On Wed, Oct 08, 2008 at 03:55:36PM +0400, Dmitry Kurochkin wrote: > Hello! > > I am not able to reproduce the issue. So it is not tested. But it is > pretty simple so I do not think it introduces any new bugs (and > hopefully fixes one :)).
It looks pretty good to me. My only (pretty minor) concern would be with the use of copyFile. This is not an atomic operation, which means that if its target is a hash file in _darcs/patches etc, then there will be a possibility of introducing a hash failure if we (or another darcs) happens to read that file while it's being written. I think a better approach would be to first copyFile to (newname++"-new") and then renameFile to newname. But since I think this should go into the release, I won't make this change myself, since I want to review it... and asking you to make this change has the effect that someone other than myself must decide that it's reasonable. Another possibility (rather than copyFile + renameFile) would be to create a hard link. This should be safe for any sort of downloaded file in darcs. But this requires a bit more code, and would have to fall back on copyFile + renameFile if the link-creation fails (as would happen on windows or if the two target names are on different filesystems, or on certain filesystem types. David _______________________________________________ darcs-users mailing list [email protected] http://lists.osuosl.org/mailman/listinfo/darcs-users
