
> > the conditions for each variant. ?(I suppose that
> > sysdeps/unix/sysv/linux/x86/Makefile would list the variants and the
> > conditions for each one, but .../{i386,x86_64/x32,x86_64/64}/Makefile
> > would each list a setting for the default variant.)
>
> If I understand it correctly, that is close to what I suggested:
>
> The only issue is ld=soname. I can try to remove the soname part
> and use abi-ld-soname in CPU/Makefile instead.
>
> Did I miss something?
I am suggesting an extra level of indirection. All the sonames live in
shlib-versions files, as now. The shlib-versions files for a
configuration need to include the sonames for every ABI variant supported
by installed headers for that configuration (hence using /x86/ directories
for them, for example). The key in shlib-versions files becomes an ABI
variant name rather than a pattern matching triplets.
Whatever approach is used, it is desirable for each soname to be specified
in exactly one place - meaning in a file in a /x86/ directory rather than
a file in a directory only used at all for a single ABI.
--
Joseph S. Myers
joseph@codesourcery.com