This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: PowerPC LE configury
- From: Alan Modra <amodra at gmail dot com>
- To: Roland McGrath <roland at hack dot frob dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Thu, 3 Oct 2013 11:40:35 +0930
- Subject: Re: PowerPC LE configury
- Authentication-results: sourceware.org; auth=none
- References: <20130809050118 dot GG3294 at bubble dot grove dot modra dot org> <20130830232101 dot 0FAEF2C060 at topped-with-meat dot com> <20130916055932 dot GA15467 at bubble dot grove dot modra dot org> <20130924230825 dot EFE462C099 at topped-with-meat dot com> <20130928110037 dot GD30589 at bubble dot grove dot modra dot org> <20131001180641 dot AF1C02C08E at topped-with-meat dot com>
On Tue, Oct 01, 2013 at 11:06:41AM -0700, Roland McGrath wrote:
> > On Tue, Sep 24, 2013 at 04:08:25PM -0700, Roland McGrath wrote:
> > > That has some of the essential characteristics of what I want to see,
> > > but details like the abilist variable are too kludgey.
> > >
> > > I think it makes most sense to first follow through with the ideas Joseph
> > > and I discussed previously about revamping shlib-versions. Then the new
> > > scheme for abilist files will be a straightforward derivation of that.
> >
> > Please reconsider. You're asking me to implement changes to the build
> > machinery that affect all targets. That really is the purview of a
> > build or global maintainer. While I might be able to do the work
> > you're asking of me, I know how this dance goes. Invariably the poor
> > schmuck (that'd be me) who is asked to implement some maintainer's
> > vision for the future gets it all wrong. After some spins around the
> > review loop, the maintainer is left wondering why he didn't just
> > implement the feature himself.
>
> I was certainly not asking you to implement this all by yourself.
Thanks, I can live with that. What I'll do is commit this part
* configure.in: Map powerpc64le and powerpcle to base_machine/machine.
* configure: Regenerate.
* nptl/shlib-versions: Powerpc*le starts at 2.19.
* shlib-versions: Likewise.
and leave out the abilist files and the Makefile abilist changes that
you don't like. This will mean "make check" failures for LE, but at
least people will be able to build glibc from the official sources for
powerpc64le-linux.
Actually, I think I made a mistake in choosing 2.19 as the base
version. At the time I was thinking that 2.19 will be the first
official release for powerpc64le glibc, but we are going to be using
glibc before that point. So if interfaces change between now and the
official 2.19 release, we won't be able to provide compatibility
functions that use current interfaces. Thus the versioning ought to
start at 2.18.
Does that all sound reasonable?
Note to anyone already using powerpc64le glibc, I'd suggest keeping
shlib-versions as you have it at the moment, since changing base
symbol version will require recompiling the world. We're going to
have one of those moments soon anyway, when we change over to the new
ABI without function descriptors.
--
Alan Modra
Australia Development Lab, IBM