[rfc] [patch] speeding up bzr log with a tree delta iterator

[Denys Duchier]

Aaron Bentley <aaron.bentley at utoronto.ca> writes:

> I prefer to avoid putting in unused (and presumably untested) code.

ok. I removed the keyword arg.

> But it would be good if we could nuke the semantics for 0 and None, and
> use Branch.revision_id_to_revno instead.

no, I still have to be able to handle revisions outside the current history.

>> [in_store] was introduced specifically to handle the case requested by jam of a
>> lone revid not in the branch's history, and it is used in (builtin.py)cmd_log.
>
> It's shocking that we didn't already have this functionality.  I still
> think throwing NoSuchRevision when you don't have to is a bit excessive.
>  You'd have to call it something other than in_store, then.  But this is
> fine.

I didn't choose the name.  There already was a RevisionSpec.in_store: in_store,
in_branch, and in_history were all one and the same.  I simply specialized it in
RevisionSpec_revid to implement the semantics that was actually needed.

> The main remaining issue I have is that your latest code still has
> public Branch methods that use None to indicate 'no revision specified'.

ok, I made it private for the time being

> The main ways I can see fixing this are
> 1. Define NO_REVISION_SPECIFIED and test for that, instead
> 2. Make it a private function in log.py
>
> You could also carry through with making 'null:' the only revision-id
> for the null revision, but that's a lot of effort for this kind of change.

I think I'd like to tackle the null revision issue after this patch has gone
in.

> There are also some integration issues, but I'm happy to fix those myself.

Thanks.  that's very kind :-)  the above fixes have been pushed.

Cheers,

--Denys