* gh-112529: Implement GC for free-threaded builds
This implements a mark and sweep GC for the free-threaded builds of
CPython. The implementation relies on mimalloc to find GC tracked
objects (i.e., "containers").
The iterator returned by ElementTree.iterparse() may hold on to a file
descriptor. The reference cycle prevented prompt clean-up of the file
descriptor if the returned iterator was not exhausted.
This improves the implementation in gh-106643.
Previously, venv passed "(<prompt>) " to the activation scripts, but we want
to provide the original value so that users can inspect it in the
$VIRTUAL_ENV_PROMPT env var.
Note: Lib/venv/scripts/common/Activate.ps1 surrounded the prompt value with
parens a second time, so no change was necessary in that file.
Always set a _MainThread as a main thread after os.fork() is called from
a thread started not by the threading module.
A new _MainThread was already set as a new main thread after fork if
threading.current_thread() was not called for a foreign thread before fork.
Now, if it was called before fork, the implicitly created _DummyThread will
be turned into _MainThread after fork.
It fixes, in particularly, an incompatibility of _DummyThread with
the threading shutdown logic which relies on the main thread
having tstate_lock.
Co-authored-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
The terminal CR -> NL mapping setting should be inherited in cbreak mode as OSes do not specify altering it as part of their stty cbreak mode definition.
* Fix writing the retrieved binary file to stdout.
* Add a newline after writing warnings to stderr.
* Fix a TypeError if the netrc file doesn't contain a host/default entry.
* Improve the usage message.
Some socket tests related to sending file descriptors cause a file descriptor leak on macOS, all of them tests that send one or more descriptors than cannot be received on the read end. This appears to be a platform bug.
This PR skips those tests when doing a refleak test run to avoid hiding other problems.
I believe I added this while chasing some performance of hash functions
when I first created hashlib. It hasn't been used since, is frankly
trivial, and not a test.
Use `_make_child_entry()` rather than `_make_child_relpath()` to retrieve
path objects for directories to visit. This saves the allocation of one
path object per directory in user subclasses of `PathBase`, and avoids a
second loop.
This trick does not apply when walking top-down, because users can affect
the walk by modifying *dirnames* in-place.
A side effect of this change is that, in bottom-up mode, subdirectories of
each directory are visited in reverse order, and that this order doesn't
match that of the names in *dirnames*. I suspect this is fine as the
order is arbitrary anyway.
Allow `os.PathLike` objects to be passed as patterns to `pathlib.Path.glob()` and `rglob()`. (It's already possible to use them in `PurePath.match()`)
While we're in the area:
- Allow empty glob patterns in `PathBase` (but not `Path`)
- Speed up globbing in `PathBase` by generating paths with trailing slashes only as a final step, rather than for every intermediate directory.
- Simplify and speed up handling of rare patterns involving both `**` and `..` segments.
* gh-104522: Fix test_subprocess failure when build Python in the root home directory
EPERM is raised when setreuid() fails.
EACCES is set in execve() when the test user has not access to sys.executable.
gh-113205: test_multiprocessing.test_terminate: Test the API works on threadpools
Threads can't be forced to terminate (without potentially corrupting too much
state), so the expected behaviour of `ThreadPool.terminate` is to wait for
the currently executing tasks to finish.
The entire test was skipped in GH-110848 (0e9c364f4a).
Instead of skipping it entirely, we should ensure the API eventually succeeds:
use a shorter timeout.
For the record: on my machine, when the test is un-skipped, the task manages to
start in about 1.5% cases.
Only set filename to cwd if it was caused by failed chdir(cwd).
_fork_exec() now returns "noexec:chdir" for failed chdir(cwd).
Co-authored-by: Robert O'Shea <PurityLake@users.noreply.github.com>
Tkinter is a fact, not necessarily a feature.
Reorganize editor key bindings in a logical order
and remove those that do not work, at least on Windows.
Improve shell bindings list.
If *trackfd* is False, the file descriptor specified by *fileno*
will not be duplicated.
Co-authored-by: Erlend E. Aasland <erlend@python.org>
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Small performance improvement of getproxies_environment() when there are many environment variables. In a benchmark with 5k environment variables not related to proxies, and 5 specifying proxies, we get a 10% walltime improvement.
`PyComplex_RealAsDouble()`/`PyComplex_ImagAsDouble` now try to convert
an object to a `complex` instance using its `__complex__()` method
before falling back to the ``__float__()`` method.
PyComplex_ImagAsDouble() also will not silently return 0.0 for
non-complex types anymore. Instead we try to call PyFloat_AsDouble()
and return 0.0 only if this call is successful.
This allows users of the `pathlib-abc` PyPI package to use `posixpath` or
`ntpath` as a path module in versions of Python lacking
`os.path.splitroot()` (3.11 and before).
Path modules provide a subset of the `os.path` API, specifically those
functions needed to provide `PurePathBase` functionality. Each
`PurePathBase` subclass references its path module via a `pathmod` class
attribute.
This commit adds a new `PathModuleBase` class, which provides abstract
methods that unconditionally raise `UnsupportedOperation`. An instance of
this class is assigned to `PurePathBase.pathmod`, replacing `posixpath`.
As a result, `PurePathBase` is no longer POSIX-y by default, and
all its methods raise `UnsupportedOperation` courtesy of `pathmod`.
Users who subclass `PurePathBase` or `PathBase` should choose the path
syntax by setting `pathmod` to `posixpath`, `ntpath`, `os.path`, or their
own subclass of `PathModuleBase`, as circumstances demand.
Fix a bug in the regex used for parsing a string input to the `fractions.Fraction` constructor. That bug led to an inconsistent exception message being given for some inputs.
---------
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
Increase the backlog for multiprocessing.connection.Listener` objects created
by `multiprocessing.manager` and `multiprocessing.resource_sharer` to
significantly reduce the risk of getting a connection refused error when creating
a `multiprocessing.connection.Connection` to them.
It's wrong for the `PurePathBase` methods to rely so much on `__str__()`.
Instead, they should treat the raw path(s) as opaque objects and leave the
details to `pathmod`.
This commit adds a `PurePathBase._raw_path` property and uses it through
many of the other ABC methods. These methods are all redefined in
`PurePath` and `Path`, so this has no effect on the public classes.
On Windows, `os.path.isabs()` now returns `False` when given a path that
starts with exactly one (back)slash. This is more compatible with other
functions in `os.path`, and with Microsoft's own documentation.
Also adjust `pathlib.PureWindowsPath.is_absolute()` to call
`ntpath.isabs()`, which corrects its handling of partial UNC/device paths
like `//foo`.
Co-authored-by: Jon Foster <jon@jon-foster.co.uk>
Make zipfile.ZipInfo.compress_level public.
A property is used to retain the behavior of the ._compresslevel.
People constructing zipfile.ZipInfo instances to pass into existing APIs to control per-file compression levels already treat this as public, there was never a reason for it not to be.
I used the more modern name compress_level instead of compresslevel as the keyword argument on other ZipFile APIs is called to be consistent with compress_type and a general long term preference of not runningwordstogether without a separator in names.
sqlite3.Connection.iterdump now ensures that foreign key support is
disabled before dumping the database schema, if there is any foreign key
violation.
Co-authored-by: Erlend E. Aasland <erlend@python.org>
test_configdialog.HighPageTest.test_highlight_target_text_mouse fails
if a line of the Highlight tab text sample is not visible. If so, bbox()
in click_char() returns None and the unpacking iteration fails.
This occurred on a Devuan Linux system. Fix by moving the
'see character' call inside click_char, just before the bbox call.
Also, reduce the click_char calls to just one per tag name and
replace the other nested function with a dict comprehension.
* improve the assert for test_one_environment_variable
* skip some test in test_subprocess when python is configured with shared
* also skip the test if AddressSanitizer is enabled
---------
Co-authored-by: Steve Dower <steve.dower@microsoft.com>
On Windows, set _O_NOINHERIT flag on file descriptors
created by os.pipe() and io.WindowsConsoleIO.
Add test_pipe_spawnl() to test_os.
Co-authored-by: Zackery Spytz <zspytz@gmail.com>
Previously the C implementation of pickle.Pickler and pickle.Unpickler
classes did not have such methods and they could only be used if
they were overloaded in subclasses or set as instance attributes.
Fixed calling super().persistent_id() and super().persistent_load() in
subclasses of the C implementation of pickle.Pickler and pickle.Unpickler
classes. It no longer causes an infinite recursion.
If other exception was raised during exiting an expired
asyncio.timeout() block, insert TimeoutError in the exception context
just above the CancelledError.
winfo_id() converts the result of "winfo id" command to integer, but
"winfo pathname" command requires an argument to be a hexadecimal number
on Win64.
When an `StopIteration` raises into `asyncio.Future`, this will cause
a thread to hang. This commit address this by not raising an exception
and silently transforming the `StopIteration` with a `RuntimeError`,
which the caller can reconstruct from `fut.exception().__cause__`
Refuse to guess what a user means when they initialise a pathlib ABC
without any positional arguments. In mainline pathlib it's normalised to
`.`, but in the ABCs this guess isn't appropriate; for example, the path
type may not represent the current directory as `.`, or may have no concept
of a "current directory" at all.
Apply pathlib's normalization and performance tuning in `pathlib.PurePath`, but not `pathlib._abc.PurePathBase`.
With this change, the pathlib ABCs do not normalize away alternate path separators, empty segments, or dot segments. A single string given to the initialiser will round-trip by default, i.e. `str(PurePathBase(my_string)) == my_string`. Implementors can set their own path domain-specific normalization scheme by overriding `__str__()`
Eliminating path normalization makes maintaining and caching the path's parts and string representation both optional and not very useful, so this commit moves the `_drv`, `_root`, `_tail_cached` and `_str` slots from `PurePathBase` to `PurePath`. Only `_raw_paths` and `_resolving` slots remain in `PurePathBase`. This frees the ABCs from the burden of some of pathlib's hardest-to-understand code.
Restore full battle-tested implementations of `PurePath.[is_]relative_to()`. These were recently split up in 3375dfe and a15a773.
In `PurePathBase`, add entirely new implementations based on `_stack`, which itself calls `pathmod.split()` repeatedly to disassemble a path. These new implementations preserve features like trailing slashes where possible, while still observing that a `..` segment cannot be added to traverse an empty or `.` segment in *walk_up* mode. They do not rely on `parents` nor `__eq__()`, nor do they spin up temporary path objects.
Unfortunately calling `pathmod.relpath()` isn't an option, as it calls `abspath()` and in turn `os.getcwd()`, which is impure.
Implement `parts` using `_stack`, which itself calls `pathmod.split()`
repeatedly. This avoids use of `_tail`, which will be moved to `PurePath`
shortly.
Test the following features for _ctypes types:
- disallow instantiation
- inheritance (MRO)
- immutability
- type name
The following _ctypes types are tested:
- Array
- CField
- COMError
- PyCArrayType
- PyCFuncPtrType
- PyCPointerType
- PyCSimpleType
- PyCStructType
- Structure
- Union
- UnionType
- _CFuncPtr
- _Pointer
- _SimpleCData
Co-authored-by: Erlend E. Aasland <erlend.aasland@protonmail.com>
Determine the support of the Kyiv timezone by checking the result of
astimezone() which uses the system tz database and not the one
populated by zoneinfo.
- Add `__slots__` to dummy path classes.
- Return namedtuple rather than `os.stat_result` from `DummyPath.stat()`.
- Reduce maximum symlink count in `DummyPathWithSymlinks.resolve()`.
`PurePathBase` does not define `__eq__()`, and so we have no business checking path equality in `test_eq_common` and `test_equivalences`. The tests only pass at the moment because we define the test class's `__eq__()` for use elsewhere.
Also move `test_parse_path_common` into the main pathlib test suite. It exercises a private `_parse_path()` method that will be moved to `PurePath` soon.
Lastly move a couple more tests concerned with optimisations and path normalisation.
It can be used to set the location of a .python_history file
---------
Co-authored-by: Levi Sabah <0xl3vi@gmail.com>
Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>
Allow extraction (no-op) of a "/" folder in a zipfile, they are commonly added by some archive creation tools.
Co-authored-by: Erlend E. Aasland <erlend@python.org>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
Remove LibreSSL specific workaround ifdefs from `_ssl.c` and delete the non-version-specific `_ssl_data.h` file (relevant for OpenSSL < 1.1.1, which we no longer support per PEP 644).
Co-authored-by: Christian Heimes <christian@python.org>
Co-authored-by: Gregory P. Smith <greg@krypto.org>
Replace use of `_from_parsed_parts()` with `with_segments()` in
`PurePathBase.relative_to()`, and move the assignment of `_drv`, `_root`
and `_tail_cached` slots into `PurePath.relative_to()`.
Slightly improve `pathlib.Path.glob()` tests for symlink loop handling
When filtering results, ignore paths with more than one `linkD/` segment,
rather than all paths below the first `linkD/` segment. This allows us
to test that other paths under `linkD/` are correctly returned.
Split test cases for invalid names into dedicated test methods. This will
make it easier to refactor tests for invalid name handling in ABCs later.
No change of coverage, just a change of test suite organisation.
The `DummyPurePath` and `DummyPath` test classes are simple subclasses of
`PurePathBase` and `PathBase`. This commit adds `__repr__()` methods to the
dummy classes, which makes debugging test failures less painful.
Raise auditing events in `pathlib.Path.glob()`, `rglob()` and `walk()`,
but not in `pathlib._abc.PathBase` methods. Also move generation of a
deprecation warning into `pathlib.Path` so it gets the right stack level.
This splits part of Modules/gcmodule.c of into Python/gc.c, which
now contains the core garbage collection implementation. The Python
module remain in the Modules/gcmodule.c file.
- Only attempt to figure out whether protocol members are "method members" or not if the class is marked as a runtime protocol. This information is irrelevant for non-runtime protocols; we can safely skip the risky introspection for them.
- Only do the risky getattr() calls in one place (the runtime_checkable class decorator), rather than in three places (_ProtocolMeta.__init__, _ProtocolMeta.__instancecheck__ and _ProtocolMeta.__subclasscheck__). This reduces the number of locations in typing.py where the risky introspection could go wrong.
- For runtime protocols, if determining whether a protocol member is callable or not fails, give a better error message. I think it's reasonable for us to reject runtime protocols that have members which raise strange exceptions when you try to access them. PEP-544 clearly states that all protocol member must be callable for issubclass() calls against the protocol to be valid -- and if a member raises when we try to access it, there's no way for us to figure out whether it's a callable member or not!
* gh-53502: Fixes for tests in gh-113363
* Use 32-bit compatible date in test_dump_naive_datetime_with_aware_datetime_option
* Saving non-aware datetimes will use the old behaviour regardless of the aware_datimetime setting
The length field of StgDictObject for Structure class contains now
the total number of items in ffi_type_pointer.elements (excluding
the trailing null).
The old behavior of using the number of elements in the parent class can
cause the array to be truncated when it is copied, especially when there
are multiple layers of subclassing.
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Run expensive tests for walking and globbing from `test_pathlib` but not
`test_pathlib_abc`. The ABCs are not as tightly optimised as the classes
in top-level `pathlib`, and so these tests are taking rather a long time on
some buildbots. Coverage of the main `pathlib` classes should suffice.
Rework error handling in the C preprocessor helper. Instead of monkey-
patching the cpp.Monitor.fail() method from within clinic.py, rewrite
cpp.py to use a subclass of the ClinicError exception. As a side-effect,
ClinicError is moved into Tools/clinic/libclinic/errors.py.
Yak-shaving in preparation for putting cpp.py into libclinic.
The `pathlib._abc` module will be made available as a PyPI backport
supporting Python 3.8+. The `warnings._deprecated()` function was only
added last year, and it's private from an external package perspective, so
here we switch to `warnings.warn()` instead.
Do not use the locale-specific default encoding in `PathBase.read_text()`
and `write_text()`. Locale settings shouldn't influence the operation of
these base classes, which are intended mostly for implementing rich paths
on *nonlocal* filesystems.
Previously some error messages complained about incompatible
combinations of directives that are not contained in the format string.
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
This fixes a divergence between the Python and C implementations of pickle
for protocol 0, such that it pickle.py fails to re-use the first pickled
representation of strings involving characters that have to be escaped.
In case the spawned process is setuid, we may not be able to send
signals to it, in which case our .kill() call will raise
PermissionError.
Ignore that in order to avoid .close() raising an exception. Hopefully
the process will exit as a result of receiving EOF on its stdin.
Move the following global helpers into libclinic:
- format_escape()
- normalize_snippet()
- wrap_declarations()
Also move strip_leading_and_trailing_blank_lines() and make it internal to libclinic.
Previously it worked differently if dst is a symbolic link:
it modified the permission bits of dst itself rather than the file
it points to if follow_symlinks is true or src is not a symbolic link,
and did nothing if follow_symlinks is false and src is a symbolic link.
Also document similar changes in shutil.copystat().
Split up clinic.py by establishing libclinic as a support package for
Argument Clinic. Get rid of clinic.py globals by either making them
class members, or by putting them into libclinic.
- Move INCLUDE_COMMENT_COLUMN to BlockPrinter
- Move NO_VARARG to CLanguage
- Move formatting helpers to libclinic
- Move some constants to libclinic (and annotate them as Final)
Replace the internal accumulator APIs by using lists of strings and join().
Yak-shaving for separating out formatting code into a separate file.
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Change the value of `pathlib._abc.PurePathBase.pathmod` from `os.path` to
`posixpath`.
User subclasses of `PurePathBase` and `PathBase` previously used the host
OS's path syntax, e.g. backslashes as separators on Windows. This is wrong
in most use cases, and likely to catch developers out unless they test on
both Windows and non-Windows machines.
In this patch we change the default to POSIX syntax, regardless of OS. This
is somewhat arguable (why not make all aspects of syntax abstract and
individually configurable?) but an improvement all the same.
This change has no effect on `PurePath`, `Path`, nor their subclasses. Only
private APIs are affected.
Sam Gross
Tian Gao
AN Long
Irit Katriel
Michael Droettboom
Steve Dower
Mark Shannon
Serhiy Storchaka
Brett Cannon
Nikita Sobolev
Fabio Zadrozny
Jim Porter
Gregory P. Smith
Ronald Oussoren
Hugo van Kemenade
Barney Gale
Skip Montanaro
keithasaurus
Phillip Schanely
Christopher Chavez
kcatss
Terry Jan Reedy
Petr Viktorin
Ethan Furman
buermarc
Sheidan
Thomas Wouters
Radislav Chugunov
Shantanu
Victor Stinner
Mike Zimin
Jérome Perrin
Zackery Spytz
Raphaël Marinier
Kamil Turek
Sergey B Kirpichev
Kirill Podoprigora
Crowthebird
Alois Klink
Ken Jin
Pablo Galindo Salgado
Mariusz Felisiak
Peter Lazorchak
Jamie Phan
Stefano Rivera
Vinay Sajip
Kristján Valur Jónsson
Rami
Grigoriev Semyon
Alex Waygood
Guido van Rossum
wookie184
Filip Łapkiewicz
Itamar Oren
Jeffrey Kintscher
Samet YASLAN
Ankit Kumar Pandey
Erlend E. Aasland
Stanley
Gordon P. Hemsley
Xu Song
Jeff Allen
Yilei Yang
Allison Karlitskaya