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.
inspect.isfunction() processes both inspect.isfunction(func) and
inspect.isfunction(partial(func, arg)) correctly but some other functions in the
inspect module (iscoroutinefunction, isgeneratorfunction and isasyncgenfunction)
lack this functionality. This commits adds a new check in the mentioned functions
in the inspect module so they can work correctly with arbitrarily nested partial
functions.
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>
* Add Py_STATIC_INLINE() macro to declare a "static inline" function.
If the compiler supports it, try to always inline the function even if no
optimization level was specified.
* Modify pydtrace.h to use Py_STATIC_INLINE() when WITH_DTRACE is
not defined.
* Add an unit test on Py_DECREF() to make sure that
_Py_NegativeRefcount() reports the correct filename.
tracemalloc now tries to update the traceback when an object is
reused from a "free list" (optimization for faster object creation,
used by the builtin list type for example).
Changes:
* Add _PyTraceMalloc_NewReference() function which tries to update
the Python traceback of a Python object.
* _Py_NewReference() now calls _PyTraceMalloc_NewReference().
* Add an unit test.
* Use _PyUnicode_Copy in sanitize_isoformat_str
* Use repr in fromisoformat error message
This reverses commit 67b74a98b2 per Serhiy Storchaka's suggestion:
I suggested to use %R in the error message because including the raw
string can be confusing in the case of empty string, or string
containing trailing whitespaces, invisible or unprintable characters.
We agree that it is better to change both the C and pure Python versions
to use repr.
* Retain non-sanitized dtstr for error printing
This does not create an extra string, it just holds on to a reference to
the original input string for purposes of creating the error message.
* PEP 7 fixes to from_isoformat
* Separate handling of Unicode and other errors
In the initial implementation, errors other than encoding errors would
both raise an error indicating an invalid format, which would not be
true for errors like MemoryError.
* Drop needs_decref from _sanitize_isoformat_str
Instead _sanitize_isoformat_str returns a new reference, even to the
original string.
For builtin types with builtin subclasses, help() on the type now shows up
to 4 of the subclasses. This partially replaces the exception hierarchy
information previously displayed in Python 2.7.
path_error() uses GetLastError() on Windows, but some os functions
are implemented via CRT APIs which report errors via errno.
This may result in raising OSError with invalid error code (such
as zero).
Introduce posix_path_error() function and use it where appropriate.
If buffering=1 is specified for open() in binary mode, it is silently
treated as buffering=-1 (i.e., the default buffer size).
Coupled with the fact that line buffering is always supported in Python 2,
such behavior caused several issues (e.g., bpo-10344, bpo-21332).
Warn that line buffering is not supported if open() is called with
binary mode and buffering=1.
The reprlib code was copied here instead of importing reprlib. I'm not sure if we really need to avoid the import, but since I expect dataclasses to be more common that reprlib, it seems wise. Plus, the code is small.
Adding `max_num_fields` to `cgi.FieldStorage` to make DOS attacks harder by
limiting the number of `MiniFieldStorage` objects created by `FieldStorage`.
It is now guarantied that children of xml.etree.ElementTree.Element
are Elements (at least in C implementation). Previously methods
__setitem__(), __setstate__() and __deepcopy__() could be used for
adding non-Element children.
When runnint test_logging with --huntrleaks after commit
18fb1fb943, test_out_of_order fails
to raise ValueError due to the fact that the new test
test_out_of_order_with_dollar_style mutates the out_of_order
dictionary. Even if the test copies the dictionary first, the mutation
is done in a very deep level so the original one is also affected.
Fix test_cmd_line_script.test_nonexisting_script(): the test must not
rely on sys.executable, since main.c uses config->program which can
be different than sys.executable in many cases (for example, on macOS
when using the framework).
Allow annotated global names in the module namespace after the symbol is
declared as global. Previously, only symbols annotated before they are declared
as global (i.e. inside a function) were allowed. This change allows symbols to be
declared as global before the annotation happens in the global scope.
With macOS framework builds, test case test_nonexisting_script in
test_nonexisting_script fails because the test case assumes that
the file name in sys.executable will appear in the error message.
For macOS framework builds, sys.executable is the file name of the
stub launcher and its file name bears no relationship to the file
name of the actual python executable. For now, skip the test in
this case.
Unconditional forcing of ``CHECKED_HASH`` invalidation was introduced in
3.7.0 in bpo-29708. The change is bad, as it unconditionally overrides
*invalidation_mode*, even if it was passed as an explicit argument to
``py_compile.compile()`` or ``compileall``. An environment variable
should *never* override an explicit argument to a library function.
That change leads to multiple test failures if the ``SOURCE_DATE_EPOCH``
environment variable is set.
This changes ``py_compile.compile()`` to only look at
``SOURCE_DATE_EPOCH`` if no explicit *invalidation_mode* was specified.
I also made various relevant tests run with explicit control over the
value of ``SOURCE_DATE_EPOCH``.
While looking at this, I noticed that ``zipimport`` does not work
with hash-based .pycs _at all_, though I left the fixes for
subsequent commits.