Re: [PATCH] add the ability to create and retrieve delta objects
- From
Linus Torvalds <torvalds@osdl.org>
- Date
- May 3, 2005, 14:48 UTC
- Message-ID
- <Pine.LNX.4.58.0505030742330.3594@ppc970.osdl.org>
- In-Reply-To
- <Pine.LNX.4.62.0505030344170.14033@localhost.localdomain>
On Tue, 3 May 2005, Nicolas Pitre wrote:
> > Yep, that's what I've done last weekend (and just made it actually > work since people are getting interested).
I have to say that it looks uncommonly simple. Also, afaik, this should still work with the current fsck, it's just that because fsck doesn't understand the linkages, the error reporting won't be as good as it could be (I'd _much_ rather see "delta failed in object xxxxx" than "unable to read xxxxxx").
Now, one thing I like about this approach is that the actual delta _generation_ can be done off-line, and independently of anything else. Which means that the performance paths I care about (commit etc) are largely unaffected, and you can "deltify" a git archive overnight or something.
In fact, it means that you might even be able to use some fairly expensive "search for the best blob object to delta against", including very much a intelligent rename search (ie "oh, this is a new object, let's see if any of the old deleted objects generate a good delta"), but you might even go back more than one generation.
Hmm. How nasty are those scripts?
Linus