bpo-6721: When os.fork() was called while another thread holds a logging lock, the child process may deadlock when it tries to log. This fixes that by acquiring all logging locks before fork and releasing them afterwards.
A regression test that fails before this change is included.
Within the new unittest itself: There is a small _potential_ due to mixing of fork and a thread in the child process if the parent's thread happened to hold a non-reentrant library call lock (malloc?) when the os.fork() happens. buildbots and time will tell if this actually manifests itself in this test or not. :/ A functionality test that avoids that would be a challenge.
An alternate test that isn't trying to produce the deadlock itself but just checking that the release and acquire calls are made would be the next best alternative if so.
This causes the tearDown code to only unimport the test modules specifically created as part of each test via the self.mkhier method rather than abusing test.support.modules_setup() and the scary test.support.modules_cleanup() code.
https://bugs.python.org/issue34200
Returning EINTR from pthread semaphore or lock acquisition is an optional POSIX
feature. musl does not provide this feature, so some threadsignal tests fail
when Python is built against it.
There's no good way to test for musl, so we skip if we're on Linux and not using
glibc pthreads.
Also, hedge in the threading documentation about when we can provide interrupts
from lock acquisition.
Store a weak reference to stream readerfor breaking strong references
It breaks the strong reference loop between reader and protocol and allows to detect and close the socket if the stream is deleted (garbage collected)
When subprocess.Popen() stdin= stdout= or stderr= handles are specified
and appear in pass_fds=, don't close the original fds after dup'ing them.
This implementation and unittest primarily came from @izbyshev (see the PR)
See also b89b52f284
This also removes the old manual p2cread, c2pwrite, and errwrite closing logic
as inheritable flags and _close_open_fds takes care of that properly today without special treatment.
This code is within child_exec() where it is the only thread so there is no
race condition between the dup and _Py_set_inheritable_async_safe call.
The recursive frame pruning code always undercounted the number of elided frames
by one. That is, in the "[Previous line repeated N more times]" message, N would
always be one too few. Near the recursive pruning cutoff, one frame could be
silently dropped. That situation is demonstrated in the OP of the bug report.
The fix is to start the identical frame counter at 1.
* Make sure that when some of the tests in test_smtplib fail, the allocated threads
and sockets are not leaked.
* Use support.join_thread() instead of thread.join() to avoid infinite blocks.
Some methods of the SMTP class use mutable default arguments. Specially
`send_message` is affected as it mutates one of the args by appending items
to it, which has side effects on further calls.
* Add %T format to PyUnicode_FromFormatV(), and so to
PyUnicode_FromFormat() and PyErr_Format(), to format an object type
name: equivalent to "%s" with Py_TYPE(obj)->tp_name.
* Replace Py_TYPE(obj)->tp_name with %T format in unicodeobject.c.
* Add unit test on %T format.
* Rename unicode_fromformat_write_cstr() to
unicode_fromformat_write_utf8(), to make the intent more explicit.
* Replace "master process" with "parent process"
* Replace "master option mappings" with "main option mappings"
* Replace "master pattern object" with "main pattern object"
* ssl: replace "master" with "server"
* And some other similar changes
Fail `test_semaphore_tracker_sigint` if no warnings are expected and one is received.
Fix race condition when the child receives SIGINT before it can register signal handlers for it.
The race condition occurs when the parent calls
`_semaphore_tracker.ensure_running()` (which in turn spawns the
semaphore_tracker using `_posixsubprocess.fork_exec`), the child
registers the signal handlers and the parent tries to kill the child.
What seem to happen is that in some slow systems, the parent sends the
signal to kill the child before the child protects against the signal.
* Fix Tools/clinic/clinic_test.py: add missing
FakeClinic.destination_buffers attribute and pass a file argument
to Clinic().
* Rename Tools/clinic/clinic_test.py to Lib/test/test_clinic.py:
add temporary Tools/clinic/ to sys.path to import the clinic
module.
Co-Authored-By: Pablo Galindo <pablogsal@gmail.com>
* Add _testcapi.get_coreconfig() to get the _PyCoreConfig of the
interpreter
* test.pythoninfo now gets the core configuration using
_testcapi.get_coreconfig()
Sometimes some versions of the shared libraries that are part of the
traceback are compiled in optimised mode and the Program Counter (PC)
is not present, not allowing gdb to walk the frames back. When this
happens, the Python bindings of gdb raise an exception, making the
test impossible to succeed.
Update all test certs and keys to use future proof crypto settings:
* 3072 bit RSA keys
* SHA-256 signature
Signed-off-by: Christian Heimes <christian@python.org>
Modify TestPosixSpawn to run Python using -I and -S options.
Disable site module to avoid side effects. For example, on Fedora 28,
if the HOME environment variable is not set, site._getuserbase()
calls pwd.getpwuid() which opens /var/lib/sss/mc/passwd, but then
leaves the file open which makes test_close_file() to fail.
Add support for the "surrogatepass" error handler in
PyUnicode_DecodeFSDefault() and PyUnicode_EncodeFSDefault()
for the UTF-8 encoding.
Changes:
* _Py_DecodeUTF8Ex() and _Py_EncodeUTF8Ex() now support the
surrogatepass error handler (_Py_ERROR_SURROGATEPASS).
* _Py_DecodeLocaleEx() and _Py_EncodeLocaleEx() now use
the _Py_error_handler enum instead of "int surrogateescape" to pass
the error handler. These functions now return -3 if the error
handler is unknown.
* Add unit tests on _Py_DecodeLocaleEx() and _Py_EncodeLocaleEx()
in test_codecs.
* Rename get_error_handler() to _Py_GetErrorHandler() and expose it
as a private function.
* _freeze_importlib doesn't need config.filesystem_errors="strict"
workaround anymore.
Py_DecodeLocale() and Py_EncodeLocale() now use the UTF-8 encoding on
Windows if Py_LegacyWindowsFSEncodingFlag is zero.
pymain_read_conf() now sets Py_LegacyWindowsFSEncodingFlag in its
loop, but restore its value at exit.
_PyCoreConfig_Read() is now responsible to choose the filesystem
encoding and error handler. Using Py_Main(), the encoding is now
chosen even before calling Py_Initialize().
_PyCoreConfig.filesystem_encoding is now the reference, instead of
Py_FileSystemDefaultEncoding, for the Python filesystem encoding.
Changes:
* Add filesystem_encoding and filesystem_errors to _PyCoreConfig
* _PyCoreConfig_Read() now reads the locale encoding for the file
system encoding.
* PyUnicode_EncodeFSDefault() and PyUnicode_DecodeFSDefaultAndSize()
now use the interpreter configuration rather than
Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors
global configuration variables.
* Add _Py_SetFileSystemEncoding() and _Py_ClearFileSystemEncoding()
private functions to only modify Py_FileSystemDefaultEncoding and
Py_FileSystemDefaultEncodeErrors in coreconfig.c.
* _Py_CoerceLegacyLocale() now takes an int rather than
_PyCoreConfig for the warning.
Standard streams like sys.stdout now use the "surrogateescape" error
handler, instead of "strict", on the POSIX locale (when the C locale is not
coerced and the UTF-8 Mode is disabled).
Add tests on sys.stdout.errors with LC_ALL=POSIX.
Python now gets the locale encoding with C code to initialize the encoding
of standard streams like sys.stdout. Moreover, the encoding is now
initialized to the Python codec name to get a normalized encoding name and
to ensure that the codec is loaded. The change avoids importing
_bootlocale and _locale modules at startup by default.
When the PYTHONIOENCODING environment variable only contains an encoding,
the error handler is now is now set explicitly to "strict".
Rename also get_default_standard_stream_error_handler() to
get_stdio_errors().
Reduce the buffer to format the "cpXXX" string (Windows locale encoding).
* The UTF-8 Mode is now also enabled by the "POSIX" locale, not only
by the "C" locale.
* On FreeBSD, Py_DecodeLocale() and Py_EncodeLocale() now also forces
the ASCII encoding if the LC_CTYPE locale is "POSIX", not only if
the LC_CTYPE locale is "C".
* test_utf8_mode.test_cmd_line() checks also that the command line
arguments are decoded from UTF-8 when the the UTF-8 Mode is enabled
with POSIX locale or C locale.
Make mixed-type `%` and `//` operations involving `Fraction` and `float` objects behave like all other mixed-type arithmetic operations: first the `Fraction` object is converted to a `float`, then the `float` operation is performed as normal. This fixes some surprising corner cases, like `Fraction('1/3') % inf` giving a NaN.
Thanks Elias Zamaria for the patch.
The current C implementations **crash** if the input includes a surrogate
Unicode code point, which is not possible to encode in UTF-8.
Important notes:
1. It is possible to pass a non-UTF-8 string as a separator to the
`.isoformat()` methods.
2. The pure-Python `datetime.fromisoformat()` implementation accepts
strings with a surrogate as the separator.
In `datetime.fromisoformat()`, in the special case of non-UTF-8 separators,
this implementation will take a performance hit by making a copy of the
input string and replacing the separator with 'T'.
Co-authored-by: Alexey Izbyshev <izbyshev@ispras.ru>
Co-authored-by: Paul Ganssle <paul@ganssle.io>
Introduce a configure check for strsignal(3) which defines HAVE_STRSIGNAL for
signalmodule.c. Add some common signals on HP-UX. This change applies for
Windows and HP-UX.
Read from data socket to avoid "[SSL] shutdown while in init" exception
during shutdown of the dummy server.
Signed-off-by: Christian Heimes <christian@python.org>
<!-- issue-number: [bpo-34391](https://www.bugs.python.org/issue34391) -->
https://bugs.python.org/issue34391
<!-- /issue-number -->
os.readlink() now accepts path-like and bytes objects on Windows.
Previously, support for path-like and bytes objects was only
implemented on Unix.
This commit also merges Unix and Windows implementations of
os.readlink() in one function and adds basic unit tests to increase
test coverage of the function.
Downstream vendors have started to deprecate weak keys. Update all RSA keys
and DH params to use at least 2048 bits.
Finite field DH param file use RFC 7919 values, generated with
certtool --get-dh-params --sec-param=high
Signed-off-by: Christian Heimes <christian@python.org>
Multiprocessing test_timeout() now accepts a delta of 100 ms instead
of just 50 ms, since the test failed with 135.8 ms instead of the
expected 200 ms.
ZipFile can zip files older than 1980-01-01 and newer than 2107-12-31 using
a new strict_timestamps parameter at the cost of setting the timestamp
to the limit.
* Fix integer overflow in os.readv(), os.writev(), os.preadv()
and os.pwritev() and in os.sendfile() with headers or trailers
arguments (on BSD-based OSes and MacOS).
* Fix sending the part of the file in os.sendfile() on MacOS.
Using the trailers argument could cause sending more bytes from
the input file than was specified.
Thanks Ned Deily for testing on 32-bit MacOS.
* help(hashlib) didn't work because of incorrect module name in blake2b and
blake2s classes.
* Constructors blake2*(), sha3_*(), shake_*() and keccak_*() incorrectly
accepted keyword argument "string" for binary data, but documented as
accepting the "data" keyword argument. Now this parameter is positional-only.
* Keyword-only parameters in blake2b() and blake2s() were not documented as
keyword-only.
* Default value for some parameters of blake2b() and blake2s() was None,
which is not acceptable value.
* The length argument for shake_*.digest() was wrapped out to 32 bits.
* The argument for shake_128.digest() and shake_128.hexdigest() was not
positional-only as intended.
* TypeError messages for incorrect arguments in all constructors sha3_*(),
shake_*() and keccak_*() incorrectly referred to sha3_224.
Also made the following enhancements:
* More accurately specified input and result types for strings, bytes and
bytes-like objects.
* Unified positional parameter names for update() and constructors.
* Improved formatting.
* The hash of BuiltinMethodType instances no longer depends on the hash
of __self__. It depends now on the hash of id(__self__).
* The hash and equality of ModuleType and MethodWrapperType instances no
longer depend on the hash and equality of __self__. They depend now on
the hash and equality of id(__self__).
* MethodWrapperType instances no longer support ordering.
Various asyncio internals expect that the default executor is a
`ThreadPoolExecutor`, so deprecate passing anything else to
`loop.set_default_executor()`.
* bpo-34239: Convert test_bz2 to use tempfile
test_bz2 currently uses the test.support.TESTFN functionality which creates a temporary file local to the test directory named around the pid.
This can give rise to race conditions where tests are competing with each other to delete and recreate the file.
This change converts the tests to use tempfile.mkstemp which gives a different file every time from the system's temp area
* Inline cmdline_get_env_flags() into config_read_env_vars():
_PyCoreConfig_Read() now reads much more environment variables like
PYTHONVERBOSE.
* Allow to override faulthandler and allocator even if dev_mode=1.
PYTHONMALLOC is now the priority over PYTHONDEVMODE.
* Fix _PyCoreConfig_Copy(): copy also install_signal_handlers,
coerce_c_locale and coerce_c_locale_warn
* _PyCoreConfig.install_signal_handlers default is now 1: install
signals by default
* Fix also a compiler warning: don't define _PyPathConfig type twice.
Enable and fix SMTPUTF8SimTests in test_smtplib.
The tests for SMTPUTF8SimTests in test_smtplib.py were not actually
being run because test_smtplib was still using the 'test_main' pattern,
and the class was never added to test_main.
Additionally, one of the tests needed to be moved to the non-UTF8 server
class because it relies on the server not being UTF-8 compatible (and it
had a bug in in).
On Windows, passing a negative value to local results in an OSError because localtime_s on Windows does not support negative timestamps. Unfortunately this means that fold detection for timestamps between 0 and max_fold_seconds will result in this OSError since we subtract max_fold_seconds from the timestamp to detect a fold. However, since we know there haven't been any folds in the interval [0, max_fold_seconds) in any timezone, we can hackily just forego fold detection for this time range on Windows.
* Fix case-sensitive comparison
test_nt_helpers assumed that two versions of a Windows path could be compared case-sensitively. This is not the case, and the difference can be triggered (apparently) by running the test on a path somewhere below a Junction.
Add more fields to _PyCoreConfig:
* _check_hash_pycs_mode
* bytes_warning
* debug
* inspect
* interactive
* legacy_windows_fs_encoding
* legacy_windows_stdio
* optimization_level
* quiet
* unbuffered_stdio
* user_site_directory
* verbose
* write_bytecode
Changes:
* Remove pymain_get_global_config() and pymain_set_global_config()
which became useless. These functions have been replaced by
_PyCoreConfig_GetGlobalConfig() and
_PyCoreConfig_SetGlobalConfig().
* sys.flags.dont_write_bytecode value is now restricted to 1 even if
-B option is specified multiple times on the command line.
* PyThreadState_Clear() now uses the config from the current
interpreter rather than using global Py_VerboseFlag
When Python is installed on Windows, python -m test test_tools failed
because it tried to run Tools\scripts\2to3.py which requires an
argument. Skip this script. On other platforms or on Windows but when
run from source code (not installed), the script is called "2to3"
instead of "2to.py" and so was already skipped.
Modify also the unit test to unload all modules which have been
loaded by the test.
The test failed on my laptop because the busy loop took 15.9 ms
whereas the test expects at least 20 ms. Modify test_process_time()
as test_thread_time() has been modified recently: only require 15 ms
instead of 20 ms.
Py_Main() can again be called after Py_Initialize(), as in Python
3.6. The new configuration is ignored, except of
_PyMainInterpreterConfig.argv which is used to update sys.argv.
Increase the timeout: give timeout x 4 instead of timeout x 2 to
threads to wait until the Event is set, but reduce the sleep from 500
ms to 250 ms. So the test should be more reliable and faster!
On Windows, sometimes test_signal.test_warn_on_full_buffer() fails to
fill the socketpair buffer. In that case, the C signal handler
succeed to write into the socket, it doesn't log the expected send
error, and so the test fail.
On Windows, the test now uses a timeout of 50 ms to fill the
socketpair buffer to fix this race condition.
Other changes:
* Begin with large chunk size to fill the buffer to speed up the
test.
* Add error messages to assertion errors to more easily identify
which assertion failed.
* Don't set the read end of the socketpair as non-blocking.
test_signal.test_socket(): On Windows, sometimes even if the C signal handler
succeed to write the signal number into the write end of the socketpair, the
test fails with a BlockingIOError on the non-blocking read.recv(1) because the
read end of the socketpair didn't receive the byte yet.
Fix the race condition on Windows by setting the read end as blocking.
`_PyUnicode_TransformDecimalAndSpaceToASCII()` missed trailing NUL char.
It caused buffer overflow in `_Py_string_to_number_with_underscores()`.
This bug is introduced in 9b6c60cb.
The test failed on AMD64 Debian root 3.x buildbot because the busy
loop of 100 ms only increased time.thread_time() by 19.9 ms which is
smaller than 20 ms. Modify the test to tolerate a delta of at least
15 ms instead of 20 ms.
* Always return bytes from _HackedGetData.get_data().
Ensure the imp.load_source shim always returns bytes by reopening the file in
binary mode if needed. Hash-based pycs have to receive the source code in bytes.
It's tempting to change imp.get_suffixes() to always return 'rb' as a mode, but
that breaks some stdlib tests and likely 3rdparty code, too.
Gregory P. Smith
Andrew Svetlov
Michael Felt
Benjamin Peterson
Oren Milman
Ethan Furman
Raymond Hettinger
orlnub123
Victor Stinner
guoci
Zackery Spytz
Daniel Hahler
Rahul Jha
Matthias Bussonnier
Berker Peksag
Tal Einat
Sergey Fedoseev
Serhiy Storchaka
Pablo Galindo
Pablo Aguiar
Marcel Plch
Naris R
Christian Heimes
Carl Meyer
Przemysław Spodymek
Elias Zamaria
Vladimir Matveev
Paul Ganssle
Michael Osipov
Alexey Izbyshev
Vadim Pushtaev
GPery
Alex Grönholm
Marcel Plch
costypetrisor
twisteroid ambassador
Elvis Pranskevichus
Bo Bayles
Mickaël Schoentgen
Franz Wöllert
Tim Golden
Dong-hee Na
chason
Ammar Askar
Alexander Marshalov
Xtreak
jdemeyer
INADA Naoki
Derek B. Kim
Bradley Laney