Re: [PATCH 2/4] revisions passed to cherry-pick should be in "default" order

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 15, 2012, 18:39 UTC
Message-ID: <7vr4r857au.fsf@alter.siamese.dyndns.org>
In-Reply-To: <CAOeW2eFK+cKt9Tnh5oe74dU+f8rOOTaWk3KvE2rtUpgcOeDD7g@mail.gmail.com>

Martin von Zweigbergk <martin.von.zweigbergk@gmail.com> writes:

> Makes sense, I'll try to implement it that way. I was afraid that
> we would need to call prepare_revision_walk() once first and then
> if we afterwards find out that we should not walk, we would need
> to call it again without the reverse option.

Show 11 quoted lines

> But after looking at
> how rev_info.reverse is used, it seem like it's only used in
> get_revision(), so we can leave it either on or off during the
> prepare_revision_walk() and the and set appropriately before
> calling get_revision(), like so:
>
>   init_revisions(&revs);
>   revs.no_walk = REVISION_WALK_NO_WALK_UNSORTED;
>   setup_revisions(...);
>   prepare_revision_walk(&revs);
>   revs.reverse = !revs.no_walk;

Sorry, but I do not understand why you frutz with "reverse" after prepare, and not before.

I think you can just set no_walk and let setup_revisions() turn it off upon seeing a range (this happens in add_pending_object()). After setup_revisions() returns, if no_walk is still set, you only got individual refs without ranges, so no reversing required.

You also need to be careful about "revert" that shares the code; when reverting range A..C in your example, you want to undo C and then B, and you do not want to reverse them.