Asymptotically faster (O(n log n)) str->int for very large strings, leveraging the faster multiplication scheme in the C-coded `_decimal` when available. This is used instead of the current Karatsuba-limited method starting at 2 million digits.
Lots of opportunity remains for fine-tuning. Good targets include changing BYTELIM, and possibly changing the internal output base (from 256 to a higher number of bytes).
Doing this was substantial work, and many of the new lines are actually comments giving correctness proofs. The obvious approaches sticking to integers were too slow to be useful, so this is doing variable-precision decimal floating-point arithmetic. Much faster, but worst-possible rounding errors have to be wholly accounted for, using as little precision as possible.
Special thanks to Serhiy Storchaka for asking many good questions in his code reviews!
Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>
Co-authored-by: sstandre <43125375+sstandre@users.noreply.github.com>
Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>
Co-authored-by: Nice Zombies <nineteendo19d0@gmail.com>
Suppress all `OSError` exceptions from `pathlib.Path.exists()` and `is_*()`
rather than a selection of more common errors as we do presently. Also
adjust the implementations to call `os.path.exists()` etc, which are much
faster on Windows thanks to GH-101196.
Follow-up of gh-101693. The previous DeprecationWarning is replaced with
raising sqlite3.ProgrammingError.
Co-authored-by: Hugo van Kemenade <1324225+hugovk@users.noreply.github.com>
Remove support for supplying additional positional arguments to
`PurePath.relative_to()` and `is_relative_to()`. This has been deprecated
since Python 3.12.
_PyWeakref_ClearRef was previously exposed in the public C-API, although
it begins with an underscore and is not documented. It's used by a few
C-API extensions. There is currently no alternative public API that can
replace its use.
_PyWeakref_ClearWeakRefsExceptCallbacks is the only thread-safe way to
use _PyWeakref_ClearRef in the free-threaded build. This exposes the C
symbol, but does not make the API public.
Some embedders and extensions include parts of the internal API. The
pycore_mimalloc.h file is transitively include by a number of other
internal headers. This avoids include errors for code that was
already including those headers.
The `list_preallocate_exact` function did not zero initialize array
contents. In the free-threaded build, this could expose uninitialized
memory to concurrent readers between the call to
`list_preallocate_exact` and the filling of the array contents with
items.
Callbacks registered in the tkinter module now take arguments as
various Python objects (int, float, bytes, tuple), not just str.
To restore the previous behavior set tkinter module global wantobject to 1
before creating the Tk object or call the wantobject() method of the Tk object
with argument 1.
Calling it with argument 2 restores the current default behavior.
Fix an edge case in `binascii.a2b_base64` strict mode, where
excessive padding was not detected when no padding is necessary.
Co-authored-by: Terry Jan Reedy <tjreedy@udel.edu>
Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>
Rationale
=========
argparse performs a complex formatting of the usage for argument grouping
and for line wrapping to fit the terminal width. This formatting has been
a constant source of bugs for at least 10 years (see linked issues below)
where defensive assertion errors are triggered or brackets and paranthesis
are not properly handeled.
Problem
=======
The current implementation of argparse usage formatting relies on regular
expressions to group arguments usage only to separate them again later
with another set of regular expressions. This is a complex and error prone
approach that caused all the issues linked below. Special casing certain
argument formats has not solved the problem. The following are some of
the most common issues:
- empty `metavar`
- mutually exclusive groups with `SUPPRESS`ed arguments
- metavars with whitespace
- metavars with brackets or paranthesis
Solution
========
The following two comments summarize the solution:
- https://github.com/python/cpython/issues/82091#issuecomment-1093832187
- https://github.com/python/cpython/issues/77048#issuecomment-1093776995
Mainly, the solution is to rewrite the usage formatting to avoid the
group-then-separate approach. Instead, the usage parts are kept separate
and only joined together at the end. This allows for a much simpler
implementation that is easier to understand and maintain. It avoids the
regular expressions approach and fixes the corresponding issues.
This closes the following GitHub issues:
- #62090
- #62549
- #77048
- #82091
- #89743
- #96310
- #98666
These PRs become obsolete:
- #15372
- #96311
Add missing import to code that handles too large files and offsets.
Use list, not tuple, for a mutable sequence.
Add tests to prevent similar mistakes.
---------
Co-authored-by: Gregory P. Smith [Google LLC] <greg@krypto.org>
Co-authored-by: Kirill Podoprigora <kirill.bast9@mail.ru>
This change makes sure all extension/builtin modules have their init function run first by the main interpreter before proceeding with import in the original interpreter (main or otherwise). This means when the import of a single-phase init module fails in an isolated subinterpreter, it won't tie any global state/callbacks to the subinterpreter.
Add the ability to enable/disable the GIL at runtime, and use that in
the C module loading code.
We can't know before running a module init function if it supports
free-threading, so the GIL is temporarily enabled before doing so. If
the module declares support for running without the GIL, the GIL is
later disabled. Otherwise, the GIL is permanently enabled, and will
never be disabled again for the life of the current interpreter.
Now, such classes will no longer require changes in Python 3.13 in the normal case.
The test suite for robotframework passes with no DeprecationWarnings under this PR.
I also added a new DeprecationWarning for the case where `_field_types` exists
but is incomplete, since that seems likely to indicate a user mistake.
Add _PyType_LookupRef and use incref before setting attribute on type
Makes setting an attribute on a class and signaling type modified atomic
Avoid adding re-entrancy exposing the type cache in an inconsistent state by decrefing after type is updated
* Add PhotoImage.read() to read an image from a file.
* Add PhotoImage.data() to get the image data.
* Add background and grayscale parameters to PhotoImage.write().
* Add the PhotoImage method copy_replace() to copy a region
from one image to other image, possibly with pixel zooming and/or
subsampling.
* Add from_coords parameter to PhotoImage methods copy(), zoom() and subsample().
* Add zoom and subsample parameters to PhotoImage method copy().
The designated initializer syntax in static inline functions in pycore_backoff.h
causes problems for C++ or MSVC users who aren't yet using C++20.
While internal, pycore_backoff.h is included (indirectly, via pycore_code.h)
by some key 3rd party software that does so for speed.
This is *not* sufficient for the final 3.13 release, but it will do for beta 1:
- What's new entry
- Updated changelog entry (news blurb)
- Mention the proxy for f_globals in the datamodel and Python frame object docs
This doesn't have any C API details (what's new refers to the PEP).
For converting large ints to strings, CPython invokes a function in _pylong.py,
which uses the decimal module to implement an asymptotically waaaaay
sub-quadratic algorithm. But if the C decimal module isn't available, CPython
uses _pydecimal.py instead. Which in turn frequently does str(int). If the int
is very large, _pylong ends up doing the work, which in turn asks decimal to do
"big" arithmetic, which in turn calls str(big_int), which in turn ... it can
become infinite mutual recursion.
This change introduces a different int->str function that doesn't use decimal.
It's asymptotically worse, "Karatsuba time" instead of quadratic time, so
still a huge improvement. _pylong switches to that when the C decimal isn't
available. It is also used for not too large integers (less than 450_000 bits),
where it is faster (up to 2 times for 30_000 bits) than the asymptotically
better implementation that uses the C decimal.
Co-authored-by: Tim Peters <tim.peters@gmail.com>
* Initial stab.
* Test the tentative fix. Hangs "forever" without this change.
* Move the new test to a better spot.
* New comment to explain why _convert_to_str allows any poewr of 10.
* Fixed a comment, and fleshed out an existing test that appeared unfinished.
* Added temporary asserts. Or maybe permanent ;-)
* Update Lib/_pydecimal.py
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
* Remove the new _convert_to_str().
Serhiy and I independently concluded that exact powers of 10
aren't possible in these contexts, so just checking the
string length is sufficient.
* At least for now, add the asserts to the other block too.
* 📜🤖 Added by blurb_it.
---------
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
This PR adds the ability to enable the GIL if it was disabled at
interpreter startup, and modifies the multi-phase module initialization
path to enable the GIL when loading a module, unless that module's spec
includes a slot indicating it can run safely without the GIL.
PEP 703 called the constant for the slot `Py_mod_gil_not_used`; I went
with `Py_MOD_GIL_NOT_USED` for consistency with gh-104148.
A warning will be issued up to once per interpreter for the first
GIL-using module that is loaded. If `-v` is given, a shorter message
will be printed to stderr every time a GIL-using module is loaded
(including the first one that issues a warning).
The function returns `True` or `False` depending on whether the GIL is
currently enabled. In the default build, it always returns `True`
because the GIL is always enabled.
Now inspect.signature() supports references to the module globals in
parameter defaults on methods in extension modules. Previously it was
only supported in functions. The workaround was to specify the fully
qualified name, including the module name.
Add "Raw" variant of PyTime functions:
* PyTime_MonotonicRaw()
* PyTime_PerfCounterRaw()
* PyTime_TimeRaw()
Changes:
* Add documentation and tests. Tests release the GIL while calling
raw clock functions.
* py_get_system_clock() and py_get_monotonic_clock() now check that
the GIL is hold by the caller if raise_exc is non-zero.
* Reimplement "Unchecked" functions with raw clock functions.
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Account for `add_stopiteration_handler` pushing a block for `async with`.
To allow generator functions that previously almost hit the `CO_MAXBLOCKS`
limit by nesting non-async blocks, the limit is increased by 1.
This increase allows one more block in non-generator functions.
Add code and config for a minimal Android app, and instructions to build and run it.
Improve Android build instructions in general.
Add a tool subcommand to download the Gradle wrapper (with its binary blob). Android
studio must be downloaded manually (due to the license).
The code for Tier 2 is now only compiled when configured
with `--enable-experimental-jit[=yes|interpreter]`.
We drop support for `PYTHON_UOPS` and -`Xuops`,
but you can disable the interpreter or JIT
at runtime by setting `PYTHON_JIT=0`.
You can also build it without enabling it by default
using `--enable-experimental-jit=yes-off`;
enable with `PYTHON_JIT=1`.
On Windows, the `build.bat` script supports
`--experimental-jit`, `--experimental-jit-off`,
`--experimental-interpreter`.
In the C code, `_Py_JIT` is defined as before
when the JIT is enabled; the new variable
`_Py_TIER2` is defined when the JIT *or* the
interpreter is enabled. It is actually a bitmask:
1: JIT; 2: default-off; 4: interpreter.
* Allow to specify the signature of custom callable instances of extension
type by the __text_signature__ attribute.
* Specify signatures of operator.attrgetter, operator.itemgetter, and
operator.methodcaller instances.
While properties like IPv6Address.is_private account for IPv4-mapped
IPv6 addresses, such as for example:
>>> ipaddress.ip_address("192.168.0.1").is_private
True
>>> ipaddress.ip_address("::ffff:192.168.0.1").is_private
True
...the same doesn't currently apply to the is_loopback property:
>>> ipaddress.ip_address("127.0.0.1").is_loopback
True
>>> ipaddress.ip_address("::ffff:127.0.0.1").is_loopback
False
At minimum, this inconsistency between different properties is
counter-intuitive. Moreover, ::ffff:127.0.0.0/104 is for all intents and
purposes a loopback address, and should be treated as such.
sqlite3.iterdump() depends on the row factory returning resulting rows
as tuples; it will fail with custom row factories like for example a
dict factory.
With this commit, we explicitly reset the row factory of the cursor used
by iterdump(), so we always get predictable results. This does not
affect the row factory of the parent connection.
Co-authored-by: Mariusz Felisiak <felisiak.mariusz@gmail.com>
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Barney Gale
Tim Peters
Donghee Na
Xie Yanbo
Kirill Podoprigora
Brandt Bucher
Serhiy Storchaka
Erlend E. Aasland
Michał Górny
Tian Gao
Jacob Walls
Savannah Ostrowski
Hugo van Kemenade
Petr Viktorin
Jelle Zijlstra
mpage
Sam Gross
Dobatymo
Steve Dower
Radislav Chugunov
Nikita Sobolev
Raymond Hettinger
Thomas Wouters
Alex Waygood
chrysn
Youfu Zhang
Ali Hamdan
John Sirois
Eric Snow
Brett Simmers
James Hilton-Balfe
Jeong, YunWon
Dino Viehland
Thomas Grainger
Pieter Eendebak
Terry Jan Reedy
Pablo Galindo Salgado
Guido van Rossum
Ryan Batchelder
Davide Rizzo
Irit Katriel
Carl Meyer
Alexander Kanavin
infohash
Raphael Gaschignard
Sergey B Kirpichev
Crowthebird
Nice Zombies
Victor Stinner
Malcolm Smith
Harmen Stoppels
Anthony Shaw
Shantanu
Henrik Tunedal
Cheryl Sabella
Faidon Liambotis
Joe Jevnik
Seth Michael Larson