homepage

update to autoconf2.65

see http://mail.python.org/pipermail/python-dev/2010-April/099639.html 

I don't think Python 2.7 should upgrade to a newer autoconf version at this point. For 3.2, we could try it out.

here is the diff for configure.in. The __EXTENSIONS__ macro is defined by 2.65 for Solaris, removed this explicit definition. running a test build

r80474: Replace AC_HELP_STRING with AS_HELP_STRING

r80475: s/AC_AIX/AC_USE_SYSTEM_EXTENSIONS/

r80478 converts all obsolete AC_TRY_* macros to AC_*_IFELSE, the outcome is ... nothing (but whitespace changes):
$ svn diff --diff-cmd diff -x -uEwB configure
Index: configure
===================================================================
--- configure	(Revision 80475)
+++ configure	(Arbeitskopie)
@@ -1,5 +1,5 @@
 #! /bin/sh
-# From configure.in Revision: 80474 .
+# From configure.in Revision: 80475 .
 # Guess values for system-dependent variables and create Makefiles.
 # Generated by GNU Autoconf 2.65 for python 3.2.
 #
running configure with the old and the new version shows no diffs (besides config.log).

r80481: configure.in: Avoid autoconf warning: Assume C89 semantics that
RETSIGTYPE is always void (issue #8510).
Keep the definition, although the python code itself doesn't use it anymore.

r80483: Makefile.pre.in (autoconf): Call autoconf/autoheader with -Wall to help the configure script to stay warning free.

You don't need to add source in case of 
 AC_COMPILE_IFELSE([AC_LANG_SOURCE([[]])] .. 
when the the test is ..main() { return 0 ;} ...
, posted long time ago as part of issue3754 ( 	minimal cross-compilation support for configure ).
Also thanks for fixing extra comma in after replacement of AC_TRY_COMPILE .
This part is save to be added to 2.7 build.

> You don't need to add source in case of 
> AC_COMPILE_IFELSE([AC_LANG_SOURCE([[]])] .. 
> when the the test is ..main() { return 0 ;} ...
is there a version of the cross configure patch for 3.x?

the update to 2.65 on the py3k branch only did show whitespace changes, applying these changes to the trunk.
r80964: require autoconf-2.65

r80965: Replace AC_HELP_STRING with AS_HELP_STRING
r80966: s/AC_AIX/AC_USE_SYSTEM_EXTENSIONS/
r80969: convert all obsolete AC_TRY_* macros to AC_*_IFELSE
r80970: Avoid autoconf warning: Assume C89 semantics that RETSIGTYPE is always void

r81004: Remove extra closing bracket and comma introduced in r80969. (This was causing misdetection of the OS X 10.5 SDK on Linux and OS X, and a test_platform failure on OS X.)

I don't have access to source so my comment will be based only on diffs from above mentioned revisions.
r80969:
- AC_LANG_PROGRAM ... with main function in body . It is legal as I don't know C compiler that fail to compile code like ...int main() { int main() {...} ...} but is better to left body empty. Counted three times. All those cases tests for compiler flags and I think that AC_LANG_SOURCE is enough.
- body include #include "confdefs.h" - may be buggy (before) and now AC_LANG_PROGRAM will add all defines. Counted any times. I think that is save to remove #include "confdefs.h" from all those test cases.

Roumen: good catch! Indeed, nested functions aren't legal in standard C, and the test for -fno-strict-aliasing now fails on my OS X 10.6.3 machine, where it used to pass:
 checking whether gcc accepts -fno-strict-aliasing... no
config.log contains:
 configure:5400: checking whether gcc accepts -fno-strict-aliasing
 configure:5418: gcc -fno-strict-aliasing -c -g -O2 conftest.c >&5
 conftest.c: In function 'main':
 conftest.c:27: error: nested functions are disabled, use -fnested-functions to re-enable
 configure:5418: $? = 1
 configure: failed program was:
 ...
Fixed these cases above and removed the extra '#include "confdefs.h"' lines (which seem harmless, but unnecessary) in r81077 (trunk).

Merged to py3k, and fixed up an additional nested 'int main()', in r81078.

BTW, it looks as though the nested functions were introduced in r76030; nothing to do with Matthias's autoconf 2.65 update.

Just double checked that r80969 didn't introduce any other non-whitespace changes: apart from the now-fixed OS X 10.5 SDK issue), the only other non-whitespace change was:
@@ -7018,7 +7023,7 @@
 int
 main ()
 {
-long double x; x = (long double)0.;
+long double x; x = (long double)0;
 ;
 return 0;
 }
which looks harmless to me.

Closing. Roumen, please open a new issue for any other configure issues that aren't related to this autoconf update.