Am Mittwoch, den 10.06.2009, 17:52 +0200 schrieb Adrian Buehlmann: > > On 10.06.2009 16:51, Steve Borho wrote: > > On Wed, Jun 10, 2009 at 6:59 AM, Adrian Buehlmann<[email protected]> wrote: > >> On 10.06.2009 13:23, Simon Heimberg wrote: > >>> # HG changeset patch > >>> # User Simon Heimberg <[email protected]> > >>> # Date 1243410000 -7200 > >>> # Node ID a703c9b518d5fe94002f30bd5b5b7ae73792016e > >>> # Parent 62587c4f0e577faca23d03048c3ad6a070660001 > >>> shlib: use repo.opener with atomic file for writing status file
snip > >> This makes us susceptible to the problem: windows can't > >> rename a file that is open for reading. And shellext *is* > >> constantly reading. > > > > Is writing to an open file safer than renaming? Or you just don't > > feel the complexity is warranted? > > If the shell extension has the file open the rename call will > fail, that is, the rpc server process will crash with a traceback. > > Writing to a file that is open for reading is no problem, > but renaming is. On unix you can rename a file that is > open for reading, on Windows a rename call for a file that > has been opened by fopen will fail. > > As such the patch doesn't solve anything but introduces > a new problem. Renaming an open file is possible. It is not possible to rename a file to an open file in windows. This problem is solved in mercurial (see the comment in util.rename). No new problem is introduced. This change is not because I ran into a problem but because I was thinking about the design. Greetings, Simon ------------------------------------------------------------------------------ Crystal Reports - New Free Runtime and 30 Day Trial Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment. http://p.sf.net/sfu/businessobjects _______________________________________________ Tortoisehg-develop mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/tortoisehg-develop
