Replace os.kill() with proc.kill() which catchs PermissionError.
Rewrite _kill_with_event():
* Use subprocess context manager ("with proc:").
* Use sleeping_retry() to wait until the child process is ready.
* Replace SIGINT with proc.kill() on error.
* Replace 10 seconds with SHORT_TIMEOUT to wait until the process is
ready.
* Replace 0.5 seconds with SHORT_TIMEOUT to wait for the process
exit.
"make regen-pegen" now creates a temporary file called "parser.c.new"
instead of "parser.new.c". Previously, if "make clinic" was run in
parallel with "make regen-all", clinic may try but fail to open
"parser.new.c" if the temporay file was removed in the meanwhile.
Increase support.LOOPBACK_TIMEOUT from 5 to 10 seconds. Also increase
the timeout depending on the --timeout option. For example, for a
test timeout of 40 minutes (ARM Raspbian 3.x), use LOOPBACK_TIMEOUT
of 20 seconds instead of 5 seconds before.
Fix a deadlock in test_socket when server fails with a timeout but
the client is still running in its thread. Don't hold a lock to call
cleanup functions in doCleanups(). One of the cleanup function waits
until the client completes, whereas the client could deadlock if it
called addCleanup() in such situation.
doCleanups() is called when the server completed, but the client can
still be running in its thread especially if the server failed with a
timeout. Don't put a lock on doCleanups() to prevent deadlock between
addCleanup() called in the client and doCleanups() waiting for
self.done.wait of ThreadableTest._setUp().
This adds a new field 'state' to PyThreadState that can take on one of three values: _Py_THREAD_ATTACHED, _Py_THREAD_DETACHED, or _Py_THREAD_GC. The "attached" and "detached" states correspond closely to acquiring and releasing the GIL. The "gc" state is current unused, but will be used to implement stop-the-world GC for --disable-gil builds in the near future.
test_builtin and test_socketserver no longer use signal.alarm() to
implement a watchdog with a hardcoded timeout (2 and 60 seconds).
Python test runner regrtest has two watchdogs: faulthandler and
timeout on running worker processes. Tests using short hardcoded
timeout can fail on slowest buildbots just because the timeout is too
short.
* Use `FindFirstFile` Win32 API to fix a bug where `ntpath.realpath()`
breaks out of traversing a series of paths where a (handled)
`ERROR_ACCESS_DENIED` or `ERROR_SHARING_VIOLATION` occurs.
* Update docs to reflect that `ntpath.realpath()` eliminates MS-DOS
style names.
When using worker processes (-jN) with --verbose3 option, regrtest
can now display the worker output even if a worker process does
crash. Previously, sys.stdout and sys.stderr were replaced and so
the worker output was lost on a crash.
We do the following:
* add a per-interpreter XID registry (PyInterpreterState.xidregistry)
* put heap types there (keep static types in _PyRuntimeState.xidregistry)
* clear the registries during interpreter/runtime finalization
* avoid duplicate entries in the registry (when _PyCrossInterpreterData_RegisterClass() is called more than once for a type)
* use Py_TYPE() instead of PyObject_Type() in _PyCrossInterpreterData_Lookup()
The per-interpreter registry helps preserve isolation between interpreters. This is important when heap types are registered, which is something we haven't been doing yet but I will likely do soon.
This is a temporary solution. The full fix may involve serializing the traceback in some form.
(FYI, I merged this yesterday and the reverted it due to buildbot failures. See gh-110248.)
CFunctionFullTests now also runs "bt" command before "py-bt-full",
similar to CFunctionTests which also runs "bt" command before
"py-bt". So test_gdb can skip the test if patterns like "?? ()" are
found in the gdb output.
dataclasses.replace() now raises TypeError instead of ValueError if
specify keyword argument for a field declared with init=False or miss keyword
argument for required InitVar field.
The test had an instability issue due to the ordering of the dummy
queue operation and the real wakeup pipe operations. Both primitives
are thread safe but not done atomically as a single update and may
interleave arbitrarily. With the old order of operations this can lead
to an incorrect state where the dummy queue is full but the wakeup
pipe is empty. By swapping the order in clear() I think this can no
longer happen in any possible operation interleaving (famous last
words).
In Python/bytecodes.c, you now write
```
DEOPT_IF(condition);
```
The code generator expands this to
```
DEOPT_IF(condition, opcode);
```
where `opcode` is the name of the unspecialized instruction.
This works inside macro expansions too.
**CAVEAT:** The entire `DEOPT_IF(condition)` statement must be on a single line.
If it isn't, the substitution will fail; an error will be printed by the code generator
and the C compiler will report some errors.
Add PyThreadState_GetUnchecked() function: similar to
PyThreadState_Get(), but don't issue a fatal error if it is NULL. The
caller is responsible to check if the result is NULL. Previously,
this function was private and known as _PyThreadState_UncheckedGet().
