[darcs-users] darcs patch: resolve issue1043: fix bug in mergeAfterConflicting.

[Jason Dagit]

On Fri, Sep 26, 2008 at 2:45 PM, David Roundy <droundy at darcs.net> wrote:

> On Fri, Sep 26, 2008 at 10:31:48PM +0100, Eric Kow wrote:
> > On Fri, Sep 26, 2008 at 17:01:49 -0400, David Roundy wrote:
> > > Here's a bug fix for issue1043, which was a pretty scary bug in the
> > > conflict-handling code for the darcs-2 formalism.  I had failed to
> > > take into account a possible scenario (which I wrongly thought was
> > > impossible).  I've fixed that by switching to using in
> > > mergeAfterConflicting a variant of commute which doesn't allow
> > > conflicting patches to commute, which seems to solve the problem.
> >
> > So, I haven't looked at this yet.
> > Do you reckon it's safe to put it in the release?
>
> I believe it should be.  Or rather, the code without this fix is
> definitely not safe to put into the release, so I think we're better
> off with the fix than without it.  It definitely fixes this particular
> problem (and yes, I was able to understand the problem enough to be
> confident of that).  I'm concerned that there might be similar bugs
> elsewhere in the code, but don't have the time for a full audit.


Thanks for putting in a big comment in front of mergeAfterConflicting.  I
think that is quite nice.

This sounds like a tricky bug.

Maybe we can convince Florent to apply this and retry his test case to see
if those other not-audited places trigger.

Thanks David!

Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.osuosl.org/pipermail/darcs-users/attachments/20080926/98c71510/attachment-0001.htm