Victor Stinner [Tue, 22 Jan 2019 20:18:05 +0000 (21:18 +0100)]
bpo-35713: Split _Py_InitializeCore into subfunctions (GH-11650)
* Split _Py_InitializeCore_impl() into subfunctions: add multiple pycore_init_xxx() functions
* Preliminary sys.stderr is now set earlier to get an usable
sys.stderr ealier.
* Move code into _Py_Initialize_ReconfigureCore() to be able to call
it from _Py_InitializeCore().
* Split _PyExc_Init(): create a new _PyBuiltins_AddExceptions()
function.
* Call _PyExc_Init() earlier in _Py_InitializeCore_impl()
and new_interpreter() to get working exceptions earlier.
* _Py_ReadyTypes() now returns _PyInitError rather than calling
Py_FatalError().
* Misc code cleanup
* The PyByteArray_Init() and PyByteArray_Fini() functions have been
removed. They did nothing since Python 2.7.4 and Python 3.2.0, were
excluded from the limited API (stable ABI), and were not
documented.
* Move "_PyXXX_Init()" and "_PyXXX_Fini()" declarations from
Include/cpython/pylifecycle.h to
Include/internal/pycore_pylifecycle.h. Replace
"PyAPI_FUNC(TYPE)" with "extern TYPE".
* _PyExc_Init() now returns an error on failure rather than calling
Py_FatalError(). Move macros inside _PyExc_Init() and undefine them
when done. Rewrite macros to make them look more like statement:
add ";" when using them, add "do { ... } while (0)".
* _PyUnicode_Init() now returns a _PyInitError error rather than call
Py_FatalError().
* Move stdin check from _PySys_BeginInit() to init_sys_streams().
* _Py_ReadyTypes() now returns a _PyInitError error rather than
calling Py_FatalError().
Lucas Cimon [Tue, 22 Jan 2019 16:15:01 +0000 (17:15 +0100)]
bpo-35720: Fixing a memory leak in pymain_parse_cmdline_impl() (GH-11528)
When the loop in the pymain_read_conf function in this same file
calls pymain_init_cmdline_argv() a 2nd time, the pymain->command
buffer of wchar_t is overriden and the previously allocated memory
is never freed.
Ivan Levkivskyi [Tue, 22 Jan 2019 11:18:22 +0000 (11:18 +0000)]
bpo-33416: Add end positions to Python AST (GH-11605)
The majority of this PR is tediously passing `end_lineno` and `end_col_offset` everywhere. Here are non-trivial points:
* It is not possible to reconstruct end positions in AST "on the fly", some information is lost after an AST node is constructed, so we need two more attributes for every AST node `end_lineno` and `end_col_offset`.
* I add end position information to both CST and AST. Although it may be technically possible to avoid adding end positions to CST, the code becomes more cumbersome and less efficient.
* Since the end position is not known for non-leaf CST nodes while the next token is added, this requires a bit of extra care (see `_PyNode_FinalizeEndPos`). Unless I made some mistake, the algorithm should be linear.
* For statements, I "trim" the end position of suites to not include the terminal newlines and dedent (this seems to be what people would expect), for example in
```python
class C:
pass
pass
```
the end line and end column for the class definition is (2, 8).
* For `end_col_offset` I use the common Python convention for indexing, for example for `pass` the `end_col_offset` is 4 (not 3), so that `[0:4]` gives one the source code that corresponds to the node.
* I added a helper function `ast.get_source_segment()`, to get source text segment corresponding to a given AST node. It is also useful for testing.
An (inevitable) downside of this PR is that AST now takes almost 25% more memory. I think however it is probably justified by the benefits.
Victor Stinner [Mon, 21 Jan 2019 09:24:12 +0000 (10:24 +0100)]
bpo-35772: Fix test_tarfile on ppc64 (GH-11606)
Fix sparse file tests of test_tarfile on ppc64le with the tmpfs
filesystem.
Fix the function testing if the filesystem supports sparse files:
create a file which contains data and "holes", instead of creating a
file which contains no data.
tmpfs effective block size is a page size (tmpfs lives in the page
cache). RHEL uses 64 KiB pages on aarch64, ppc64 and ppc64le, only
s390x and x86_64 use 4 KiB pages, whereas the test punch holes of
4 KiB.
It previously deleted Window => Zoom Height by mistake.
(Zoom Height is now on the Options menu). On Mac, the settings
dialog is accessed via Preferences on the IDLE menu.
Nick Coghlan [Thu, 17 Jan 2019 10:41:29 +0000 (20:41 +1000)]
bpo-35486: Note Py3.6 import system API requirement change (GH-11540)
While the introduction of ModuleNotFoundError was fully backwards
compatible on the import API consumer side, folks providing alternative
implementations of `__import__` need to make an update to be
forward compatible with clients that start relying on the new subclass.
Victor Stinner [Tue, 15 Jan 2019 23:02:35 +0000 (00:02 +0100)]
bpo-35537: subprocess uses os.posix_spawn in some cases (GH-11452)
The subprocess module can now use the os.posix_spawn() function
in some cases for better performance. Currently, it is only used on macOS
and Linux (using glibc 2.24 or newer) if all these conditions are met:
* executable path contains a directory
* close_fds=False
* preexec_fn, pass_fds, cwd, stdin, stdout, stderr
and start_new_session parameters are not set
Christian Heimes [Tue, 15 Jan 2019 22:47:42 +0000 (23:47 +0100)]
bpo-35746: Fix segfault in ssl's cert parser (GH-11569)
Fix a NULL pointer deref in ssl module. The cert parser did not handle CRL
distribution points with empty DP or URI correctly. A malicious or buggy
certificate can result into segfault.
Signed-off-by: Christian Heimes <christian@python.org>
https://bugs.python.org/issue35746
asyncio.ProactorEventLoop now catchs and logs send errors when the
self-pipe is full: BaseProactorEventLoop._write_to_self() now catchs
and logs OSError exceptions, as done by
BaseSelectorEventLoop._write_to_self().
IocpProactor.close() now uses time to decide when to log: wait 1
second before the first log, then log every second. Log also the
number of seconds since close() was called.
Previously, calling the strftime() method on a datetime object with a
trailing '%' in the format string would result in an exception. However,
this only occured when the datetime C module was being used; the python
implementation did not match this behavior. Datetime is now PEP-399
compliant, and will not throw an exception on a trailing '%'.
Serhiy Storchaka [Sat, 12 Jan 2019 08:30:35 +0000 (10:30 +0200)]
bpo-35552: Fix reading past the end in PyUnicode_FromFormat() and PyBytes_FromFormat(). (GH-11276)
Format characters "%s" and "%V" in PyUnicode_FromFormat() and "%s" in PyBytes_FromFormat()
no longer read memory past the limit if precision is specified.
Serhiy Storchaka [Sat, 12 Jan 2019 06:26:34 +0000 (08:26 +0200)]
bpo-35719: Optimize multi-argument math functions. (GH-11527)
Use the fast call convention for math functions atan2(),
copysign(), hypot() and remainder() and inline unpacking
arguments. This sped up them by 1.3--2.5 times.
Eric Snow [Fri, 11 Jan 2019 21:26:55 +0000 (14:26 -0700)]
bpo-35423: Stop using the "pending calls" machinery for signals. (gh-10972)
This change separates the signal handling trigger in the eval loop from the "pending calls" machinery. There is no semantic change and the difference in performance is insignificant.
The change makes both components less confusing. It also eliminates the risk of changes to the pending calls affecting signal handling. This is particularly relevant for some upcoming pending calls changes I have in the works.
Serhiy Storchaka [Fri, 11 Jan 2019 16:01:42 +0000 (18:01 +0200)]
bpo-35582: Argument Clinic: Optimize the "all boring objects" case. (GH-11520)
Use _PyArg_CheckPositional() and inlined code instead of
PyArg_UnpackTuple() and _PyArg_UnpackStack() if all parameters
are positional and use the "object" converter.
Fix memory leaks in asyncio ProactorEventLoop on overlapped operation
failures.
Changes:
* Implement the tp_traverse slot in the _overlapped.Overlapped type
to help to break reference cycles and identify referrers in the
garbage collector.
* Always clear overlapped on failure: not only set type to
TYPE_NOT_STARTED, but release also resources.
Victor Stinner [Thu, 10 Jan 2019 10:24:40 +0000 (11:24 +0100)]
asyncio: __del__() keep reference to warnings.warn (GH-11491)
* asyncio: __del__() keep reference to warnings.warn
The __del__() methods of asyncio classes now keep a strong reference
to the warnings.warn() to be able to display the ResourceWarning
warning in more cases. Ensure that the function remains available if
instances are destroyed late during Python shutdown (while module
symbols are cleared).
* Rename warn parameter to _warn
"_warn" name is a hint that it's not the regular warnings.warn()
function.
Victor Stinner [Thu, 10 Jan 2019 10:23:26 +0000 (11:23 +0100)]
IocpProactor: prevent modification if closed (GH-11494)
* _wait_for_handle(), _register() and _unregister() methods of
IocpProactor now raise an exception if closed
* Add "closed" to IocpProactor.__repr__()
* Simplify IocpProactor.close()
As in title, expose C `raise` function as `raise_function` in `signal` module. Also drop existing `raise_signal` in `_testcapi` module and replace all usages with new function.
test_asyncio/test_sendfile.py now resets the event loop policy using
tearDownModule() as done in other tests, to prevent a warning when
running tests on Windows.
Xtreak [Mon, 7 Jan 2019 15:09:14 +0000 (20:39 +0530)]
bpo-35560: Remove assertion from format(float, "n") (GH-11288)
Fix an assertion error in format() in debug build for floating point
formatting with "n" format, zero padding and small width. Release build is
not impacted. Patch by Karthikeyan Singaravelan.
Stefan Behnel [Wed, 2 Jan 2019 12:22:06 +0000 (13:22 +0100)]
bpo-35588: Speed up mod, divmod and floordiv operations for Fraction type (#11322)
* bpo-35588: Implement mod and divmod operations for Fraction type by spelling out the numerator/denominator calculation, instead of instantiating and normalising Fractions along the way. This speeds up '%' and divmod() by 2-3x.
* bpo-35588: Also reimplement Fraction.__floordiv__() using integer operations to make it ~4x faster.
* Improve code formatting.
Co-Authored-By: scoder <stefan_ml@behnel.de>
* bpo-35588: Fix return type of divmod(): the result of the integer division should be an integer.
* bpo-35588: Further specialise __mod__() and inline the original helper function _flat_divmod() since it's no longer reused.
* bpo-35588: Add some tests with large numerators and/or denominators.
* bpo-35588: Use builtin "divmod()" function for implementing __divmod__() in order to simplify the implementation, even though performance results are mixed.
* Rremove accidentally added empty line.
* bpo-35588: Try to provide more informative output on test failures.