Re: RFC: Generate lib-names.h and stubs.h from ABI-LIST

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Thu, May 24, 2012 at 2:43 PM, Roland McGrath <roland@hack.frob.com> wrote:
> It's odd that everything says "abi-list" instead of just "abi".
> It was "syscall-list" because bits/syscall.h is a list of syscalls.
> Nothing we generate involves a list of ABIs.

I will change it to abi.

> Some of the comments need cleanup both for typos/grammar and for
> mention of stuff specific to the Linux syscall list where they should
> just talk about the generic purpose of generated headers that need
> different contents for each ABI.

I will update and I can use some help here.

> I don't like the *-ld-soname variables and also don't see why they're
> needed. ?If you need those, then why don't you need one for every
> library? ?Nothing should assume that all the other sonames match
> across all the related ABIs--there's no reason they necessarily should.
>
> Anyway, that information should not be repeated in makefiles. ?It's
> already in shlib-versions and everything should just get it from
> there. ?The makefiles can give for each ABI the tuple to match in
> shlib-versions, akin to the ABI entries in shlib-versions.

The problem is x86-64 build won't see sysdeps/x86_64/x32/shlib-versions
and x32 build won't see sysdeps/x86_64/64/shlib-versions.  i386 build
won't see neither.  How does it work?

> A branch with intermediate changes and merge commits is slightly
> annoying to review. ?When a change actually goes in, it should be one
> clean commit. ?Please clean up the change and post it here as a single
> patch in the usual form for further review.
>

I will do that.

Thanks.

-- 
H.J.