STINNER Victor added the comment: Ok, initial patches are attached. Let describe them a little bit.
- pymem_debugcheckgil-2.patch: I don't think that this patch can be commited before the "bootstrap" issue is solved (Python doesn't start in debug mode with this patch when -W or -X command line option is used). I wrote it to check that other patches are correct (check if the GIL is held when PyMem_Malloc is called). - py_finalize.patch: should be safe - malloc_init.patch: change _Py_char2wchar() API, the result must now be freed by PyMem_RawFree() instead of PyMem_Free(). the change does not hurt in release mode (both functions are just wrapper to free()), but may break in debug mode (because python checks that PyMem_RawFree() is called on a buffer allocated by PyMem_RawMalloc()) in extension modules using _Py_char2wchar(). I would like to _Py_char2wchar() API (not to solve this issue, just because it is very useful for applications embedding Python), so changing its name would avoid a crash (applications would get a compilation or link error instead). Other change of the patch: replace free() and PyMem_Free() with PyMem_RawFree(), should be safe. - malloc_modules.patch: I replaced many malloc() with PyMem_Malloc(), which is not safe. All these patches must be reviewed carefully to check if the GIL is held or not, and tested with pymem_debugcheckgil-2.patch (on Windows too! some patched functions os posixmodule.c are only compiled on Windows). ---------- _______________________________________ Python tracker <rep...@bugs.python.org> <http://bugs.python.org/issue18203> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
