[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: nasty commit bug

From: Karl Fogel <kfogel_at_newton.ch.collab.net>
Date: 2002-07-15 19:15:39 CEST

Philip Martin <philip@codematters.co.uk> writes:
> ra_dav looks to be at fault.
>
> 1. Check out a working copy.
> 2. Modify a file.
> 3. Run the svn client under gdb.
> 4. Set a break point at svn_wc_process_committed.
> 5. Commit and when the breakpoint is reached kill the client.
> 6. Remove the lock manually or use cleanup (this is necessary because
> since r2506 the client will no longer ignore the lock).
> 7. Run the commit again.
>
> When using ra_local the second commit fails, with transaction out of
> date. When using ra_dav the second commit succeeds :-(

Holy cow. Well, that's an Alpha bug, then :-).

First of all, thanks a million Philip for the intense debugging
efforts you've been putting in here.

So...

What we had was in interaction between two independent bugs: the
libsvn_wc code was (for a while) in a state where it would ignore a
lock encountered during commit -- and simultaneously, there is a bug
in ra_dav that allowed a commit to proceed against an out-of-date
file.

I'm going to work on the ra_dav bug now -- if it's what it seems to
be, it needs to be fixed before Alpha.

As for revs 2452, 2470, and 2504: I'm leaning toward reverting them
for Alpha. With this ra_dav thing to debug, there just isn't going to
be time to thoroughly review even the changes already committed or
posted, let alone to finish issue #749. And if the issue isn't going
to be finished, there's no compelling reason to have its beginnings in
the code when we release Alpha; in fact, it could even be sort of
confusing to have it in that state, if the Alpha code gets a lot of
new eyeballs as we hope.

I'll start debugging the ra_dav thing now, Ben or I will probably
revert 2452 and friends later today. Speak now or forever hold your
peace :-).

It need not be said -- but I'll say it anyway -- that this is *not* a
comment on the quality of any of those revisions, it's just an issue
of risk control and API consistency for a major release.

-K

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Jul 15 19:27:33 2002

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.