bug#25359: autoreconf no longer suffices

[Glenn Morris <rgm@gnu.org>]

Eli Zaretskii wrote:

>> I think it is wrong to make autogen.sh compulsory in this way.
>
> Why do you think it's wrong?

Perhaps it's just my opinion. It was the original intent of autogen.sh
to be just a convenience wrapper for sanity checks + autoreconf.

If this is no longer true, here are some places that need to be changed:

1) autogen.sh error output says: "If you know that the required versions
are in your PATH, but this script has made an error, then you can simply
run autoreconf -fi -I m4 instead of this script."

(INSTALL.REPO used to say this too, before the IMO unrelated a3f989d
removed it.)

2) make-dist with update=yes uses autoreconf

3) top-level Makefile needs updating as per bug#25372

4) autogen.sh should ensure the outputs are newer than the inputs.
nt/gnulib.mk is only tested for existence, not freshness.
Since we use autoreconf -f, I think you can simply unconditionally
create nt/gnulib.mk every time.

All in all, it would make lifer simpler if plain automake would still
work, but I can see it might be difficult to implement.

>> (Obviously the missing file is irrelevant anyway on non MS-Windows
>> platforms.)
>
> Evidently, it isn't irrelevant, otherwise the error would have never
> happened, right?

Eg "touch nt/gnulib.mk" works just as well on non-MS platforms, where
the contents of the file are completely irrelevant.