STINNER Victor added the comment:
> Victor, you really shouldn't submit code that you don't understand.
In fact I do understand overlapped_dealloc.patch, it's "quite simple". On
Windows XP, if you have exit Python with pending overlapped operations, you
must not release the read or write buffer of the operation, or the process may
crash because Windows tries to read from or write into freed memory. The trick
is to not release the read/write buffer (but close the Windows handle of the
overlapped operation) if the operation is pending and python is exiting.
On more recent version, you can safetly cancel a pending operations, so the
issue is specific to Windows XP (and older, but we don't support Windows 2000
anymore, right?).
Martin: I didn't understand your suggestion on Rietveld:
"Why is this the default case? I think it should be the ERROR_IO_INCOMPLETE
case, because only this error handled. Any other error should IMO be raised as
is, and the resources still be released."
I wrote a patch just to ask you if it is the change you suggested, and you
still didn't answer to this question ("is overlapped_dealloc-2.patch what you
wanted?")
I'm not sure that it is safe to release the memory if the state is different
than ERROR_IO_INCOMPLETE. Well, it is safer to never release memory at exit,
but it may leeak memory. So maybe Martin's suggestion can be delayed after
Python 3.4 (to give more time to buildbots to test it), but Richard's fix
(overlapped_dealloc.patch) can be applied (on _overlapped) to Python 3.4?
--
Sorry for mixing two different issues in the same issue, in fact
overlapped_dealloc.patch is just the follow up of #19565 for the _overlapped
module. I didn't expect so long discussions for such patch :-(
----------
versions: -Python 3.4
_______________________________________
Python tracker <[email protected]>
<http://bugs.python.org/issue20414>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
