Re: AC_FUNC_* cross-compilation guesses

[Bruno Haible]

Ralf Wildenhues wrote:
> > In particular, it's bad to guess "no" for embedded glibc
> > systems because that increases the size of compiled executables.
> 
> But that holds for Autoconf's AC_FUNC_STRNLEN as well.  Shouldn't it
> have a more optimistic cross compilation guess, too?

Autoconf has traditionally used a pessimistic approach to cross compilation
guesses. If the AC_FUNC_STRNLEN macro in Autoconf was to use a more precise
cross compilation guess, there would indeed be no need to duplicate this code
in gnulib. Here is a proposed patch.


2010-08-28  Bruno Haible  <address@hidden>

        AC_FUNC_STRNLEN: more realistic cross-compilation guess
        * lib/autoconf/functions.m4 (AC_FUNC_STRNLEN): Require
        AC_CANONICAL_HOST. When cross-compiling, guess it works everywhere
        except on AIX.

--- lib/autoconf/functions.m4.orig      Sat Aug 28 15:05:43 2010
+++ lib/autoconf/functions.m4   Sat Aug 28 15:04:59 2010
@@ -1673,6 +1673,7 @@
 AN_FUNCTION([strnlen], [AC_FUNC_STRNLEN])
 AC_DEFUN([AC_FUNC_STRNLEN],
 [AC_REQUIRE([AC_USE_SYSTEM_EXTENSIONS])dnl
+AC_REQUIRE([AC_CANONICAL_HOST]) dnl for cross-compiles
 AC_CACHE_CHECK([for working strnlen], ac_cv_func_strnlen_working,
 [AC_RUN_IFELSE([AC_LANG_PROGRAM([AC_INCLUDES_DEFAULT], [[
 #define S "foobar"
@@ -1692,7 +1693,11 @@
 ]])],
               [ac_cv_func_strnlen_working=yes],
               [ac_cv_func_strnlen_working=no],
-              [ac_cv_func_strnlen_working=no])])
+              [# Guess no on AIX systems, yes otherwise.
+               case "$host_os" in
+                 aix*) ac_cv_func_strnlen_working=no;;
+                 *)    ac_cv_func_strnlen_working=yes;;
+               esac])])
 test $ac_cv_func_strnlen_working = no && AC_LIBOBJ([strnlen])
 ])# AC_FUNC_STRNLEN