This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Remove configuration name patterns from shlib-versions
- From: Roland McGrath <roland at hack dot frob dot com>
- To: "Joseph S. Myers" <joseph at codesourcery dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Thu, 11 Sep 2014 15:03:38 -0700 (PDT)
- Subject: Re: Remove configuration name patterns from shlib-versions
- Authentication-results: sourceware.org; auth=none
- References: <Pine dot LNX dot 4 dot 64 dot 1409111633270 dot 17850 at digraph dot polyomino dot org dot uk>
I've forgotten about whatever change made this not break the multi-variant
(e.g. biarch) cases for generating gnu/lib-names.h. It appears that now
operates by assuming that most SONAMEs will be the same for all the ABI
variants, and requires that abi-variant-libfoo-soname variables be set in
sysdeps/.../Makefile for the ones that vary. I'm not particularly happy
with that. There should only be one flavor of file where SONAME strings
appear in the sources.
I think we need a more thoroughgoing overhaul of the shlib-versions
picture. I believe we already have consensus that all such things should
be organized purely via the sysdeps/ hierarchy rather than with any config
tuple matching. So this change could well be an appropriate incremental
step along the path to the clean overhaul. But I'd prefer to see a design
for that end state before we just whittle away with incremental changes
like this one. As is, it just feels to me like it's making something
baroque slightly differently baroque (and perhaps moreso), and not clearly
easier to grok than the status quo that (some) people already grok.
As to the tunnel-vision review of the patch per se, the only problem I see
is that the comments in shlib-versions still describe things in ways that
don't really make sense any more, e.g. "loose patterns at the end of the
list can give defaults."
Thanks,
Roland