The concrete PyDict_* API is used to interact with PyInterpreterState.modules in a number of places. This isn't compatible with all dict subclasses, nor with other Mapping implementations. This patch switches the concrete API usage to the corresponding abstract API calls.
We also add a PyImport_GetModule() function (and some other helpers) to reduce a bunch of code duplication.
* Add Py_UNREACHABLE() as an alias to abort().
* Use Py_UNREACHABLE() instead of assert(0)
* Convert more unreachable code to use Py_UNREACHABLE()
* Document Py_UNREACHABLE() and a few other macros.
A bunch of code currently uses PyInterpreterState.modules directly instead of PyImport_GetModuleDict(). This complicates efforts to make changes relative to sys.modules. This patch switches to using PyImport_GetModuleDict() uniformly. Also, a number of related uses of sys.modules are updated for uniformity for the same reason.
Note that this code was already reviewed and merged as part of #1638. I reverted that and am now splitting it up into more focused parts.
PR #1638, for bpo-28411, causes problems in some (very) edge cases. Until that gets sorted out, we're reverting the merge. PR #3506, a fix on top of #1638, is also getting reverted.
* Drop warnoptions from PyInterpreterState.
* Drop xoptions from PyInterpreterState.
* Don't set warnoptions and _xoptions again.
* Decref after adding to sys.__dict__.
* Drop an unused macro.
* Check sys.xoptions *before* we delete it.
This undoes a853a8ba78 except for the pytime.c
parts. We want to continue to allow IEEE 754 doubles larger than FLT_MAX to be
rounded into finite floats. Tests were added to very this behavior.
* 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).
f_trace_lines: enable/disable line trace events
f_trace_opcodes: enable/disable opcode trace events
These are intended primarily for testing of the interpreter
itself, as they make it much easier to emulate signals
arriving at unfortunate times.
* 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).
According to the comment, there was previously a call to PyObject_IsSubclass() involved which could fail, but since it was replaced with a call to PyType_IsSubtype(), it can no longer fail.
Use sys.modules.get() in the "with _ModuleLockManager(name):" block
to protect the dictionary key with the module lock and use an atomic
get to prevent race condition.
Remove also _bootstrap._POPULATE since it was unused
(_bootstrap_external now has its own _POPULATE object), add a new
_SENTINEL object instead.
* Rewrite importlib _get_module_lock(): it is now responsible to hold
the imp lock directly.
* _find_and_load() now holds the module lock to check if name is in
sys.modules to prevent a race condition
* bpo-30832: Remove own implementation for thread-local storage
CPython has provided the own implementation for thread-local storage
(TLS) on Python/thread.c, it's used in the case which a platform has
not supplied native TLS. However, currently all supported platforms
(NT and pthreads) have provided native TLS and defined the
Py_HAVE_NATIVE_TLS macro with unconditional in any case.
* bpo-30832: replace NT with Windows
* bpo-30832: change to directive chain
* bpo-30832: remove comemnt which making no sense
- On some versions of FreeBSD, setting the "UTF-8" locale
succeeds, but a subsequent "nl_langinfo(CODESET)" fails
- adding a check for this in the coercion logic means that
coercion will happen on systems where this check succeeds,
and will be skipped otherwise
- that way CPython should automatically adapt to changes in
platform behaviour, rather than needing a new release to
enable coercion at build time
- this also allows UTF-8 to be re-enabled as a coercion
target, restoring the locale coercion behaviour on Mac OS X
`PYTHONFRAMEWORK` is defined in `Makefile` and it shoulnd't be used
in `pyconfig.h`.
`sysconfig.py --generate-posix-vars` reads config vars from Makefile
and `pyconfig.h`. Conflicting variables should be avoided.
Especially, string config variables in Makefile are unquoted, but
in `pyconfig.h` are keep quoted. So it should be private (starts with
underscore).
* Improve signal delivery
Avoid using Py_AddPendingCall from signal handler, to avoid calling signal-unsafe functions.
* Remove unused function
* Improve comments
* Add stress test
* Adapt for --without-threads
* Add second stress test
* Add NEWS blurb
* Address comments @haypo
Based on patch by Victor Stinner.
Add private C API function _PyUnicode_AsUnicode() which is similar to
PyUnicode_AsUnicode(), but checks for null characters.
* bpo-30765: Avoid blocking when PyThread_acquire_lock() is asked not to lock
This is especially important if PyThread_acquire_lock() is called reentrantly
(for example from a signal handler).
* Update 2017-06-26-14-29-50.bpo-30765.Q5iBmf.rst
* Avoid core logic when taking the mutex failed
* bpo-30183: Fixes HP-UX cc compilation error in pytime.c
HP-UX does not support the CLOCK_MONOTONIC identifier, and will fail to
compile:
"Python/pytime.c", line 723: error #2020: identifier
"CLOCK_MONOTONIC" is undefined
const clockid_t clk_id = CLOCK_MONOTONIC;
Add a new section for __hpux that calls 'gethrtime()' instead of
'clock_gettime()'.
* bpo-30183: Removes unnecessary return
- removes PY_WARN_ON_C_LOCALE build time flag
- locale coercion and compatibility warnings are now always compiled
in, but are off by default
- adds PYTHONCOERCECLOCALE=warn runtime option to aid in
debugging potentially locale related compatibility problems
Due to not-yet-resolved test failures on *BSD systems (including
Mac OS X), this also temporarily disables UTF-8 as a locale coercion
target, and skips testing the interpreter's behavior in the POSIX locale.
- new PYTHONCOERCECLOCALE config setting
- coerces legacy C locale to C.UTF-8, C.utf8 or UTF-8 by default
- always uses C.UTF-8 on Android
- uses `surrogateescape` on stdin and stdout in the coercion
target locales
- configure option to disable locale coercion at build time
- configure option to disable C locale warning at build time
The function '_PyArg_ParseStack()' and
'_PyArg_UnpackStack' were failing (with error
"XXX() takes Y argument (Z given)") before
the function '_PyArg_NoStackKeywords()' was called.
Thus, the latter did not raise its more meaningful
error : "XXX() takes no keyword arguments".
'invalid character in identifier' now is raised instead of
'f-string: empty expression not allowed' if a subexpression contains
only whitespaces and they are not accepted by Python parser.
Error messages when pass keyword arguments to some builtins that
don't support keyword arguments contained double parenthesis: "()()".
The regression was introduced by bpo-30534.
Fix a reference in subinterpreters, like test_callbacks_leak() of
test_atexit.
warnoptions is a list used to pass options from the command line to
the sys module constructor. Before this change, the list was shared
by multiple interpreter which is not the expected behaviour. Each
interpreter should have their own independent mutable world.
This change duplicates the list in each interpreter. So each
interpreter owns its own list, so each interpreter can clear its own
list.
* bpo-16500: Allow registering at-fork handlers
* Address Serhiy's comments
* Add doc for new C API
* Add doc for new Python-facing function
* Add NEWS entry + doc nit
* Improves test_underpth_nosite_file to reveal why it fails.
* Enable building with Windows 10 SDK.
* Fix WinSDK detection
* Fix initialization on Windows when a ._pth file exists.
* Fix tabs
* Adds comment about Py_GetPath call.
PEP 432 specifies a number of large changes to interpreter startup code, including exposing a cleaner C-API. The major changes depend on a number of smaller changes. This patch includes all those smaller changes.
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.
Python/thread_foobar.h is the template code that is threading adaptation
for new platforms, also hasn't been used on actual platforms.
Python/thread_*.h give concrete examples of adaptation instead of the
template code.
is_valid_fd() now uses fstat() instead of dup() on macOS to return 0
on a pipe when the other side of the pipe is closed. fstat() fails
with EBADF in that case, whereas dup() succeed.
* 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().
When LOAD_METHOD is used for calling C mehtod, PyMethodDescrObject
was passed to profilefunc from 5566bbb.
But lsprof traces only PyCFunctionObject. Additionally, there can be
some third party extension which assumes passed arg is
PyCFunctionObject without calling PyCFunction_Check().
So make PyCFunctionObject from PyMethodDescrObject when
tstate->c_profilefunc is set.
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.
bpo-29463 added optional "docstring" field to 4 AST types.
While it is optional, it breaks backward compatibility because AST constructor
requires number of positional argument is same to number of fields.
AST types accepts empty arguments, and incomplete keyword arguments.
But it's not big problem because field can be filled after creation, and checked when compiling.
So stop requiring complete set of fields for positional arguments too.
When you use `'%s' % SubClassOfStr()`, where `SubClassOfStr.__rmod__` exists, the reverse operation is ignored as normally such string formatting operations use the `PyUnicode_Format()` fast path. This patch tests for subclasses of `str` first and picks the slow path in that case.
Patch by Martijn Pieters.
* bpo-29463: Add docstring field to some AST nodes.
ClassDef, ModuleDef, FunctionDef, and AsyncFunctionDef has docstring
field for now. It was first statement of there body.
* fix document. thanks travis!
* doc fixes
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
* Move all functions to call objects in a new Objects/call.c file.
* Rename fast_function() to _PyFunction_FastCallKeywords().
* Copy null_error() from Objects/abstract.c
* Inline type_error() in call.c to not have to copy it, it was only
called once.
* Export _PyEval_EvalCodeWithName() since it is now called
from call.c.
* Move all functions to call objects in a new Objects/call.c file.
* Rename fast_function() to _PyFunction_FastCallKeywords().
* Copy null_error() from Objects/abstract.c
* Inline type_error() in call.c to not have to copy it, it was only
called once.
* Export _PyEval_EvalCodeWithName() since it is now called
from call.c.
* Replace PyArg_ParseTupleAndKeywords() with _PyArg_ParseStackAndKeywords()
which is more efficient to parse keywords, since it decodes only keywords
(char*) from UTF-8 once, instead of decoding at each call.
* METH_FASTCALL avoids the creation of a temporary tuple to pass positional
arguments.
Patch written by INADA Naoki, pushed by Victor Stinner.
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.
Issue #29227: Inline call_function() into _PyEval_EvalFrameDefault() using
Py_LOCAL_INLINE to reduce the stack consumption.
It reduces the stack consumption, bytes per call, before => after:
test_python_call: 1152 => 1040 (-112 B)
test_python_getitem: 1008 => 976 (-32 B)
test_python_iterator: 1232 => 1120 (-112 B)
=> total: 3392 => 3136 (- 256 B)
Copy and then adapt Python/random.c from default branch. Difference between 3.5
and default branches:
* Python 3.5 only uses getrandom() in non-blocking mode: flags=GRND_NONBLOCK
* If getrandom() fails with EAGAIN: py_getrandom() immediately fails and
remembers that getrandom() doesn't work.
* Python 3.5 has no _PyOS_URandomNonblock() function: _PyOS_URandom()
works in non-blocking mode on Python 3.5
* dev_urandom() now calls py_getentropy(). Prepare the fallback to support
getentropy() failure and falls back on reading from /dev/urandom.
* Simplify dev_urandom(). pyurandom() is now responsible to call getentropy()
or getrandom(). Enhance also dev_urandom() and pyurandom() documentation.
* getrandom() is now preferred over getentropy(). The glibc 2.24 now implements
getentropy() on Linux using the getrandom() syscall. But getentropy()
doesn't support non-blocking mode. Since getrandom() is tried first, it's not
more needed to explicitly exclude getentropy() on Solaris. Replace:
"if defined(HAVE_GETENTROPY) && !defined(sun)"
with "if defined(HAVE_GETENTROPY)"
* Enhance py_getrandom() documentation. py_getentropy() now supports ENOSYS,
EPERM & EINTR
The glibc now implements getentropy() on Linux using the getrandom() syscall.
But getentropy() doesn't support non-blocking mode.
Since getrandom() is tried first, it's not more needed to explicitly exclude
getentropy() on Solaris. Replace:
if defined(HAVE_GETENTROPY) && !defined(sun)
with
if defined(HAVE_GETENTROPY)
Issue #28870: Add a new _PY_FASTCALL_SMALL_STACK constant, size of "small
stacks" allocated on the C stack to pass positional arguments to
_PyObject_FastCall().
_PyObject_Call_Prepend() now uses a small stack of 5 arguments (40 bytes)
instead of 8 (64 bytes), since it is modified to use _PY_FASTCALL_SMALL_STACK.
Special thanks to INADA Naoki for pushing the patch through
the last mile, Serhiy Storchaka for reviewing the code, and to
Victor Stinner for suggesting the idea (originally implemented
in the PyPy project).
The PEP 523 modified PyEval_EvalFrameEx(): it's now an indirection to
interp->eval_frame().
Inline the call in performance critical code. Leave PyEval_EvalFrame()
unchanged, this function is only kept for backward compatibility.
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 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.
Issue #28915: Py_ssize_t type is better for indexes. The compiler might emit
more efficient code for i++. Py_ssize_t is the type of a PyTuple index for
example.
Replace also "int endchar" with "char endchar".
Issue #28838: Rename parameters of the "calls" functions of the Python C API.
* Rename 'callable_object' and 'func' to 'callable': any Python callable object
is accepted, not only Python functions
* Rename 'method' and 'nameid' to 'name' (method name)
* Rename 'o' to 'obj'
* Move, fix and update documentation of PyObject_CallXXX() functions
in abstract.h
* Update also the documentaton of the C API (update parameter names)
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.
Handling zero-argument super() in __init_subclass__ and
__set_name__ involved moving __class__ initialisation to
type.__new__. This requires cooperation from custom
metaclasses to ensure that the new __classcell__ entry
is passed along appropriately.
The initial implementation of that change resulted in abruptly
broken zero-argument super() support in metaclasses that didn't
adhere to the new requirements (such as Django's metaclass for
Model definitions).
The updated approach adopted here instead emits a deprecation
warning for those cases, and makes them work the same way they
did in Python 3.5.
This patch also improves the related class machinery documentation
to cover these details and to include more reader-friendly
cross-references and index entries.
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.
Issue #28799:
* Remove the PyEval_GetCallStats() function.
* Deprecate the untested and undocumented sys.callstats() function.
* Remove the CALL_PROFILE special build
Use the sys.setprofile() function, cProfile or profile module to profile
function calls.
Issue #28782: Fix a bug in the implementation ``yield from`` when checking
if the next instruction is YIELD_FROM. Regression introduced by WORDCODE
(issue #26647).
Reviewed by Serhiy Storchaka and Yury Selivanov.
Issue #28691: Fix warn_invalid_escape_sequence(): handle correctly
DeprecationWarning raised as an exception. First clear the current exception to
replace the DeprecationWarning exception with a SyntaxError exception.
Unit test written by Serhiy Storchaka.
When Python is not compiled with PGO, the performance of Python on call_simple
and call_method microbenchmarks depend highly on the code placement. In the
worst case, the performance slowdown can be up to 70%.
The GCC __attribute__((hot)) attribute helps to keep hot code close to reduce
the risk of such major slowdown. This attribute is ignored when Python is
compiled with PGO.
The following functions are considered as hot according to statistics collected
by perf record/perf report:
* _PyEval_EvalFrameDefault()
* call_function()
* _PyFunction_FastCall()
* PyFrame_New()
* frame_dealloc()
* PyErr_Occurred()
Antoine Pitrou
Eric Snow
Barry Warsaw
Oren Milman
Benjamin Peterson
Han Lee
Serhiy Storchaka
Nick Coghlan
Łukasz Langa
INADA Naoki
Stefan Krah
Yuan Chao Chou
Ville Skyttä
scoder
Zackery Spytz
Segev Finer
Stéphane Wirtel
Victor Stinner
Masayuki Yamamoto
Adrian Wielgosik
Dino Viehland
haney
ericvsmith
Sylvain
Amit Kumar
Steve Dower
Zachary Ware
Jason Fried
Nathaniel J. Smith
Xiang Zhang
Michael Seifert
Daniel Birnstiel
Ned Deily
Yury Selivanov
Matthias Bussonnier
Martijn Pieters
Matthieu Dartiailh
Yen Chi Hsuan
INADA Naoki
Martin Panter
Xavier de Gaye
Yury Selivanov
Raymond Hettinger