Re: AC_CHECK_FUNCS always succeeds on AIX with Autoconf 2.56

[Martin Frydl]

Hi,

  I've looked into this problem and created patch. In fact I've only 
put back the original code while maintaining the current one, i.e. `f' 
is assigned both in global initialization and also in main(). I've tried 
this fix on Linux with gcc 2.96, Sun with gcc 2.95 and SunCC 7.0, HP-UX 
with gcc 3.0.4 and AIX with gcc 3.1.1 and IBM C/C++ 3.6.6. I'm not able 
to check it with unbundled cc A.05.36 for ia64 so someone probably 
should check it (this configuration is why the old test was changed).

        * lib/autoconf/c.m4 (AC_LANG_FUNC_LINK_TRY(C)): Put f=$1 also
        to main function to avoid optimizations on AIX.

  Regards
    Martin

P.S. Please reply directly to my e-mail since I'm not in 
autoconf-patches list.

Martin Frydl wrote:
> Hello,
>
>   I've put AC_CHECK_FUNCS(stricmp) to configure.in and it always 
> succeeds on AIX with both IBM compiler and gcc. The test looks something 
> like this:
>
> char strncasecmp ();
> char (*f) () = strncasecmp;
>
> int main () {
>   return 0;
> }
>
>   When this is compiled and linked, no error is reported. But when test 
> from Autoconf 2.52 is used, everything works (i.e. reports error since 
> stricmp is not available):
>
> char stricmp ();
> char (*f) ();
>
> int main () {
>   f = stricmp;
>   return 0;
> }
>
>   It looks like the first test is somehow "optimized" by compiler and 
> thus is linkable. What about putting back the original test? Why was it 
> changed?
>
>   My configuration:
>     AIX 4.3.3
>     IBM C/C++ 3.6.6
>     gcc 3.1.1
>
>   Thanks
>     Martin Frydl
? aix-func-check.patch
Index: c.m4
===================================================================
RCS file: /cvsroot/autoconf/autoconf/lib/autoconf/c.m4,v
retrieving revision 1.172
diff -u -r1.172 c.m4
--- c.m4        13 Nov 2002 23:34:07 -0000      1.172
+++ c.m4        19 Nov 2002 12:24:36 -0000
@@ -170,6 +170,9 @@
 # unbundled cc A.05.36 for ia64 under +O3, presumably on the basis that
 # there's no need to do that store if the program is about to exit.
 #
+# Assignment f=$1 put also to main() because compilers on AIX seem to suffer
+# similar optimization if only global assignment is used.
+#
 m4_define([AC_LANG_FUNC_LINK_TRY(C)],
 [AC_LANG_PROGRAM(
 [/* System header to define __stub macros and hopefully few prototypes,
@@ -194,6 +197,15 @@
 #ifdef __cplusplus
 }
 #endif
+],
+[/* The GNU C library defines this for functions which it implements
+    to always fail with ENOSYS.  Some functions are actually named
+    something starting with __ and the normal name is an alias.  */
+#if defined (__stub_$1) || defined (__stub___$1) 
+choke me
+#else 
+f = $1;
+#endif 
 ])])