[RFC] Store all important data in KnitData

Robert Collins robertc at robertcollins.net
Mon Oct 29 18:26:50 GMT 2007


So I've looked at your branch somewhat..

bb:resubmit


On Fri, 2007-10-26 at 17:24 -0500, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> The attached patch changes how we store the data in packs. Basically, the data
> that used to only be stored in the KnitIndex is now also stored in the KnitData
> itself.

rather than passing in a factory for data, less just pass in a data
object and replace the access_method parameter. Otherwise it just kinda
grows and grows.

Another thing to do is to separate out in the knit code the compression
parents and the logical parents based graph facility.

Another thing to do is to remove the graph from inventory 'knits' and
instead use just the compression parents information.

It needs a new format - lets use --experimental and
--experimental-subtrees again.

Rather than KnitDataV2 I propose KnitDataZLib

> There are a couple pieces that I'm not happy with (the InterKnit needs a better
> way to detect that it needs to do the conversion). Also, as long as the Index
> has data redundantly stored, it should probably be compared against the other
> when extracting.

Agreed.

> Data storage wise, the difference is actually pretty small. I converted all of
> bzr.dev and with normal packs it is 63MB, with this extra data it is 65MB. I
> attribute this to the gzip compression. I would expect even less effect if
> packs were annotated, since it is very likely that the parent string already
> exists in the knit data.
> 
> I don't know that we want to do this before we convert over to packs, but it
> would mean we don't need yet another 'bzr upgrade'.
> 
> I'll probably also play with switching from gzip to using straight zlib, which
> should be slightly faster and smaller.

-Rob
-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20071030/569279d8/attachment.pgp 


More information about the bazaar mailing list