Now all platforms use a value for the "EXT_SUFFIX" build variable derived
from SOABI (for instance in FreeBSD, "EXT_SUFFIX" is now ".cpython-310d.so"
instead of ".so"). Previously only Linux, Mac and VxWorks were using a value
for "EXT_SUFFIX" that included "SOABI".
Co-authored-by: Pablo Galindo <pablogsal@gmail.com>
This is invalid in C99 and later and is an error with some compilers
(e.g. clang in Xcode 12), and can thus cause configure checks to
produce incorrect results.
As [bpo-38443]() says the error message from configure when specifying --enable-universalsdk with a set of architectures that is not supported by the compiler is not very helpful. This PR explicitly checks if the compiler works and bails out if it doesn't.
As AIX 5.3 and below do not support thread_cputime, it was decided in
https://bugs.python.org/issue40680 to require AIX 6.1 and above. This
commit removes workarounds for — and references to — older, unsupported
AIX versions.
Co-authored-by: Lawrence D’Anna <lawrence_danna@apple.com>
* Add support for macOS 11 and Apple Silicon (aka arm64)
As a side effect of this work use the system copy of libffi on macOS, and remove the vendored copy
* Support building on recent versions of macOS while deploying to older versions
This allows building installers on macOS 11 while still supporting macOS 10.9.
* bpo-35823: subprocess: Use vfork() instead of fork() on Linux when safe
When used to run a new executable image, fork() is not a good choice
for process creation, especially if the parent has a large working set:
fork() needs to copy page tables, which is slow, and may fail on systems
where overcommit is disabled, despite that the child is not going to
touch most of its address space.
Currently, subprocess is capable of using posix_spawn() instead, which
normally provides much better performance. However, posix_spawn() does not
support many of child setup operations exposed by subprocess.Popen().
Most notably, it's not possible to express `close_fds=True`, which
happens to be the default, via posix_spawn(). As a result, most users
can't benefit from faster process creation, at least not without
changing their code.
However, Linux provides vfork() system call, which creates a new process
without copying the address space of the parent, and which is actually
used by C libraries to efficiently implement posix_spawn(). Due to sharing
of the address space and even the stack with the parent, extreme care
is required to use vfork(). At least the following restrictions must hold:
* No signal handlers must execute in the child process. Otherwise, they
might clobber memory shared with the parent, potentially confusing it.
* Any library function called after vfork() in the child must be
async-signal-safe (as for fork()), but it must also not interact with any
library state in a way that might break due to address space sharing
and/or lack of any preparations performed by libraries on normal fork().
POSIX.1 permits to call only execve() and _exit(), and later revisions
remove vfork() specification entirely. In practice, however, almost all
operations needed by subprocess.Popen() can be safely implemented on
Linux.
* Due to sharing of the stack with the parent, the child must be careful
not to clobber local variables that are alive across vfork() call.
Compilers are normally aware of this and take extra care with vfork()
(and setjmp(), which has a similar problem).
* In case the parent is privileged, special attention must be paid to vfork()
use, because sharing an address space across different privilege domains
is insecure[1].
This patch adds support for using vfork() instead of fork() on Linux
when it's possible to do safely given the above. In particular:
* vfork() is not used if credential switch is requested. The reverse case
(simple subprocess.Popen() but another application thread switches
credentials concurrently) is not possible for pure-Python apps because
subprocess.Popen() and functions like os.setuid() are mutually excluded
via GIL. We might also consider to add a way to opt-out of vfork() (and
posix_spawn() on platforms where it might be implemented via vfork()) in
a future PR.
* vfork() is not used if `preexec_fn != None`.
With this change, subprocess will still use posix_spawn() if possible, but
will fallback to vfork() on Linux in most cases, and, failing that,
to fork().
[1] https://ewontfix.com/7
Co-authored-by: Gregory P. Smith [Google LLC] <gps@google.com>
Since c19c5a6, AIX builds have defaulted to using dynload_shlib over
dynload_aix when dlopen is available. This function has been available
since AIX 4.3, which went out of support in 2003, the same year the
previously referenced commit was made. It has been nearly 20 years
since a version of AIX has been supported which has not used
dynload_shlib so there's no reason to keep this legacy code around.
close_range(2) should be preferred at all times if it's available, otherwise we'll use closefrom(2) if available with a fallback to fdwalk(3) or plain old loop over fd range in order of most efficient to least.
[note that this version does check for ENOSYS, but currently ignores all other errors]
Automerge-Triggered-By: @pablogsal
This commit removes the old parser, the deprecated parser module, the old parser compatibility flags and environment variables and all associated support code and documentation.
"make install" now uses the PLATLIBDIR variable for the destination
lib-dynload/ directory when ./configure --with-platlibdir is used.
Update --with-platlibdir comment in configure.
This reverts commit 0da5466650.
The commit is causing make failures on a FreeBSD buildbot.
Due to the imminent 3.9.0b1 cutoff, revert this commit for
now pending further investigation.
Add support to the configure script for OBJC and OBJCXX command line options so that the macOS builds can use the clang compiler for the macOS-specific Objective C source files. This allows third-party compilers, like GNU gcc, to be used to build the rest of the project since some of the Objective C system header files are not compilable by GNU gcc.
Co-authored-by: Jeffrey Kintscher <websurfer@surf2c.net>
Co-authored-by: Terry Jan Reedy <tjreedy@udel.edu>
This is the initial implementation of PEP 615, the zoneinfo module,
ported from the standalone reference implementation (see
https://www.python.org/dev/peps/pep-0615/#reference-implementation for a
link, which has a more detailed commit history).
This includes (hopefully) all functional elements described in the PEP,
but documentation is found in a separate PR. This includes:
1. A pure python implementation of the ZoneInfo class
2. A C accelerated implementation of the ZoneInfo class
3. Tests with 100% branch coverage for the Python code (though C code
coverage is less than 100%).
4. A compile-time configuration option on Linux (though not on Windows)
Differences from the reference implementation:
- The module is arranged slightly differently: the accelerated module is
`_zoneinfo` rather than `zoneinfo._czoneinfo`, which also necessitates
some changes in the test support function. (Suggested by Victor
Stinner and Steve Dower.)
- The tests are arranged slightly differently and do not include the
property tests. The tests live at test/test_zoneinfo/test_zoneinfo.py
rather than test/test_zoneinfo.py or test/test_zoneinfo/__init__.py
because we may do some refactoring in the future that would likely
require this separation anyway; we may:
- include the property tests
- automatically run all the tests against both pure Python and C,
rather than manually constructing C and Python test classes (similar
to the way this works with test_datetime.py, which generates C
and Python test cases from datetimetester.py).
- This includes a compile-time configuration option on Linux (though not
on Windows); added with much help from Thomas Wouters.
- Integration into the CPython build system is obviously different from
building a standalone zoneinfo module wheel.
- This includes configuration to install the tzdata package as part of
CI, though only on the coverage jobs. Introducing a PyPI dependency as
part of the CI build was controversial, and this is seen as less of a
major change, since the coverage jobs already depend on pip and PyPI.
Additional changes that were introduced as part of this PR, most / all of
which were backported to the reference implementation:
- Fixed reference and memory leaks
With much debugging help from Pablo Galindo
- Added smoke tests ensuring that the C and Python modules are built
The import machinery can be somewhat fragile, and the "seamlessly falls
back to pure Python" nature of this module makes it so that a problem
building the C extension or a failure to import the pure Python version
might easily go unnoticed.
- Adjustments to zoneinfo.__dir__
Suggested by Petr Viktorin.
- Slight refactorings as suggested by Steve Dower.
- Removed unnecessary if check on std_abbr
Discovered this because of a missing line in branch coverage.
Add --with-experimental-isolated-subinterpreters build option to
configure: better isolate subinterpreters, experimental build mode.
When used, force the usage of the libc malloc() memory allocator,
since pymalloc relies on the unique global interpreter lock (GIL).
This fixes a regression introduced in bpo-38960.
When DFLAGS was empty, "$DFLAGS" results in an empty argument ("").
Without the quotes, an empty variable will be ignored by the shell.
Add --with-platlibdir option to the configure script: name of the
platform-specific library directory, stored in the new sys.platlitdir
attribute. It is used to build the path of platform-specific dynamic
libraries and the path of the standard library.
It is equal to "lib" on most platforms. On Fedora and SuSE, it is
equal to "lib64" on 64-bit systems.
Co-Authored-By: Jan Matějek <jmatejek@suse.com>
Co-Authored-By: Matěj Cepl <mcepl@cepl.eu>
Co-Authored-By: Charalampos Stratakis <cstratak@redhat.com>
Setting `-D_XOPEN_SOURCE=700` on HP-UX causes system functions such as chroot to be undefined. This change stops `_XOPEN_SOURCE` begin set on HP-UX
Co-authored-by: Benjamin Peterson <benjamin@python.org>
The os.putenv() and os.unsetenv() functions are now always available.
On non-Windows platforms, Python now requires setenv() and unsetenv()
functions to build.
Remove putenv_dict from posixmodule.c: it's not longer needed.
If setenv() C function is available, os.putenv() is now implemented
with setenv() instead of putenv(), so Python doesn't have to handle
the environment variable memory.
Provides a richer platform tag for AIX that we expect to be sufficient for PEP 425
binary distribution identification. Any backports to earlier Python versions will be
handled via setuptools.
Patch by Michael Felt.
Fix stdatomic.h header check for ICC compiler: the ICC implementation
lacks atomic_uintptr_t type which is needed by Python.
Test:
* atomic_int and atomic_uintptr_t types
* atomic_load_explicit() and atomic_store_explicit()
* memory_order_relaxed and memory_order_seq_cst constants
But don't test ATOMIC_VAR_INIT(): it's not used in Python.
This changeset increases the default size of the stack
for threads on macOS to the size of the stack
of the main thread and reenables the relevant
recursion test.
Reduce the number of unit tests run for the PGO generation task. This
speeds up the task by a factor of about 15x. Running the full unit test
suite is slow. This change may result in a slightly less optimized build
since not as many code branches will be executed. If you are willing to
wait for the much slower build, the old behavior can be restored using
'./configure [..] PROFILE_TASK="-m test --pgo-extended"'. We make no
guarantees as to which PGO task set produces a faster build. Users who
care should run their own relevant benchmarks as results can depend on
the environment, workload, and compiler tool chain.
Under some conditions the earlier fix for bpo-18075, "Infinite recursion
tests triggering a segfault on Mac OS X", now causes failures on macOS
when attempting to change stack limit with resource.setrlimit
resource.RLIMIT_STACK, like regrtest does when running the test suite.
The reverted change had specified a non-default stack size when linking
the python executable on macOS. As of macOS 10.14.4, the previous
code causes a hard failure when running tests, although similar
failures had been seen under some conditions under some earlier
systems. Reverting the change to the interpreter stack size at link
time helped for release builds but caused some tests to fail when
built --with-pydebug. Try the opposite approach: continue to build
the interpreter with an increased stack size on macOS and remove
the failing setrlimit call in regrtest initialization. This will
definitely avoid the resource.RLIMIT_STACK error and should have
no, or fewer, side effects.
Matti Picus
pxinwr
Joshua Root
Ronald Oussoren
Pablo Galindo
Kevin Adler
Christian Heimes
Alexey Izbyshev
Petr Viktorin
Kyle Evans
Stefan Krah
wasiher
Lysandros Nikolaou
Victor Stinner
Ned Deily
Paul Ganssle
karl ding
Zackery Spytz
Ian Norton
David Carlier
Anthony Shaw
Michael Felt
Benjamin Peterson
Vinay Sajip
Jesús Cea
Doyle Rowland
Jakub Kulík
Sergey Fedoseev
Neil Schemenauer