If we have a chain of generators/coroutines that are 'yield from'ing
each other, then resuming the stack works like:
- call send() on the outermost generator
- this enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- which calls send() on the next generator
- which enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- ...etc.
However, every time we enter _PyEval_EvalFrameDefault, the first thing
we do is to check for pending signals, and if there are any then we
run the signal handler. And if it raises an exception, then we
immediately propagate that exception *instead* of starting to execute
bytecode. This means that e.g. a SIGINT at the wrong moment can "break
the chain" – it can be raised in the middle of our yield from chain,
with the bottom part of the stack abandoned for the garbage collector.
The fix is pretty simple: there's already a special case in
_PyEval_EvalFrameEx where it skips running signal handlers if the next
opcode is SETUP_FINALLY. (I don't see how this accomplishes anything
useful, but that's another story.) If we extend this check to also
skip running signal handlers when the next opcode is YIELD_FROM, then
that closes the hole – now the exception can only be raised at the
innermost stack frame.
This shouldn't have any performance implications, because the opcode
check happens inside the "slow path" after we've already determined
that there's a pending signal or something similar for us to process;
the vast majority of the time this isn't true and the new check
doesn't run at all.
Before, it was possible to get the following sequence of
events (especially on Windows, where the C-level signal handler for
SIGINT is run in a separate thread):
- SIGINT arrives
- trip_signal is called
- trip_signal writes to the wakeup fd
- the main thread wakes up from select()-or-equivalent
- the main thread checks for pending signals, but doesn't see any
- the main thread drains the wakeup fd
- the main thread goes back to sleep
- trip_signal sets is_tripped=1 and calls Py_AddPendingCall to notify
the main thread the it should run the Python-level signal handler
- the main thread doesn't notice because it's asleep
This has been causing repeated failures in the Trio test suite:
https://github.com/python-trio/trio/issues/119
when there are no more `await` or `yield (from)` before return in coroutine,
cancel was ignored.
example:
async def coro():
asyncio.Task.current_task().cancel()
return 42
...
res = await coro() # should raise CancelledError
Compiled regular expression objects with the re.LOCALE flag no longer
depend on the locale at compile time. Only the locale at matching
time affects the result of matching.
FileIO.seek() and FileIO.tell() method now set the internal seekable
attribute to avoid one syscall on open() (in buffered or text mode).
The seekable property is now also more reliable since its value is
set correctly on memory allocation failure.
bpo-28769 changed PyUnicode_AsUTF8() return type from const char* to
char* in Python 3.7, but tm_zone field type of the tm structure is
char* on FreeBSD.
Cast PyUnicode_AsUTF8() to char* in gettmarg() to fix the warning:
Modules/timemodule.c:443:20: warning: assigning to 'char *'
from 'const char *' discards qualifiers
timegm() return type is time_t, not int. Use time_t to prevent the
following compiler warning on Windows:
timemodule.c: warning C4244: '=': conversion from 'time_t' to 'int',
possible loss of data
* bpo-30125: Cleanup faulthandler.c
* Use size_t type for iterators
* Add { ... }
* bpo-30125: Fix faulthandler.disable() on Windows
On Windows, faulthandler.disable() now removes the exception handler
installed by faulthandler.enable().
Only define the get_zone() and get_gmtoff() private functions in the
time module if these functions are needed to initialize the module.
The change fixes the following warnings on AIX:
Modules/timemodule.c:1175:1: warning: 'get_gmtoff' defined but not used [-Wunused-function]
Modules/timemodule.c:1164:1: warning: 'get_zone' defined but not used [-Wunused-function]
* Remove conditional on free of `dps`, since `dps` is now allocated for
all versions of OpenSSL
* Remove call to `x509_check_ca` since it was only used to cache
the `crldp` field of the certificate
CRL_DIST_POINTS_free is available in all supported versions of OpenSSL
(recent 0.9.8+) and LibreSSL.
* Implement math.remainder.
* Fix markup for arguments; use double spaces after period.
* Mark up function reference in what's new entry.
* Add comment explaining the calculation in the final branch.
* Fix out-of-order entry in whatsnew.
* Add comment explaining why it's good enough to compare m with c, in spite of possible rounding error.
Fix the use of recursion in itertools.chain.from_iterable. Using recursion
is unnecessary, and can easily cause stack overflows, especially when
building in low optimization modes or with Py_DEBUG enabled.
Element.getiterator() and the html parameter of XMLParser() were
deprecated only in the documentation (since Python 3.2 and 3.4 correspondintly).
Now using them emits a deprecation warning.
* Don’t need check_warnings any more.
There was few cases of using literal 0 instead of NULL in the context of
pointers. While this was a legitimate C code, using NULL rather than 0 makes
the code clearer.
* bpo-6532: Make the thread id an unsigned integer.
From C API side the type of results of PyThread_start_new_thread() and
PyThread_get_thread_ident(), the id parameter of
PyThreadState_SetAsyncExc(), and the thread_id field of PyThreadState
changed from "long" to "unsigned long".
* Restore a check in thread_get_ident().
* Add _PyObject_HasFastCall()
* partial_call() now avoids temporary tuple to pass positional
arguments if the callable supports the FASTCALL calling convention
for positional arguments.
* Fix also a performance regression in partial_call() if the callable
doesn't support FASTCALL.
Directory and zipfile execution previously added
the parent directory of the directory or zipfile
as sys.path[0] and then subsequently overwrote
it with the directory or zipfile itself.
This caused problems in isolated mode, as it
overwrote the "stdlib as a zip archive" entry
in sys.path, as the parent directory was
never added.
The attempted fix to that issue in bpo-29319
created the opposite problem in *non*-isolated
mode, by potentially leaving the parent
directory on sys.path instead of overwriting it.
This change fixes the root cause of the problem
by removing the whole "add-and-overwrite" dance
for sys.path[0], and instead simply never adds
the parent directory to sys.path in the first
place.
* bpo-26121: Use C library implementation for math functions:
tgamma(), lgamma(), erf() and erfc().
* Don't use tgamma() and lgamma() from libc on OS X.
sys.version and the platform module python_build(),
python_branch(), and python_revision() functions now use
git information rather than hg when building from a repo.
Based on original patches by Brett Cannon and Steve Dower.
* init commit, with initial tests for from_param and fields __set__ and __get__, and some additions to from_buffer and from_buffer_copy
* added the rest of tests and patches. probably only a first draft.
* removed trailing spaces
* replace ctype with ctypes in error messages
* change back from ctypes instance to ctype instance
The curses module used mkstemp() + fopen() to create a temporary file in
/tmp. The /tmp directory does not exist on Android. The tmpfile()
function simplifies the task a lot. It creates a temporary file in a
correct directory, takes care of cleanup and returns FILE*.
tmpfile is supported on all platforms (C89, POSIX 2001, Android,
Windows).
Signed-off-by: Christian Heimes <christian@python.org>
sock_addr_t is used to define the minimum size of any socket address on
the stack. Let's make sure that an AF_ALG address always fits. Coverity
complains because in theory, AF_ALG might be larger than any of the other
structs. In practice it already fits.
Closes Coverity CID 1398948, 1398949, 1398950
Signed-off-by: Christian Heimes <christian@python.org>
* Fixed bpo-29565: Corrected ctypes passing of large structs by value.
Added code and test to check that when a structure passed by value
is large enough to need to be passed by reference, a copy of the
original structure is passed. The callee updates the passed-in value,
and the test verifies that the caller's copy is unchanged. A similar
change was also added to the test added for bpo-20160 (that test was
passing, but the changes should guard against regressions).
* Reverted unintended whitespace changes.
bltinmodule.c: Added in b744ba1 and no longer necessary since d64e8a7
posixmodule.c: Added in d1cd4d4 and no longer necessary since efb00c0
pythonrun.c: Added in 73d538b and no longer necessary since d600951
sysmodule.c: Added in 5467d4c and no longer necessary since a2c17c5
Issue #29452: Use METH_FASTCALL calling convention for index(), insert() and
rotate() methods of collections.deque to avoid the creation a temporary tuple
to pass position arguments. Speedup on deque methods:
* d.rotate(): 1.10x faster
* d.rotate(1): 1.24x faster
* d.insert(): 1.18x faster
* d.index(): 1.24x faster
Issue #29300: Rename struct.unpack() second parameter from "inputstr" to
"buffer", and use the Py_buffer type.
Fix also unit tests on struct.unpack() which passed a Unicode string instead of
a bytes string as struct.unpack() second parameter. The purpose of
test_trailing_counter() is to test invalid format strings, not to test the
buffer parameter.
* The struct module now requires contiguous buffers.
* Convert most functions and methods of the _struct module to Argument Clinic
* Use "Py_buffer" type for the "buffer" argument. Argument Clinic is
responsible to create and release the Py_buffer object.
* Use "PyStructObject *" type for self to avoid explicit conversions.
* Add an unit test on the _struct.Struct.unpack_from() method to test passing
arguments as keywords.
* Rephrase docstrings.
* Rename "fmt" argument to "format" in docstrings and the documentation.
As a side effect, functions and methods which used METH_VARARGS calling
convention like struct.pack() now use the METH_FASTCALL calling convention
which avoids the creation of temporary tuple to pass positional arguments and
so is faster. For example, struct.pack("i", 1) becomes 1.56x faster (-36%)::
$ ./python -m perf timeit \
-s 'import struct; pack=struct.pack' 'pack("i", 1)' \
--compare-to=../default-ref/python
Median +- std dev: 119 ns +- 1 ns -> 76.8 ns +- 0.4 ns: 1.56x faster (-36%)
Significant (t=295.91)
Patch co-written with Serhiy Storchaka.
Issue #29286. Run Argument Clinic to get the new faster METH_FASTCALL calling
convention for functions using "boring" positional arguments.
Manually fix _elementtree: _elementtree_XMLParser_doctype() must remain
consistent with the clinic code.
* Indent versionchanged at method level, not class level
* Mark up ``--help`` to avoid generating an en dash
* Use forward slash in Unix command line with a dollar sign ($) prompt
Fix time_hash() function: replace DATE_xxx() macros with TIME_xxx() macros.
Before, the hash function used a wrong value for microseconds if fold is set
(equal to 1).
Issue #28920: Replace _PyObject_CallMethodId(obj, meth, "O", arg) with
_PyObject_CallMethodIdObjArgs(obj, meth, arg, NULL) to avoid
_PyObject_CallMethodId() special case when arg is a tuple.
If arg is a tuple, _PyObject_CallMethodId() unpacks the tuple: obj.meth(*arg).
Issue #28915: Replace PyObject_CallFunction() with
PyObject_CallFunctionObjArgs() when the format string was only made of "O"
formats, PyObject* arguments.
PyObject_CallFunctionObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() in various modules when the format string was
only made of "O" formats, PyObject* arguments.
_PyObject_CallMethodIdObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() when the format string was only made of "O"
formats, PyObject* arguments.
_PyObject_CallMethodIdObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() when the format string was only made of "O"
formats, PyObject* arguments.
_PyObject_CallMethodIdObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() when the format string was only made of "O"
formats, PyObject* arguments.
_PyObject_CallMethodIdObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() in unpickle(). _PyObject_CallMethodIdObjArgs()
avoids the creation of a temporary tuple and doesn't have to parse a format
string.
Replace _PyObject_CallMethodId() with _PyObject_GetAttrId()+PyObject_Call() for
the second call since it requires to "unpack" a tuple.
Add also a check in the type of the second parameter (state): it must be a
tuple.
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() when the format string was only made of "O"
formats, PyObject* arguments.
_PyObject_CallMethodIdObjArgs() avoids the creation of a temporary tuple and
doesn't have to parse a format string.
Issue #28915: Use PyObject_Call() to pass a tuple of positional arguments,
instead of relying on _PyObject_CallMethodId() weird behaviour to unpack the
tuple.
Issue #28915: Avoid calling _PyObject_CallMethodId() with "(...)" format to
avoid the creation of a temporary tuple: use Py_BuildValue() with
_PyObject_CallMethodIdObjArgs().
Issue #28915: Replace _PyObject_CallMethodId() with
_PyObject_CallMethodIdObjArgs() when the format string only use the format 'O'
for objects, like "(O)".
_PyObject_CallMethodIdObjArgs() avoids the code to parse a format string and
avoids the creation of a temporary tuple.
Replace
_PyObject_CallArg1(func, arg)
with
PyObject_CallFunctionObjArgs(func, arg, NULL)
Using the _PyObject_CallArg1() macro increases the usage of the C stack, which
was unexpected and unwanted. PyObject_CallFunctionObjArgs() doesn't have this
issue.
Issue #28858: The change b9c9691c72c5 introduced a regression. It seems like
_PyObject_CallArg1() uses more stack memory than
PyObject_CallFunctionObjArgs().
Replace
PyObject_CallFunction(func, "O", arg)
and
PyObject_CallFunction(func, "O", arg, NULL)
with
_PyObject_CallArg1(func, arg)
Replace
PyObject_CallFunction(func, NULL)
with
_PyObject_CallNoArg(func)
_PyObject_CallNoArg() and _PyObject_CallArg1() are simpler and don't allocate
memory on the C stack.
* 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.
Remove aliases from the C module. Always implement bisect() and insort()
aliases in bisect.py
Remove also the "# backward compatibility" command, there is no plan to
deprecate nor remove these aliases. When keys are equal, it makes sense to use
bisect.bisect() and bisect.insort().
Issue #28727: Regular expression patterns, _sre.SRE_Pattern objects created by
re.compile(), become comparable (only x==y and x!=y operators). This change
should fix the issue #18383: don't duplicate warning filters when the warnings
module is reloaded (thing usually only done in unit tests).
Nathaniel J. Smith
Xiang Zhang
INADA Naoki
Serhiy Storchaka
Antoine Pitrou
Victor Stinner
Eijebong
bladebryan
Benjamin Peterson
Olivier Vielpeau
Mark Dickinson
Andrew Nester
Angus Hollands
Louie Lu
T. Wouters
Sylvain
Christophe Zeitouny
Michael Seifert
Oren Milman
Nick Coghlan
Ned Deily
Aviv Palivoda
Yury Selivanov
orenmn
Donald Stufft
Christian Heimes
Berker Peksag
Louie Lu
Vinay Sajip
Yen Chi Hsuan
Hiroki Noda
Maximilian Hils
Steve Dower
doko@ubuntu.com
Martin Panter
Mark Dickinson
Gregory P. Smith
Raymond Hettinger
Stefan Krah
Antoine Pitrou
INADA Naoki
Yury Selivanov