[MERGE] Remove knit caches

[Robert Collins]

On Mon, 2006-06-26 at 10:21 -0400, Aaron Bentley wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Collins wrote:
> > On Sun, 2006-06-25 at 19:21 -0400, Aaron Bentley wrote:
> 
> >>Therefore, I propose that we get rid of the knit caches.  It seems to be
> >>best to do them at a higher level, instead.
> > 
> > 
> > I put them in for conversions from weaves - multiple inserts of non-knit
> > data.
> 
> Oh, okay.  I did try to have some discussion before doing this, but no
> one realized that this was what they were intended to accelerate.

Thats fine - I was at UDS with 0 spare cycles, so I saw the message go
by but had no time to reply to it.

> > The win there was to take some operations from hours down to minutes,
> > particularly for inventory insertions.
> > 
> > I'm happy for them to be taken out, as I was never particularly happy
> > about them being in, but the conversion path should be kept fast.
> 
> As a general API, the cache behaved badly when KnitVersionedFiles were
> used in the obvious way.  I'm not sure what level you're using knits at.
>  If you're able to predict all the texts/records you'll need, then we
> probably don't need to change the API.

Conversion of a tree with 8000 inventories has this access pattern:
insert  rev 0
insert rev1 reads rev0
insert rev2 reads rev0 rev1
insert rev3 reads rev0 rev1 rev2
...
insert rev51 reads rev50
insert rev52 reads rev50 rev51
...

etc.

The counts from 0 to 50 accumulate up the lines of left-most ancestor
history, so 'cache the last 50 inserts' is not sufficient to get optimal
hit rate, but is better than no caching as such.

> If your use case does require this caching functionality, I think it
> would be best to make it something explicitly requested, to avoid
> gotchas in the  general case.  Perhaps this could be accomplished by
> subclassing _KnitData.

This might be a reasonable approach.

Cheers,
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: 191 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20060627/96e5ea63/attachment.pgp