| msg93728 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-07 23:01 |
| The strxfrm function in the locale module can potentially raise a MemoryError. The failing malloc is in Modules/_localemodule.c, line 291. This is because the variable n0 of type Py_ssize_t is passed to PyArg_ParseTuple, which expects an int when PY_SSIZE_T_CLEAN is not defined. Patch attached, which also fixes an unrelated memory leak. |
|
|
| msg94079 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-15 08:55 |
| Could someone triage this? And review my simple fix? I think it should be a release blocker. It seems to me to be a problem on all 64-bit non-Windows platforms, or any platform if using UCS-4. The 4 high-order bytes of an 8-byte Py_ssize_t variable are not initialized. Thanks. |
|
|
| msg94085 - (view) |
Author: Antoine Pitrou (pitrou) *  |
Date: 2009-10-15 11:51 |
| Nice catch. I wonder why not all of our modules are PY_SSIZE_T_CLEAN in py3k. This function is not actually tested, could you add a test to the regression suite? |
|
|
| msg94092 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-15 15:13 |
| All right. The function strcoll also isn't tested (except for issue #3303). I'll look into that one, too. |
|
|
| msg94186 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-17 19:29 |
| I've added the tests. I found that on Windows, strxfrm has been unavailable in py3k since r61306, because of an omission in PC/pyconfig.h. (cf. patch) In determining whether I should first test for the existence of strcoll (or strxfrm) in the locale module, as was done in TestMiscellaneous, I found this is no longer needed in py3k since r61339. (cf. patch) (In python2 such checks are still necessary.) |
|
|
| msg94242 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-19 15:33 |
| The tests in TestEnUSCollation I added don't work for all encodings (e.g. Asian ones, or ASCII). Now checked for encodings which are known to work. Found and fixed a bug (result not returned) in getpreferredencoding in Lib/locale.py. This test is skipped on Mac, since r30377. I've tested it on 10.3 (PPC), and it passed, so I enabled it again. I did, however, find that FreeBSD and Mac OS X have known bugs in wcscoll and wcsxfrm: http://www.freebsd.org/cgi/man.cgi?query=wcsxfrm (cf. patch) |
|
|
| msg94254 - (view) |
Author: Antoine Pitrou (pitrou) * |
Date: 2009-10-19 19:23 |
| Nice work, thanks! One detail: when giving you credit, who should I mention? just "egreen"? |
|
|
| msg94255 - (view) |
Author: Derk Drukker (drukker) |
Date: 2009-10-19 19:35 |
| I've entered my real name. |
|
|
| msg94256 - (view) |
Author: Antoine Pitrou (pitrou) * |
Date: 2009-10-19 19:35 |
| Ok, thanks. |
|
|
| msg94258 - (view) |
Author: Antoine Pitrou (pitrou) * |
Date: 2009-10-19 19:49 |
| The patch is now committed to py3k (r75539) and 3.1 (r75541), closing. |
|
|