The previous implementation used execv(2) to run the real interpreter, which means that
you cannot use the arch(1) tool to select the architecture you want to use for a
universal build because that only affects the python/pythonw wrapper and not the actual
interpreter.
The new version uses posix_spawnv with a number of OSX-specific options that ensure that
the real interpreter is started using the same CPU architecture as the wrapper, and that
means that 'arch -ppc python' now actually works.
I've also changed the way that the wrapper looks for the framework: it is now linked to
the framework rather than hardcoding the framework path. This should make it easier to
provide pythonw support in tools like virtualenv.
properly, and use this to skip test_multiprocessing on platforms
where sem_open raises a signal. This should fix some FreeBSD buildbot
failures for test_multiprocessing.
On MacOSX 10.6 the CoreFoundation framework must be initialized on the main
thread, the constructor function in that framework will cause an SIGABRT when
it is called on any other thread.
Because a number of extension link (indirectly) to CoreFoundation and the
Python core itself didn't the interpreter crashed when importing some
extensions, such as _locale, on a secondary thread.
This fix ensures that Python is linked to CoreFoundation on OSX, which results
in the CoreFoundation constructor being called when Python is loaded. This
does not require code changes.
- add double endianness detection to configure script
- add configure-time check to see whether we can use inline
assembly to get and set x87 control word in configure script
- add functions to get and set x87 control word in Python/pymath.c
- add pyport.h logic to determine whether it's safe to use the
short float repr or not
to that README file with some explanation.
* Be more strict in the configure script: complain loudly when the user has
specified invalid combinations of OSX-specific configure arguments.
The error message refers to the Mac/README file for more information.
issue #2937. This information can be helpful for diagnosing platform-
specific problems in math and cmath. The result of the test also
serves as a fairly reliable indicator of whether the x87 floating-point
instructions (as opposed to SSE2) are in use on Intel x86/x86_64 systems.
to get the correct completion_matches function to avoid crashes on
x86_64 (amd64).
I don't have OpenBSD to test myself. I tested that it does not break
anything on linux. It is simple.
This patch adds a new configure argument on OSX:
--with-universal-archs=[32-bit|64-bit|all]
When used with the --enable-universalsdk option this controls which
CPU architectures are includes in the framework. The default is 32-bit,
meaning i386 and ppc. The most useful alternative is 'all', which includes
all 4 CPU architectures supported by MacOS X (i386, ppc, x86_64 and ppc64).
This includes limited support for the Carbon bindings in 64-bit mode as well,
limited because (a) I haven't done extensive testing and (b) a large portion
of the Carbon API's aren't available in 64-bit mode anyway.
I've also duplicated a feature of Apple's build of python: setting the
environment variable 'ARCHFLAGS' controls the '-arch' flags used for building
extensions using distutils.
information about the processor on the Debian/alpha
buildbot. (I'm still trying to track down the cause
of the test_math failures for this machine.) This
checkin will be reverted within the next 48 hours.
This introduces a new configure option: --with-framework-name=NAME
(defaulting to 'Python'). This allows you to install several copies
of the Python framework with different names (such as a normal build
and a debug build).
the compiler. This should(?) help to fix failures in test_math
and test_cmath on Linux/alpha.
Also add configure message reporting the result of uname -m, as
a debugging aid.
apparently because tanh(-0.) loses the sign of zero on that platform.
If true, this is a bug in FreeBSD.
Added a configure test to verify this. I still need to figure out
how best to deal with this failure.
The patch also adds acosh, asinh, atanh, log1p and copysign to all platforms. Finally it fixes differences between platforms like different results or exceptions for edge cases. Have fun :)
The patch adds wrappers for the Linux epoll syscalls and the BSD kqueue syscalls. Thanks to Thomas Herve and the Twisted people for their support and help.
TODO: Finish documentation documentation
-no-cpp-precomp, and -mno-fused-madd from configure.
* r22183 added -no-cpp-precomp, which
http://gcc.gnu.org/ml/gcc/2005-12/msg00368.html claims hasn't been
needed since gcc-3.1.
* r25607 added -Wno-long-double to avoid a warning in
Include/objimpl.h (issue 525481). The long double is still there,
but OSX 10.4's gcc no longer warns about it.
* r33666 fixed issue 775892 on OSX 10.3 by adding -mno-fused-madd,
which changed the sign of some float 0s. Tim Peters said it wasn't
a real issue anyway, and it no longer causes test failures.
Fixes issue #1779871.
The bundled libffi copy is now in sync with the recently released
libffi3.0.4 version, apart from some small changes to
Modules/_ctypes/libffi/configure.ac.
I gave up on using libffi3 files on os x.
Instead, static configuration with files from pyobjc is used.
Georg Brandl has added fchmod() and fchown(). I've contributed lchown but I'm not able to test it on Linux. However it should be available on Mac and some other flavors of Unix.
I've made a quick test of fchmod() and fchown() on my system. They are working as expected.
Benjamin Peterson
Ronald Oussoren
Mark Dickinson
Antoine Pitrou
Martin v. Löwis
Gregory P. Smith
Georg Brandl
Alexandre Vassalotti
Tarek Ziadé
Jesse Noller
Neal Norwitz
Christian Heimes
Thomas Heller
Brett Cannon
Jeffrey Yasskin
Matthias Klose
Skip Montanaro