[PATCH] Makefile.am: don't confuse CFLAGS and CFLAGS_FOR_BUILD

[Thomas De Schampheleire]

Hi Dmitry,

On Wed, Dec 7, 2016 at 4:18 PM, Dmitry V. Levin <ldv at altlinux.org> wrote:
> On Wed, Dec 07, 2016 at 02:58:35PM +0300, Dmitry V. Levin wrote:
>> On Tue, Dec 06, 2016 at 09:06:08PM +0100, Thomas De Schampheleire wrote:
>> > On Wed, Nov 16, 2016 at 8:27 PM, Dmitry V. Levin wrote:
>> [...]
>> > > The correct fix is going to be more complex, e.g.
>> > > - move all gl_WARN_ADD/WARN_CFLAGS related code from configure.ac to a
>> > >   separate m4 macro, say st_WARN_CFLAGS;
>> > > - call st_WARN_CFLAGS in configure.ac;
>> > > - modify AX_PROG_CC_FOR_BUILD to
>> > >   + pushdef WARN_CFLAGS to WARN_CFLAGS_FOR_BUILD,
>> > >   + call st_WARN_CFLAGS,
>> > >   + popdef WARN_CFLAGS back,
>> > >   + AC_SUBST WARN_CFLAGS_FOR_BUILD;
>> > > - add WARN_CFLAGS_FOR_BUILD to AM_CFLAGS_FOR_BUILD.
>> >
>> > Thanks for the suggestion.
>> > I tried implementing it but got stuck. What I observe is that the
>> > second invocation of ST_WARN_CFLAGS is using cached results, which
>> > should not happen because the compiler can be different. Looking at
>> > the definition of gl_WARN_ADD, I think that the cache_id of the
>> > AC_CACHE_CHECK call in gl_COMPILER_OPTION_IF should contain CC or a
>> > similar variable, so that caching does not happen.
>> >
>> > I'm sending what I have for now below.
>>
>> Yes, it appeared to be even more complicated than I expected.
>> I've pushed a tentative fix to
>> https://github.com/strace/strace/commits/ldv/WARN_CFLAGS_FOR_BUILD
>> Please give it a try.
>>
>> I'm not quite happy with the change I made in m4/warnings.m4, though.
>> The change of gl_Flags is just an ad hoc solution, and the tearing
>> gl_UNKNOWN_WARNINGS_ARE_ERRORS off gl_WARN_ADD is not something
>> I'd like to accept on gnulib side.
>
> I've pushed there a less hackish change to m4/warnings.m4; if the whole
> thing works for you, I'm fine with pushing it on to master.

I tested commit 045228095418e2cc1469021ffa46cc9d7faa6f02 in the
failure case and confirm that it works fine now: the configure script
takes two passes for the warning flags, and sees that the host
compiler gcc 4.1 does not respect the flags, while the more recent
target compiler does.

That commit was the last version, right?

Thanks,
Thomas