Store the current running task on the thread state, it makes it thread safe for the free-threading build and while improving performance as there is no lock contention, this effectively makes it lock free.
When accessing the current task of the current running loop in current thread, no locking is required and can be acessed without locking.
In the rare case of accessing current task of a loop running in a different thread, the stop the world pauses is used in free-threading builds to stop all other running threads and find the task for the specified loop.
This also makes it easier for external introspection to find the current task, and now it will be always correct.
GitHub discovered it was an issue with the underlying hardware
and have migrated all the runners to a different arm sku (dpdsv5).
This reverts commit 80b9e79d84.
* Implement C recursion protection with limit pointers
* Remove calls to PyOS_CheckStack
* Add stack protection to parser
* Make tests more robust to low stacks
* Improve error messages for stack overflow
Add cast macros for date and time objects
Fix UBSan failures for `PyDateTime_Delta`, `PyDateTime_Date`, `PyDateTime_TZInfo`, `PyDateTime_TimeZone`, `PyDateTime_Time`, `PyDateTime_DateTime`, `PyDateTime_IsoCalendarDate`
Fix up naming
Make tuple iteration more thread-safe, and actually test concurrent iteration of tuple, range and list. (This is prep work for enabling specialization of FOR_ITER in free-threaded builds.) The basic premise is:
Iterating over a shared iterable (list, tuple or range) should be safe, not involve data races, and behave like iteration normally does.
Using a shared iterator should not crash or involve data races, and should only produce items regular iteration would produce. It is not guaranteed to produce all items, or produce each item only once. (This is not the case for range iteration even after this PR.)
Providing stronger guarantees is possible for some of these iterators, but it's not always straight-forward and can significantly hamper the common case. Since iterators in general aren't shared between threads, and it's simply impossible to concurrently use many iterators (like generators), better to make sharing iterators without explicit synchronization clearly wrong.
Specific issues fixed in order to make the tests pass:
- List iteration could occasionally fail an assertion when a shared list was shrunk and an item past the new end was retrieved concurrently. There's still some unsafety when deleting/inserting multiple items through for example slice assignment, which uses memmove/memcpy.
- Tuple iteration could occasionally crash when the iterator's reference to the tuple was cleared on exhaustion. Like with list iteration, in free-threaded builds we can't safely and efficiently clear the iterator's reference to the iterable (doing it safely would mean extra, slow refcount operations), so just keep the iterable reference around.
Entries may be added or removed from `sys.meta_path` concurrently. For
example, setuptools temporarily adds and removes the `distutils` finder from
the beginning of the list. The local copy ensures that we don't skip over any
entries.
Some packages modify `sys.modules` during import. For example, `collections`
inserts the entry for `collections.abc` into `sys.modules` during import. We
need to ensure that we re-check `sys.modules` *after* the parent module is
fully initialized.
In the Python implementation, "Z" was allowed where only "+" or "-" should be allowed in time zone specifiers. In the C implementation, ":" was allowed as a separator between the whole and fractional portion of times (seconds). These have both been forbidden and the error messages harmonized.
The test should use the correct idiom for starting the task, `loop._run_once` is private API which should not be used directly, instead use `asyncio.sleep(0)` for 1 event loop cycle.
Add the following private methods to `pathlib.Path.info`:
- `_posix_permissions()`: the POSIX file permissions (`S_IMODE(st_mode)`)
- `_file_id()`: the file ID (`(st_dev, st_ino)`)
- `_access_time_ns()`: the access time in nanoseconds (`st_atime_ns`)
- `_mod_time_ns()`: the modify time in nanoseconds (`st_mtime_ns`)
- `_bsd_flags()`: the BSD file flags (`st_flags`)
- `_xattrs()`: the file extended attributes as a list of key, value pairs,
or an empty list if `listxattr()` or `getxattr()` fail in an ignorable
way.
These methods replace `LocalCopyReader.read_metadata()`, and so we can
delete the `CopyReader` and `LocalCopyReader` classes. Rather than reading
metadata via `source._copy_reader.read_metadata()`, we instead call
`source.info._posix_permissions()`, `_access_time_ns()`, etc.
Preserving metadata is only supported for local-to-local copies at the
moment. To support copying metadata between arbitrary `ReadablePath` and
`WritablePath` objects, we'd need to make the new methods public and
documented.
Co-authored-by: Petr Viktorin <encukou@gmail.com>
* gh-129701: Fix a data race in `intern_common` in the free threaded build
* Use a mutex to avoid potentially returning a non-immortalized string,
because immortalization happens after the insertion into the interned
dict.
* Use `Py_DECREF()` calls instead of `Py_SET_REFCNT(s, Py_REFCNT(s) - 2)`
for thread-safety. This code path isn't performance sensistive, so
just use `Py_DECREF()` unconditionally for simplicity.
Fix some UBSan failures for `dequeobject`, `dequeiterobject`, `defdictobject` and `tuplegetterobject`.
We also perform some cleanup by suppressing unused return values and renaming the
unused argument in `METH_NOARGS` methods to `dummy` for semantic purposes.
Fix UBSan failures for `PyBaseExceptionObject`, `PyStopIterationObject`, `PySystemExitObject`, `PyImportErrorObject`, `PyOSErrorObject`, `PyNameErrorObject`, `PyAttributeErrorObject`, `PySyntaxErrorObject`, `KeyError`, `UnicodeError*`, `PyBaseExceptionGroupObject`
Remove redundant casts
Align the naming convention `Py[...]_CAST(op)` where only an assert-only type check is performed.
Remove broken singledispatchmethod caching introduced in gh-85160.
Achieve the same performance using different optimization.
* Add more tests.
* Fix issues with __module__ and __doc__ descriptors.
Remove the caching `_is_case_sensitive()` function.
The cache used to speed up `PurePath.[full_]match()` and `Path.[r]glob()`,
but that's no longer the case - these methods use
`self.parser is posixpath` to determine case sensitivity.
This makes the `pathlib._abc` module a little easier to backport to Python
3.8, where `functools.cache()` is unavailable.
