From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeffrey A Law To: Craig Burley Cc: oliva@dcc.unicamp.br, pfeifer@dbai.tuwien.ac.at, egcs@cygnus.com Subject: Re: Inconsistance in snapshots repository Date: Thu, 16 Apr 1998 23:59:00 -0000 Message-id: <6435.892776422@hurl.cygnus.com> References: <199804162331.TAA25018@melange.gnu.org> X-SW-Source: 1998-04/msg00669.html In message < 199804162331.TAA25018@melange.gnu.org >you write: > I'm CVS-unaware, but, generally, is this really guaranteed to provide > a coherent snapshot to anyone, anytime, including if they try using > it while a snapshot is being produced (and the "moving tag" updated)? Nope. But what it does do for the CVS folks is allow them to upgrade to whatever the current snapshot happens to be without knowing the exact name of the tag. I think that was the idea behind the suggestion. The tag doesn't move until I make the snapshot available. And the window where things can get out of sync is relatively small; directories are locked as the tag is added/moved. [ I'm not a CVS expert - there may or may not be things in CVS to prevent the following "problems". ] To get an out of sync tree you'd have to start after the tag process and somehow get ahead of the tag process (might be able to happen if for example you don't have some subdirs and can "leap frog" over the tag process while the tag process walks through the subdirs. Another possibility would be an update starting after tagging, then both the update & tag block waiting on the same lock. The updater might be able to get the lock before the tagger and get an inconsistent tree. The opposite of those cases might be able to happen too. In both situations another cvs update ought to fix any inconsistencies. I don't consider these serious prolems (if they exist at all), but others may disagree. jeff