[RFC]combine find_ids_across_trees and tree._iter_changes

[Aaron Bentley]

Robert Collins wrote:
> Aaron, for dirstate, we can do single-pass diff generation straight from
> file paths - correctly - without needing a call to
> find_ids_across_trees: I've got a dirstate tuned version of that call,
> but rather than change the api to return paths and ids, I'd like to just
> avoid invoking it at all - in the common case it just gathers all the
> ids from all trees, which isn't very helpful at all as we end up
> considering all ids anyway.
> 
> Is this ok with you ? If not do you have a proposed alternative ?

I'm fine with providing a combined call, but leaving the existing
functionality in place.  If you want to remove the existing
functionality, my main objection is that it prevents us from doing more
precise selection, and from using uninvolved trees for finding ids.  For
example, doing "bzr diff -r 5..6 foo.py" will match a file that is named
"foo.py" in the current tree, even if it was not named that in revision
5 or 6.

Aaron