Re: Bug#369396: Regression: problems with macro calling AC_PROG_CC and AC_C_BIGENDIAN

[Ralf Wildenhues]

* Ralf Wildenhues wrote on Mon, May 29, 2006 at 05:38:12PM CEST:
> * Mikael Magnusson wrote on Mon, May 29, 2006 at 05:06:13PM CEST:
> > 
> > If you call AC_PROG_CC and AC_C_BIGENDIAN in a macro, the checks for
> > unistd.h and some other headers will fail. See attached test case.

OK, this is ugly.

AC_C_BIGENDIAN requires (through some indirect macro chain) both
AC_PROG_CC and AC_LANG_PREPROC(C).  Now, since you've explicitly called
AC_PROG_CC before, that requirement is already deemed fulfilled by the
stacking algorithm.  The AC_LANG_PREPROC(C) requirement is not yet
fulfilled, so that macro is expanded right before the expansion of
TEST_INIT.  Boing: AC_PROG_CC will be expanded after it.  :-(

One solution is to write the AC_PROG_CC as a requirement, so that it
will be expanded outside, like this:

AC_DEFUN([TEST_INIT],[
AC_REQUIRE([AC_PROG_CC])
echo test2
AC_C_BIGENDIAN
echo test3
])

echo 2
TEST_INIT
echo 3
AC_CHECK_HEADER([unistd.h], [], [AC_MSG_ERROR([unistd.h not available?])])

AC_OUTPUT

Cheers,
Ralf