After hacking the configure script to work around the issues http://bugs.python.org/issue4025 and http://bugs.python.org/issue1633863
the build still fails:

building 'fcntl' extension
xlc_r -DNDEBUG -O -I. -I/home/u0006584/Python-2.6/./Include -I. -
IInclude -I./Include -I/home/u0006584/Python-2.6/Include -
I/home/u0006584/Python-2.6 -c /home/u0006584/Python-
2.6/Modules/fcntlmodule.c -o build/temp.aix-6.1-
2.6/home/u0006584/Python-2.6/Modules/fcntlmodule.o
./Modules/ld_so_aix xlc_r -bI:Modules/python.exp build/temp.aix-6.1-
2.6/home/u0006584/Python-2.6/Modules/fcntlmodule.o -o build/lib.aix-6.1-
2.6/fcntl.so
ld: 0711-317 ERROR: Undefined symbol: .flock
ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more
information.
*** WARNING: renaming "fcntl" since importing it failed: 0509-022
Cannot load module build/lib.aix-6.1-2.6.
0509-026 System error: A file or directory in the path name does
not exist.
error: No such file or directory
make: 1254-004 The error code from the last command is 1.

(Full transcript attached)

I'm mystified and I can't log in to the AIX box right now.

This does not appear to be the same as http://bugs.python.org/issue1756343 and http://bugs.python.org/issue1694442 which both appear to be the same,
but documenting a differen AIX build problem.

This occurs because it must link with -lbsd on AIX 6.1. If you hand
link the failing module by adding -lbsd it builds quietly.

If you build the prerequisites for the _tkinter module on AIX 6.1, you
will also encounter an error becuase this module is not linked with the
-lXext library. Again, you can hand link it with that library.

Notice that the modules ignore these two libraries when they are passed
to the python build. The modules should build against the libraries
that are mentioned in the --with-libs=LIBS argument to python configure.

I will eventually figure out how to modify setup.py so that these two
failures no longer occur on AIX 6.1.

This also happens with Python 2.5.2 (not the latest 2.5 series, I know)
on AIX 6.1.

I have now confirmed that the fix described here1 works as desired. I
don't know if this is proper or not but it matches what inkblotter said.

This very same problem happens (with Python-2.6.2) on AIX5.3 now too, after upgrading to:
$ oslevel -s
5300-08-09-1013

Unlike before (comparing with old build logs), this AIX5.3 now provides flock() in <sys/file.h> and libbsd.a[shr.o] like AIX6.1.

Interesting enough, /usr/lib/libbsd.a contains 32bit shared objects only, so -lbsd does not help in 64bit mode (don't know if python actually supports 64bit on AIX). I don't have an AIX6.1 to check this.

Because of this, upgrading "checking for flock" from compile- to link-check (eventually trying -lbsd a second time) might help?

Ohw, looking on another machine, being AIX5.3 TL6:
$ oslevel -s
5300-06-00-0000

Here flock() is provided in libbsd.a(shr.o) (32bit only) too, but it isn't declared in any header-file. So that recent AIX5.3 patchset just adds the flock() declaration, not the function itself (is already there).

Hi,

Here is a patch that solves this problem. It was tested with Python 2.6.6 on AIX 6.1.

The same problem applies to Python 2.7 and 3.x, but since the syntax has been changed in configure.in for Python 2.7 and 3.x, I need to adapt a little bit my patch (coming soon).

regards

Here is the version of the patch for Python 2.7.
I had to change AC_TRY_COMPILE to AC_COMPILE_IFELSE and AC_TRY_LINK to AC_LINK_IFELSE.

The syntax in configure.in is the same between Python 2.7 and the trunk. It is also the same between Python 2.6.6 and Python 3.1.2.

So this patch can be applied on the trunk and the patch Python-2.6.6_flock_AIX.diff can be applied on Python 3.1.2.

While I've not tested these patches myself, I do think there still is an unhandled case: building a 64bit python.
As libbsd.a contains 32bit objects only, there is no 64bit flock().
So AC_CHECK_LIB(bsd,flock) may still fail, which should result in have_flock=no then.
Thank you!

[rant: grrr, AIX is really a crappy platform; no consistency between releases; there are dozens of obvious bugs like that... anyway...]

The fact that there is no 64 bits objects in libbsd.a on AIX 6.1 is a documented bug:
http://www-01.ibm.com/support/docview.wss?uid=isg1IZ45155

They recommend to install patch APAR IZ45155.

I suppose there is nothing more we can do if the function is not available on the system. Maybe we could display a warning or error message and suggest to upgrade to a more recent patchlevel if we are on AIX and flock cannot be found?

Though if the installation is relatively up to date, there should be no problem.

Well, I suppose if flock needs libbsd on AIX (which is quite weird in itself), and AIX fails to ship libbsd for some executable formats, then it's AIX's problem.

While I do agree this being an AIX bug, it is not a blocker here:
fcntl extension does not _depend_ on flock, it just does _prefer_ flock:
If not available, fcntl extension uses something else (fcntl IIRC), as it did (even without linking libbsd) before the AIX5.3-patch, which just added flock to the headers (the implementation in libbsd already has been there before the patch).
The problem after that AIX5.3-patch was that the compile-check found flock, and then fcntl extension failed to link due to missing libbsd.
So it's just for how to detect flock, where the most safe variant is both the compile- and link-check. When one of them fails, flock should not be flagged as available and the alternative gets used.

Here is a new test for flock. HAVE_FLOCK is defined if we can link a C application calling flock, or if flock is defined in libbsd.
FLOCK_NEEDS_LIBBSD is also defined in the second case.

AC_MSG_CHECKING(for flock)
have_flock=no
AC_TRY_LINK([
#include "confdefs.h" 
#include <sys/file.h>
], [void* p = flock; flock(0, 0)],
  [AC_CHECK_LIB(bsd,flock, [
    AC_DEFINE(FLOCK_NEEDS_LIBBSD, 1, Define if flock needs to be linked with bsd library.)
    have_flock=yes
    ])],
  [have_flock=yes]
)
if test "$have_flock" = yes ; then
  AC_DEFINE(HAVE_FLOCK, 1, Define if you have the 'flock' function.)
fi
AC_MSG_RESULT($have_flock)

I think that this new test would also cover your case with a broken AIX libbsd? [I haven't actually tried to compile it yet, if the behavior is OK I will test it in both autoconf formats and provide some new patches tomorrow]

I've just merged Sébastien's patches in r84584 (3.x), r84585 (2.7) and r84586 (3.1). If you'd like to improve the fixes (as per your latest message), please say so, otherwise I'll close the issue.

(and thanks for your contributions!)

Sorry to be pedantic - but it looks like (don't have a build environment at hand at the moment) it will fail to build as 64bit on the not-so-up-to-date AIX with missing 64bit flock implementation:
There, the compile-check will succeed and define HAVE_FLOCK, but the link-check will fail - which really should reset have_flock to 'no'.
IMO, HAVE_FLOCK should be defined when _both_ checks succeed only, not upon the compile-check only as it is done now.
Thank you!

Sorry to be pedantic - but it looks like (don't have a build
environment at hand at the moment) it will fail to build as 64bit on
the not-so-up-to-date AIX with missing 64bit flock implementation:
There, the compile-check will succeed and define HAVE_FLOCK, but the
link-check will fail - which really should reset have_flock to 'no'.
IMO, HAVE_FLOCK should be defined when _both_ checks succeed only, not
upon the compile-check only as it is done now.

It might be better indeed. Patches against the current SVN branches are
welcome (I don't have an AIX system to test them on, and besides I'm not
an autoconf expert).

Using this patch based on current py3k branch I've been able now to build the fcntl module for 32bit aix5.3 (with flock) and 64bit aix5.3 (without flock), both using an old gcc-3.3.6, as well as for 64bit linux.
If necessary, I should be able to create a similar patch for the other branches tomorrow.

I inverted the actions in the test I proposed in msg115768 (quick note from home, should have tested at work before posting). It should be:

AC_MSG_CHECKING(for flock)
have_flock=no
AC_TRY_LINK([
#include "confdefs.h" 
#include <sys/file.h>
], [void* p = flock; flock(0, 0)],
  [have_flock=yes],
  [AC_CHECK_LIB(bsd,flock, [
    AC_DEFINE(FLOCK_NEEDS_LIBBSD, 1, Define if flock needs to be linked with bsd library.)
    have_flock=yes
    ])]
)
if test "$have_flock" = yes ; then
  AC_DEFINE(HAVE_FLOCK, 1, Define if you have the 'flock' function.)
fi
AC_MSG_RESULT($have_flock)

Mickael, does that new test works for you? I am not sure I follow your explanation: there is no more 'compile-check' (AC_TRY_COMPILE) in the new test I provided; just a 'link-check' (AC_TRY_LINK - which tests that compilation and link did work).
HAVE_FLOCK is undefined by default.
If the link-check succeeds, then HAVE_FLOCK is defined.
If it fails, then we try to look for flock in libbsd; if that test works, then HAVE_FLOCK is defined as well as FLOCK_NEEDS_LIBBSD.

In the case of broken AIX 64 bits, the link-check will fail (no flock in default C libraries); then the test to find flock in libbsd will also fail (no 64 bits objets in libbsd). So HAVE_FLOCK will be undefined, which is what you expect.

m4/autoconf is quite unintuitive so I may have made a mistake, or it may be easily misunderstood. But I intend my test to define HAVE_FLOCK only if compilation _and_ linking works, like you expect.

Sébastien, while this gives expected results on the AIX box here, it still has one subtle problem:
AC_TRY_LINK may fail due to missing declaration in <sys/file.h> _or_ due to missing implementation in libc.
The subsequent AC_CHECK_LIB won't fail when the implementation is in libbsd but the declaration is missing - this is the case on not-so-recent AIX5.3 (and the reason python worked there before).

Another minor nit is configure's output when libbsd is needed:

checking for flock... checking for flock in -lbsd... yes
yes

Anyway - I've hacked up another variant (with recent autoconf manual at hand), which really should cover all known situations now (additionally using cache-checks):

AC_CACHE_CHECK([for flock declaration], [ac_cv_flock_decl],
[AC_COMPILE_IFELSE(
[AC_LANG_PROGRAM(
[#include <sys/file.h>],
[void* p = flock]
)],
[ac_cv_flock_decl=yes],
[ac_cv_flock_decl=no]
)
])
if test "x${ac_cv_flock_decl}" = xyes; then
AC_CHECK_FUNCS(flock,,
AC_CHECK_LIB(bsd,flock,
[AC_DEFINE(HAVE_FLOCK)
AC_DEFINE(FLOCK_NEEDS_LIBBSD, 1, Define if flock needs to be linked with bsd library.)
])
)
fi

Thank you anyway!

Thanks Michael, your new test looks fine to me.

I wanted to test it on my AIX system with branch py3k but I had some other issues (bpo-9799), but I will test it tomorrow.

I tested Michael test in msg115868 on my AIX system (AIX 6.1 with flock correctly present in libbsd) with py3k and it works fine:

checking for flock declaration... yes
checking for flock... no
checking for flock in -lbsd... yes

[103/344] test_fcntl
= OK

I attach it as a patch against py3k.

Ok, thank you. The improvement has been committed in all three branches (3.2, 3.1, 2.7).