On Wed, Sep 12, 2018 at 02:23:42PM -0700, Junio C Hamano wrote:
> > diff --git a/commit-reach.c b/commit-reach.c
> > index 86715c103c..6de72c6e03 100644
> > --- a/commit-reach.c
> > +++ b/commit-reach.c
> > @@ -544,20 +544,31 @@ int can_all_from_reach_with_flag(struct object_array *from,
> > {
> > struct commit **list = NULL;
> > int i;
> > + int nr_commits;
> > int result = 1;
> >
> > ALLOC_ARRAY(list, from->nr);
> > + nr_commits = 0;
> > for (i = 0; i < from->nr; i++) {
> > - list[i] = (struct commit *)from->objects[i].item;
> > + struct object *from_one = from->objects[i].item;
> >
> > - if (parse_commit(list[i]) ||
> > - list[i]->generation < min_generation)
> > - return 0;
> > + from_one = deref_tag(the_repository, from_one,
> > + "a from object", 0);
> > + if (!from_one || from_one->type != OBJ_COMMIT) {
> > + from->objects[i].item->flags |= assign_flag;
>
> I wondered why this is not futzing with "from_one->flags"; by going
> back to the original from->objects[] array, the code is setting the
> flags on the original tag object and not the non-commit object that
> was pointed at by the tag.
Note that from_one may even be NULL.
> > + continue;
> > + }
> > +
> > + list[nr_commits] = (struct commit *)from_one;
> > + if (parse_commit(list[nr_commits]) ||
> > + list[nr_commits]->generation < min_generation)
> > + return 0; /* is this a leak? */
> > + nr_commits++;
> > }
>
> In the original code, the flags bits were left unchanged if the loop
> terminated by hitting a commit whose generation is too young (or
> parse_commit() returns non-zero). With this updated code, flags bit
> can be modified before the code notices the situation and leave the
> function, bypassing the "cleanup" we see below that clears the
> "assign_flag" bits.
>
> Would it be a problem that we return early without cleaning up?
>
> Even if we do not call this early return, the assign_flag bits added
> to the original tag in from->objects[i].item won't be cleaned in
> this new code, as "cleanup:" section now loops over the list[] that
> omits the object whose flags was smudged above before the "continue".
>
> Would it be a problem that we leave the assign_flags without
> cleaning up?
Yeah, I hadn't thought about the bit cleanup when making my original
suggestion. In the original code (before 4fbcca4eff), I think we did set
flags as we iterated through the loop, and we could still do an early
return when we hit "!reachable(...)". But I don't see any cleanup of
assign_flag there at all.
So I guess I'm pretty confused about what the semantics are supposed to
be.
-Peff
