autoconf-2.52 and configure.in "fragments"

[Adam J. Richter]

        glibc (at least glibc-2.2.3 through current glibc-cvs) maintains
a bunch of little configure "fragment" files in subdirectories that are
sourced by the parent configure script.  Each of these configure
"fragments" is generated from a matching configure.in by running
autoconf individually on that "fragment" configure.in.

        Many thanks to Roland McGrath for correcting my original
misunderstanding of this situation.

        The problem is that autoconf-2.52 seems to require a number
of definitions that are normally created by calling AC_INIT, but
having each of these "fragment" configure.in files call AC_INIT
results in the answers to previous configuration tests done by
the parent configure script being forgotten, presumably because AC_INIT
clears some variables or files.

        I would like to propose that AC_INIT be restructured like so:

AC_DEFUN(AC_INIT, [
        AC_BEGIN_FRAGMENT($1)
        AC_ERASE_ANSWERS
        ])

AC_DEFUN(AC_BEGIN_FRAGMENT, [
         ...code to AC_REQUIRE or otherwise define whatever macros are needed...
         AC_LANG_PUSH(C)
        ])

AC_DEFUN(AC_ERASE_ANSWERS, [
        ...current code that clears files and variables that AC_INIT clears...
])

AC_DEFUN(AC_END_FRAGMENT, [
        AC_LANG_POP(C)
        ])

        Under this scheme, configure.in "fragments" could be portably written
with

        AC_BEGIN_FRAGMENT(configure.in)
        ...
        AC_END_FRAGMENT


        and the main configure.in script, would still start with
AC_INIT(configure.in).

        Am I duplicating someone else's effort?  If not, does this
look good?  Should I develop a patch?

Adam J. Richter     __     ______________   4880 Stevens Creek Blvd, Suite 104
address@hidden     \ /                  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."