[Python-Dev] windows file closing race condition?

[R. David Murray]

On Fri, 06 Sep 2013 13:21:35 +0100, Tim Golden <mail at timgolden.me.uk> wrote:
> On 06/09/2013 12:50, Richard Oudkerk wrote:
> > On 06/09/2013 11:23am, Tim Golden wrote:
> >> On 06/09/2013 11:14, Antoine Pitrou wrote:
> >>> Le Fri, 06 Sep 2013 08:58:06 +0100,
> >>> Tim Golden <mail at timgolden.me.uk> a écrit :
> >>>>
> >>>> What should Python do?
> >>>
> >>> Maybe using FILE_SHARE_DELETE could help?
> >>> http://bugs.python.org/issue15244
> >>
> >> I don't think so. It's the use of FILE_SHARE_DELETE (by other programs,
> >> eg Virus Checkers) that typically causes the problem. IOW, the
> >> sequence is:
> >>
> >> * [Some Other Prog] takes FILE_SHARE_DELETE handle, allowing other
> >> programs to delete the file even while this handle is still open
> >>
> >> * Python calls DeleteFile -- succeeds if the only open handles are
> >> FILE_SHARE_DELETE; carries on
> >>
> >> * File has apparently disappeared but still has open handles
> >>
> >> * Any attempt to create a file of the same name or to delete a
> >> containing directory fail because the file is still open, even though
> >> it's successfully been deleted.
> >>
> >> * (Some time later) [Some Other Prog] closes its handle and the file is
> >> now completely gone
> >>
> >>
> >> Unless I'm missing something, there's nothing Python can do to help here
> >> apart from the sort of delay-retry dance which test.support uses and
> >> which I'm advocating against as a core feature.
> > 
> > Instead of deleting, the file could be moved to a temporary name in the
> > root directory (or some other permanent directory on the same drive) and
> > then deleted.  That would allow the directory to be closed even if a
> > FILE_SHARE_DELETE handle is still open for the file.
> > 
> 
> True, but then you're into determining a temporary name somewhere on the
> same volume if possible and avoiding collisions etc. Again, I don't
> think this is something we need to be doing by default in core Python.

What about moving the test.support delete helper to somewhere in
unittest, since this will come up in pretty much every test suite that
runs on Windows?

--David