This patch appropriately marks known false alarms in the
small object allocator when address sanity checking is
enabled (patch contributed by Dhiru Kholia).
The setobject freelist was consuming memory but not providing much value.
Even when a freelisted setobject was available, most of the setobject
fields still needed to be initialized and the small table still required
a memset(). This meant that the custom freelisting scheme for sets was
providing almost no incremental benefit over the default Python freelist
scheme used by _PyObject_Malloc() in Objects/obmalloc.c.
Issue #18942: sys._debugmallocstats() output was damaged on Windows.
_PyDebugAllocatorStats() called PyOS_snprintf() with a %zd format
code, but MS doesn't support that code. Interpolated
PY_FORMAT_SIZE_T in place of the "z".
_PyDebugAllocatorStats() called PyOS_snprintf() with a %zd format
code, but MS doesn't support that code. Interpolated
PY_FORMAT_SIZE_T in place of the "z".
Modern processors tend to make consecutive memory accesses cheaper than
random probes into memory.
Small sets can fit into L1 cache, so they get less benefit. But they do
come out ahead because the consecutive probes don't probe the same key
more than once and because the randomization step occurs less frequently
(or not at all).
For the open addressing step, putting the perturb shift before the index
calculation gets the upper bits into play sooner.
"Issue #18408: PyObject_Str(), PyObject_Repr() and type_call() now fail with an
assertion error if they are called with an exception set (PyErr_Occurred()).
As PyEval_EvalFrameEx(), they may clear the current exception and so the caller
looses its exception."
The Gdb prettyprint plugin depended on the dummy object being displayable.
Other solutions besides a unicode object are possible. For now, get it
back up and running.
The identity checks in lookkey() need to be there to prevent the dummy
object from leaking through Py_RichCompareBool() into user code in the
rare circumstance where the dummy's hash value exactly matches the hash
value of the actual key being looked up.
Remove an unused early-out test from the critical path for
dict and set lookups.
When the strings already have matching lengths, kinds, and hashes,
there is no additional information gained by checking the first
characters (the probability of a mismatch is already known to
be less than 1 in 2**64).
Letting the compiler decide how to optimize the multiply by five
gives it the freedom to make better choices for the best technique
for a given target machine.
For example, GCC on x86_64 produces a little bit better code:
Old-way (3 steps with a data dependency between each step):
shrq $5, %r13
leaq 1(%rbx,%r13), %rax
leaq (%rax,%rbx,4), %rbx
New-way (3 steps with no dependency between the first two steps
which can be run in parallel):
leaq (%rbx,%rbx,4), %rax # i*5
shrq $5, %r13 # perturb >>= PERTURB_SHIFT
leaq 1(%r13,%rax), %rbx # 1 + perturb + i*5
- replace 'long int' / 'long' by 'int'
- fix capitalization of "Python" in PyLong_AsUnsignedLong
- "is too large" -> "too large", for consistency with other messages.
PyStructSequence_InitType() except that it has a return value (0 on success,
-1 on error).
* PyStructSequence_InitType2() now raises MemoryError on memory allocation failure
* Fix also some calls to PyDict_SetItemString(): handle error
assertion error if they are called with an exception set (PyErr_Occurred()).
As PyEval_EvalFrameEx(), they may clear the current exception and so the caller
looses its exception.
_PyDict_GetItemId() is more efficient: it only builds the Unicode string once.
Identifiers (dictionary keys) are now created at Python initialization, and if
the creation failed, Python does exit with a fatal error.
Before, PyDict_GetItemString() failure was not handled: structseq_new() could
call PyObject_GC_NewVar() with a negative size, and structseq_dealloc() could
also crash.
It was easy to miss the call to type->tp_init because it was done in a long
conditional expression. Split the long expression in multiple lines to make the
debug step by step easier.
and _PyUnicode_HAS_WSTR_MEMORY() macros
These macros are called in unicode_dealloc(), whereas the unicode object can be
"inconsistent" if the creation of the object failed.
For example, when unicode_subtype_new() fails on a memory allocation,
_PyUnicode_CheckConsistency() fails with an assertion error because data is
NULL.
type->tp_str must not point to slot_tp_str() if type has no __str__ attribute,
so there is no reason for slot_tp_str() to fallback on slot_tp_str() on lookup
error. Moreover, calling PyErr_Clear() may hide a real bug like MemoryError.
If __str__ attribute is removed, slots must be updated (which is done by
type_setattro()).
Don't mark old extra memory dead before calling realloc(). realloc() can fail
and realloc() must not touch the original buffer on failure.
So mark old extra memory dead only on success if the new buffer did not move
(has the same address).
* Replace malloc() with PyMem_RawMalloc()
* Replace PyMem_Malloc() with PyMem_RawMalloc() where the GIL is not held.
* _Py_char2wchar() now returns a buffer allocated by PyMem_RawMalloc(), instead
of PyMem_Malloc()
Add new enum:
* PyMemAllocatorDomain
Add new structures:
* PyMemAllocator
* PyObjectArenaAllocator
Add new functions:
* PyMem_RawMalloc(), PyMem_RawRealloc(), PyMem_RawFree()
* PyMem_GetAllocator(), PyMem_SetAllocator()
* PyObject_GetArenaAllocator(), PyObject_SetArenaAllocator()
* PyMem_SetupDebugHooks()
Changes:
* PyMem_Malloc()/PyObject_Realloc() now always call malloc()/realloc(), instead
of calling PyObject_Malloc()/PyObject_Realloc() in debug mode.
* PyObject_Malloc()/PyObject_Realloc() now falls back to
PyMem_Malloc()/PyMem_Realloc() for allocations larger than 512 bytes.
* Redesign debug checks on memory block allocators as hooks, instead of using C
macros
CID 983320: Resource leak (RESOURCE_LEAK)
CID 983321: Resource leak (RESOURCE_LEAK)
leaked_storage: Variable substring going out of scope leaks the storage it points to.
* Add a new PyMemAllocators structure
* New functions:
- PyMem_RawMalloc(), PyMem_RawRealloc(), PyMem_RawFree(): GIL-free memory
allocator functions
- PyMem_GetRawAllocators(), PyMem_SetRawAllocators()
- PyMem_GetAllocators(), PyMem_SetAllocators()
- PyMem_SetupDebugHooks()
- _PyObject_GetArenaAllocators(), _PyObject_SetArenaAllocators()
* Add unit test for PyMem_Malloc(0) and PyObject_Malloc(0)
* Add unit test for new get/set allocators functions
* PyObject_Malloc() now falls back on PyMem_Malloc() instead of malloc() if
size is bigger than SMALL_REQUEST_THRESHOLD, and PyObject_Realloc() falls
back on PyMem_Realloc() instead of realloc()
* PyMem_Malloc() and PyMem_Realloc() now always call malloc() and realloc(),
instead of calling PyObject_Malloc() and PyObject_Realloc() in debug mode
ImportError.
The exception is raised by import when a module could not be found.
Technically this is defined as no viable loader could be found for the
specified module. This includes ``from ... import`` statements so that
the module usage is consistent for all situations where import
couldn't find what was requested.
This should allow for the common idiom of::
try:
import something
except ImportError:
pass
to be updated to using ModuleNotFoundError and not accidentally mask
ImportError messages that should propagate (e.g. issues with a
loader).
This work was driven by the fact that the ``from ... import``
statement needed to be able to tell the difference between an
ImportError that simply couldn't find a module (and thus silence the
exception so that ceval can raise it) and an ImportError that
represented an actual problem.
The result type is int, return -1 to avoid a compiler warning (cast Py_ssize_t
to int). PyObject_Size() can only fail with -1, and anyway a constructor
should return -1 on error, not an arbitrary negative number.
attributes to None.
The long-term goal is for people to be able to rely on these
attributes existing and checking for None to see if they have been
set. Since import itself sets these attributes when a loader does not
the only instances when the attributes are None are from someone
overloading __import__() and not using a loader or someone creating a
module from scratch.
This patch also unifies module initialization. Before you could have
different attributes with default values depending on how the module
object was created. Now the only way to not get the same default set
of attributes is to circumvent initialization by calling
ModuleType.__new__() directly.
Antoine Pitrou
Nick Coghlan
Raymond Hettinger
Georg Brandl
Tim Peters
Ethan Furman
Serhiy Storchaka
Victor Stinner
Mark Dickinson
Victor Stinner
Christian Heimes
Brett Cannon
Martin v. Löwis
Benjamin Peterson
Guido van Rossum
Alexandre Vassalotti
Ezio Melotti