I'm not sure I understand what you mean. I was talking about Fossil, which uses SQLite under the hood, which would then maintain the index whenever data in it is modified.

Scale. Microsoft couldn't make a SQL VCS work. Maybe it's because they tried to use SQL Server instead of PostgreSQL, I dunno :) but IIRC the amounts of metadata involved essentially meant that they had no hope of doing git-like cloning: it was too much to clone.

You would need to implement a persistent datastructure, but without refcounts, you would have to do a gc. I guess you could do a refcount tree and only RC at the last common node.

Really atime is the devil. Funny thing is, is that all modern higher level systems have some form of data access logging, which is what you want anyway. So why not form some easily compressible event log?