Re: [all-arch][PATCH 00/19] Fix -Wundef warning in SHLIB_COMPAT

[Siddhesh Poyarekar <siddhesh at redhat dot com>]

On Fri, Aug 22, 2014 at 04:01:21PM +0000, Joseph S. Myers wrote:
> Do you mean some previous patch in the series broke those exports and 
> patch 20/20 restored them?  The exports appear in libc.abilist so I 
> presume are currently working as they should, or at least were working 
> recently.  General principles of bisectability mean things should be 
> designed to stay working after each patch rather than patch 20 fixing 
> something broken by an earlier patch.

I don't think any patches in my series broke it, but I haven't done a
per-patch analysis for all architectures; I only did that for x86_64.
For everything else I did comparisons at master, 5/20 and 20/20.  In
any case, I'm redoing those bits to reduce the impact on other bits.

> (I think this series is far too big and risky to go in during the freeze, 
> as illustrated by these architecture-specific issues being found; it would 
> much better go in after the release.  It's possible the *other* -Wundef 
> warnings have simple enough fixes to go in during the freeze.)

The other warning fix (i.e. the _POSIX_* stuff) looks like it'll be
complicated as well.  The new warning I discovered on one of the other
architectures (i.e. ABI_libm_GLIBC_2_19 and similar) may also need a
design rethink if simply using ifdef is not enough.

Another way to look at these issues would be that it's quite easy to
find issues resulting from these changes, so a bit of quick iteration
ought to close it out for 2.20.  However, I think Roland's concern was
that these warnings shouldn't stay untouched because there wasn't
enough of a push to fix them.  Given that I have committed to them and
also made some progress, maybe that concern is now moot.

Siddhesh

Attachment: pgplSBXuGSfto.pgp
Description: PGP signature