If a checkout fails mid-way, or is interrupted, we
have a 2 big problems.
1. The working copy is not "aware" that it's
incomplete. An 'svn up' only repairs missing data
because it notices that something is in its
entries file but not on disk. But an incomplete
checkout has an entries file that is downright
*wrong*. If the entries file is missing data, the
wc is screwed forevermore. An 'update' won't do a
thing, because it has no idea what the proper
entries list should be. Somehow, we need create
working copies in an "invalid" state from birth;
when the checkout truly finishes, the whole wc
should be loggily shifted into "valid" state.
2. Once we accomplish point #1, it would be nice
if we could somehow make failed checkouts
"continue" from where they left off when the
checkout is re-run. CVS is able to do this, and
we can't. This is a design issue that needs
discussion. In a nutshell, the svn client always
assumes that it has a valid entries file, while
the cvs client never trusts the entries file -- it
says "I have these entries, what do I need?"
...Whereas svn will simply report nothing but diffs.
Critical SVN-1011 Update failed due to timeout, produces obstructing object in wc
