* [3.6] bpo-30703: Improve signal delivery (GH-2415)
* Improve signal delivery
Avoid using Py_AddPendingCall from signal handler, to avoid calling signal-unsafe functions.
* Remove unused function
* Improve comments
* Add stress test
* Adapt for --without-threads
* Add second stress test
* Add NEWS blurb
* Address comments @haypo.
(cherry picked from commit c08177a1cc)
* bpo-30796: Fix failures in signal delivery stress test (#2488)
* bpo-30796: Fix failures in signal delivery stress test
setitimer() can have a poor minimum resolution on some machines,
this would make the test reach its deadline (and a stray signal
could then kill a subsequent test).
* Make sure to clear the itimer after the test
* bpo-29512: Add test.bisect, bisect failing tests (#2452)
Add a new "python3 -m test.bisect" tool to bisect failing tests.
It can be used to find which test method(s) leak references, leak
files, etc.
(cherry picked from commit 84d9d14a1f)
* bpo-30776: regrtest: reduce memleak false positive (#2484)
Only report a leak if each run leaks at least one memory block.
(cherry picked from commit beeca6e1e5)
* bpo-30280: asyncio now cleans up threads
asyncio base TestCase now uses threading_setup() and
threading_cleanup() of test.support to cleanup threads.
* asyncio: Fix TestBaseSelectorEventLoop cleanup
bpo-30280: TestBaseSelectorEventLoop of
test.test_asyncio.test_selector_events now correctly closes the event
loop: cleanup its executor to not leak threads.
Don't override the close() method of the event loop, only override
the_close_self_pipe() method.
(cherry picked from commit b903067462)
* Clear potential ref cycle between Process and Process target
Besides Process.join() not being called, this was an indirect cause of bpo-30775.
The threading module already does this.
* Add issue reference.
(cherry picked from commit 79d37ae979)
Based on patch by Victor Stinner.
Add private C API function _PyUnicode_AsUnicode() which is similar to
PyUnicode_AsUnicode(), but checks for null characters..
(cherry picked from commit f7eae0adfc)
* bpo-30523: regrtest --list-cases --match (#2401)
* regrtest --list-cases now supports --match and --match-file options.
Example: ./python -m test --list-cases -m FileTests test_os
* --list-cases now also sets support.verbose to False to prevent
messages to stdout when loading test modules.
* Add support._match_test() private function.
(cherry picked from commit ace56d5836)
* bpo-30764: regrtest: add --fail-env-changed option (#2402)
* bpo-30764: regrtest: change exit code on failure
* Exit code 2 if failed tests ("bad")
* Exit code 3 if interrupted
* bpo-30764: regrtest: add --fail-env-changed option
If the option is set, mark a test as failed if it alters the
environment, for example if it creates a file without removing it.
(cherry picked from commit 63f54c6893)
* bpo-30776: reduce regrtest -R false positives (#2422)
* Change the regrtest --huntrleaks checker to decide if a test file
leaks or not. Require that each run leaks at least 1 reference.
* Warmup runs are now completely ignored: ignored in the checker test
and not used anymore to compute the sum.
* Add an unit test for a reference leak.
Example of reference differences previously considered a failure
(leak) and now considered as success (success, no leak):
[3, 0, 0]
[0, 1, 0]
[8, -8, 1]
(cherry picked from commit 48b5c422ff)
bpo-30764, bpo-29335: test_child_terminated_in_stopped_state() of
test_subprocess now uses support.SuppressCrashReport() to prevent the
creation of a core dump on FreeBSD.
(cherry picked from commit cdee3f14f7)
The current regex based splitting produces a wrong result. For example::
http://abc#@def
Web browsers parse that URL as ``http://abc/#@def``, that is, the host
is ``abc``, the path is ``/``, and the fragment is ``#@def``.
(cherry picked from commit 90e01e50ef)
* test_normalization fails if download fails
bpo-29887. The test is still skipped if "-u urlfetch" option is not
passed to regrtest (python3 -m test -u urlfetch test_normalization).
* Fix ResourceWarning in test_normalization
bpo-29887: Fix ResourceWarning in test_normalization if tests are
interrupted by CTRL+c.
(cherry picked from commit 722a3af092)
the original logic was just comparing the network address
but this is wrong because if the network address is equal then
we need to compare the ip address for breaking the tie
add more ip_interface comparison tests.
(cherry picked from commit 7bd8d3e794)
The public cyrus.andrew.cmu.edu IMAP server (port 993) doesn't accept
TLS connection using our self-signed x509 certificate. Remove the two
tests which are already skipped.
Write a new test_certfile_arg_warn() unit test for the certfile
deprecation warning.
(cherry picked from commit b18563da88)
* bpo-30649: test_os tolerates 50 ms delta for utime (#2156)
On Windows, tolerate a delta of 50 ms instead of 20 ms in
test_utime_current() and test_utime_current_old() of test_os.
On other platforms, reduce the delta from 20 ms to 10 ms.
(cherry picked from commit c94caca65c)
* bpo-30649: Revert utime delta in test_os (#2176)
PPC64 Fedora 3.x buildbot requires at least a delta of 14 ms: revert
the utime delta to 20 ms.
I tried 10 ms, but test_os failed on the PPC64 Fedora 3.x buildbot.
(cherry picked from commit 3402f72688)
* bpo-24484: Avoid race condition in multiprocessing cleanup
The finalizer registry can be mutated while inspected by multiprocessing
at process exit.
* Use test.support.start_threads()
* Add Misc/NEWS.
(cherry picked from commit 1eb6c0074d)
Add a test to check the current MAGIC_NUMBER against the
expected number for the release if the current release is
at candidate or final level. On test failure, describe to
the developer the procedure for changing the magic number.
This ensures that pre-merge CI will automatically pick up
on magic number changes in maintenance releases (and
explain why those are problematic), rather than relying on
all core developers to be aware of the implications of
such changes.
* Move co_extra_freefuncs to interpreter state to avoid crashes in
multi-threaded scenarios involving deletion of code objects
* Don't require that extra be zero initialized
* Build test list instead of defining empty test class
* Ensure extra is always assigned on success
* Keep the old fields in the thread state object, just don't use them
Add new linked list of code extra objects on a per-interpreter basis
so that interpreter state size isn't changed
* Rename __PyCodeExtraState_Get and add comment about it going away in 3.7
Fix sort order of import's in test_code.py
* Remove an extraneous space
* Remove docstrings for comments
* Touch up formatting
* Fix casing of coextra local
* Fix casing of another variable
* Prefix PyCodeExtraState with __ to match C API for getting it
* Update NEWS file for bpo-30604
This PR contains two updates to typing module:
- Support ContextManager on all versions (original PR by Jelle Zijlstra).
- Add generic AsyncContextManager..
(cherry picked from commit 29fda8db16)
* bpo-29406: asyncio SSL contexts leak sockets after calling close with certain servers (#409)
(cherry picked from commit a608d2d5a7)
* [3.6] bpo-29406: asyncio SSL contexts leak sockets after calling close with certain servers (GH-409)
* asyncio SSL contexts leak sockets after calling close with certain servers
* cleanup _shutdown_timeout_handle on _fatal_error.
(cherry picked from commit a608d2d5a7)
contextlib.AbstractContextManager now supports anti-registration
by setting __enter__ = None or __exit__ = None, following the pattern
introduced in bpo-25958..
(cherry picked from commit 57161aac5e)
If we have a chain of generators/coroutines that are 'yield from'ing
each other, then resuming the stack works like:
- call send() on the outermost generator
- this enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- which calls send() on the next generator
- which enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- ...etc.
However, every time we enter _PyEval_EvalFrameDefault, the first thing
we do is to check for pending signals, and if there are any then we
run the signal handler. And if it raises an exception, then we
immediately propagate that exception *instead* of starting to execute
bytecode. This means that e.g. a SIGINT at the wrong moment can "break
the chain" – it can be raised in the middle of our yield from chain,
with the bottom part of the stack abandoned for the garbage collector.
The fix is pretty simple: there's already a special case in
_PyEval_EvalFrameEx where it skips running signal handlers if the next
opcode is SETUP_FINALLY. (I don't see how this accomplishes anything
useful, but that's another story.) If we extend this check to also
skip running signal handlers when the next opcode is YIELD_FROM, then
that closes the hole – now the exception can only be raised at the
innermost stack frame.
This shouldn't have any performance implications, because the opcode
check happens inside the "slow path" after we've already determined
that there's a pending signal or something similar for us to process;
the vast majority of the time this isn't true and the new check
doesn't run at all..
(cherry picked from commit ab4413a7e9)
'invalid character in identifier' now is raised instead of
'f-string: empty expression not allowed' if a subexpression contains
only whitespaces and they are not accepted by Python parser.
(cherry picked from commit 2e9cd58)
* Fix bpo-30584
* Adding a comment mentionning the bpo and explaining what is the identifier
* Add Denis Osipov to Misc/ACKS
(cherry picked from commit 897bba7563)
At the time when an abstract base class' __init_subclass__ runs,
ABCMeta.__new__ has not yet finished running, so in the presence of
__init_subclass__, inspect.isabstract() can no longer depend only on
TPFLAGS_IS_ABSTRACT.
(cherry picked from commit fcfe80ec25)
Many metaclasses in the standard library don't play nice with
__init_subclass__. This bug makes ABCMeta in particular with
__init_subclass__, which is an 80/20 solution for me personally.
AFAICT, a general solution to this problem requires updating all
metaclasses in the standard library to make sure they pass **kwargs to
type.__new__, whereas this PR only fixes ABCMeta. For context, see
https://bugs.python.org/issue29581.
* added a test combining ABCMeta and __init_subclass__
* Added NEWS item
(cherry picked from commit bd583ef985)
* [3.6] bpo-29581: Make ABCMeta.__new__ pass **kwargs to type.__new__ (GH-527)
Many metaclasses in the standard library don't play nice with
__init_subclass__. This bug makes ABCMeta in particular with
__init_subclass__, which is an 80/20 solution for me personally.
AFAICT, a general solution to this problem requires updating all
metaclasses in the standard library to make sure they pass **kwargs to
type.__new__, whereas this PR only fixes ABCMeta. For context, see
https://bugs.python.org/issue29581.
* added a test combining ABCMeta and __init_subclass__
* Added NEWS item.
(cherry picked from commit bd583ef985)
* **kwargs -> ``kwargs`` in attempts to fix the Travis build.
* Quote the **kwargs
Serhiy Storchaka
Yury Selivanov
Victor Stinner
Nick Coghlan
Antoine Pitrou
Dong-hee Na
Mariatta
Dino Viehland
Denis Osipov
Antoine Pietri
Nate