[MERGE][#354036] Server-side (sink-side) fix for missing inventory parents

[Andrew Bennetts]

Robert Collins wrote:
> BB may not have noticed this.

It did notice.  When it's back up you can see it at:
<http://bundlebuggy.aaronbentley.com/project/bzr/request/%3C20090422071133.GE10211%40steerpike.home.puzzling.org%3E>

> On Wed, 2009-04-22 at 17:11 +1000, Andrew Bennetts wrote:
[...]
> 
> It would be nice to put an upper time limit on how long we have the
> client do this - it is an extra round trip that the sink should be
> telling us anyway.

Yeah.  Perhaps we can make it conditional on the remote side being < 1.15, if
we can get at the medium...

> > @@ -2818,6 +2829,10 @@
> >              new_missing =
> > graph_index.external_references(ref_list_num=1)
> > 
> > new_missing.difference_update(self.get_parent_map(new_missing))
> >              self._missing_compression_parents.update(new_missing)
> > +        if self._external_parent_refs is not None:
> > +            for node in graph_index.iter_all_entries():
> > +                self._external_parent_refs.update(node[3][0])
> > +                self._external_parent_refs.discard(node[1])
> 
> Because the code is a little opaque, a one-liner here about what these 4
> lines do would be good.

Will do.

> ^ unnecessary VWS

Oops, thanks.

-Andrew.