Re: PATCH: Move include/stubs-biarch.h to sysdeps/generic

[Roland McGrath <roland at hack dot frob dot com>]

I think some cleanup of this stuff would be fine.  I definitely would
prefer to see the cleanup done first, with an eye to making sure it is
adequate for the x32 case, and then related x32 bits going in only when
they can be done the clean new way.

I really don't recall my original rationale for having a single
shlib-versions file with regexps in it rather than letting that
information accumulate from sysdeps files like everything else does.

Anyway, this can be done piecemeal.

I don't really see why we have scripts/data/ at all.  Why don't we just
have c++-types.data and localplt.data files in sysdeps directories?
Then we have no existing use for abi-name at all.

AFAICT the only way in which sparc is actually special for the abi-name
issue is that there is part of $cpu that isn't relevant.  That's
actually true of others like i?86 and sh*, but those win now because
base_machine is right while for sparc it doesn't distinguish 32/64.
So if it were sysdeps-based, they'd all be equally fine.

The biarch stuff (both gnu/stubs.h and gnu/lib-names.h) can be handled
with new generic support for multi-variant generated headers.

Then perhaps we can change the shlib-versions scheme to be sysdeps-based
rather than pattern-based.

There might be some merit in having a layer of explicit ABI moniker to
avoid slightly different configurations accidentally winding up with
slightly different ABI details.  But it needs more consideration.


Thanks,
Roland