Josh Pieper <jjp@pobox.com> writes:
> Philip Martin wrote:
> > The original problem generated some dev list discussion between me and
> > Karl, and some code from Karl. As I recall the problem was that an
> > interrupted update caused the wcprops and the text-base/revision to be
> > out-of-sync (I can't remember which one was wrong). I don't know
> > whether the wcprops code is still fragile, or whether Karl's change
> > fixed it.
>
> Ok, I did some testing over ra_dav and found no problems. It passes
> the whole test-suite, plus I did quite a bit of checkouts/updating and
> interrupting without any issues. I also looked at issue #1136, which
> I think is what you were referring to here. I was unable to reproduce
> that failure with the new code either.
You wouldn't generally see the problem in normal testing, you'd only
see it in interrupted operations over ra_dav and other extraordinary
circumstances.
Let me dredge up what I can from the musty caverns of memory...
Well, hmm, this might be resolved issue #1136, in which case what I
said above isn't true -- you *can* get a bug in normal operations.
The patch in that issue helps explain why wcprops are "special"
sometimes.
> One last concern. Do we need to be worried about compatability? This
> change would prevent trunk from running the log files from older
> working copies since their names have changed. Should I hack in
> something to try "log" before trying "log.1" or can we just accept
> that amount of change?
The only operation this could affect would be 'svn cleanup', since log
files shouldn't be present when any other operation starts. So the
scenario here is that someone
1. does some operation that leaves the working copy in need of cleanup
2. upgrades their subversion :-)
3. runs the next operation, is told they must run 'svn cleanup' first
4. tries 'svn cleanup'
Step 2 seems unlikely, though I suppose it could happen to someone.
They can always downgrade temporarily to solve it, though, so there's
a workaround. So I don't think we really need to consider log
behavior to be part of the "working copy format"; no need to worry
about compatibility here, IMHO.
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon May 24 17:17:45 2004