about the delta technique used in Bazaar

Robert Collins robertc at robertcollins.net
Wed Jul 3 01:36:32 UTC 2013

On 3 July 2013 13:25, Martin Pool <mbp at sourcefrog.net> wrote:
> On 3 July 2013 05:03, Bo Chen <chen at chenirvine.org> wrote:
>> Thanks. Martin.
>> I am also curious on some more specific aspects:
>> 1) in this group, it only includes different versions of the same file, or
>> can include versions from other files?
> I think it can include other files.

It can both at the plumbing layer and the policy layer.

>> 2) how to choose the base for delta?
> This isn't really a system where there is a delta base.

There kindof is - there's there first text which is stored verbatim,
vs the second and subsequent texts which are stored as diffs against
the block being output. What there really isn't is a delta-chain,
because we don't build deltas against other texts, but against the
output representation.

It's more efficient to store deletes than adds in a group, so we take
the most recent version of each text when we put it into the group.
Arguably we could take the largest text rather than the newest.


Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Cloud Services

More information about the bazaar mailing list