Re: [RFC PATCH] Not computing changed path filter for root commits

> Taylor Blau <me@ttaylorr.com> writes:
> > This only happens when we return REV_TREE_NEW from a call to
> > `rev_compare_tree(revs, p, commit, nth_parent)`. But we'll only get
> > REV_TREE_NEW back if
> >
> >     repo_get_commit_tree(the_repository, p);
> >
> > returns NULL. But when we call rev_same_tree_as_empty(revs, p) in the
> > REV_TREE_NEW case, we return early as follows:
> >
> >     struct tree *t1 = repo_get_commit_tree(revs, p);
> >     if (!t1)
> >       return 0;
> >
> > So we won't even consult the Bloom filter in that case, since t1 is NULL
> > for the same reason as what caused rev_compare_tree() to return
> > REV_TREE_NEW in the first place.
> >
> > I am still dumbfounded by how we would ever get REV_TREE_NEW in the
> > first place, but if we did, I think we would be OK here.
> >
> > Thanks,
> > Taylor
>
> Ah, good point. Your patch in
> https://lore.kernel.org/git/ZQnmTXUO94%2FQy8mq@nand.local/ looks good to
> me, then.