Re: Checksum mismatch after canceling conflict resolution

On 11/2/07, Hyrum K. Wright <hyrum_wright@mail.utexas.edu> wrote:
> Hyrum K. Wright wrote:
> > David Glasser wrote:
> >> On 11/2/07, Hyrum K. Wright <hyrum_wright@mail.utexas.edu> wrote:
> >>> I've been spending some time this morning playing around with
> >>> interactive conflict resolution, and ran across this little gem:
> >>>
> >>> hwright@spock:~/test/greek.wc$ svn up
> >>> Conflict discovered in 'iota'.
> >>> Select: (p)ostpone, (d)iff, (e)dit, (h)elp for more options : h
> >>> (p)ostpone - mark the conflict to be resolved later
> >>> (d)iff - show all changes made to merged file
> >>> (e)dit - change merged file in an editor
> >>> (i)nteractive - prompt for each conflicting hunk
> >>> (r)esolved - accept merged version of file
> >>> (m)ine - accept my version of file
> >>> (t)heirs - accept their version of file
> >>> (l)aunch - use third-party tool to resolve conflict
> >>> (h)elp - show this list
> >>>
> >>> Select: (p)ostpone, (d)iff, (e)dit, (h)elp for more options : d
> >>> --- .svn/text-base/iota.svn-base Fri Nov 2 09:35:14 2007
> >>> +++ .svn/tmp/tempfile.tmp Fri Nov 2 09:35:20 2007
> >>> @@ -1 +1,5 @@
> >>> -This is the file 'iota'.
> >>> +<<<<<<< .mine
> >>> +This is the file 'iota'. asdf
> >>> +=======
> >>> +This is the file 'iota'. 1234
> >>> +>>>>>>> .r2
> >>> Select: (p)ostpone, (d)iff, (e)dit, (r)esolved, (h)elp for more options
> >>> : subversion/svn/main.c:934: (apr_err=200015)
> >>> svn: Caught signal
> >
> > This is where I hit ctrl-C.
> >
> >>> hwright@spock:~/test/greek.wc$ svn up
> >>> At revision 2.
> >>> hwright@spock:~/test/greek.wc$ svn up -r1
> >>> subversion/libsvn_wc/update_editor.c:2032: (apr_err=155017)
> >>> svn: Checksum mismatch for '.svn/text-base/iota.svn-base'; expected:
> >>> '0a18339a4d1bddab170f4811a8ef6384', actual:
> >>> '2d18c5e57e84c5b8a5e9a6e13fa394dc'
> >>> hwright@spock:~/test/greek.wc$ ls
> >>> A iota
> >>> hwright@spock:~/test/greek.wc$ svn up -r0
> >>> D A
> >>> D iota
> >>> Updated to revision 0.
> >>>
> >>> As a result of canceling the conflict resolver in operation, it appears
> >>> that some metadata in the wc wasn't updated correctly. I couldn't
> >>> update to r1, but I could update to r0 and then back to r1.
> >>>
> >>> Is this intended behavior?
> >> Absolutely not. The wc should never be corrupted.
> >>
> >> Do you have an automated reproduction script (perhaps one that at one
> >> point echoes "hit ctrl-C during the next prompt")?
> >
> > Sure, I can create such a script. In the mean time, I've noted above
> > where I hit ctrl-C.
>
> This triggers the problem without using ctrl-C:
> [Assumes greek.wc and greek-2.wc are checked out greek trees from the
> same repository.]
>
> echo "asdf" >> greek.wc/iota
> svn ci greek.wc/iota -m 'asdf'
> echo "1243" >> greek-2.wc/iota
>
> # This causes a premature abort
> echo "d\n" | svn up greek-2.wc
>
> # WC-corruption here
> svn up greek-2.wc -r 1

Want to know something hilarious?

If you give an EOF or a SIGINT to the "svn up greek-2.wc" during the
conflict prompt, you end up with a corrupted wc. (In this case, the
entries file believes that iota is at r2 even though both its
text-base and checksum are still for r1.)

If you just kill -9, though, everything is happy; the wc is locked,
but "svn cleanup" fixes that. And no corruption :)

So somewhere in our error handling or pool cleanup chain there is a
"helpful" function that loggily bumps the revnum on iota. Now to find
it...

--dave

-- 
David Glasser | glasser_at_davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org