Each interpreter now has its own tuple free lists:
* Move tuple numfree and free_list arrays into PyInterpreterState.
* Define PyTuple_MAXSAVESIZE and PyTuple_MAXFREELIST macros in
pycore_interp.h.
* Add _Py_tuple_state structure. Pass it explicitly to tuple_alloc().
* Add tstate parameter to _PyTuple_ClearFreeList()
* Each interpreter now has its own empty tuple singleton.
When Python is built with experimental isolated interpreters, a
garbage collection now does nothing in an isolated interpreter.
Temporary workaround until subinterpreters stop sharing Python
objects.
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 the functions PyObject_GC_IsTracked and PyObject_GC_IsFinalized to the public API to allow to query if Python objects are being currently tracked or have been already finalized by the garbage collector respectively.
Convert the PyObject_GET_WEAKREFS_LISTPTR() macro to a function to
hide implementation details: the macro accessed directly to the
PyTypeObject.tp_weaklistoffset member.
Add _PyObject_GET_WEAKREFS_LISTPTR() static inline function to the
internal C API.
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
collect() should not get an exception, but it does, logging the
exception is enough. Override sys.unraisablehook to decide how to
handle unraisable exceptions.
Py_FatalError() should be avoided whenever possible.
The comment about the collection rules for the permanent generation was
incorrectly referenced by a comment in gcmodule.c (the comment has been
moved long ago into a header file). Moving the comment into the relevant
code helps with readability and avoids broken references.
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
* 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.
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".
It should be impossible for an untracked object to have the collecting
flag set. Back when state was stored in gc_refs, it obviously was
impossible (gc_refs couldn't possibly have a positive & negative value
simultaneously). While the _implementation_ of "state" has gotten much
more complicated, it's still _logically_ just as impossible.
* Misc gc code & comment cleanups.
validate_list: there are two temp flags polluting pointers, but this checked only one. Now it checks both, and verifies that the list head's pointers are not polluted.
move_unreachable: repaired incoherent comments. Added new comments. Cleared the pollution of the unreachable list head's 'next' pointer (it was expedient while the function was running, but there's no excuse for letting this damage survive the function's end).
* Update Modules/gcmodule.c
Co-Authored-By: Pablo Galindo <Pablogsal@gmail.com>
Currently if any finalizer invoked during garbage collection resurrects any object, the gc gives up and aborts the collection. Although finalizers are assured to only run once per object, this behaviour of the gc can lead to an ever-increasing memory situation if new resurrecting objects are allocated in every new gc collection.
To avoid this, recompute what objects among the unreachable set need to be resurrected and what objects can be safely collected. In this way, resurrecting objects will not block the collection of other objects in the unreachable set.
subtract_refs() now pass the parent object to visit_decref() which
pass it to _PyObject_ASSERT(). So if the "is freed" assertion fails,
the parent is used in debug trace, rather than the freed object. The
parent object is more likely to contain useful information. Freed
objects cannot be inspected are are displayed as "<object at xxx is
freed>" with no other detail.
In debug mode, PyObject_GC_Track() now calls tp_traverse() of the
object type to ensure that the object is valid: test that objects
visited by tp_traverse() are valid.
Fix pyexpat.c: only track the parser in the GC once the parser is
fully initialized.
bpo-36389, bpo-38376: The _PyObject_CheckConsistency() function is
now also available in release mode. For example, it can be used to
debug a crash in the visit_decref() function of the GC.
Modify the following functions to also work in release mode:
* _PyDict_CheckConsistency()
* _PyObject_CheckConsistency()
* _PyType_CheckConsistency()
* _PyUnicode_CheckConsistency()
Other changes:
* _PyMem_IsPtrFreed(ptr) now also returns 1 if ptr is NULL
(equals to 0).
* _PyBytesWriter_CheckConsistency() now returns 1 and is only used
with assert().
* Reorder _PyObject_Dump() to write safe fields first, and only
attempt to render repr() at the end.
Fix a bug due to the interaction of weakrefs and the cyclic garbage
collector. We must clear any weakrefs in garbage in order to prevent
their callbacks from executing and causing a crash.
In debug mode, visit_decref() now calls _PyObject_IsFreed() to ensure
that the object is not freed. If it's freed, the program fails with
an assertion error and Python dumps informations about the freed
object.
gc used several PySys_WriteStderr() calls to write stats.
It caused stats mixed up when stderr is shared by multiple
processes like this:
gc: collecting generation 2...
gc: objects in each generation: 0 0gc: collecting generation 2...
gc: objects in each generation: 0 0 126077 126077
gc: objects in permanent generation: 0
gc: objects in permanent generation: 0
gc: done, 112575 unreachable, 0 uncollectablegc: done, 112575 unreachable, 0 uncollectable, 0.2223s elapsed
, 0.2344s elapsed
It is now allowed to add new fields at the end of the PyTypeObject struct without having to allocate a dedicated compatibility flag in tp_flags.
This will reduce the risk of running out of bits in the 32-bit tp_flags value.
* sys.unraisablehook: add 'err_msg' field to UnraisableHookArgs.
* Use _PyErr_WriteUnraisableMsg() in _ctypes _DictRemover_call()
and gc delete_garbage().
* Add a 'runtime' variable to Py_FinalizeEx() rather than working
directly on the global variable _PyRuntime
* Add a 'runtime' parameter to _PyGC_Fini(), _PyGILState_Fini()
and call_ll_exitfuncs()
* Add _PyObject_ASSERT_FROM() and _PyObject_ASSERT_FAILED_MSG()
macros.
* PyObject_GC_Track() now calls _PyObject_ASSERT_FAILED_MSG(),
instead of Py_FatalError(), if the object is already tracked, to
dump more information on error.
* _PyObject_GC_TRACK() no longer checks if the object is already
tracked at runtime, use an assertion instead for best performances;
PyObject_GC_Track() still checks at runtime.
* pycore_object.h now includes pycore_pystate.h.
* Convert _PyObject_GC_TRACK() and _PyObject_GC_UNTRACK() macros to
inline functions.
Replace assert() with _PyObject_ASSERT() in Modules/gcmodule.c
to dump the faulty object on assertion failure to ease debugging.
Fix also indentation of a large comment.
Initial patch written by David Malcolm.
Co-Authored-By: David Malcolm <dmalcolm@redhat.com>
Changes:
* Add _PyObject_AssertFailed() function.
* Add _PyObject_ASSERT() and _PyObject_ASSERT_WITH_MSG() macros.
* gc_decref(): replace assert() with _PyObject_ASSERT_WITH_MSG() to
dump the faulty object if the assertion fails.
_PyObject_AssertFailed() calls:
* _PyMem_DumpTraceback(): try to log the traceback where the object
memory has been allocated if tracemalloc is enabled.
* _PyObject_Dump(): log repr(obj).
* Py_FatalError(): log the current Python traceback.
_PyObject_AssertFailed() uses _PyObject_IsFreed() heuristic to check
if the object memory has been freed by a debug hook on Python memory
allocators.
Initial patch written by David Malcolm.
Co-Authored-By: David Malcolm <dmalcolm@redhat.com>
* Failure in adding to gc.garbage is no longer fatal.
* An exception in tp_clear() no longer lead to crash (though tp_clear() should not leave exceptions).
* Fixed a leak when the GC fails to add an object with __del__ into
the gc.garbage list.
* PyGC_Collect() can now be called when an exception is set and
preserves it.
* Fixed an undefined behavior with comparing a dead pointer with NULL.
* Use wider types (int => Py_ssize_t) to avoid integer overflows.
* Fix gc.get_freeze_count(): use Py_ssize_t type rather than int, since gc_list_size() returns a Py_ssize_t.
Freeze all the objects tracked by gc - move them to a permanent generation
and ignore all the future collections. This can be used before a POSIX
fork() call to make the gc copy-on-write friendly or to speed up collection.
* 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).
* 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.
Tested on macOS 10.11 dtrace, Ubuntu 16.04 SystemTap, and libbcc.
Largely based by an initial patch by Jesús Cea Avión, with some
influence from Dave Malcolm's SystemTap patch and Nikhil Benesch's
unification patch.
Things deliberately left out for simplicity:
- ustack helpers, I have no way of testing them at this point since
they are Solaris-specific
- PyFrameObject * in function__entry/function__return, this is
SystemTap-specific
- SPARC support
- dynamic tracing
- sys module dtrace facility introspection
All of those might be added later.
threading.Lock.acquire(), threading.RLock.acquire() and socket operations now
use a monotonic clock, instead of the system clock, when a timeout is used.
Issue #21435: Segfault in gc with cyclic trash
Changed the iteration logic in finalize_garbage() to tolerate objects vanishing
from the list as a side effect of executing a finalizer.
PyObject_Calloc(), _PyObject_GC_Calloc(). bytes(int) and bytearray(int) are now
using ``calloc()`` instead of ``malloc()`` for large objects which is faster
and use less memory (until the bytearray buffer is filled with data).
Clarified the "At the moment" wording, and added the get_stats entry in the
module summary that Serhiy noted was missing at the end of issue 16351.
Given that pydoc lists all the function docstrings, I'm not sure that module
summary section is actually needed; but, it is probably better to address that
when the module is converted to use Argument Clinic. In the meantime we
should keep the list complete.
Victor Stinner
Hai Shi
Pablo Galindo
Petr Viktorin
Tim Peters
Neil Schemenauer
Victor Stinner
Inada Naoki
Jeroen Demeyer
Antoine Pitrou
Quan Tian
INADA Naoki
Serhiy Storchaka
Yury Selivanov
Victor Stinner
brainfvck
Eric Snow
Łukasz Langa
Antoine Pitrou
Tim Peters
R David Murray
Gregory P. Smith