Storage internals: UUID

[Daniel Carrera]

On Wednesday, June 06, 2012 at 11:05 AM, John Arbash Meinel <john at arbash-meinel.com> wrote:

> There are interesting benefits to this, but Bazaar did explicitly make
> the choice that we validate the snapshot more than the full ancestry.
> (you can't have a revision which is mentioned but not converted if you
> have chained hashes.)

I personally like the idea of the history being included in the hash, but I can see myself being flexible on this point if the hash still takes measures to ensure that the entire revision is correct and not just the patch or the delta. For example, suppose you made a hash that included:

1. The entire contents of the source tree.
2. Metadata like author, date, branch, etc.
3. The non-hash ID of both the present revision and the parents.

It seems to me that this might have assurances comparable to what Monotone, Hg and Git do, while keeping the benefits of Bzr's metthod: If you wanted to interoperate with SVN, you would only need the ID of the immediate parents, and not their hashes. So there is no chaining of hashes. The problem I see is that it might be very slow to take a hash of an entire repository. At the same time, I really don't think it's enough to take a hash of just the deltas because it could be applied to a different tree without your necessarily noticing (imagine a clever and malicious attacker).

Cheers,
Daniel.