When multiplying lists and tuples by `n`, increment each element's refcount, by `n`, just once.
Saves `n-1` increments per element, and allows for a leaner & faster copying loop.
Code by sweeneyde (Dennis Sweeney).
This change is strictly renames and moving code around. It helps in the following ways:
* ensures type-related init functions focus strictly on one of the three aspects (state, objects, types)
* passes in PyInterpreterState * to all those functions, simplifying work on moving types/objects/state to the interpreter
* consistent naming conventions help make what's going on more clear
* keeping API related to a type in the corresponding header file makes it more obvious where to look for it
https://bugs.python.org/issue46008
Keep track of whether unsafe_tuple_compare() calls are resolved by the very
first tuple elements, and adjust strategy accordingly. This can significantly
cut the number of calls made to the full-blown PyObject_RichCompareBool(),
and especially when duplicates are rare.
Co-authored-by: Łukasz Langa <lukasz@langa.pl>
Freelists for object structs can now be disabled. A new ``configure``
option ``--without-freelists`` can be used to disable all freelists
except empty tuple singleton. Internal Py*_MAXFREELIST macros can now
be defined as 0 without causing compiler warnings and segfaults.
Signed-off-by: Christian Heimes <christian@python.org>
For list.sort(), replace our ad hoc merge ordering strategy with the principled, elegant,
and provably near-optimal one from Munro and Wild's "powersort".
* Add Py_TPFLAGS_SEQUENCE and Py_TPFLAGS_MAPPING, add to all relevant standard builtin classes.
* Set relevant flags on collections.abc.Sequence and Mapping.
* Use flags in MATCH_SEQUENCE and MATCH_MAPPING opcodes.
* Inherit Py_TPFLAGS_SEQUENCE and Py_TPFLAGS_MAPPING.
* Add NEWS
* Remove interpreter-state map_abc and seq_abc fields.
Pass the current interpreter (interp) rather than the current Python
thread state (tstate) to internal functions which only use the
interpreter.
Modified functions:
* _PyXXX_Fini() and _PyXXX_ClearFreeList() functions
* _PyEval_SignalAsyncExc(), make_pending_calls()
* _PySys_GetObject(), sys_set_object(), sys_set_object_id(), sys_set_object_str()
* should_audit(), set_flags_from_config(), make_flags()
* _PyAtExit_Call()
* init_stdio_encoding()
* etc.
No longer use deprecated aliases to functions:
* Replace PyMem_MALLOC() with PyMem_Malloc()
* Replace PyMem_REALLOC() with PyMem_Realloc()
* Replace PyMem_FREE() with PyMem_Free()
* Replace PyMem_Del() with PyMem_Free()
* Replace PyMem_DEL() with PyMem_Free()
Modify also the PyMem_DEL() macro to use directly PyMem_Free().
In debug mode, ensure that free lists are no longer used after being
finalized. Set numfree to -1 in finalization functions
(eg. _PyList_Fini()), and then check that numfree is not equal to -1
before using a free list (e.g list_dealloc()).
Each interpreter now has its own list free list:
* Move list numfree and free_list into PyInterpreterState.
* Add _Py_list_state structure.
* Add tstate parameter to _PyList_ClearFreeList()
and _PyList_Fini().
* Remove "#ifdef EXPERIMENTAL_ISOLATED_SUBINTERPRETERS".
* _PyGC_Fini() clears gcstate->garbage list which can be stored in
the list free list. Call _PyGC_Fini() before _PyList_Fini() to
prevent leaking this list.
When Python is built with experimental isolated interpreters, disable
the list free list.
Temporary workaround until this cache is made per-interpreter.
Remove the following function from the C API:
* PyAsyncGen_ClearFreeLists()
* PyContext_ClearFreeList()
* PyDict_ClearFreeList()
* PyFloat_ClearFreeList()
* PyFrame_ClearFreeList()
* PyList_ClearFreeList()
* PySet_ClearFreeList()
* PyTuple_ClearFreeList()
Make these functions private, move them to the internal C API and
change their return type to void.
Call explicitly PyGC_Collect() to free all free lists.
Note: PySet_ClearFreeList() did nothing.
Add _PyIndex_Check() function to the internal C API: fast inlined
verson of PyIndex_Check().
Add Include/internal/pycore_abstract.h header file.
Replace PyIndex_Check() with _PyIndex_Check() in C files of Objects
and Python subdirectories.
This implements things like `list[int]`,
which returns an object of type `types.GenericAlias`.
This object mostly acts as a proxy for `list`,
but has attributes `__origin__` and `__args__`
that allow recovering the parts (with values `list` and `(int,)`.
There is also an approximate notion of type variables;
e.g. `list[T]` has a `__parameters__` attribute equal to `(T,)`.
Type variables are objects of type `typing.TypeVar`.
Speed up calls to list() by using the PEP 590 vectorcall
calling convention. Patch by Mark Shannon.
Co-authored-by: Mark Shannon <mark@hotpy.org>
Co-authored-by: Dong-hee Na <donghee.na92@gmail.com>
The bulk of this patch was generated automatically with:
for name in \
PyObject_Vectorcall \
Py_TPFLAGS_HAVE_VECTORCALL \
PyObject_VectorcallMethod \
PyVectorcall_Function \
PyObject_CallOneArg \
PyObject_CallMethodNoArgs \
PyObject_CallMethodOneArg \
;
do
echo $name
git grep -lwz _$name | xargs -0 sed -i "s/\b_$name\b/$name/g"
done
old=_PyObject_FastCallDict
new=PyObject_VectorcallDict
git grep -lwz $old | xargs -0 sed -i "s/\b$old\b/$new/g"
and then cleaned up:
- Revert changes to in docs & news
- Revert changes to backcompat defines in headers
- Nudge misaligned comments
…nctions with asserts
The actual overflow can never happen because of the following:
* The size of a list can't be greater than PY_SSIZE_T_MAX / sizeof(PyObject*).
* The size of a pointer on all supported plaftorms is at least 4 bytes.
* ofs is positive and less than the list size at the beginning of each iteration.
https://bugs.python.org/issue35091
Add new trashcan macros to deal with a double deallocation that could occur when the `tp_dealloc` of a subclass calls the `tp_dealloc` of a base class and that base class uses the trashcan mechanism.
Patch by Jeroen Demeyer.
There is already a `Py_ssize_t i` defined at function scope that is used
for similar loops. By removing the local `int i` declaration that `i` is
used, which has the appropriate type.
The accu.h header is no longer part of the Python C API: it has been
moved to the "internal" headers which are restricted to Python
itself.
Replace #include "accu.h" with #include "pycore_accu.h".
The list() constructor isn't taking full advantage of known input
lengths or length hints. This commit makes the constructor
pre-size and not over-allocate when the input size is known (the
input collection implements __len__). One on the main advantages is
that this provides 12% difference in memory savings due to the difference
between overallocating and allocating exactly the input size.
For efficiency purposes and to avoid a performance regression for small
generators and collections, the size of the input object is calculated using
__len__ and not __length_hint__, as the later is considerably slower.
METH_NOARGS functions need only a single argument but they are cast
into a PyCFunction, which takes two arguments. This triggers an
invalid function cast warning in gcc8 due to the argument mismatch.
Fix this by adding a dummy unused argument.
Py_Main() now handles two more -X options:
* -X showrefcount: new _PyCoreConfig.show_ref_count field
* -X showalloccount: new _PyCoreConfig.show_alloc_count field
* group the (stateful) runtime globals into various topical structs
* consolidate the topical structs under a single top-level _PyRuntimeState struct
* add a check-c-globals.py script that helps identify runtime globals
Other globals are excluded (see globals.txt and check-c-globals.py).
Issue #28858: The change b9c9691c72c5 introduced a regression. It seems like
_PyObject_CallArg1() uses more stack memory than
PyObject_CallFunctionObjArgs().
* PyObject_CallFunctionObjArgs(func, NULL) => _PyObject_CallNoArg(func)
* PyObject_CallFunctionObjArgs(func, arg, NULL) => _PyObject_CallArg1(func, arg)
PyObject_CallFunctionObjArgs() allocates 40 bytes on the C stack and requires
extra work to "parse" C arguments to build a C array of PyObject*.
_PyObject_CallNoArg() and _PyObject_CallArg1() are simpler and don't allocate
memory on the C stack.
This change is part of the fastcall project. The change on listsort() is
related to the issue #23507.
SHOW_ALLOC_COUNT or SHOW_TRACK_COUNT macros is now off by default. It can
be re-enabled using the "-X showalloccount" option. It now outputs to stderr
instead of stdout.
Dennis Sweeney
Eric Snow
Christian Heimes
Tim Peters
Christian Clauss
Serhiy Storchaka
Mark Shannon
Brandt Bucher
Victor Stinner
Jeong Ukjae
Sergey Fedoseev
Guido van Rossum
Petr Viktorin
Andy Lester
Dong-hee Na
Inada Naoki
Zackery Spytz
HongWeipeng
Victor Stinner
Jeroen Demeyer
Tim Hoffmann
Alexey Izbyshev
Rémi Lapeyre
sth
Pablo Galindo
Raymond Hettinger
Siddhesh Poyarekar
Eddie Elizondo
Victor Stinner
embg
stratakis
Xiang Zhang
Martin Panter
Benjamin Peterson
Mark Dickinson
Raymond Hettinger