Jim Meyering wrote:
> While the dependency of AC_EGREP_CPP on AC_PROG_EGREP
> is unavoidable, there are others that are indeed unnecessary,
> since they serve only to work around a bug in autoconf-2.57,
> which is no longer relevant, since we (gnulib-tool) now
> requires 2.59 or newer.
> ...
> Now we require at least 2.59, and the work-around induced an
> unwarranted circular dependency in grep's build process.
> The following .m4 files pulled in the AC_PROG_EGREP requirement,
> yet $EGREP was never used, and configuring grep on a system without
> an existing grep program would thus fail unnecessarily.
> * m4/alloca.m4 (gl_FUNC_ALLOCA): Remove work-around.
> * m4/dirfd.m4 (gl_FUNC_DIRFD): Likewise.
> * m4/mmap-anon.m4 (gl_FUNC_MMAP_ANON): Likewise.
> * m4/strtoimax.m4 (gl_FUNC_STRTOIMAX): Likewise.
> * m4/strtoumax.m4 (gl_FUNC_STRTOUMAX): Likewise.
> Prompted by the report, http://savannah.gnu.org/bugs/?31646
> by Florin Iucha.
There are three separate issues here:
- The AC_REQUIRE([AC_PROG_EGREP]) in gnulib macros.
These should not be needed any more since autoconf 2.54, because the
definition of AC_EGREP_CPP includes this requirement already. It is
possible that the bug that was observed back then (in 2002 or earlier)
was due to wrong handling of AC_REQUIRE, but that was also fixed in
autoconf by Eric about two years ago.
- The AC_REQUIRE([AC_PROG_CPP]) in gnulib macros.
This _probably_ can removed as well, because AC_EGREP_CPP does
AC_LANG_PREPROC_REQUIRE() which should require AC_LANG_PREPROC(C),
and the latter is defined as
AC_DEFUN([AC_LANG_PREPROC(C)],
[AC_REQUIRE([AC_PROG_CPP])])
- The issue of the bug reporter is unrelated. It is simply a consequence
of how AC_PROG_EGREP is implemented. But any use of AC_EGREP_CPP needs
AC_PROG_EGREP anyway.
So, your patch is OK, but the ChangeLog entry and commit message should
IMO be rewritten to not mention Florin Iucha's bug report.
Bruno
