PATCH: Re: Do we really want to use i586 code for i686?

[Roland McGrath]

> You seem to ignore my bug report:
> 
> http://sources.redhat.com/ml/libc-alpha/2001-11/msg00052.html

No, I just don't agree with your conclusions.  That is a bug in
linuxthreads that should be fixed.  One way to fix it would be to move the
contents of linuxthreads/sysdeps/i386/i586/ to a differently-named subdir
that is referenced in linuxthreads/sysdeps/i386/i586/Implies and
linuxthreads/sysdeps/i386/i586/i686/Implies.

Another way would be to add a different flavor of Implies file (or a marker
within the file, or whatever) for the Implies-within-add-on semantics.
linuxthreads has two Implies files.  For
linuxthreads/sysdeps/unix/sysv/linux/Implies, the current behavior is
right, and your changes would break (e.g.) the aio code.  For
linuxthreads/sysdeps/i386/i586/, the current behavior does not match the
author's apparent intent.

> My point is no addon should change libc. Pthread seems to be a special
> case. But it should be treated as such.

The sole purpose of add-ons is to change libc.
Something that is truly separate belongs in its own package.