Optimizing tree building

Aaron Bentley aaron.bentley at utoronto.ca
Fri Jun 8 01:10:46 BST 2007

Hash: SHA1

Robert Collins wrote:
> On Thu, 2007-06-07 at 19:01 -0400, Aaron Bentley wrote:
>> I really don't know where this will lead to.  But even if we don't
>> move
>> to a blob format for storing revision data, the index problem has to
>> be
>> solved.
> Because of ghost filling the index is not in strict topological order.
> So the best we can do is walk backwards from the end in increasing read
> sizes. That is doable if I recall the results last time I looked.

For the local-disk case, this probably works.  But for the network case,
it seems likely to make things slower, and I wouldn't want to be the
cause of that.

> I think that is a reasonable short term thing to do; I dont think we
> want to tweak the knit disk format though, not if we think we can get a
> container based repository up and running.

I figure if we do get a container-based repository, we'll still need
indexes, no?  That makes me think it might be more productive to work on
a new way of handling inventories than to optimize the current one.

> I'm working up a needs document for the repository core; similar to the
> dirstate one, which should bring together the analysis from london in a
> coherent form - expect that mid next week.

Okay.  Looking forward to it.

Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org


More information about the bazaar mailing list