Possible performance improvement

[Martin Pool]

That could be a good idea.

I was wondering about keeping a limited-size cache that can cover all
file versions, not just the inventory, indexed by the file's SHA-1. 
Once we read the revision file we'll know the inventory's hash, and it
avoids needing to store the revision-id.  (On the other hand then we'd
need to always load the revision object, so perhaps your way is best.)

Keeping it as a pickle might well be faster.  I was also wondering if
a pickle would be faster than the format currently used for the
hash-cache.

--
Martin