Re: [PATCH] Make AC_COMPUTE_INT public

[Ralf Wildenhues]

Hi Stepan,

* Stepan Kasal wrote on Tue, Jun 20, 2006 at 12:45:37PM CEST:
> 
> My personal preference is to put the documenting of this feature off,
> too.  We are very close to a release, and I'm not sure whether we
> want to commit to this interface.  Maybe I'm just overcautions.

My thinking is: there are a dozen packages out there _actively using_
_AC_COMPUTE_INT already.  It's bad that third-parties use internal
macros.  And this usage won't go away just because we're not documenting
the interface.

I don't have a strong opinion either way here.  But the above tells
me it's better to have a defined interface.  Plus, we've (Paul has)
agreed on having this published, several months ago.  (OTOH, surely
there's an argument for punishing Paolo for his bad timing... ;-)

> Is the AU_ALIAS always safe?  What if a project contains
> AC_COMPUTE_INT in its aclocal.m4?

Then they are well into undefined country.  Same as with any other
Autoconf macro names a third party may have overridden.

But sure, in practice that would be a problem.

> I underrstand that Ralf has concerns about m4-2.0 (and wants the
> chocolate), but I would be happy if even this was postponed and 2.60
> were released right now.

I have already applied your patch.  I thought long about it (which
clearly doesn't mean anything, really), I decided I like it clearly
better than the other suggestions.  I could not find a package that
it breaks.  And I will probably bite and create 2.61 if it does.

I will post a patch for the datarootdir issue (which is in effect a
regression) and the APR issue (which isn't but would enable the APR
people to stay away from using internal interfaces), and let others
decide on whether they should be applied or not.

Cheers, and thank you, Stepan, for pushing in the right direction!
Ralf