From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Fri, 15 Oct 2010 10:49:00 GMT
Subject: Re: [PATCH] compat: add memrchr()
Message-ID: <AANLkTi=YN41n-BnVNo3HsnxNxQNBX=Ev-upmM0N49uOZ@mail.gmail.com>
In-Reply-To: <20101015060654.GD21830@burratino>

On Fri, Oct 15, 2010 at 06:06, Jonathan Nieder <jrnieder@gmail.com> wrote:
> Ævar Arnfjörð Bjarmason wrote:
>
>> Maybe something like this for configure.ac:
>>
>>     AC_CHECK_LIB([c], [memchr],
>>     [HAVE_MEMRCHR=YesPlease],
>>     [HAVE_MEMRCHR=])
>>     AC_SUBST(HAVE_MEMRCHR)
>>
>> And documentation with the other HAVE_* variables at the top of the
>> Makefile?
>
> Hmm, the BSDs and plan 9 have an memrchr() apparently.   Any idea for
> taking advantage of that (the makefile support part) that's less ugly
> than this?

Check out my "Makefile & configure: add a NO_FNMATCH flag" and
"Makefile & configure: add a NO_FNMATCH_CASEFOLD flag" patches already
in pu.

Maybe it should be a NO_* flag since that's what we use when we expect
sanity, e.g. we have NO_REGEXP=, NO_CURL and NO_GETTEXT.

Then you just need NO_MEMRCHR=UnfortunatelyYes entries in the big
if/else block for those platforms that don't have it.

But memrchr() is a GNU extension so it should probably be a HAVE_*. I
don't know.

> If we miss a platform, that's no big deal.  The 1-char-at-a-time
> loop is not so slow, and the "#define memrchr gitmemrchr" ensures
> that it would not conflict with the libc version.

Yup, probably best as a HAVE_*. In any case the autoconf probe I added
should work when memrchr is spelled properly.