Currently, during runtime destruction, `_PyImport_Cleanup` is clearing the interpreter state before clearing out the modules themselves. This leads to a segfault on modules that rely on the module state to clear themselves up.
For example, let's take the small snippet added in the issue by @DinoV :
```
import _struct
class C:
def __init__(self):
self.pack = _struct.pack
def __del__(self):
self.pack('I', -42)
_struct.x = C()
```
The module `_struct` uses the module state to run `pack`. Therefore, the module state has to be alive until after the module has been cleared out to successfully run `C.__del__`. This happens at line 606, when `_PyImport_Cleanup` calls `_PyModule_Clear`. In fact, the loop that calls `_PyModule_Clear` has in its comments:
> Now, if there are any modules left alive, clear their globals to minimize potential leaks. All C extension modules actually end up here, since they are kept alive in the interpreter state.
That means that we can't clear the module state (which is used by C Extensions) before we run that loop.
Moving `_PyInterpreterState_ClearModules` until after it, fixes the segfault in the code snippet.
Finally, this updates a test in `io` to correctly assert the error that it now throws (since it now finds the io module state). The test that uses this is: `test_create_at_shutdown_without_encoding`. Given this test is now working is a proof that the module state now stays alive even when `__del__` is called at module destruction time. Thus, I didn't add a new tests for this.
https://bugs.python.org/issue38076
Move the following functions from the public C API to the internal C
API:
* _PyDebug_PrintTotalRefs(),
* _Py_PrintReferenceAddresses()
* _Py_PrintReferences()
PyThreadState.on_delete is a callback used to notify Python when a
thread completes. _thread._set_sentinel() function creates a lock
which is released when the thread completes. It sets on_delete
callback to the internal release_sentinel() function. This lock is
known as Threading._tstate_lock in the threading module.
The release_sentinel() function uses the Python C API. The problem is
that on_delete is called late in the Python finalization, when the C
API is no longer fully working.
The PyThreadState_Clear() function now calls the
PyThreadState.on_delete callback. Previously, that happened in
PyThreadState_Delete().
The release_sentinel() function is now called when the C API is still
fully working.
Replace a few Py_FatalError() calls if tstate is NULL with
assert(tstate != NULL) in ceval.c.
PyEval_AcquireThread(), PyEval_ReleaseThread() and
PyEval_RestoreThread() must never be called with a NULL tstate.
* Add DICT_UPDATE and DICT_MERGE bytecodes. Use them for ** unpacking.
* Remove BUILD_MAP_UNPACK and BUILD_MAP_UNPACK_WITH_CALL, as they are now unused.
* Update magic number for ** unpacking opcodes.
* Update dis.rst to incorporate new bytecodes.
* Add blurb entry.
* Add three new bytecodes: LIST_TO_TUPLE, LIST_EXTEND, SET_UPDATE. Use them to implement star unpacking expressions.
* Remove four bytecodes BUILD_LIST_UNPACK, BUILD_TUPLE_UNPACK, BUILD_SET_UNPACK and BUILD_TUPLE_UNPACK_WITH_CALL opcodes as they are now unused.
* Update magic number and dis.rst for new bytecodes.
* bpo-39336: Allow setattr to fail on modules which aren't assignable
When attaching a child module to a package if the object in sys.modules raises an AttributeError (e.g. because it is immutable) it causes the whole import to fail. This now allows immutable packages to exist and an ImportWarning is reported and the AttributeError exception is ignored.
Python-ast.h contains a macro named Yield that conflicts with the Yield macro
in Windows system headers. While Python-ast.h has an "undef Yield" directive
to prevent this, it means that Python-ast.h must be included before Windows
header files or we run into a re-declaration warning. In commit c96be811fa
an include for pycore_pystate.h was added which indirectly includes Windows
header files. In this commit we re-order the includes to fix this warning.
* Reorder the __aenter__ and __aexit__ checks for async with
* Add assertions for async with body being skipped
* Swap __aexit__ and __aenter__ loading in the documentation
Break up COMPARE_OP into four logically distinct opcodes:
* COMPARE_OP for rich comparisons
* IS_OP for 'is' and 'is not' tests
* CONTAINS_OP for 'in' and 'is not' tests
* JUMP_IF_NOT_EXC_MATCH for checking exceptions in 'try-except' statements.
This adds a new function named _PyErr_GetExcInfo() that is a variation of the
original PyErr_GetExcInfo() taking a PyThreadState as its first argument.
That function allows to retrieve the exceptions information of any Python
thread -- not only the current one.
The fix changes copy_location() to require an extra node from which to extract the end location, and fixing all 5 call sites.
https://bugs.python.org/issue39235
When producing the bytecode of exception handlers with name binding (like `except Exception as e`) we need to produce a try-finally block to make sure that the name is deleted after the handler is executed to prevent cycles in the stack frame objects. The bytecode associated with this try-finally block does not have source lines associated and it was causing problems when the tracing functionality was running over it.
All keywords should first be checked for pointer identity. Only
after that failed for all keywords (unlikely) should unicode
equality be used.
The original code would call unicode equality on any non-matching
keyword argument. Meaning calling it often e.g. when a function
has many kwargs but only the last one is provided.
Each Python subinterpreter now has its own "small integer
singletons": numbers in [-5; 257] range.
It is no longer possible to change the number of small integers at
build time by overriding NSMALLNEGINTS and NSMALLPOSINTS macros:
macros should now be modified manually in pycore_pystate.h header
file.
For now, continue to share _PyLong_Zero and _PyLong_One singletons
between all subinterpreters.
When parsing an "elif" node, lineno and col_offset of the node now point to the "elif" keyword and not to its condition, making it consistent with the "if" node.
https://bugs.python.org/issue39031
Automerge-Triggered-By: @pablogsal
In Python 3.9.0a1, sys.argv[0] was made an asolute path if a filename
was specified on the command line. Revert this change, since most
users expect sys.argv to be unmodified.
new_interpreter() now calls _PySys_Create() to create a new sys
module isolated from the main interpreter. It now calls
_PySys_InitCore() and _PyImport_FixupBuiltin().
init_interp_main() now calls _PySys_InitMain().
new_interpreter() now calls _PyBuiltin_Init() to create the builtins
module and calls _PyImport_FixupBuiltin(), rather than using
_PyImport_FindBuiltin(tstate, "builtins").
pycore_init_builtins() is now responsible to initialize
intepr->builtins_copy: inline _PyImport_Init() and remove this
function.
If _PyImport_FixupExtensionObject() is called from a subinterpreter,
leave extensions unchanged and don't copy the module dictionary
into def->m_base.m_copy.
Remove BEGIN_FINALLY, END_FINALLY, CALL_FINALLY and POP_FINALLY bytecodes. Implement finally blocks by code duplication.
Reimplement frame.lineno setter using line numbers rather than bytecode offsets.
Remove PyMethod_ClearFreeList() and PyCFunction_ClearFreeList()
functions: the free lists of bound method objects have been removed.
Remove also _PyMethod_Fini() and _PyCFunction_Fini() functions.
* Add GCState type for readability
* gcmodule.c now gets its gcstate from tstate
* _PyGC_DumpShutdownStats() now expects tstate rather than runtime
* Rename "state" to "gcstate" for readability: to avoid confusion
between "state" and "tstate" for example.
* collect() now only expects tstate: it gets gcstate from tstate.
* Pass tstate to _PyErr_xxx() functions
Clear the current thread later in the Python finalization.
* The PyInterpreterState_Delete() function is now responsible
to call PyThreadState_Swap(NULL).
* The tstate_delete_common() function is now responsible to clear the
"autoTSSKey" thread local storage and it only clears it once the
thread state is fully cleared. It allows to still get the current
thread from TSS in tstate_delete_common().
* Factorize code in common between Py_FinalizeEx() and
Py_EndInterpreter().
* Py_EndInterpreter() now also calls _PyWarnings_Fini().
* Call _PyExc_Fini() and _PyGC_Fini() later in the finalization.
The PyFPE_START_PROTECT() and PyFPE_END_PROTECT() macros are empty:
they have been doing nothing for the last year (since commit
735ae8d139), so stop using them.
Add PyInterpreterState.runtime field: reference to the _PyRuntime
global variable. This field exists to not have to pass runtime in
addition to tstate to a function. Get runtime from tstate:
tstate->interp->runtime.
Remove "_PyRuntimeState *runtime" parameter from functions already
taking a "PyThreadState *tstate" parameter.
_PyGC_Init() first parameter becomes "PyThreadState *tstate".
* Add _PyObject_VectorcallTstate() function: similar to
_PyObject_Vectorcall(), but with tstate parameter
* Add tstate parameter to _PyObject_MakeTpCall()
bpo-3605, bpo-38733: Optimize _PyErr_Occurred(): remove "tstate ==
NULL" test.
Py_FatalError() no longer calls PyErr_Occurred() if called without
holding the GIL. So PyErr_Occurred() no longer has to support
tstate==NULL case.
_Py_CheckFunctionResult(): use directly _PyErr_Occurred() to avoid
explicit "!= NULL" test.
Additional note: the `method_check_args` function in `Objects/descrobject.c` is written in such a way that it applies to all kinds of descriptors. In particular, a future re-implementation of `wrapper_descriptor` could use that code.
CC @vstinner @encukou
https://bugs.python.org/issue37645
Automerge-Triggered-By: @encukou
* Add _Py_EnterRecursiveCall() and _Py_LeaveRecursiveCall() which
require a tstate argument.
* Pass tstate to _Py_MakeRecCheck() and _Py_CheckRecursiveCall().
* Convert Py_EnterRecursiveCall() and Py_LeaveRecursiveCall() macros
to static inline functions.
_PyThreadState_GET() is the most efficient way to get the tstate, and
so using it with _Py_EnterRecursiveCall() and
_Py_LeaveRecursiveCall() should be a little bit more efficient than
using Py_EnterRecursiveCall() and Py_LeaveRecursiveCall() which use
the "slower" PyThreadState_GET().
Provide Py_EnterRecursiveCall() and Py_LeaveRecursiveCall() as
regular functions for the limited API. Previously, there were defined
as macros, but these macros didn't work with the limited API which
cannot access PyThreadState.recursion_depth field.
Remove _Py_CheckRecursionLimit from the stable ABI.
Add Include/cpython/ceval.h header file.
The symbol table handing of PEP572's assignment expressions is not resolving correctly the scope of some variables in presence of global/nonlocal keywords in conjunction with comprehensions.
* _Py_FindEnvConfigValue() now returns a string allocated
by PyMem_RawMalloc().
* calculate_init() now decodes VPATH macro.
* Add calculate_open_pyenv() function.
* Add substring() and joinpath2() functions.
* Fix add_exe_suffix()
And a few cleanup changes.
Following symbolic links is now limited to 40 attempts, just to
prevent loops.
Add subfunctions:
* Add resolve_symlinks()
* Add calculate_argv0_path_framework()
* Add calculate_which()
* Add calculate_program_macos()
Fix also _Py_wreadlink(): readlink() result type is Py_ssize_t, not
int.
For now, we'll rely on the fact that the config structures aren't covered by the stable ABI.
We may revisit this in the future if we further explore the idea of offering a stable embedding API.
(cherry picked from commit bdace21b76)
Fix warnings options priority: PyConfig.warnoptions has the highest
priority, as stated in the PEP 587.
* Document options order in PyConfig.warnoptions documentation.
* Make PyWideStringList_INIT macro private: replace "Py" prefix
with "_Py".
* test_embed: add test_init_warnoptions().
Add a new struct_size field to PyPreConfig and PyConfig structures to
allow to modify these structures in the future without breaking the
backward compatibility.
* Replace private _config_version field with public struct_size field
in PyPreConfig and PyConfig.
* Public PyPreConfig_InitIsolatedConfig() and
PyPreConfig_InitPythonConfig()
return type becomes PyStatus, instead of void.
* Internal _PyConfig_InitCompatConfig(),
_PyPreConfig_InitCompatConfig(), _PyPreConfig_InitFromConfig(),
_PyPreConfig_InitFromPreConfig() return type becomes PyStatus,
instead of void.
* Remove _Py_CONFIG_VERSION
* Update the Initialization Configuration documentation.
* Py_InitializeFromConfig() now writes PyConfig path configuration to
the global path configuration (_Py_path_config).
* Add test_embed.test_get_pathconfig().
* Fix typo in _PyWideStringList_Join().
* Add test_embed.test_init_setpath_config(): test Py_SetPath()
with PyConfig.
* test_init_setpath() and test_init_setpythonhome() no longer call
Py_SetProgramName(), but use the default program name.
* _PyPathConfig: isolated, site_import and base_executable
fields are now only available on Windows.
* If executable is set explicitly in the configuration, ignore
calculated base_executable: _PyConfig_InitPathConfig() copies
executable to base_executable.
* Complete path config documentation.
Py_SetPath() now sets sys.executable to the program full path
(Py_GetProgramFullPath()), rather than to the program name
(Py_GetProgramName()).
Fix also memory leaks in pathconfig_set_from_config().
* _PyConfig_InitPathConfig() now starts by copying the global path
configuration, and then override values set in PyConfig.
* _PyPathConfig_Calculate() implementations no longer override
_PyPathConfig fields which are already computed. For example,
if _PyPathConfig.prefix is not NULL, leave it unchanged.
* If Py_SetPath() has been called, _PyConfig_InitPathConfig() doesn't
call _PyPathConfig_Calculate() anymore.
* _PyPathConfig_Calculate() no longer uses PyConfig,
except to initialize PyCalculatePath structure.
* pathconfig_calculate(): remove useless temporary
"_PyPathConfig new_config" variable.
* calculate_module_search_path(): remove hack to workaround memory
allocation failure, call Py_FatalError() instead.
* Fix get_program_full_path(): handle memory allocation failure.
* If Py_SetPath() has been called, _PyConfig_InitPathConfig() now
uses its value.
* Py_Initialize() now longer copies path configuration from PyConfig
to the global path configuration (_Py_path_config).
The DLL path is not computed from any user configuration and cannot
be configured by PyConfig. Instead, add a new _Py_dll_path global variable.
Remove _PyConfig_SetPathConfig(): replaced with _PyPathConfig_Init().
Py_Initialize() now longer sets the "global path configuration",
but only initialize _Py_dll_path.
Add _PyRuntimeState.preinitializing field: set to 1 while
Py_PreInitialize() is running.
_PyRuntimeState: rename also pre_initialized field to preinitialized.
In ArgumentClinic, value "NULL" should now be used only for unrepresentable default values
(like in the optional third parameter of getattr). "None" should be used if None is accepted
as argument and passing None has the same effect as not passing the argument at all.
All call sites pass NULL for `recode_encoding`, so this path is
completely untested. That's been true since before Python 3.0.
It adds significant complexity to this logic, so it's best to
take it out.
All call sites now have a literal NULL, and that's been true since
commit 768921cf3 eliminated a conditional (`foo ? bar : NULL`) at
the call site in Python/ast.c where we're parsing a bytes literal.
But even before then, that condition `foo` had been a constant
since unadorned string literals started meaning Unicode, in commit
572dbf8f1 aka v3.0a1~1035 .
The `unicode` parameter is already unused, so mark it as unused too.
The code that acted on it was also taken out before Python 3.0, in
commit 8d30cc014 aka v3.0a1~1031 .
The function (PyBytes_DecodeEscape) is exposed in the API, but it's
never been documented.
The >=, checking whether a module index was in already in the module-by-index list, needed to be strict.
Also, fold nested ifs into one and fix some bad spacing.
Summary: This mostly migrates Python-ast.c to PEP384 and removes all statics from the whole file. This modifies the generator itself that generates the Python-ast.c. It leaves in the usage of _PyObject_LookupAttr even though it's not fully PEP384 compatible (this could always be shimmed in by anyone who needs it).
bpo-37151: remove special case for PyCFunction from PyObject_Call
Alse, make the undocumented function PyCFunction_Call an alias
of PyObject_Call and deprecate it.
Relative imports use resolve_name to get the absolute target name,
which first seeks the current module's absolute package name from the globals:
If __package__ (and __spec__.parent) are missing then
import uses __name__, truncating the last segment if
the module is a submodule rather than a package __init__.py
(which it guesses from whether __path__ is defined).
The __name__ attempt should fail if there is no parent package (top level modules),
if __name__ is '__main__' (-m entry points), or both (scripts).
That is, if both __name__ has no subcomponents and the module does not seem
to be a package __init__ module then import should fail.
* Rename PyThreadState_DeleteCurrent()
to _PyThreadState_DeleteCurrent()
* Move it to the internal C API
Co-Authored-By: Carol Willing <carolcode@willingconsulting.com>
* Use the 'p' format unit instead of manually called PyObject_IsTrue().
* Pass boolean value instead 0/1 integers to functions that needs boolean.
* Convert some arguments to boolean only once.
With `symtable_visit_expr` now correctly adjusting the recursion depth for named
expressions, `symtable_handle_namedexpr` should be leaving it alone.
Also adds a new check to `PySymtable_BuildObject` that raises `SystemError`
if a successful first symbol analysis pass fails to keep the stack depth
accounting clean.
- drop TargetScopeError in favour of raising SyntaxError directly
as per the updated PEP 572
- comprehension iteration variables are explicitly local, but
named expression targets in comprehensions are nonlocal or
global. Raise SyntaxError as specified in PEP 572
- named expression targets in the outermost iterable of a
comprehension have an ambiguous target scope. Avoid resolving
that question now by raising SyntaxError. PEP 572
originally required this only for cases where the bound name
conflicts with the iteration variable in the comprehension,
but CPython can't easily restrict the exception to that case
(as it doesn't know the target variable names when visiting
the outermost iterator expression)
PyConfig_Read() is now responsible to handle early calls to
PySys_AddXOption() and PySys_AddWarnOption().
Options added by PySys_AddXOption() are now handled the same way than
PyConfig.xoptions and command line -X options.
For example, PySys_AddXOption(L"faulthandler") enables faulthandler
as expected.
empty_argv is no longer static in Python 3.8, but it is declared in
a temporary scope, whereas argv keeps a reference to it.
empty_argv memory (allocated on the stack) is reused by
make_sys_argv() code which is inlined when using gcc -O3.
Define empty_argv in PySys_SetArgvEx() body, to ensure
that it remains valid for the whole lifetime of
the PySys_SetArgvEx() call.
bpo-37834: Normalise handling of reparse points on Windows
* ntpath.realpath() and nt.stat() will traverse all supported reparse points (previously was mixed)
* nt.lstat() will let the OS traverse reparse points that are not name surrogates (previously would not traverse any reparse point)
* nt.[l]stat() will only set S_IFLNK for symlinks (previous behaviour)
* nt.readlink() will read destinations for symlinks and junction points only
bpo-1311: os.path.exists('nul') now returns True on Windows
* nt.stat('nul').st_mode is now S_IFCHR (previously was an error)
Fix codecs.lookup() to normalize the encoding name the same way
than encodings.normalize_encoding(), except that codecs.lookup()
also converts the name to lower case.
The fact that keyword names are strings is now part of the vectorcall and `METH_FASTCALL` protocols. The biggest concrete change is that `_PyStack_UnpackDict` now checks that and raises `TypeError` if not.
CC @markshannon @vstinner
https://bugs.python.org/issue37540
DeprecationWarning will continue to be emitted for invalid escape
sequences in string and bytes literals just as it did in 3.7.
SyntaxWarning may be emitted in the future. But per mailing list
discussion, we don't yet know when because we haven't settled on how to
do so in a non-disruptive manner.
(Applies 4c5b6bac24 to the master branch).
(This is https://github.com/python/cpython/pull/15142 for master/3.9)
https://bugs.python.org/issue32912
Automerge-Triggered-By: @gpshead
Imports now raise `TypeError` instead of `ValueError` for relative import failures. This makes things consistent between `builtins.__import__` and `importlib.__import__` as well as using a more natural import for the failure.
https://bugs.python.org/issue37444
Automerge-Triggered-By: @brettcannon
This changeset increases the default size of the stack
for threads on macOS to the size of the stack
of the main thread and reenables the relevant
recursion test.
Nested BinOp instances (e.g. a+b+c) had a wrong col_offset for the
second BinOp (e.g. 2 instead of 0 in the example). Fix it by using the
correct st node to copy the line and col_offset from in ast.c.
Michael Felt
Victor Stinner
Brandt Bucher
Andy Lester
Zackery Spytz
Eddie Elizondo
Hai Shi
David Carlier
Mark Shannon
Pablo Galindo
Dino Viehland
Niklas Fiekas
Ammar Askar
Géry Ogam
Julien Danjou
Dong-hee Na
Guido van Rossum
Alex Henrie
Anthony Wee
Anthony Sottile
Benjamin Peterson
Jendrik Seipp
Ned Batchelder
Batuhan Taşkaya
Sebastian Berg
Lysandros Nikolaou
Xtreak
Inada Naoki
Steve Dower
AMIR
xdegaye
Jeroen Demeyer
Max Bernstein
Serhiy Storchaka
Hansraj Das
Vinay Sajip
Joannah Nanjekye
Victor Stinner
Raymond Hettinger
Greg Price
Dino Viehland
T. Wouters
Jason R. Coombs
Jeroen Demeyer
Ben Lewis
Jordon Xu
Min ho Kim
Sergey Fedoseev
Rémi Lapeyre
Nick Coghlan
shireenrao
Gregory P. Smith
Ngalim Siregar
Ronald Oussoren
Minmin Gong
Carl Friedrich Bolz-Tereick