bug#13514: unintentional regression with AC_CONFIG_MACRO_DIR in 1.13.1

[Pavel Raiskup]

>>> [..skip..]
>>> Maybe the current error should be demoted to a warning, to allow us to
>>> be more forward-compatible with such possible future scenarios?
>>
>> I vote for this approach.
>>
>I'm starting to be convinced myself that it is the best one.
>
> But then again, mulling over the issue a little longer, I've come to realize
> that autoreconf takes care of re-executing aclocal after (and if) libtoolize
> is run, which makes my previous concerns about the "wrong execution order"
> mostly moot, and might give a valid use case for not having an error nor a
> warning if the "primary" m4 macro dir doesn't exist.
>
> So, what should we do in the end?

As we probably don't want to run 'aclocal --install' for the first run
during 'autoreconf' (there could be unintentional file substitutions of m4
files in general, or are there other problems?), (1) I would let there
warning in aclocal and (2) I would create this directory by autoreconf
because this tool specifies where libtool should put .m4 files.
(There is not possible to not run aclocal without '-I m4' as this is
specified in Makefile.am by ACLOCAL_AMFLAGS).  Eric, would you agree with
this solution?  Should I try to prepare patch for this?

I would let the warning there just because of older autoreconfs.  And
for situations when user specifies bad ACLOCAL_AMFLAGS inclusions (to
inform him).

Pavel