1. Remove conditions already checked by assert()
2. Remove object_init() call that effectively creates an empty tuple and
checks that this tuple is empty
Currently, if an arg value escapes (into the closure for an inner function) we end up allocating two indices in the fast locals even though only one gets used. Additionally, using the lower index would be better in some cases, such as with no-arg `super()`. To address this, we update the compiler to fix the offsets so each variable only gets one "fast local". As a consequence, now some cell offsets are interspersed with the locals (only when an arg escapes to an inner function).
https://bugs.python.org/issue43693
* Specialize LOAD_ATTR with LOAD_ATTR_SLOT and LOAD_ATTR_SPLIT_KEYS
* Move dict-common.h to internal/pycore_dict.h
* Add LOAD_ATTR_WITH_HINT specialized opcode.
* Quicken in function if loopy
* Specialize LOAD_ATTR for module attributes.
* Add specialization stats
This is the same fix as for PyFrame_LocalsToFast() in gh-26609, but applied to PyFrame_FastToLocalsWithError(). (It should have been in that PR.)
https://bugs.python.org/issue43693
This was reverted in GH-26596 (commit 6d518bb) due to some bad memory accesses.
* Add the MAKE_CELL opcode. (gh-26396)
The memory accesses have been fixed.
https://bugs.python.org/issue43693
This moves logic out of the frame initialization code and into the compiler and eval loop. Doing so simplifies the runtime code and allows us to optimize it better.
https://bugs.python.org/issue43693
These were reverted in gh-26530 (commit 17c4edc) due to refleaks.
* 2c1e258 - Compute deref offsets in compiler (gh-25152)
* b2bf2bc - Add new internal code objects fields: co_fastlocalnames and co_fastlocalkinds. (gh-26388)
This change fixes the refleaks.
https://bugs.python.org/issue43693
* Add co_firstinstr field to code object.
* Implement barebones quickening.
* Use non-quickened bytecode when tracing.
* Add NEWS item
* Add new file to Windows build.
* Don't specialize instructions with EXTENDED_ARG.
* Revert "bpo-43693: Compute deref offsets in compiler (gh-25152)"
This reverts commit b2bf2bc1ec.
* Revert "bpo-43693: Add new internal code objects fields: co_fastlocalnames and co_fastlocalkinds. (gh-26388)"
This reverts commit 2c1e2583fd.
These two commits are breaking the refleak buildbots.
Merges locals and cells into a single array.
Saves a pointer in the interpreter and means that we don't need the LOAD_CLOSURE opcode any more
https://bugs.python.org/issue43693
A number of places in the code base (notably ceval.c and frameobject.c) rely on mapping variable names to indices in the frame "locals plus" array (AKA fast locals), and thus opargs. Currently the compiler indirectly encodes that information on the code object as the tuples co_varnames, co_cellvars, and co_freevars. At runtime the dependent code must calculate the proper mapping from those, which isn't ideal and impacts performance-sensitive sections. This is something we can easily address in the compiler instead.
This change addresses the situation by replacing internal use of co_varnames, etc. with a single combined tuple of names in locals-plus order, along with a minimal array mapping each to its kind (local vs. cell vs. free). These two new PyCodeObject fields, co_fastlocalnames and co_fastllocalkinds, are not exposed to Python code for now, but co_varnames, etc. are still available with the same values as before (though computed lazily).
Aside from the (mild) performance impact, there are a number of other benefits:
* there's now a clear, direct relationship between locals-plus and variables
* code that relies on the locals-plus-to-name mapping is simpler
* marshaled code objects are smaller and serialize/de-serialize faster
Also note that we can take this approach further by expanding the possible values in co_fastlocalkinds to include specific argument types (e.g. positional-only, kwargs). Doing so would allow further speed-ups in _PyEval_MakeFrameVector(), which is where args get unpacked into the locals-plus array. It would also allow us to shrink marshaled code objects even further.
https://bugs.python.org/issue43693
The PyType_Ready() function now raises an error if a type is defined
with the Py_TPFLAGS_HAVE_GC flag set but has no traverse function
(PyTypeObject.tp_traverse).
* Move up the comment about fields using in hashing/comparision.
* Group the fields more clearly.
* Add co_ncellvars and co_nfreevars.
* Raise ValueError if nlocals != len(varnames), rather than aborting.
Fix a regression in type() when a metaclass raises an exception. The
C function type_new() must properly report the exception when a
metaclass constructor raises an exception and the winner class is not
the metaclass.
Fix a crash at Python exit when a deallocator function removes the
last strong reference to a heap type.
Don't read type memory after calling basedealloc() since
basedealloc() can deallocate the type and free its memory.
_PyMem_IsPtrFreed() argument is now constant.
* Remove 'zombie' frames. We won't need them once we are allocating fixed-size frames.
* Add co_nlocalplus field to code object to avoid recomputing size of locals + frees + cells.
* Move locals, cells and freevars out of frame object into separate memory buffer.
* Use per-threadstate allocated memory chunks for local variables.
* Move globals and builtins from frame object to per-thread stack.
* Move (slow) locals frame object to per-thread stack.
* Move internal frame functions to internal header.
These are passed and called as PyCFunction, however they are defined here without the (ignored) args parameter.
This works fine in some C compilers, but fails in webassembly or anything else that has strict function pointer call type checking.
"Zero cost" exception handling.
* Uses a lookup table to determine how to handle exceptions.
* Removes SETUP_FINALLY and POP_TOP block instructions, eliminating (most of) the runtime overhead of try statements.
* Reduces the size of the frame object by about 60%.
check_set_special_type_attr() and type_set_annotations()
now check for immutable flag (Py_TPFLAGS_IMMUTABLETYPE).
Co-authored-by: Victor Stinner <vstinner@python.org>
The PyStdPrinter_Type type now uses the
Py_TPFLAGS_DISALLOW_INSTANTIATION flag to disallow instantiation,
rather than seting a tp_init method which always fail.
Write also unit tests for PyStdPrinter_Type.
Add a new Py_TPFLAGS_DISALLOW_INSTANTIATION type flag to disallow
creating type instances: set tp_new to NULL and don't create the
"__new__" key in the type dictionary.
The flag is set automatically on static types if tp_base is NULL or
&PyBaseObject_Type and tp_new is NULL.
Use the flag on the following types:
* _curses.ncurses_version type
* _curses_panel.panel
* _tkinter.Tcl_Obj
* _tkinter.tkapp
* _tkinter.tktimertoken
* _xxsubinterpretersmodule.ChannelID
* sys.flags type
* sys.getwindowsversion() type
* sys.version_info type
Update MyStr example in the C API documentation to use
Py_TPFLAGS_DISALLOW_INSTANTIATION.
Add _PyStructSequence_InitType() function to create a structseq type
with the Py_TPFLAGS_DISALLOW_INSTANTIATION flag set.
type_new() calls _PyType_CheckConsistency() at exit.
* 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.
While working on another issue, I noticed two minor nits in the C implementation of the module object. Both are related to getting a module's name.
First, the C function module_dir() (module.__dir__) starts by ensuring the module dict is valid. If the module dict is invalid, it wants to format an exception using the name of the module, which it gets from PyModule_GetName(). However, PyModule_GetName() gets the name of the module from the dict. So getting the name in this circumstance will never succeed.
When module_dir() wants to format the error but can't get the name, it knows that PyModule_GetName() must have already raised an exception. So it leaves that exception alone and returns an error. The end result is that the exception raised here is kind of useless and misleading: dir(module) on a module with no __dict__ raises SystemError("nameless module"). I changed the code to actually raise the exception it wanted to raise, just without a real module name: TypeError("<module>.__dict__ is not a dictionary"). This seems more useful, and would do a better job putting the programmer who encountered this on the right track of figuring out what was going on.
Second, the C API function PyModule_GetNameObject() checks to see if the module has a dict. If m->md_dict is not NULL, it calls _PyDict_GetItemIdWithError(). However, it's possible for m->md_dict to be None. And if you call _PyDict_GetItemIdWithError(Py_None, ...) it will *crash*.
Unfortunately, this crash was due to my own bug in the other branch. Fixing my code made the crash go away. I assert that this is still possible at the API level.
The fix is easy: add a PyDict_Check() to PyModule_GetNameObject().
Unfortunately, I don't know how to add a unit test for this. Having changed module_dir() above, I can't find any other interfaces callable from Python that eventually call PyModule_GetNameObject(). So I don't know how to trick the runtime into reproducing this error.
Since both these changes are minor--each entails only a small edit to only one line--I didn't bother with a news item.
Change class and module objects to lazy-create empty annotations dicts on demand. The annotations dicts are stored in the object's `__dict__` for backwards compatibility.
Accessing the following attributes will now fire PEP 578 style audit hooks as ("object.__getattr__", obj, name):
* PyTracebackObject: tb_frame
* PyFrameObject: f_code
* PyGenObject: gi_code, gi_frame
* PyCoroObject: cr_code, cr_frame
* PyAsyncGenObject: ag_code, ag_frame
Add an AUDIT_READ attribute flag aliased to READ_RESTRICTED.
Update obsolete flag documentation.
* Add length parameter to PyLineTable_InitAddressRange and doen't use sentinel values at end of table. Makes the line number table more robust.
* Update PyCodeAddressRange to match PEP 626.
Introduce Py_TPFLAGS_IMMUTABLETYPE flag for immutable type objects, and
modify PyType_Ready() to set it for static types.
Co-authored-by: Victor Stinner <vstinner@python.org>
To improve the user experience understanding what part of the error messages associated with SyntaxErrors is wrong, we can highlight the whole error range and not only place the caret at the first character. In this way:
>>> foo(x, z for z in range(10), t, w)
File "<stdin>", line 1
foo(x, z for z in range(10), t, w)
^
SyntaxError: Generator expression must be parenthesized
becomes
>>> foo(x, z for z in range(10), t, w)
File "<stdin>", line 1
foo(x, z for z in range(10), t, w)
^^^^^^^^^^^^^^^^^^^^
SyntaxError: Generator expression must be parenthesized
Add pycore_moduleobject.h internal header file with static inline
functions to access module members:
* _PyModule_GetDict()
* _PyModule_GetDef()
* _PyModule_GetState()
These functions don't check at runtime if their argument has a valid
type and can be inlined even if Python is not built with LTO.
_PyType_GetModuleByDef() uses _PyModule_GetDef().
Replace PyModule_GetState() with _PyModule_GetState() in the
extension modules, considered as performance sensitive:
* _abc
* _functools
* _operator
* _pickle
* _queue
* _random
* _sre
* _struct
* _thread
* _winapi
* array
* posix
The following extensions are now built with the Py_BUILD_CORE_MODULE
macro defined, to be able to use the internal pycore_moduleobject.h
header: _abc, array, _operator, _queue, _sre, _struct.
PyType_Ready() now ensures that a type MRO cannot be empty.
_PyType_GetModuleByDef() no longer checks "i < PyTuple_GET_SIZE(mro)"
at the first loop iteration to optimize the most common case, when
the argument is the defining class.
_PyType_GetModuleByDef() no longer checks if types are heap types.
_PyType_GetModuleByDef() must only be called on a heap type created
by PyType_FromModuleAndSpec() or on its subclasses.
type_ready_mro() ensures that a static type cannot inherit from a
heap type.
When printing NameError raised by the interpreter, PyErr_Display
will offer suggestions of simmilar variable names in the function that the exception
was raised from:
>>> schwarzschild_black_hole = None
>>> schwarschild_black_hole
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NameError: name 'schwarschild_black_hole' is not defined. Did you mean: schwarzschild_black_hole?
When printing AttributeError, PyErr_Display will offer suggestions of similar
attribute names in the object that the exception was raised from:
>>> collections.namedtoplo
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: module 'collections' has no attribute 'namedtoplo'. Did you mean: namedtuple?
* Rename functions
* Only pass type parameter to "add_xxx" functions.
* Clarify the role of the type_ready_inherit_as_structs() function.
* Move type_dict_set_doc() code to call it in type_ready_fill_dict().
* Split PyType_Ready() into sub-functions.
* type_ready_mro() now checks if bases are static types earlier.
* Check tp_name earlier, in type_ready_checks().
* Add _PyType_IsReady() macro to check if a type is ready.
Add the Py_Is(x, y) function to test if the 'x' object is the 'y'
object, the same as "x is y" in Python. Add also the Py_IsNone(),
Py_IsTrue(), Py_IsFalse() functions to test if an object is,
respectively, the None singleton, the True singleton or the False
singleton.
* Split type_new() into into many small functions.
* Add type_new_ctx structure to pass variables between subfunctions.
* Initialize some PyTypeObject and PyHeapTypeObject members earlier
in type_new_alloc().
* Rename variables to more specific names.
* Add "__weakref__" identifier for type_new_visit_slots().
* Factorize code to convert a method to a classmethod
(__init_subclass__ and __class_getitem__).
* Add braces to respect PEP 7.
* Move variable declarations where the variables are initialized.
Static methods (@staticmethod) and class methods (@classmethod) now
inherit the method attributes (__module__, __name__, __qualname__,
__doc__, __annotations__) and have a new __wrapped__ attribute.
Changes:
* Add a repr() method to staticmethod and classmethod types.
* Add tests on the @classmethod decorator.
* Handle check for sending None to starting generator and coroutine into bytecode.
* Document new bytecode and make it fail gracefully if mis-compiled.
The limited C API is now supported if Python is built in debug mode
(if the Py_DEBUG macro is defined). In the limited C API, the
Py_INCREF() and Py_DECREF() functions are now implemented as opaque
function calls, rather than accessing directly the PyObject.ob_refcnt
member, if Python is built in debug mode and the Py_LIMITED_API macro
targets Python 3.10 or newer. It became possible to support the
limited C API in debug mode because the PyObject structure is the
same in release and debug mode since Python 3.8 (see bpo-36465).
The limited C API is still not supported in the --with-trace-refs
special build (Py_TRACE_REFS macro).
Reorganize pycore_interp_init() to initialize singletons before the
the first PyType_Ready() call. Fix an issue when Python is configured
using --without-doc-strings.
* Use instruction offset, rather than bytecode offset. Streamlines interpreter dispatch a bit, and removes most EXTENDED_ARGs for jumps.
* Change some uses of PyCode_Addr2Line to PyFrame_GetLineNumber
When printing stats, move radix tree info to its own section.
Restore that the breakdown of bytes in arenas exactly accounts for the total of arena bytes allocated.
Add an assert so that invariant doesn't break again.
* Remove m68k-specific hack from ascii_decode
On m68k, alignments of primitives is more relaxed, with 4-byte and
8-byte types only requiring 2-byte alignment, thus using sizeof(size_t)
does not work. Instead, use the portable alternative.
Note that this is a minimal fix that only relaxes the assertion and the
condition for when to use the optimised version remains overly strict.
Such issues will be fixed tree-wide in the next commit.
NB: In C11 we could use _Alignof(size_t) instead, but for compatibility
we use autoconf.
* Optimise string routines for architectures with non-natural alignment
C only requires that sizeof(x) is a multiple of alignof(x), not that the
two are equal. Thus anywhere where we optimise based on alignment we
should be using alignof(x) not sizeof(x).
This is more annoying than it would be in C11 where we could just use
_Alignof(x) (and alignof(x) in C++11), but since we still require only
C99 we must plumb the information all the way from autoconf through the
various typedefs and defines.
The radix tree approach is a relatively simple and memory sanitary
alternative to the old (slightly) unsanitary address_in_range().
To disable the radix tree map, set a preprocessor flag as follows:
-DWITH_PYMALLOC_RADIX_TREE=0.
Co-authored-by: Tim Peters <tim.peters@gmail.com>
See [PEP 597](https://www.python.org/dev/peps/pep-0597/).
* Add `-X warn_default_encoding` and `PYTHONWARNDEFAULTENCODING`.
* Add EncodingWarning
* Add io.text_encoding()
* open(), TextIOWrapper() emits EncodingWarning when encoding is omitted and warn_default_encoding is enabled.
* _pyio.TextIOWrapper() uses UTF-8 as fallback default encoding used when failed to import locale module. (used during building Python)
* bz2, configparser, gzip, lzma, pathlib, tempfile modules use io.text_encoding().
* What's new entry
Remove the compiler functions using "struct _mod" type, because the
public AST C API was removed:
* PyAST_Compile()
* PyAST_CompileEx()
* PyAST_CompileObject()
* PyFuture_FromAST()
* PyFuture_FromASTObject()
These functions were undocumented and excluded from the limited C API.
Rename functions:
* PyAST_CompileObject() => _PyAST_Compile()
* PyFuture_FromASTObject() => _PyFuture_FromAST()
Moreover, _PyFuture_FromAST() is no longer exported (replace
PyAPI_FUNC() with extern). _PyAST_Compile() remains exported for
test_peg_generator.
Remove also compatibility functions:
* PyAST_Compile()
* PyAST_CompileEx()
* PyFuture_FromAST()
The common case going through _PyType_Lookup is to have a cache hit. There are some small tweaks that can make this a little cheaper:
* The name field identity is used for a cache hit and is kept alive by the cache. So there's no need to read the hash code o the name - instead, the address can be used as the hash.
* There's no need to check if the name is cachable on the lookup either, it probably is, and if it is, it'll be in the cache.
* If we clear the version tag when invalidating a type then we don't actually need to check for a valid version tag bit.
* Remove an assertion which required CO_NEWLOCALS and CO_OPTIMIZED
code flags. It is ok to call this function on a code with these
flags set.
* Fix reference counting on builtins: remove Py_DECREF().
Fix regression introduced in the
commit 46496f9d12.
Add also a comment to document that _PyEval_BuiltinsFromGlobals()
returns a borrowed reference.
Python no longer fails at startup with a fatal error if a command
line argument contains an invalid Unicode character.
The Py_DecodeLocale() function now escapes byte sequences which would
be decoded as Unicode characters outside the [U+0000; U+10ffff]
range.
Use MAX_UNICODE constant in unicodeobject.c.
Implement an enhanced variant of Crochemore and Perrin's Two-Way string searching algorithm, which reduces worst-case time from quadratic (the product of the string and pattern lengths) to linear. This applies to forward searches (like``find``, ``index``, ``replace``); the algorithm for reverse searches (like ``rfind``) is not changed.
Co-authored-by: Tim Peters <tim.peters@gmail.com>
* No longer save/restore the current exception. It is no longer used
with an exception raised.
* No longer clear the current exception on error: it's now up to the
caller.
For some mysterious reason we have PySet_Check, PyFrozenSet_Check, PyAnySet_Check, PyAnySet_CheckExact and PyFrozenSet_CheckExact but no PySet_CheckExact.
The types.FunctionType constructor now inherits the current builtins
if the globals dictionary has no "__builtins__" key, rather than
using {"None": None} as builtins: same behavior as eval() and exec()
functions.
Defining a function with "def function(...): ..." in Python is not
affected, globals cannot be overriden with this syntax: it also
inherits the current builtins.
PyFrame_New(), PyEval_EvalCode(), PyEval_EvalCodeEx(),
PyFunction_New() and PyFunction_NewWithQualName() now inherits the
current builtins namespace if the globals dictionary has no
"__builtins__" key.
* Add _PyEval_GetBuiltins() function.
* _PyEval_BuiltinsFromGlobals() now uses _PyEval_GetBuiltins() if
builtins cannot be found in globals.
* Add tstate parameter to _PyEval_BuiltinsFromGlobals().
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.
Sergey Fedoseev
Eric Snow
Gabriele N. Tornetta
Mark Shannon
Mark Dickinson
Tim Peters
Serhiy Storchaka
Pablo Galindo
Victor Stinner
E-Paine
Inada Naoki
Joe Marshall
Erlend Egeberg Aasland
Jakub Kulík
larryhastings
Steve Dower
Ryan Hileman
Hristo Venev
Ken Jin
Raymond Hettinger
Dennis Sweeney
Jessica Clarke
Neil Schemenauer
Joshua Bronson
Dino Viehland
Skip Montanaro
junyixie
Jozef Grajciar
Brandt Bucher
Alex