Daniel Berlin wrote:
>On Sat, 9 Feb 2002, Branko [UTF-8] ÄŒibej wrote:
>
>>Daniel Berlin wrote:
>>
>>>On Fri, 8 Feb 2002, Daniel Berlin wrote:
>>>
>>>>One thing we should also be doing is fulltext compression. Rather than
>>>>give up, or store full fulltexts, we could at least delta them against an
>>>>empty string.
>>>>Since vdelta is based off lempel-ziv, it'll give you some compression
>>>>(better than compress, worse than gzip, according to the paper).
>>>>
>>>>Right now, we just leave it completely fulltext.
>>>>
>>>Strike that.
>>>
>>>I forgot the compression comes mainly from their encoding/folding, rather
>>>than anything else.
>>>Since we do neither, ...
>>>
>>Not at all. Vdelta itself compresses. That has nothing to do with the
>>encoding of the block-copy instructions.
>>
>
>Yes,yes, i remembered it runs over the target too.
>
>But turning everything into a delta (fulltexts = delta against empty
>string) would mean we'd never need to worry
>about fulltexts. If fulltexts simply became deltas against the empty
>string, then everything becomes a delta combination problem.
>
>We'd effectively just make sure delta chains don't become too long for a
>given revision, by combining deltas until the chain of deltas necessary
>for a given revision was < N or something.
>
>Not sure if that's better or worse.
>
Oh, runnung a compress-only vdelta at checkpoints would definitely be
better than storing fulltexts.
--
Brane Čibej <brane_at_xbc.nu> http://www.xbc.nu/brane/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 21 14:37:06 2006