Re: 13-m4-undivert-culprit.patch

[Akim Demaille]

| > testsuite.log:
| > @@ -0,0 +1,3 @@
| > +/usr/local/bin/m4: configure.in: 3401: Non-numeric argument to `divert': 
NOTICE
| > +/usr/local/bin/m4: configure.in: 3401: Non-numeric argument to `divert': 
NOTICE
| > +/usr/local/bin/m4: configure.in: 3401: Non-numeric argument to `divert': 
NOTICE
| > 
| > Aaaarg :(  Yet it starts again: people use internals without asking.
| > 
| > I understand there can be needs we don't fulfilf, but you have to tell
| > us.  I would have provided Libtool with a _macro_, definitely not a
| > diversion name.
| 
| Mae Culpa.  I seem to remember this was based on some stuff I had been
| carrying around for a couple of years that I pasted into libtool.m4 a
| while back without thinking carefully enough.  Sorry about that.

No problem :)

| > Anyway, now that I see what it is used for, it is clear to me that
| > pseudo `echo' is something that M4sh must provide (and as a matter of
| > fact I had removed NOTICE precisely to introduce M4SH-INIT, a very
| > early diversion used by M4sh to output its initialization stuff before
| > Autoconf's).  So that diversion name will no longer be needed.  And I
| > guess there are many other sh idioms to move from Libtool to M4sh.
| 
| Okay.  I think that libtool's requirement for a non-backslash
| intepreting echo is quite unusual -- are you planning to migrate that
| functionality from libtool.m4 into m4sh.m4?

Yes, I do.  It is not unusual and actual Autoconf is in bad need for
such a feature.  For instance we know that our smally heredocs for
AC_DEFINE and so on are extremely costly.  But we can't take any
risks, hence heredocs.  With a good echo, we're done, and then it
speeeeeeeeeeeeeds again.

There were impressive figures about GCC's configure slowness on some
arch precisely because of these heredocs.