This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Possible issue with hurd: Fix abi-tag, following ba90e05
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Samuel Thibault <samuel dot thibault at gnu dot org>
- Cc: Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>, "GNU C. Library" <libc-alpha at sourceware dot org>
- Date: Mon, 9 Feb 2015 18:31:08 +0000
- Subject: Re: Possible issue with hurd: Fix abi-tag, following ba90e05
- Authentication-results: sourceware.org; auth=none
- References: <54D8F6F6 dot 9010309 at linux dot vnet dot ibm dot com> <20150209180920 dot GW3105 at type dot bordeaux dot inria dot fr>
On Mon, 9 Feb 2015, Samuel Thibault wrote:
> Hello,
>
> Adhemerval Zanella, le Mon 09 Feb 2015 16:05:42 -0200, a écrit :
> > Since ae20c9a17cd95b6f367209c2a9075587616ceb42 my powerpc builds can not load
> > older binaries using the new loader,
>
> Ergl.
>
> So I guess it's actually ba90e05052ce57db51e3cb18978614fd0db5c7ef which
> was overly optimistic about possibly removing the gnu -> gnu-gnu
> transformation. Joseph, do you know the details?
Maybe - the comment referred to shlib-versions, and the libc-abis uses
were unaffected, but actually this transformation was significant for
host_os / config-os uses in abi-tags as well.
I think the right fix would be to move abi-tags away from a shared file -
instead of taking the first matching entry from a shared file, take the
unique matching entry from the first abi-tags file found in sysdeps
directories (so you'd have sysdeps/unix/sysv/linux/abi-tags and
sysdeps/mach/hurd/abi-tags, missing the present first column, and the
toplevel file might just have comments documenting the format and the
allocated OS numbers but not being used to determine which applies to the
current glibc build).
But if the comment about "earlier entries taking precedence over later
entries" is accurate, .*-.*-gnu.* should have no effects on any system for
which .*-.*-linux.* matches.
(Of course, a simpler fix might be to restore the host_os transformation
but with a more accurate comment.)
--
Joseph S. Myers
joseph@codesourcery.com