annotate & stacking

[Robert Collins]

On Mon, 2008-06-23 at 14:09 -0500, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Collins wrote:
> | So,
> |     knit.py has an Annotator object that looks inside knits to perform
> | annotations. This appears to duplicate a bunch of logic and also looks
> | inside the knit at the _index object etc.
> |
> | To stack cleanly, I need something that can live within the knit, and
> | then build on top of the public VersionedFiles interface (or indeed
> | something that lives solely outside and using the public interface).
> |
> | John - I know you have been working on annotate() recently; are we are a
> | position to remove annotate() from VersionedFiles, or otherwise tweak
> | things to meet the constraints above?
> |
> | -Rob
> 
> There is quite a bit that is knit/pack specific about _KnitAnnotator.
> There are parts that could be made more generic, but it certainly
> expects knit-deltas (for example).
> 
> It would certainly be possible to write an Annotator class that would
> just process fulltexts and have it be fairly simple. However, I'm
> guessing it would be quite a bit slower. Being able to re-use the
> existing deltas for left-hand parents is a pretty big win for a lot of
> the test files I was using.
> 
> However, if we go with a full "policy" based system, then we may not
> even be able to use the left-hand blocks, and if the underlying storage
> uses xdelta, it won't have something we can use either. (xdelta is much
> more of a gzip across 2 texts, and throw out the first set, so you get
> stuff like RLE, etc.)
> 
> So... I would probably go with a generic Annotator that just works on
> full texts and ancestries, and implement a custom one that can take
> advantage of more intimate knowledge of the layout of knits/packs.
> 
> Also, some stuff is pulled into the annotator, so that it can be smart
> about which pieces need to be cached, and what can be freed, etc.

So, I think its fine for the hunks that are *in* the local knit to be
directly examined (though get_record_stream allows them to be accessed
without using private functions). But once the knit doesn't have the
content locally is where the problem arose.

As you're working on annotation already, is this something we can
parallelise?

-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/20080624/45abadbd/attachment.pgp