This is the implementation of PEP683
Motivation:
The PR introduces the ability to immortalize instances in CPython which bypasses reference counting. Tagging objects as immortal allows up to skip certain operations when we know that the object will be around for the entire execution of the runtime.
Note that this by itself will bring a performance regression to the runtime due to the extra reference count checks. However, this brings the ability of having truly immutable objects that are useful in other contexts such as immutable data sharing between sub-interpreters.
* The majority of the monitoring code is in instrumentation.c
* The new instrumentation bytecodes are in bytecodes.c
* legacy_tracing.c adapts the new API to the old sys.setrace and sys.setprofile APIs
We can revisit the options for keeping it global later, if desired. For now the approach seems quite complex, so we've gone with the simpler isolation solution in the meantime.
https://github.com/python/cpython/issues/100227
This reverts commit 87be8d9.
This approach to keeping the interned strings safe is turning out to be too complex for my taste (due to obmalloc isolation). For now I'm going with the simpler solution, making the dict per-interpreter. We can revisit that later if we want a sharing solution.
This is effectively two changes. The first (the bulk of the change) is where we add _Py_AddToGlobalDict() (and _PyRuntime.cached_objects.main_tstate, etc.). The second (much smaller) change is where we update PyUnicode_InternInPlace() to use _Py_AddToGlobalDict() instead of calling PyDict_SetDefault() directly.
Basically, _Py_AddToGlobalDict() is a wrapper around PyDict_SetDefault() that should be used whenever we need to add a value to a runtime-global dict object (in the few cases where we are leaving the container global rather than moving it to PyInterpreterState, e.g. the interned strings dict). _Py_AddToGlobalDict() does all the necessary work to make sure the target global dict is shared safely between isolated interpreters. This is especially important as we move the obmalloc state to each interpreter (gh-101660), as well as, potentially, the GIL (PEP 684).
https://github.com/python/cpython/issues/100227
* Eliminate all remaining uses of Py_SIZE and Py_SET_SIZE on PyLongObject, adding asserts.
* Change layout of size/sign bits in longobject to support future addition of immortal ints and tagged medium ints.
* Add functions to hide some internals of long object, and for setting sign and digit count.
* Replace uses of IS_MEDIUM_VALUE macro with _PyLong_IsCompact().
Moving it valuable with a per-interpreter GIL. However, it is also useful without one, since it allows us to identify refleaks within a single interpreter or where references are escaping an interpreter. This becomes more important as we move the obmalloc state to PyInterpreterState.
https://github.com/python/cpython/issues/102304
The essentially eliminates the global variable, with the associated benefits. This is also a precursor to isolating this bit of state to PyInterpreterState.
Folks that currently read _Py_RefTotal directly would have to start using _Py_GetGlobalRefTotal() instead.
https://github.com/python/cpython/issues/102304
When __getattr__ is defined, python with try to find an attribute using _PyObject_GenericGetAttrWithDict
find nothing is reasonable so we don't need an exception, it will hurt performance.
Add `MS_WINDOWS_DESKTOP`, `MS_WINDOWS_APPS`, `MS_WINDOWS_SYSTEM` and `MS_WINDOWS_GAMES` preprocessor definitions to allow switching off functionality missing from particular API partitions ("partitions" are used in Windows to identify overlapping subsets of APIs).
CPython only officially supports `MS_WINDOWS_DESKTOP` and `MS_WINDOWS_SYSTEM` (APPS is included by normal desktop builds, but APPS without DESKTOP is not covered). Other configurations are a convenience for people building their own runtimes.
`MS_WINDOWS_GAMES` is for the Xbox subset of the Windows API, which is also available on client OS, but is restricted compared to `MS_WINDOWS_DESKTOP`. These restrictions may change over time, as they relate to the build headers rather than the OS support, and so we assume that Xbox builds will use the latest available version of the GDK.
Some incompatible changes had gone in, and the "ignore" lists weren't properly undated. This change fixes that. It's necessary prior to enabling test_check_c_globals, which I hope to do soon.
Note that this does include moving last_resort_memory_error to PyInterpreterState.
https://github.com/python/cpython/issues/90110
Found some duplicate `to`s in the documentation and some code comments and fixed them.
[Misc/NEWS.d/3.12.0a1.rst](ed55c69ebd/Misc/NEWS.d/3.12.0a1.rst) also contains two duplicate `to`s, but I wasn't sure if it's ok to touch that file. Looks auto generated. I'm happy to amend the PR if requested. :)
Automerge-Triggered-By: GH:AlexWaygood
Make docstrings for `as_integer_ratio` consistent across types, and document that
the returned pair is always normalized (coprime integers, with positive denominator).
---------
Co-authored-by: Owain Davies <116417456+OTheDev@users.noreply.github.com>
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
This change is almost entirely moving code around and hiding import state behind internal API. We introduce no changes to behavior, nor to non-internal API. (Since there was already going to be a lot of churn, I took this as an opportunity to re-organize import.c into topically-grouped sections of code.) The motivation is to simplify a number of upcoming changes.
Specific changes:
* move existing import-related code to import.c, wherever possible
* add internal API for interacting with import state (both global and per-interpreter)
* use only API outside of import.c (to limit churn there when changing the location, etc.)
* consolidate the import-related state of PyInterpreterState into a single struct field (this changes layout slightly)
* add macros for import state in import.c (to simplify changing the location)
* group code in import.c into sections
*remove _PyState_AddModule()
https://github.com/python/cpython/issues/101758
* Make sure that the current exception is always normalized.
* Remove redundant type and traceback fields for the current exception.
* Add new API functions: PyErr_GetRaisedException, PyErr_SetRaisedException
* Add new API functions: PyException_GetArgs, PyException_SetArgs
Fix the behaviour of the `__sizeof__` method (and hence the results returned by `sys.getsizeof`) for subclasses of `int`. Previously, `int` subclasses gave identical results to the `int` base class, ignoring the presence of the instance dictionary.
<!-- gh-issue-number: gh-101266 -->
* Issue: gh-101266
<!-- /gh-issue-number -->
We've factored out a struct from the two PyThreadState fields. This accomplishes two things:
* make it clear that the trashcan-related code doesn't need any other parts of PyThreadState
* allows us to use the trashcan mechanism even when there isn't a "current" thread state
We still expect the caller to hold the GIL.
https://github.com/python/cpython/issues/59956
Use C long arithmetic instead of PyLong arithmetic to compute the range length, where possible.
Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
This PR fixes object allocation in long_subtype_new to ensure that there's at least one digit in all cases, and makes sure that the value of that digit is copied over from the source long.
Needs backport to 3.11, but not any further: the change to require at least one digit was only introduced for Python 3.11.
Fixes#101037.
When executing the BUILD_LIST opcode, steal the references from the stack,
in a manner similar to the BUILD_TUPLE opcode. Implement this by offloading
the logic to a new private API, _PyList_FromArraySteal(), that works similarly
to _PyTuple_FromArraySteal().
This way, instead of performing multiple stack pointer adjustments while the
list is being initialized, the stack is adjusted only once and a fast memory
copy operation is performed in one fell swoop.
Fixes behaviour where int (and subtypes like bool) __sizeof__ under-reports true size as it did not take into account the size 1 `ob_digit` array for the zero int.
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
This improves the lives of type annotation users of `float` - which type checkers implicitly treat as `int|float` because that is what most code actually wants. Before this change a `.is_integer()` method could not be assumed to exist on things annotated as `: float` due to the method not existing on both types.
* Uses a better hashing algorithm to get better dispersion and remove commutativity.
* Incorporates `co_firstlineno`, `Py_SIZE(co)`, and bytecode instructions.
* This is now the entire set of criteria used in `code_richcompare`, except for `_PyCode_ConstantKey` (which would incorporate the types of `co_consts` rather than just their values).
* move _PyRuntime.global_objects.interned to _PyRuntime.cached_objects.interned_strings (and use _Py_CACHED_OBJECT())
* rename _PyRuntime.global_objects to _PyRuntime.static_objects
(This also relates to gh-96075.)
https://github.com/python/cpython/issues/90111
builtins and extension module functions and methods that expect boolean values for parameters now accept any Python object rather than just a bool or int type. This is more consistent with how native Python code itself behaves.
* Add API to allow extensions to set callback function on creation and destruction of PyCodeObject
Co-authored-by: Ye11ow-Flash <janshah@cs.stonybrook.edu>
The implementation of __sizeof__() methods using _PyObject_SIZE() now
use an unsigned type (size_t) to compute the size, rather than a signed
type (Py_ssize_t).
Cast explicitly signed (Py_ssize_t) values to unsigned type
(Py_ssize_t).
* code_sizeof() now uses an unsigned type (size_t) to compute the result.
* Fix _PyObject_ComputedDictPointer(): cast _PyObject_VAR_SIZE() to
Py_ssize_t, rather than long: it's a different type on 64-bit Windows.
* Clarify that _PyObject_VAR_SIZE() uses an unsigned type (size_t).
* Change _PyDict_KeysSize() and shared_keys_usable_size() return type
from signed (Py_ssize_t) to unsigned (size_t) type.
* new_values() argument type is now unsigned (size_t).
* init_inline_values() now uses size_t rather than int for the 'i'
iterator variable.
* type.__sizeof__() implementation now uses unsigned (size_t) type.
Fix potential race condition in code patterns:
* Replace "Py_DECREF(var); var = new;" with "Py_SETREF(var, new);"
* Replace "Py_XDECREF(var); var = new;" with "Py_XSETREF(var, new);"
* Replace "Py_CLEAR(var); var = new;" with "Py_XSETREF(var, new);"
Other changes:
* Replace "old = var; var = new; Py_DECREF(var)"
with "Py_SETREF(var, new);"
* Replace "old = var; var = new; Py_XDECREF(var)"
with "Py_XSETREF(var, new);"
* And remove the "old" variable.
The global allocators were stored in 3 static global variables: _PyMem_Raw, _PyMem, and _PyObject. State for the "small block" allocator was stored in another 13. That makes a total of 16 global variables. We are moving all 16 to the _PyRuntimeState struct as part of the work for gh-81057. (If PEP 684 is accepted then we will follow up by moving them all to PyInterpreterState.)
https://github.com/python/cpython/issues/81057
* Adds EXIT_INTERPRETER instruction to exit PyEval_EvalDefault()
* Simplifies RETURN_VALUE, YIELD_VALUE and RETURN_GENERATOR instructions as they no longer need to check for entry frames.
We do the following:
* move the generated _PyUnicode_InitStaticStrings() to its own file
* move the generated _PyStaticObjects_CheckRefcnt() to its own file
* include pycore_global_objects.h in extension modules instead of pycore_runtime_init.h
These changes help us avoid including things that aren't needed.
https://github.com/python/cpython/issues/90868
Add PyFrame_GetVar() and PyFrame_GetVarString() functions to get a
frame variable by its name.
Move PyFrameObject C API tests from test_capi to test_frame.
* Properly decref on _pylong import error.
* Improve the error message on _pylong TypeError.
* Fix the assertion error in pydebug builds to be a TypeError.
* Tie the return value comments together.
These are minor followups to issues not caught among the reviewers on
https://github.com/python/cpython/pull/96673.
Fix subscription of type aliases containing bare generic types or types
like TypeVar: for example tuple[A, T][int] and tuple[TypeVar, T][int],
where A is a generic type, and T is a type variable.
Change FOR_ITER to have the same stack effect regardless of whether it branches or not.
Performance is unchanged as FOR_ITER (and specialized forms jump over the cleanup code).
Add Python implementations of certain longobject.c functions. These use
asymptotically faster algorithms that can be used for operations on
integers with many digits. In those cases, the performance overhead of
the Python implementation is not significant since the asymptotic
behavior is what dominates runtime. Functions provided by this module
should be considered private and not part of any public API.
Co-author: Tim Peters <tim.peters@gmail.com>
Co-author: Mark Dickinson <dickinsm@gmail.com>
Co-author: Bjorn Martinsson
Work on test coverage for `PyObject_Print` made it clear that some lines can't get executed.
Simplify the function by excluding the checks for non-string types.
Also eliminate creating a temporary bytes object.
The os module and the PyUnicode_FSDecoder() function no longer accept
bytes-like paths, like bytearray and memoryview types: only the exact
bytes type is accepted for bytes strings.
asciilib_count() is the same than ucs1lib_count(): the code is not
specialized for ASCII strings, so it's not worth it to have a
separated function. Remove asciilib_count() function.
It had to live as a global outside of PyConfig for stable ABI reasons in
the pre-3.12 backports.
This removes the `_Py_global_config_int_max_str_digits` and gets rid of
the equivalent field in the internal `struct _is PyInterpreterState` as
code can just use the existing nested config struct within that.
Adds tests to verify unique settings and configs in subinterpreters.
Fix multiplying a list by an integer (list *= int): detect the
integer overflow when the new allocated length is close to the
maximum size. Issue reported by Jordan Limor.
list_resize() now checks for integer overflow before multiplying the
new allocated length by the list item size (sizeof(PyObject*)).
This is a preliminary PR to refactor `PyLong_FromString` which is currently quite messy and has spaghetti like code that mixes up different concerns as well as duplicating logic.
In particular:
- `PyLong_FromString` now only handles sign, base and prefix detection and calls a new function `long_from_string_base` to parse the main body of the string.
- The `long_from_string_base` function handles all string validation and then calls `long_from_binary_base` or a new function `long_from_non_binary_base` to construct the actual `PyLong`.
- The existing `long_from_binary_base` function is simplified by factoring duplicated logic to `long_from_string_base`.
- The new function `long_from_non_binary_base` factors out much of the code from `PyLong_FromString` including in particular the quadratic algorithm reffered to in gh-95778 so that this can be seen separately from unrelated concerns such as string validation.
Converting a large enough `int` to a decimal string raises `ValueError` as expected. However, the raise comes _after_ the quadratic-time base-conversion algorithm has run to completion. For effective DOS prevention, we need some kind of check before entering the quadratic-time loop. Oops! =)
The quick fix: essentially we catch _most_ values that exceed the threshold up front. Those that slip through will still be on the small side (read: sufficiently fast), and will get caught by the existing check so that the limit remains exact.
The justification for the current check. The C code check is:
```c
max_str_digits / (3 * PyLong_SHIFT) <= (size_a - 11) / 10
```
In GitHub markdown math-speak, writing $M$ for `max_str_digits`, $L$ for `PyLong_SHIFT` and $s$ for `size_a`, that check is:
$$\left\lfloor\frac{M}{3L}\right\rfloor \le \left\lfloor\frac{s - 11}{10}\right\rfloor$$
From this it follows that
$$\frac{M}{3L} < \frac{s-1}{10}$$
hence that
$$\frac{L(s-1)}{M} > \frac{10}{3} > \log_2(10).$$
So
$$2^{L(s-1)} > 10^M.$$
But our input integer $a$ satisfies $|a| \ge 2^{L(s-1)}$, so $|a|$ is larger than $10^M$. This shows that we don't accidentally capture anything _below_ the intended limit in the check.
<!-- gh-issue-number: gh-95778 -->
* Issue: gh-95778
<!-- /gh-issue-number -->
Co-authored-by: Gregory P. Smith [Google LLC] <greg@krypto.org>
Integer to and from text conversions via CPython's bignum `int` type is not safe against denial of service attacks due to malicious input. Very large input strings with hundred thousands of digits can consume several CPU seconds.
This PR comes fresh from a pile of work done in our private PSRT security response team repo.
Signed-off-by: Christian Heimes [Red Hat] <christian@python.org>
Tons-of-polishing-up-by: Gregory P. Smith [Google] <greg@krypto.org>
Reviews via the private PSRT repo via many others (see the NEWS entry in the PR).
<!-- gh-issue-number: gh-95778 -->
* Issue: gh-95778
<!-- /gh-issue-number -->
I wrote up [a one pager for the release managers](https://docs.google.com/document/d/1KjuF_aXlzPUxTK4BMgezGJ2Pn7uevfX7g0_mvgHlL7Y/edit#). Much of that text wound up in the Issue. Backports PRs already exist. See the issue for links.
* gh-96132: Add some comments and minor fixes missed in the original PR
* Update Doc/using/cmdline.rst
Co-authored-by: Kumar Aditya <59607654+kumaraditya303@users.noreply.github.com>
Co-authored-by: Kumar Aditya <59607654+kumaraditya303@users.noreply.github.com>
⚠️⚠️ Note for reviewers, hackers and fellow systems/low-level/compiler engineers ⚠️⚠️
If you have a lot of experience with this kind of shenanigans and want to improve the **first** version, **please make a PR against my branch** or **reach out by email** or **suggest code changes directly on GitHub**.
If you have any **refinements or optimizations** please, wait until the first version is merged before starting hacking or proposing those so we can keep this PR productive.
- "comparison of integers of different signs" in typeobject.c
- only define static_builtin_index_is_set in DEBUG builds
- only define recreate_gil with ifdef HAVE_FORK
* Treat tp_weakref and tp_dictoffset like other opaque slots for multiple inheritance.
* Document Py_TPFLAGS_MANAGED_DICT and Py_TPFLAGS_MANAGED_WEAKREF in what's new.
- On WASI `ENOTCAPABLE` is now mapped to `PermissionError`.
- The `errno` modules exposes the new error number.
- `getpath.py` now ignores `PermissionError` when it cannot open landmark
files `pybuilddir.txt` and `pyenv.cfg`.
* Make sure that tp_dictoffset is correct with Py_TPFLAGS_MANAGED_DICT is set.
* Avoid traversing managed dict twice when subclassing class with Py_TPFLAGS_MANAGED_DICT set.
We only statically initialize for core code and builtin modules. Extension modules still create
the tuple at runtime. We'll solve that part of interpreter isolation separately.
This change includes generated code. The non-generated changes are in:
* Tools/clinic/clinic.py
* Python/getargs.c
* Include/cpython/modsupport.h
* Makefile.pre.in (re-generate global strings after running clinic)
* very minor tweaks to Modules/_codecsmodule.c and Python/Python-tokenize.c
All other changes are generated code (clinic, global strings).
An unrecognized format character in PyUnicode_FromFormat() and
PyUnicode_FromFormatV() now sets a SystemError.
In previous versions it caused all the rest of the format string to be
copied as-is to the result string, and any extra arguments discarded.
* Add test for inheriting explicit __dict__ and weakref.
* Restore 3.10 behavior for multiple inheritance of C extension classes that store their dictionary at the end of the struct.
* Store tp_weaklist on the interpreter state for static builtin types.
* Factor out _PyStaticType_GET_WEAKREFS_LISTPTR().
* Add _PyStaticType_ClearWeakRefs().
* Add a comment about how _PyStaticType_ClearWeakRefs() loops.
* Document the change.
* Update Doc/whatsnew/3.12.rst
* Fix a typo.
This is the last precursor to storing tp_subclasses (and tp_weaklist) on the interpreter state for static builtin types.
Here we add per-type storage on PyInterpreterState, but only for the static builtin types. This involves the following:
* add PyInterpreterState.types
* move PyInterpreterState.type_cache to it
* add a "num_builtins_initialized" field
* add a "builtins" field (a static array big enough for all the static builtin types)
* add _PyStaticType_GetState() to look up a static builtin type's state
* (temporarily) add PyTypeObject.tp_static_builtin_index (to hold the type's index into PyInterpreterState.types.builtins)
We will be eliminating tp_static_builtin_index in a later change.
* Add _Py_memory_repeat function to pycore_list
* Add _Py_RefcntAdd function to pycore_object
* Use the new functions in tuplerepeat, list_repeat, and list_inplace_repeat
Static builtin types are finalized by calling _PyStaticType_Dealloc(). Before this change, we were skipping finalizing such a type if it still had subtypes (i.e. its tp_subclasses hadn't been cleared yet). The problem is that types hold several heap objects, which leak if we skip the type's finalization. This change addresses that.
For context, there's an old comment (from e9e3eab0b8) that says the following:
// If a type still has subtypes, it cannot be deallocated.
// A subtype can inherit attributes and methods of its parent type,
// and a type must no longer be used once it's deallocated.
However, it isn't clear that is actually still true. Clearing tp_dict should mean it isn't a problem.
Furthermore, the only subtypes that might still be around come from extension modules that didn't clean them up when unloaded (i.e. extensions that do not implement multi-phase initialization, AKA PEP 489). Those objects are already leaking, so this change doesn't change anything in that regard. Instead, this change means more objects gets cleaned up that before.
This is the first of several precursors to storing tp_subclasses (and tp_weaklist) on the interpreter state for static builtin types.
We do the following:
* add `_PyStaticType_InitBuiltin()`
* add `_Py_TPFLAGS_STATIC_BUILTIN`
* set it on all static builtin types in `_PyStaticType_InitBuiltin()`
* shuffle some code around to be able to use _PyStaticType_InitBuiltin()
* rename `_PyStructSequence_InitType()` to `_PyStructSequence_InitBuiltinWithFlags()`
* add `_PyStructSequence_InitBuiltin()`.
Nikita Sobolev
Brett Simmers
Alex Gaynor
Eddie Elizondo
Eric Wieser
AN Long
Mark Shannon
Irit Katriel
Brandt Bucher
Eric Snow
Inada Naoki
Kumar Aditya
Thomas Krennwallner
wangxiang-hz
Max Bachmann
Carl Meyer
Raj
Michael K
Petr Viktorin
Sergey B Kirpichev
Jelle Zijlstra
Ionite
Steve Dower
Furkan Onder
Nicko van Someren
Dong-hee Na
penguin_wwy
Matthieu Dartiailh
Mark Dickinson
Erlend E. Aasland
Pieter Eendebak
Raymond Hettinger
Neil Schemenauer
L. A. F. Pereira
Géry Ogam
Shantanu
Sebastian Berg
Dennis Sweeney
Bill Fisher
yonillasky
Itamar Ostricher
Ned Batchelder
Ken Jin
Serhiy Storchaka
Victor Stinner
Ivan Savov
mpage
Hood Chatham
jonasdlindner
Kaushik Kulkarni
Gregory P. Smith
Benjamin Peterson
Pablo Galindo Salgado
MonadChains
larryhastings
Michael
Ofey Chan
LiarPrincess
Oscar Benjamin
Samuel
adphrost
philg314
Gregory P. Smith
Christian Heimes
Matthias Görgens
Ken Jin
Tim Gates
Brandt Bucher
Pamela Fox
Aivars Kalvāns
neonene