Concurrent accesses from multiple threads to the same `cell` object did not
scale well in the free-threaded build. Use `_PyStackRef` and optimistically
avoid locking to improve scaling.
With the locks around cell reads gone, some of the free threading tests were
prone to starvation: the readers were able to run in a tight loop and the
writer threads weren't scheduled frequently enough to make timely progress.
Adjust the tests to avoid this.
Co-authored-by: Donghee Na <donghee.na@python.org>
* Rename 'defined' attribute to 'in_local' to more accurately reflect how it is used
* Make death of variables explicit even for array variables.
* Convert in_memory from boolean to stack offset
* Don't apply liveness analysis to optimizer generated code
* Fix RETURN_VALUE in optimizer
When `pathlib._os.magic_open()` is called to open a path in binary mode,
raise `ValueError` if any of the *encoding*, *errors* or *newline*
arguments are given. This matches the `open()` built-in.
- Restore max field size to sys.maxsize, as in Python 3.13 & below
- PyCField: Split out bit/byte sizes/offsets.
- Expose CField's size/offset data to Python code
- Add generic checks for all the test structs/unions, using the newly exposed attrs
add a set of asserts to test.test_capi.test_bytearray
1. Assert empty bytearray object for PyByteArray_Check.
2. Assert empty bytearray object for PyByteArray_CheckExact.
3. Assert 0-size bytearray object for PyByteArray_Size.
4. Assert empty bytearray object for PyByteArray_AsString.
5. Assert concatenation of the bytearray object with itself for PyByteArray_Concat.
Adjust the tests for the `pathlib.types` module so that they can be run
against the `pathlib-abc` PyPI package, which is a backport of the module
for older Python versions.
Specifically, we add a `.support.is_pypi` switch that is false in the
stdlib and true in the pathlib-abc package. This controls which package
we import, and whether or not we run tests against `PurePath` and `Path`.
For compatibility with older Python versions, we stop using
`zipfile.ZipFile.mkdir()` and `zipfile.ZipInfo._for_archive()`.
The subinterpreter tests have data races (see gh-129824).
TSAN attempts to intercept some of the fatal signals, which can lead to
bogus reports. We could possibly handle these via TSAN_OPTIONS, but it's
simpler to just skip those tests -- they're not multithreaded anyways.
Call `urllib.request.pathname2url()` from `pathlib.Path.as_uri()`, and
deprecate the duplicate implementation in `PurePath`.
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
Deprecate the `nturl2path` module. Its functionality is merged into
`urllib.request`.
Add `tests.test_nturl2path` to exercise `nturl2path`, as it's no longer
covered by `test_urllib`.
* Make musl test skips smarter (fixes Alpine errors)
A relatively small number of tests fail when the underlying c library is
provided by musl. This was originally reported in bpo-46390 by
Christian Heimes. Among other changes, these tests were marked for
skipping in gh-31947/ef1327e3 as part of bpo-40280 (emscripten support),
but the skips were conditioned on the *platform* being emscripten (or
wasi, skips for which ere added in 9b50585e02).
In gh-131071 Victor Stinner added a linked_to_musl function to enable
skipping a test in test_math that fails under musl, like it does on a
number of other platforms. This check can successfully detect that
python is running under musl on Alpine, which was the original problem
report in bpo-46390.
This PR replaces Victor's solution with an enhancement to
platform.libc_ver that does the check more cheaply, and also gets the
version number. The latter is important because the math test being
skipped is due to a bug in musl that has been fixed, but as of this
checkin date has not yet been released. When it is, the test skip can
be fixed to check for the minimum needed version.
The enhanced version of linked_to_musl is also used to do the skips of
the other tests that generically fail under musl, as opposed to
emscripten or wasi only failures. This will allow these tests to be
skipped automatically on Alpine.
This PR does *not* enhance libc_ver to support emscripten and wasi, as
I'm not familiar with those platforms; instead it returns a version
triple of (0, 0, 0) for those platforms. This means the musl tests will
be skipped regardless of musl version, so ideally someone will add
support to libc_ver for these platforms.
* Platform tests and bug fixes.
In adding tests for the new platform code I found a bug in the old code:
if a valid version is passed for version and it is greater than the
version found for an so *and* there is no glibc version, then the
version from the argument was returned. The code changes here fix
that.
* Add support docs, including for some preexisting is_xxx's.
* Add news item about libc_ver enhancement.
* Prettify platform re expression using re.VERBOSE.
Closes#130914: Make graphlib.TopologicalSorter.prepare() idempotent
Relax the rules so that `.prepare()` can be called multiple times, provided that no work has been passed out by `.get_ready()` yet.
Email generators using email.policy.default may convert an RFC 2047
encoded-word to unencoded form during header refolding. In a structured
header, this could allow 'specials' chars outside a quoted-string,
leading to invalid address headers and enabling spoofing. This change
ensures a parsed encoded-word that contains specials is kept as an
encoded-word while the header is refolded.
[Better fix from @bitdancer.]
---------
Co-authored-by: R David Murray <rdmurray@bitdance.com>
Co-authored-by: Petr Viktorin <encukou@gmail.com>
New features:
* refactor `hashlib_helper.requires_hashdigest` in prevision of a future
`hashlib_helper.requires_builtin_hashdigest` for built-in hashes only
* add `hashlib_helper.requires_openssl_hashdigest` to request OpenSSL
hashes, assuming that `_hashlib` exists.
Refactoring:
* split hmac.copy() test by implementation
* update how algorithms are discovered for RFC test cases
* simplify how OpenSSL hash digests are requested
* refactor hexdigest tests for RFC test vectors
* typo fix: `assert_hmac_hexdigest_by_new` -> `assert_hmac_hexdigest_by_name`
Improvements:
* strengthen contract on `hmac_new_by_name` and `hmac_digest_by_name`
* rename mixin classes to better match their responsibility
Remove the `test.test_pathlib.test_pathlib_abc` test module, which was
hollowed out in previous commits. Its few remaining tests are most relevant
to `PurePath` and `Path`, so we move them into `test_pathlib`.
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
The implementation of `PyLong_FromLong()`, `PyLong_FromLongLong()` and `PyLong_FromSsize_t()`
are now handled by a common macro `PYLONG_FROM_INT` which contains fast paths depending on the
size of the integer to convert. Consequently, `PyLong_FromSsize_t()` for medium-sized integers is faster
by roughly 25%.
---------
Co-authored-by: Sergey B Kirpichev <skirpichev@gmail.com>
`concurrent.futures.Executor.map` now supports limiting the number of submitted
tasks whose results have not yet been yielded via the new `buffersize` parameter.
---------
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Modifies the behavior of the interpreter on crash under Emscripten:
1. No Python traceback shown on segfault/trap
2. The JavaScript source line is shown
The JavaScript source line is super long and completely unenlightening,
whereas the Python traceback is very helpful.
Check would fail if value would create a pseudo-member, but that member
had not yet been created. We now attempt to create a pseudo-member for
a passed-in value first.
Co-authored-by: Tomas R. <tomas.roun8@gmail.com>
Test `pathlib.types._WritablePath` in a dedicated test module. These tests
cover `WritableZipPath`, `WritableLocalPath` and `Path`, where the former
two classes are implementations of `_WritablePath` for use in tests.
Add free-threaded versions of existing specialization for FOR_ITER (list, tuples, fast range iterators and generators), without significantly affecting their thread-safety. (Iterating over shared lists/tuples/ranges should be fine like before. Reusing iterators between threads is not fine, like before. Sharing generators between threads is a recipe for significant crashes, like before.)
Test `pathlib.types._ReadablePath` in a dedicated test module. These tests
cover `ReadableZipPath`, `ReadableLocalPath` and `Path`, where the former
two classes are implementations of `_ReadablePath` for use in tests.
Test Windows-flavoured `pathlib.types._JoinablePath` in a dedicated test
module. These tests cover `LexicalWindowsPath`, `PureWindowsPath` and
`WindowsPath`, where `LexicalWindowsPath` is a simple implementation of
`_JoinablePath` for use in tests.
* gh-130959: Reject whitespace in fractions, in pure Python `fromisoformat()`
Fix the pure Python implementation of `fromisoformat()` to reject any
non-digit characters, including whitespace, in the fractional part
of time specification. This makes the behavior consistent with the C
implementation, and prevents incorrect parsing of these fractions
(e.g. `.400 ` would be misinterpreted as `.04`).
* Add the news entry
* Use a different example to fix Sphinx lint
* Apply suggestions from code review
Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Paul Ganssle <1377457+pganssle@users.noreply.github.com>
* Try fixing `:func:` ref.
---------
Co-authored-by: Peter Bierma <zintensitydev@gmail.com>
Co-authored-by: Paul Ganssle <1377457+pganssle@users.noreply.github.com>
Test Posix-flavoured `pathlib.types._JoinablePath` in a dedicated test
module. These tests cover `LexicalPosixPath`, `PurePosixPath` and
`PosixPath`, where `LexicalPosixPath` is a simple implementation of
`_JoinablePath` for use in tests.
Remove the *case_sensitive* argument from `_JoinablePath.full_match()` and
`_ReadablePath.glob()`. Using a non-native case sensitivity forces the use
of "case-pedantic" globbing, where we `iterdir()` even for non-wildcard
pattern segments. But it's hard to know when to enable this mode, as
case-sensitivity can vary by directory, so `_PathParser.normcase()` doesn't
always give the full picture. The `Path.glob()` implementation is forced to
make an educated guess, but we can avoid the issue in the ABCs by dropping
the *case_sensitive* argument.
(I probably shouldn't have added these arguments in `PurePath` and `Path`
in the first place!)
Also drop support for `_ReadablePath.glob(recurse_symlinks=False)`, which
makes recursive globbing much slower.
Test `pathlib.types._JoinablePath` in a dedicated test module. These tests
cover `LexicalPath`, `PurePath` and `Path`, where `LexicalPath` is defined
in a new `test.test_pathlib.support` package.
The test could deadlock trying join on the worker processes.
Apply the same technique as gh-130933.
Join the process before the test ends in `test_notify` as well.
The new methods simply delegate to the underlying buffer, much like the existing GzipFile.read[1] methods. This avoids extra allocations caused by the BufferedIOBase.readinto implementation previously used.
This commit also factors out a common readability check rather than copying it an additional two times.
The workload to advance the virtual timeout is too lightweight for some
platforms. As result the test goes in timeout as it never reaches the
end of the timer. By having a heavier workload, the virtual timer
advances rapidly and the SIGVTALRM is sent before the timeout.
The test could deadlock trying join on the worker processes due to a
combination of behaviors:
* The use of `assertReachesEventually` did not ensure that workers
actually woken.release() because the SyncManager's Semaphore does not
implement get_value.
* This mean that the test could finish and the variable "sleeping" would
got out of scope and be collected. This unregisters the proxy leading
to failures in the worker or possibly the manager.
* The subsequent call to `p.join()` during cleanUp therefore never
finished.
This takes two approaches to fix this:
1) Use woken.acquire() to ensure that the workers actually finish
calling woken.release()
2) At the end of the test, wait until the workers are finished, while `cond`,
`sleeping`, and `woken` are still valid.
I chose to not raise an exception here because I think it would be
confusing for module attribute access to start raising something other
than AttributeError if e.g. the cwd goes away
Without the change in moduleobject.c
```
./python.exe -m unittest test.test_import.ImportTests.test_script_shadowing_stdlib_cwd_failure
...
Assertion failed: (PyErr_Occurred()), function _PyObject_SetAttributeErrorContext, file object.c, line 1253.
```
Fix `test_list.ListTest.test_no_memory` under trace refs build
Memory allocation ends up failing in _PyRefchainTrace(), which produces
different output. Assert that we don't segfault, which is the thing
we want to test and is less brittle than checking output.
This adds two new methods to `multiprocessing`'s `ProcessPoolExecutor`:
- **`terminate_workers()`**: forcefully terminates worker processes using `Process.terminate()`
- **`kill_workers()`**: forcefully kills worker processes using `Process.kill()`
These methods provide users with a direct way to stop worker processes without `shutdown()` or relying on implementation details, addressing situations where immediate termination is needed.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Sam Gross @colesbury
Commit-message-mostly-authored-by: Claude Sonnet 3.7 (because why not -greg)
* Fix use after free in list objects
Set the items pointer in the list object to NULL after the items array
is freed during list deallocation. Otherwise, we can end up with a list
object added to the free list that contains a pointer to an already-freed
items array.
* Mark `_PyList_FromStackRefStealOnSuccess` as escaping
I think technically it's not escaping, because the only object that
can be decrefed if allocation fails is an exact list, which cannot
execute arbitrary code when it is destroyed. However, this seems less
intrusive than trying to special cases objects in the assert in `_Py_Dealloc`
that checks for non-null stackpointers and shouldn't matter for performance.
The bytecode compiler only generates a few different types of constants,
like str, int, tuple, slices, etc. Users can construct code objects with
various unusual constants, including ones that are not hashable or not
even constant.
The free threaded build previously crashed with a fatal error when
confronted with these constants. Instead, treat distinct objects of
otherwise unhandled types as not equal for the purposes of deduplication.
Fix a race condition in test_check_output_timeout() of
test_subprocess. Don't write into stdout anymore, since there is no
reliable way to synchronize the parent and the child processes.
Change the timeout from 3 seconds to 0.1 seconds, and remove
@requires_resource('walltime') decorator.
Disable pedantic check for c++03 (unlimited API)
Also add a check for c++03 *limited* API, which passes in pedantic mode
after removing a comma in the `PySendResult` declaration, and allowing
`long long`.
Add support for generating UUIDv7 objects according to RFC 9562, §5.7 [1].
The functionality is provided by the `uuid.uuid7()` function. The implementation
is based on a 42-bit counter as described by Method 1, §6.2 [2] and guarantees
monotonicity within the same millisecond.
[1]: https://www.rfc-editor.org/rfc/rfc9562.html#section-5.7
[2]: https://www.rfc-editor.org/rfc/rfc9562.html#section-6.2
---------
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Victor Stinner <vstinner@python.org>
Co-authored-by: Éric <merwok@netwok.org>
Replace hardcoded delay (100 ms) with a loop awaiting until a
condition is true: replace assertReturnsIfImplemented() with
assertReachesEventually().
Use sleeping_retry() in assertReachesEventually() to tolerate slow
buildbots and raise an exception on timeout (30 seconds).
There used to be a meaningful distinction between these modules: `pathlib`
imported `pathlib._abc` but not `pathlib.types`. This is no longer the
case (neither module is imported), so we move the ABCs as follows:
- `pathlib._abc.JoinablePath` --> `pathlib.types._JoinablePath`
- `pathlib._abc.ReadablePath` --> `pathlib.types._ReadablePath`
- `pathlib._abc.WritablePath` --> `pathlib.types._WritablePath`
* Mark functions that will never be called with # pragma: no cover
* Fix testpatch.PatchTest.test_exit_idempotent
.stop() and __exit__ have subtly different code paths, so to really test __exit__ idempotency, we need to call it specifically twice.
Since we plan to introduce a built-in implementation for HMAC based on HACL*,
it becomes important for the HMAC tests to be flexible enough to avoid code
duplication.
In addition to the new layout based on mixin classes, we extend test coverage by
also testing the `__repr__` of HMAC objects and the HMAC one-shot functions.
We also fix the import to `_sha256` which, since gh-101924, resulted in some tests being
skipped as the module is no more available (its content was moved to the `_sha2` module).
This adds two new methods to `multiprocessing`'s `ProcessPoolExecutor`:
- **`terminate_workers()`**: forcefully terminates worker processes using `Process.terminate()`
- **`kill_workers()`**: forcefully kills worker processes using `Process.kill()`
These methods provide users with a direct way to stop worker processes without `shutdown()` or relying on implementation details, addressing situations where immediate termination is needed.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Commit-message-mostly-authored-by: Claude Sonnet 3.7 (because why not -greg)
Add support for generating UUIDv6 objects according to RFC 9562, §5.6 [1].
The functionality is provided by the `uuid.uuid6()` function which takes as inputs an optional 48-bit
hardware address and an optional 14-bit clock sequence. The UUIDv6 temporal fields are ordered
differently than those of UUIDv1, thereby providing improved database locality.
[1]: https://www.rfc-editor.org/rfc/rfc9562.html#section-5.6
---------
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Co-authored-by: Victor Stinner <vstinner@python.org>
Remove the *mode*, *parents* and *exist_ok* arguments from
`WritablePath.mkdir()`. These arguments imply support for POSIX permissions
and checking for preexistence of the path or its parents, but subclasses of
`WritablePath` may not have these capabilities.
The public `Path.mkdir()` method retains these arguments.
Remove `ReadablePath` methods duplicated by `ReadablePath.info`. To be
specific, we remove `exists()`, `is_dir()`, `is_file()` and `is_symlink()`.
The public `Path` class retains these methods.
This broke tests on the 'aarch64 Fedora Stable Clang Installed 3.x' and
'AMD64 Fedora Stable Clang Installed 3.x' build bots.
This reverts commit da4899b94a.
## Filtered recursive walk
Expanding a recursive `**` segment entails walking the entire directory
tree, and so any subsequent pattern segments (except special segments) can
be evaluated by filtering the expanded paths through a regex. For example,
`glob.glob("foo/**/*.py", recursive=True)` recursively walks `foo/` with
`os.scandir()`, and then filters paths through a regex based on "`**/*.py`,
with no further filesystem access needed.
This fixes an issue where `glob()` could return duplicate results.
## Tracking path existence
We store a flag alongside each path indicating whether the path is
guaranteed to exist. As we process the pattern:
- Certain special pattern segments (`""`, `"."` and `".."`) leave the flag
unchanged
- Literal pattern segments (e.g. `foo/bar`) set the flag to false
- Wildcard pattern segments (e.g. `*/*.py`) set the flag to true (because
children are found via `os.scandir()`)
- Recursive pattern segments (e.g. `**`) leave the flag unchanged for the
root path, and set it to true for descendants discovered via
`os.scandir()`.
If the flag is false at the end, we call `lstat()` on each path to filter
out missing paths.
## Minor speed-ups
- Exclude paths that don't match a non-terminal non-recursive wildcard
pattern _prior_ to calling `is_dir()`.
- Use a stack rather than recursion to implement recursive wildcards.
- This fixes a recursion error when globbing deep trees.
- Pre-compile regular expressions and pre-join literal pattern segments.
- Convert to/from `bytes` (a minor use-case) in `iglob()` rather than
supporting `bytes` throughout. This particularly simplifies the code
needed to handle relative bytes paths with `dir_fd`.
- Avoid calling `os.path.join()`; instead we keep paths in a normalized
form and append trailing slashes when needed.
- Avoid calling `os.path.normcase()`; instead we use case-insensitive regex
matching.
## Implementation notes
Much of this functionality is already present in pathlib's implementation
of globbing. The specific additions we make are:
1. Support for `dir_fd`
2. Support for `include_hidden`
3. Support for generating paths relative to `root_dir`
This unifies the implementations of globbing in the `glob` and `pathlib`
modules.
Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
* Combine _GUARD_GLOBALS_VERSION_PUSH_KEYS and _LOAD_GLOBAL_MODULE_FROM_KEYS into _LOAD_GLOBAL_MODULE
* Combine _GUARD_BUILTINS_VERSION_PUSH_KEYS and _LOAD_GLOBAL_BUILTINS_FROM_KEYS into _LOAD_GLOBAL_BUILTINS
* Combine _CHECK_ATTR_MODULE_PUSH_KEYS and _LOAD_ATTR_MODULE_FROM_KEYS into _LOAD_ATTR_MODULE
* Remove stack transient in LOAD_ATTR_WITH_HINT
A reference loop was resulting in the `fileobj` held by the `GzipFile`
being closed before the `GzipFile`.
The issue started with gh-89550 in 3.12, but was hidden in most cases
until 3.13 when gh-62948 made it more visible.
Add two optional, traling elements in the AF_BLUETOOTH socket address tuple:
- l2_cid, to allow e.g raw LE ATT connections
- l2_bdaddr_type. To be able to connect L2CAP sockets to Bluetooth LE devices,
the l2_bdaddr_type must be set to BDADDR_LE_PUBLIC or BDADDR_LE_RANDOM.
There are a number of data races in the default build without
suppressions that are exposed by this test. Disable the test for now
under TSAN until we have suppressions or fix the data races.
The `free_work_item()` function in QSBR may call arbitrary code via
Python object destructors, which may reenter the QSBR code. Reorder
the processing of work items to be robust to reentrancy.
Also fix the TODO for the out of memory situation.
* Revert "gh-128982: Substitute regular expression in `http.cookiejar.join_header_words` for an efficient alternative (GH-128983)"
This reverts commit 56e1900681.
* Add tests
The use of PySys_GetObject() and _PySys_GetAttr(), which return a borrowed
reference, has been replaced by using one of the following functions, which
return a strong reference and distinguish a missing attribute from an error:
_PySys_GetOptionalAttr(), _PySys_GetOptionalAttrString(),
_PySys_GetRequiredAttr(), and _PySys_GetRequiredAttrString().
There are multiple data races reported when running the
InterpreterPoolMixin tests, but it's still useful to run the other
test_concurrent_futures tests under TSAN.
Add test_concurrent_futures to the TSAN test suite.
* Implement C recursion protection with limit pointers for Linux, MacOS and Windows
* Remove calls to PyOS_CheckStack
* Add stack protection to parser
* Make tests more robust to low stacks
* Improve error messages for stack overflow
In `pathlib.Path.copy()` and `move()`, return a fresh `Path` object with an
unpopulated `info` attribute, rather than a `Path` object with information
recorded *prior* to the path's creation.
Revert "GH-91079: Implement C stack limits using addresses, not counters. (GH-130007)" for now
Unfortunatlely, the change broke some buildbots.
This reverts commit 2498c22fa0.
The SyncManager provided support for various data structures such as dict, list, and queue, but oddly, not set.
This introduces support for set by defining SetProxy and registering it with SyncManager.
---
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
To support virtual terminal mode in Windows PYREPL, we need a scanner
to read over the supported escaped VT sequences.
Windows REPL input was using virtual key mode, which does not support
terminal escape sequences. This patch calls `SetConsoleMode` properly
when initializing and send sequences to enable bracketed-paste modes
to support verbatim copy-and-paste.
Signed-off-by: y5c4l3 <y5c4l3@proton.me>
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Co-authored-by: Pablo Galindo Salgado <Pablogsal@gmail.com>
Co-authored-by: Dustin L. Howett <dustin@howett.net>
Co-authored-by: wheeheee <104880306+wheeheee@users.noreply.github.com>
First, write the test name without color. Then, write the test name
and the result with color. Each test is displayed twice.
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
The `test_trace.test_coverage_ignore` test would fail if you had
`setuptools` installed, such as in `~/.local/lib/python3.xxx/site-packages/`.
Ignore everything in `sys.path` when running the test.
The code is still flawed, because it does not recognize class and static
methods, and the first argument is not removed from the signature of
bound methods, but at least it does not worse than in 3.13 and older.
Ensure colorize tests will run on dumb terminals (or environment with TERM=dumb set)
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
* 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
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.
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>
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.
Sam Gross