| msg56546 - (view) |
Author: Stephen P. Schaefer (thyrsus) |
Date: 2007-10-18 23:26 |
| The following fails with python 2.5 as built by Fedora 7: t2.py: import sys, Tkinter Tkinter.Button(text=u"").pack( ) Tkinter.mainloop( ) $ python t2.py python: ./Modules/_tkinter.c:941: AsObj: Assertion `size < size * sizeof(Tcl_UniChar)' failed. Aborted The following patch corrects the problem: --- Python-2.5-fc7/Modules/_tkinter.c 2006-08-11 22:33:36.000000000 -0400 +++ Python-2.5/Modules/_tkinter.c 2007-10-18 18:44:40.000000000 -0400 @@ -938,7 +938,7 @@ #if defined(Py_UNICODE_WIDE) && TCL_UTF_MAX == 3 Tcl_UniChar *outbuf; Py_ssize_t i; - assert(size < size * sizeof(Tcl_UniChar)); + assert(size == 0 | |
size < size * sizeof(Tcl_UniChar)); outbuf = (Tcl_UniChar*)ckalloc(size * sizeof(Tcl_UniChar)); if (!outbuf) { PyErr_NoMemory(); |
|
| msg56547 - (view) |
Author: Brett Cannon (brett.cannon) *  |
Date: 2007-10-19 00:55 |
| Patch is inlined in a comment. |
|
|
| msg56743 - (view) |
Author: Facundo Batista (facundobatista) * |
Date: 2007-10-25 13:09 |
| Really do not understand that assert. It says: assert(size < size * sizeof(Tcl_UniChar)); For that to be true, considering size to be positive, the result of sizeof needs to be greater than 0. If you modify it, and also accepts it to be 0, and considering that the result of sizeof won't be negative, what is the point of keeping the assert? |
|
|
| msg59105 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2008-01-02 22:16 |
| So what's the correct fix? Skip the whole block if size==0? |
|
|
| msg59107 - (view) |
Author: Christian Heimes (christian.heimes) * |
Date: 2008-01-02 23:08 |
| I've hit the same bug today. I vote for the removal of the assert() and a skip of the entire block if size == 0. I don't see what the assert() should test. |
|
|
| msg59158 - (view) |
Author: Stephen P. Schaefer (thyrsus) |
Date: 2008-01-03 21:03 |
| I'm not sure of the scope of the "block" to which you're referring. As patched, the code returns a Tcl_Obj representing 0 length unicode string, which appears to me to be the correct behavior, and works for the applications I use (originally developed in python 2.3 and python 2.4). I doubt that it's worth special-casing away the call for a 0 length outbuf, the skipped for loop, plus the call to free the 0 length outbuf. |
|
|
| msg59163 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2008-01-03 21:42 |
| I suspect the assert is an overflow check. But since asserts are compiled into no-ops unless we're using debug mode, I don't see the point of using an assert. Just to close the issue I'll replace it with a proper overflow check. |
|
|
| msg59164 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2008-01-03 21:49 |
| Before I make a fool out of myself, what do folks think of this patch? |
|
|
| msg59168 - (view) |
Author: Stephen P. Schaefer (thyrsus) |
Date: 2008-01-03 22:14 |
| It looks good to me (to the extent that means anything). I'm doing an rpmbuild now with your patch applied to the Fedora 7 .src.rpm, and I'll start using the result on my workstation ASAP. |
|
|
| msg59182 - (view) |
Author: Guido van Rossum (gvanrossum) * |
Date: 2008-01-03 23:55 |
| Committed revision 59685. (2.5.2 branch) Committed revision 59686. (2.6 trunk) |
|
|
| msg59192 - (view) |
Author: Stephen P. Schaefer (thyrsus) |
Date: 2008-01-04 00:35 |
| Mr. van Rossum's patch, applied to the Fedora 7 .src.rpm, is working for me. |
|
|