Re: regression in multi-threaded git-pack-index

[Jeff King <peff@peff.net>]

On Tue, Mar 19, 2013 at 05:30:34AM -0400, Jeff King wrote:

> On Tue, Mar 19, 2013 at 09:17:32AM +0100, Thomas Rast wrote:
> 
> > > but the line in question is:
> > >
> > >   if (deepest_delta < delta_obj->delta_depth)
> > >
> > > And in the debugger, both of those variables appear to have sane values
> > > (nor should either impacted by the patch you bisected to). On top of
> > > that, running with pack.threads=1 produces the same error. So I think it
> > > may be a false positive from valgrind, and unrelated to your issue.
> > 
> > I find that somewhat unlikely, for two reasons: memcheck is actually
> > quite good at finding uninitialized memory use, it just isn't that good
> > at distinguishing if it makes a difference.  Most false positives are of
> > the "loading an entire word and discarding most of it" kind.
> 
> Yes, that has been my experience with valgrind false positives, too. But
> if this is a real problem, it may be different from the OP's issue. It
> seems to trigger for me in v1.7.10, before Duy's threading patches. It
> does not seem to be in v1.7.5. I'm bisecting now.

Hmph. It bisects to Junio's d1a0ed1 (index-pack: show histogram when
emulating "verify-pack -v", 2011-06-03), which introduces those lines.
The deepest_delta variable is static, so by definition it is always
initialized to something. So I guess some objects may not have
delta_depth set. Still looking.

-Peff