Re: Autoconf and gnu-m4 dependency

[Akim Demaille <akim at epita dot fr>]

>>>>> "Marc" == Marc Espie <Marc.Espie@liafa.jussieu.fr> writes:

Hi Marc,

Marc> I believe this to be of benefit to everybody... Apart from
Marc> copyright issues, which autoconf people probably don't really
Marc> care about (or probably not in the same direction that we do),
Marc> this does mean that autoconf is robust enough to work with two
Marc> distinct implementations of m4. Since my reimplementation of
Marc> patsubst and friends is completely different from gnu-m4, it is
Marc> less likely that autoconf will depend on `undocumented features'
Marc> of gnu-m4 in the future.

First of all, I am expressing only my own feelings, and I'm not
representative of the maintenance committee.

Personally, although I perfectly understand your motivations, I have
no desire to move to another m4 than GNU m4.  We already have to spend
an incredible amount of time to face the variation that exist on
users' side, and I think this is way enough.

Supporting another m4 means that we regularly have to check that
Autoconf behaves properly on both implementations, and have to face
the differences that might arise.

Sometimes we depend upon very precise features of GNU m4, such as the
layout of --trace.  In the future, we might even rely on the modules
to improve both the robustness and the speed of Autoconf.

So yes, we do depend upon non documented features of GNU m4, but it is
precisely because we know quite well our tool, and know it is that
single one which will be used.

I understand how my point of view is open to criticism, but I'm tired
enough of the pressure downstream to want to avoid anything upstream.

Now, of course, if the committee has a different opinion...

        Akim