summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* [0.29] implement the --depfile command-line option for the "cython" tool ↵releaseEli Schwartz2022-08-056-21/+97
| | | | | (GH-4949) Backports https://github.com/cython/cython/pull/4916
* Hide a C compiler "unused argument" warning in Py3.11 where the "fast thread ↵Stefan Behnel2022-08-041-1/+1
| | | | | | state" usage is disabled. Closes https://github.com/cython/cython/issues/4948
* Prepare release of 0.29.32.0.29.32Stefan Behnel2022-07-292-1/+14
|
* Reallow capture of memoryview arguments (GH-4929)da-woods2022-07-292-8/+19
| | | | | | | | | | | It was actually OK in def functions. It only looks very dodgy: ``` __Pyx_XDEC_MEMVIEW(closure->arg) ``` This gets called twice and `INC` gets called once. However this is actually OK since XDEC really means "clear" Fixes https://github.com/cython/cython/issues/4798 for 0.29.x (completely I think)
* Fix error where "import *" tried to overwrite a macro in utility code (GH-4930)da-woods2022-07-283-2/+11
| | | Closes https://github.com/cython/cython/issues/4927
* Fix release date.Stefan Behnel2022-07-271-1/+1
|
* Prepare release of 0.29.31.0.29.31Stefan Behnel2022-07-272-2/+9
|
* [0.29] Add --module-name argument to cython command (GH-4906)h-vetinari2022-07-274-14/+138
| | | | | | | | | | Backport of https://github.com/cython/cython/pull/4548 It can be useful to specify the module name for the output file directly, rather than working it out from the enclosing file tree - particularly for out of tree build systems, like Meson. See background in https://github.com/rgommers/scipy/issues/31#issuecomment-1002662816
* Update changelog.Stefan Behnel2022-07-271-25/+40
|
* Backport "noexcept" function modifier to Cython 0.29.x (GH-4903)da-woods2022-07-273-1/+26
| | | | | As a no-op, but it parses fine. Also add some basic compile tests, and some brief documentation.
* [0.29] Add configuration for the "nogil" CPython fork (GH-4912) (GH-4914)Sam Gross2022-07-262-0/+52
|
* [0.29] Use atomic reference counting in MemoryView in more cases (GH-4912) ↵Sam Gross2022-07-263-32/+27
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | (GH-4915) This fixes a few issues in MemoryView_C.c to allow atomic reference counting to be used in more cases. - Enable GNU atomics for `__GNUC__` >= 5. Previously, GCC 5.0, 6.0, X.0 versions used lock-based reference counting due to an incorrect preprocessor check. - Typo in `__GNUC_PATCHLEVEL__` macro (missing underscores) - Enable atomics in MSVC and fix returned values. InterlockedExchangeAdd returns the *initial* value (like __sync_fetch_and_add). InterlockedIncrement returned the *resulting* value (post increment), which would have been incorrect if MSVC atomics had been enabled. Also avoids allocating a lock in MemoryView when atomics are available, which additionally fixes a thread-safety issue in the "nogil" CPython fork. * Use _InterlockedExchangeAdd intrinsic The InterlockedExchangeSubtract function isn't available in older versions of MSVC, while InterlockedExchangeAdd is available since Windows XP. The intrinsic variant (with the underscore prefix) avoids needing to include the entire Windows.h header. * Only use MSVC atomics when compiling for the "nogil" CPython fork to prevent potential breakage of existing Windows setups.
* Update macos version for github actions (#4917)da-woods2022-07-251-10/+10
| | | See https://github.com/cython/cython/issues/4913
* Minor code cleanups after 3a373e2ccdddc79202d9ed13edc85d4b95616b26.Stefan Behnel2022-07-201-3/+5
|
* Make pickle checksum calculation succeed even if one of the hash algorithms ↵Stefan Behnel2022-07-202-19/+29
| | | | | | | | is blocked at runtime. Also, pass "usedforsecurity=False" in Py3.9+ to unblock MD5 also on some security constrained systems (FIPS). Closes https://github.com/cython/cython/issues/4909
* CI: Remove "allow_failures" for Python 3.11 to make sure we notice failures ↵Ewout ter Hoeven2022-07-181-3/+3
| | | | during the release phases (GH-4780)
* Fixed over-zealous optimization of append attribute usage to ↵da-woods2022-07-163-2/+36
| | | | | "__Pyx_PyObject_Append" (GH-4834) Fixes https://github.com/cython/cython/issues/4828
* Error on memoryview argument capture on 0.29.x (GH-4849)da-woods2022-07-122-2/+33
| | | | | | | I don't believe it's easy to fix https://github.com/cython/cython/issues/4798 on 0.29.x Therefore, generate an error message that explains two possible workarounds. This at least makes sure that people don't end up with mysterious crashes.
* Add tests for NULL objects in memoryviews (GH-4871)da-woods2022-07-062-11/+49
| | | | | Follow up on https://github.com/cython/cython/pull/4859 by adding tests for memoryviews too. Additional refactoring to avoid invalid decref calls on test failures. Instead, the item is safely cleared directly before the access.
* Revert "includes/cpython: Fix newfunc to use PyObject* for args/kwargs ↵Stefan Behnel2022-07-041-1/+1
| | | | | | instead of object (#4823)" This reverts commit 7c7890348625871e6442b4b1bbd6e3e0e7c38e73.
* includes/cpython: Fix newfunc to use PyObject* for args/kwargs instead of ↵Kirill Smelkov2022-07-031-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | object (#4823) object means the argument is always non-NULL valid Python object, while PyObject* argument can be generally NULL. If the argument is indeed passed as NULL, and we declare it as object, generated code will crash while trying to incref it. Quoting https://github.com/cython/cython/issues/4822: object.pxd currently declares `newfunc` as follows: ```pyx ctypedef object (*newfunc)(cpython.type.type, object, object) # (type, args, kwargs) ``` which implies that `args` and `kwargs` are always live objects and cannot be NULL. However Python can, and does, call tp_new with either args=NULL, or kwargs=NULL or both. And in such cases this leads to segfault in automatically-generated __Pyx_INCREF for args or kw. The fix is to change `object` to `PyObject*` for both args and kwargs. Please see below for details: ```cython # cython: language_level=3 from cpython cimport newfunc, type as cpytype, Py_TYPE cdef class X: cdef int i def __init__(self, i): self.i = i def __repr__(self): return 'X(%d)' % self.i cdef newfunc _orig_tp_new = Py_TYPE(X(0)).tp_new cdef object _trace_tp_new(cpytype cls, object args, object kw): print('_trace_tp_new', cls, args, kw) return _orig_tp_new(cls, args, kw) Py_TYPE(X(0)).tp_new = _trace_tp_new x = X(123) print(x) ``` ```console (neo) (py3.venv) (g.env) kirr@deca:~/src/tools/go/pygolang$ cythonize -i x.pyx Compiling /home/kirr/src/tools/go/pygolang/x.pyx because it changed. [1/1] Cythonizing /home/kirr/src/tools/go/pygolang/x.pyx running build_ext building 'x' extension ... x86_64-linux-gnu-gcc -pthread -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O2 -Wall -g -ffile-prefix-map=/build/python3.9-RNBry6/python3.9-3.9.2=. -fstack-protector-strong -Wformat -Werror=format-security -g -fwrapv -O2 -g -ffile-prefix-map=/build/python3.9-RNBry6/python3.9-3.9.2=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -fPIC -I/home/kirr/src/wendelin/venv/py3.venv/include -I/usr/include/python3.9 -c /home/kirr/src/tools/go/pygolang/x.c -o /home/kirr/src/tools/go/pygolang/tmpqkz1r96s/home/kirr/src/tools/go/pygolang/x.o x86_64-linux-gnu-gcc -pthread -shared -Wl,-O1 -Wl,-Bsymbolic-functions -Wl,-z,relro -g -fwrapv -O2 -Wl,-z,relro -g -fwrapv -O2 -g -ffile-prefix-map=/build/python3.9-RNBry6/python3.9-3.9.2=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 /home/kirr/src/tools/go/pygolang/tmpqkz1r96s/home/kirr/src/tools/go/pygolang/x.o -o /home/kirr/src/tools/go/pygolang/x.cpython-39-x86_64-linux-gnu.so ``` ```console (neo) (py3.venv) (g.env) kirr@deca:~/src/tools/go/pygolang$ python -c 'import x' Ошибка сегментирования (стек памяти сброшен на диск) ``` ```console (neo) (py3.venv) (g.env) kirr@deca:~/src/tools/go/pygolang$ gdb python core ... Reading symbols from python... Reading symbols from /usr/lib/debug/.build-id/f9/02f8a561c3abdb9c8d8c859d4243bd8c3f928f.debug... [New LWP 218557] [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". Core was generated by `python -c import x'. Program terminated with signal SIGSEGV, Segmentation fault. #0 _Py_INCREF (op=0x0) at /usr/include/python3.9/object.h:408 408 op->ob_refcnt++; (gdb) bt 5 #0 _Py_INCREF (op=0x0) at /usr/include/python3.9/object.h:408 #1 __pyx_f_1x__trace_tp_new (__pyx_v_cls=0x7f5ce75e6880 <__pyx_type_1x_X>, __pyx_v_args=(123,), __pyx_v_kw=0x0) at /home/kirr/src/tools/go/pygolang/x.c:1986 #2 0x000000000051dd7e in type_call (type=type@entry=0x7f5ce75e6880 <__pyx_type_1x_X>, args=args@entry=(123,), kwds=kwds@entry=0x0) at ../Objects/typeobject.c:1014 #3 0x00007f5ce75df8d4 in __Pyx_PyObject_Call (func=<type at remote 0x7f5ce75e6880>, arg=(123,), kw=0x0) at /home/kirr/src/tools/go/pygolang/x.c:3414 #4 0x00007f5ce75df276 in __pyx_pymod_exec_x (__pyx_pyinit_module=<optimized out>) at /home/kirr/src/tools/go/pygolang/x.c:3017 (More stack frames follow...) (gdb) f 1 #1 __pyx_f_1x__trace_tp_new (__pyx_v_cls=0x7f5ce75e6880 <__pyx_type_1x_X>, __pyx_v_args=(123,), __pyx_v_kw=0x0) at /home/kirr/src/tools/go/pygolang/x.c:1986 1986 __Pyx_INCREF(__pyx_v_kw); ``` -> Change newfunc signature to use PyObject* instead of object to fix it. With this fix, and test example updates to account for object -> PyObject* change as follows ... --- a/x.pyx.kirr +++ b/x.pyx @@ -1,5 +1,5 @@ # cython: language_level=3 -from cpython cimport newfunc, type as cpytype, Py_TYPE +from cpython cimport newfunc, type as cpytype, Py_TYPE, PyObject cdef class X: cdef int i @@ -10,8 +10,12 @@ cdef class X: cdef newfunc _orig_tp_new = Py_TYPE(X(0)).tp_new -cdef object _trace_tp_new(cpytype cls, object args, object kw): - print('_trace_tp_new', cls, args, kw) +cdef object xobject(PyObject* x): + return "null" if x == NULL else \ + <object>x + +cdef object _trace_tp_new(cpytype cls, PyObject* args, PyObject* kw): + print('_trace_tp_new', cls, xobject(args), xobject(kw)) return _orig_tp_new(cls, args, kw) Py_TYPE(X(0)).tp_new = _trace_tp_new ... it works as expected without crashing: $ python -c 'import x' _trace_tp_new <type 'x.X'> (123,) null X(123) Fixes: https://github.com/cython/cython/issues/4822
* Fix tuple*float test on PyPyda-woods2022-07-031-2/+2
| | | | | Test added in 5c900c59d03f23f7329d6e68e114e4a277112916 PyPy gives a slightly different error message for the unsupported operation
* Updated changelogda-woods2022-07-031-0/+34
|
* BUG: Fortify object buffers against included NULLs (#4859)Sebastian Berg2022-07-032-7/+55
| | | | | | | | | | | | | | * BUG: Fortify object buffers against included NULLs While NumPy tends to not actively create object buffers initialized only with NULL (rather than filled with None), at least older versions of NumPy did do that. And NumPy guards against this. This guards against embedded NULLs in object buffers interpreting a NULL as None (and anticipating a NULL value also when setting the buffer for reference count purposes). Closes gh-4858
* Fix tuple multiplication in MergedSequenceNode (GH-4864)da-woods2022-06-282-1/+19
| | | Fixes https://github.com/cython/cython/issues/4861
* Docs: don't say cdef functions exist in module dict (#4865)da-woods2022-06-261-7/+0
| | | | Patch is against 0.29.x branch (to fix both versions of the documentation).
* Fix GCC -Wconversion warning in C utility code (GH-4854)Lisandro Dalcin2022-06-211-1/+1
|
* Fix bytearray iteration in 0.29.x (#4108)Kenrick Everett2022-06-212-0/+19
| | | By explicitly setting the result type
* BUG: fused types not subscriptable in Cython.Shadow (#4842)pfebrer2022-06-171-1/+1
|
* MNT: always require va_start to have two arguments (#4820)Thomas A Caswell2022-06-071-1/+1
| | | | | | | | | | | | | | * MNT: always require va_start to have two arguments https://github.com/python/cpython/pull/93215 chance CPython to always use the 2-input version of va_start and dropped defining HAVE_STDARG_PROTOTYPES. This resulted in the 1-argument version being used when compiling cython source which fails This makes cython also always use the 2-argument version. * Remove blank line * FIX: version gate 2-argument va_start checking to py311
* Use encoded string since `ModuleNode.full_module_name` may originate from ↵Stefan Behnel2022-05-171-1/+1
| | | | | | different sources. See https://github.com/cython/cython/pull/4764
* Merge branch '0.29.29-reg' into 0.29.xStefan Behnel2022-05-172-1/+12
|\
| * Move from deprecated find_loader() to find_spec() for setting up the import ↵Matti Picus2022-05-172-32/+14
| | | | | | | | | | time package path (GH-4764) Closes https://github.com/cython/cython/issues/4763
| * Prepare release of 0.29.30.0.29.30Stefan Behnel2022-05-172-1/+12
| |
| * Revert "Avoid acquiring the GIL at the end of nogil functions (GH-3556) ↵Stefan Behnel2022-05-175-81/+35
| | | | | | | | | | | | (GH-4749)" This reverts commit 74073417c4f7e616af0a139a55f790f1ec7fe728.
| * typoStefan Behnel2022-05-171-1/+1
| |
* | Revert "Avoid acquiring the GIL at the end of nogil functions (GH-3556) ↵Stefan Behnel2022-05-175-81/+35
| | | | | | | | | | | | (GH-4749)" This reverts commit 74073417c4f7e616af0a139a55f790f1ec7fe728.
* | Move from deprecated find_loader() to find_spec() for setting up the import ↵Matti Picus2022-05-172-32/+14
| | | | | | | | | | time package path (GH-4764) Closes https://github.com/cython/cython/issues/4763
* | typoStefan Behnel2022-05-161-1/+1
|/
* Prepare release of 0.29.29.0.29.29Stefan Behnel2022-05-162-2/+2
|
* Update changelog.Stefan Behnel2022-05-161-0/+8
|
* Remove ".__contains__" -> "PySequence_Contains" slot mappings for known ↵da-woods2022-05-162-10/+52
| | | | | | | builtin types (GH-4792) They prevent explicitly calling the base-class __contains__. Closes https://github.com/cython/cython/issues/4785
* Update changelog.Stefan Behnel2022-05-161-1/+1
|
* Fix version detection and compile/runtime Python version warning with ↵da-woods2022-05-161-4/+30
| | | | | double-digit versions such as Py3.10 (GH-4777) See https://github.com/cython/cython/issues/2542#issuecomment-1121484721
* Change IndexError to TypeError when calling unbound CyFunctions (GH-4783)da-woods2022-05-162-0/+9
|
* Fix credited author name in changelog.Stefan Behnel2022-05-131-1/+1
|
* Update changelog.Stefan Behnel2022-05-041-0/+12
|
* Adapt doctest to new exception messages in Py3.11 (says "modulo" instead of ↵Stefan Behnel2022-05-031-2/+2
| | | | "division").
* Avoid acquiring the GIL at the end of nogil functions (GH-3556) (GH-4749)Oleksandr Pavlyk2022-05-035-35/+81
| | | | | | | | | | | | Closes https://github.com/cython/cython/issues/4637 See See https://github.com/cython/cython/issues/3556 * Acquire the GIL in nogil functions only when strictly needed on function exit, e.g. for cleaning up temp variables from with-gil blocks or adding tracebacks. See https://github.com/cython/cython/issues/3554 * Make the GIL-avoidance in 7d99b0f0 actually work in nogil functions and not just nogil sections. See https://github.com/cython/cython/issues/3558
* Mark reverse operators with METHOD_COEXIST (GH-4753)da-woods2022-05-033-2/+40
| | | | | | | | | | This means that reverse operators (e.g. `__radd__`) won't be hidden by the automatic wrapper that `PyType_Ready()` produces if the forward method exists. Although they won't work as in Python, they will be possible to look up and call explicitly. This should make it easier to write code that's compatible with Cython 0.29.x and Cython 3 (where reverse operators will be full supported). Closes https://github.com/cython/cython/issues/4750
View Lorry Controller Status