The quadigraph removal didn't fix this.

I also thought of a broken sed problem. But aclocal and autom4te are
perl scripts on theses versions and don't use sed. Another thing that
cannot be the culprit.

With the '-I m4' things disabled and copying the contents of
m4/reentrant.m4 directly into the end of acinclude.m4 and removing
m4/reentrant.m4 before invoking aclocal still triggers the problem.
This scenario shows that somehow aclocal parses 4 times the
CURL_CHECK_NEED_REENTRANT_* macros in acinclude.m4 see:
http://curl.haxx.se/auto/log.cgi?id=20080730012824-29500

Oops on hte previous run it parses them three times, see:
http://curl.haxx.se/auto/log.cgi?id=20080729225738-27769

And in previous one only two times.
http://curl.haxx.se/auto/log.cgi?id=20080729211738-19764

And in previous only one:
http://curl.haxx.se/auto/log.cgi?id=20080729201738-13942

At those runs buildconf was supposed to append m4/reentrant.m4 to
acinclude.m4. I suppose that no CVS checkout was taking place and
that would explain the increased concatenation. But it was also
supposed to have removed m4/reentrant.m4 in first instance.
http://cool.haxx.se/cvs.cgi/curl/buildconf.diff?r1=1.55&r2=1.56

I can't explain that behaviour. Well, in any case...

First run http://curl.haxx.se/auto/log.cgi?id=20080729201738-13942
equally fails. Notice that the "found macro
CURL_CHECK_NEED_REENTRANT_" lines all mention acinclude.m4 and not
m4/reentrant.m4

So I reinstated the proper way of including m4/reentarnt.m4 with the
'-I m4' things.

I suppose that you have no way to knowing if all tests passed when
building gm4 1.4.5 ?

And if you build your own gm4 1.4.5 ?

By bag of tricks is exhausted for this, the same as gm4 1.4.5 memory ;-)

If it helps you to further investigate this:

* buildconf calls aclocal, which calls autom4te, which calls m4.
* aclocal is supposed to be installed with automake.
* autom4te is supposed to be installed with autoconf.

What about reinstalling autoconf and or automake while keeping gm4 1.4.5 ?

And trying another machine?

The number of 'seen' macros, the ones actually used is 165. And the
number of 'found' macros, the ones scanned in all packages, is 232. I
don't think this are too big numbers. As a matter of fact this works
on other systems.

HTH

-- 
-=[Yang]=-