bzr blame: annotate --revision takes exactly one revision identifier; ideally any arbitary range should work

If the proposed rule here is "we should make a best guess at what the user meant rather than erroring" then I don't find it very convincing. People can enter all kinds of strange strings by half-editing previous commands.

If we can think of a reason meaning for annotate on a range let's do that. Perhaps it should show a kind of annotated diff, or an annotation only of lines inserted across that range?

Yes, what I'd /expected/ to happen is that annotations would be shown for all the lines that changed in that range, with other lines blank... but I was going to the "minimal intervention" approach. I'm perfectly fine that this wouldn't be suitable for bzr-as-a-low-level-tool, but it would be useful functionality if bzr were to transition to focus on higher-level primitives.

If there is scope for looking at range-handling more generally then it might be possible to break that range handling out so that it works more consistently (eg. at the moment 'bzr log -r B..B' returns that matching 'bzr diff -r A..B' whereas 'bzr diff -r A..A' errors).

I've needed this one, and it might be a nice way to speed up "blame" command which is painfully slow on huge trees (lp:gcc).