branching inside a repository shouldn't lock it

[Martin Pool]

When doing a branch (or other fetch-based operation) we currently lock both source and destination repositories. If the two branches are in the same repository the fetch should be a no-op, and we should be able to do without the repo lock.

[Aaron Bentley (abentley)]

When we are branching, we should ensure that the last-revision of the the branch is not a ghost. In order to prevent the last-revision from being removed while we branch, we need a write lock.

[John A Meinel (jameinel)]

I think we shouldn't *double* lock the repository (causing problems on win32, etc). I think if the source is a ghost, making the target a ghost isn't any *worse*, so I think I agree with Martin that if we can do this lockless it is a good thing.

Further, use "lock_write()" doesn't actually take out a disk lock anymore. We only take one out during updating 'pack-names'. So maybe we should consider this Fix Released?

[John A Meinel (jameinel)]

I decided to go with Fix Released since pack repositories don't take out a disk lock unless a fetch actually occurs.