usertest on brisbane-core status report

John Arbash Meinel john at arbash-meinel.com
Fri Feb 13 15:21:23 GMT 2009


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ian Clatworthy wrote:
> I've spent today trying to getting usertest supporting and
> benchmarking the new formats being compared in brisbane-core.
> I haven't had a lot of luck so far unfortunately.
> 
> I've raised a bug report (https://bugs.edge.launchpad.net/bzr/+bug/328962)
> for the issue blocking development4, development4-hash16 and
> development4-hash255 benchmarking. Perhaps John can take a
> look at that?
> 

This at least sounds like a similar problem to what Robert was
encountering. Where we have delta chains that go through revisions that
don't actually refer to those texts.

It works fine for Packer (pack => pack fetching) because it only
consults the index, and pulls across everything.

It doesn't work for the generic fetch, because that iterates the
inventories to find the texts to copy, but then uses
get_record_stream(include_delta_closure=False) to actually copy them
across as delta chunks (if it used True, then the source repo would
unpack them to a fulltext, and they would get re-inserted into the
target repo, but it would be very inefficient.)

Again, we had encountered this before when upgrading knits => packs, and
we updated "bzr reconcile" to supposedly fix this case. (This was back
circa bzr-1.0.) I don't know what happened, as I seem to have the same
gaps in my repository, and Robert noticed it in his repo as well.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkmVj/MACgkQJdeBCYSNAAOIwwCfX/qtZZxRT/g/7F5fJ24+I4EX
tncAn1Hx1J9Bqg6qjd6M4huDwAQEKHBI
=i9xv
-----END PGP SIGNATURE-----



More information about the bazaar mailing list