glibc 2.2.3 + autoconf 2.52

Adam J. Richter adam@yggdrasil.com
Sun Jul 29 20:48:00 GMT 2001


	I don't see any discussion of this in the libc-alpha or
libc-hacker archives, so I'll just mention this in case the
information might be helpful.  Also, I'm a little unclear
if this belong in libc-alpha, libc-hackers or bug-glibc.  I'll
start here, and, if anyone wants to direct me to a different
forum, please feel free to do so.

	I've started trying to get glibc-2.2.3 to build with autoconf-2.52
(after autoconf has regenerated ./configure, et al).  I have attached
a patch with how far I have gotten so far, in case someone might fine
it useful to look at.  It should not be applied for reasons I will
discuss below.

	First, the stuff that might be correct are the patches to
the top level configure.in and aclocal.m4 files.  There are a few
places where autoconf-2.52 seems to need a few more things to be
quoted, and that is what these changes do.

	The part that I am sure is wrong are the changes to the
other configure.in files in sysdeps/.  The primary problem with the
original files  was that the top level configure script wants to
source a bunch of configure script "fragments" in sysdeps/, and,
later, during "make all", these script "fragments" are also
run as independent shell scripts.  Autoconf-2.52 does not like this,
because it needs some initialization that is done in AC_INIT to be
done in each of the configure.in files in sysdeps/ in order to get the
definitions of some macros to do the configure.in --> configure
creation.  Unfortunately, other parts of AC_INIT do things like
clearing variables and erasing files, so just putting
AC_INIT(configure.in) in the top of each configure.in in sysdeps/.  So,
when the top level configure script sources these subordinate scripts as
"fragments" the would clear the results of a bunch of tests if they
were changed to begin with AC_INIT.

	My changes to the configure.in files in sysdeps/ is to have them
call just the parts of AC_INIT that they seem to need.  Unfortunately,
they seem to need AC_LANG_PUSH(C), and, for some reason that I do
not understand, adding a matching AC_LANG_POP(C) at the bottom of
these scripts causes autoconf to complain.  I am looking into this.

	I think that, ultimately, the correction solution is probably
for these configure.in scripts to become separate standalone scripts,
which the top level configure script could call relatively early in the
configure process (invoking each a separate program with something like
AC_OUTPUT_SUBDIRS).  Then, the top level configure script could fish
out the results that it needs from files generated by these subordinate
configure scripts.

	Anyhow, I have attached a diff of what I have so far.  If
anyone else is working on this, I'd be interested in knowing to avoid
duplication of effort.

-- 
Adam J. Richter     __     ______________   4880 Stevens Creek Blvd, Suite 104
adam@yggdrasil.com     \ /                  San Jose, California 95129-1034
+1 408 261-6630         | g g d r a s i l   United States of America
fax +1 408 261-6631      "Free Software For The Rest Of Us."


More information about the Libc-alpha mailing list