cache for copy/rename detection

I thought about this lately. git's expensive copy/rename detection can
be run once and the results are kept somewhere in .git. Copy/rename
detection depends on tree objects and git detection algorithm. Tree
objects are immutable and git algorithm change rate is low IMO. So
cache invalidation is unlikely to happen.

Another thing I want to do with this cache is ability to manually
inject some hints to git. For example, I remove a file. Several
commits later, I revive that file in another place. I know that two
files are the same but can't tell git (and I don't think Git will
detect this pattern). With this cache, I can tell git this file is a
copy of that file.

What do you think?
--