relaxing the same-length constraint on B+Tree indices

[Robert Collins]

On Wed, 2009-03-25 at 21:04 -0500, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Collins wrote:
> > I'd like to relax the constraint that all the keys and references in a B
> > +Tree index be the same width.
> > 
> > This would remove a safeguard against bad code in knit->pack mapping,
> > but I am confident in the rest of our test suite to catch changes there.
> > 
> > Thoughts?
> > 
> > -Rob
> 
> Is this so you could combine revisions and inventories, etc in the same
> index?

revisions, inventories, search data :)

> Conceptually I'm fine with it, but I think it breaks the parser to have
> arbitrary width nodes. (I think parsing the key grabs the first N null
> bytes.)

I would probably do it as a new disk format for B+Tree indices, with a
new builder, but with [hopefully] most code shared.

-Rob
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20090326/4ce68ef3/attachment-0001.pgp