Build, release, QA
------------------
*Alpha3*
Alpha3 was released Tuesday, Aug 1.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006525.html

OSAF changed Chandler's license to the Apache 2.0 License.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006531.html

Two bugs remained open (Bug 5906: search text field in toolbar and Bug
6414: repository corruption). If other fixes accumulate, we have the
option of spinning another release. Bear pointed out this
as-yet-theoretical next release would be called "0.7alpha3.1".
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006517.html

*ASCII filenames required for SVN*
Bear reminded everyone that our use of SVN requires ASCII filenames and
directory names for optimal cross platform performance.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006527.html

Intern and SoC projects
-----------------------
*Tags and topology*
Following up on Philippe's thread about how Xun's project might address
the topology of tags, Reid pointed out that Flickr has a relationship
among tags called "clusters" (e.g. a "wasp" might cluster with "insect",
"macro", "bug", "nature", "bee").
http://lists.osafoundation.org/pipermail/chandler-dev/2006-July/006489.html

Davor gave a reference to a paper on the topic (generating hierarchies
of tags based on frequency of co-occurrence).
http://pages.stern.nyu.edu/~panos/publications/cikm2005.pdf

Xun discussed the paper and folksonomies like Flickr and delicious. He
noted that PIMs are relatively small and highly personalized; they don't
provide a good data set on their own for inferring tag hierarchies. One
option is to make use of external data: WordNet, or a model derived from
established internet categories. He noted that tag clustering (ala
Flickr) is easier, and may end up being useful to Chandler users.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006538.html

Core projects
-------------
*Repository repair*
Andi added a new "check and repair" test menu item; it rebuilds indexes
that fail check() tests. (In the future it may add other repair steps).
http://lists.osafoundation.org/pipermail/chandler-dev/2006-July/006491.html

Andi added an "undo" command line option; it rolls back repository
versions, including an option to roll back to a version that passes check().
http://lists.osafoundation.org/pipermail/chandler-dev/2006-July/006501.html

Bear asked about doing a backup. Andi obliged, adding a --backup option
that causes the repository to be backed up immediately after Chandler
opens the repository.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-July/006510.html

Andi added yet another option, "repair". This runs "check and repair" as
described above.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-July/006511.html

*Sharing performance*
Lisa did another performance analysis with the final alpha3 build. She
added graphs and correlated output to twisted.log. Lowest hanging fruit:
reuse of connections, avoiding the server needing to drop connections,
figuring out 7.5s case of live TCP connection without SSL handshake
starting. Heikki noted that we've only observed the 7s delay in Lisa's
case; he offered to help her diagnose the scenario on her computer.
http://wiki.osafoundation.org/bin/view/Journal/PublishConnectionAnalysis
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006528.html

*Incoherent floating event indexes*
Jeffrey brought up a problem with indexes for events -- in particular
events with floating timezones. If the current timezone changes,
floating events change their timezone relative to UTC, and the index may
be sorted incorrectly. Options: (a) separately indexed floating
collection and non-floating collections, (b) update event index when
timezone changes. (a) requires efficient merge of the two indexes, (b)
scales poorly for large #s of floating events.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006530.html

Jeffrey worked out a proposal with Andi: re-index after every timezone
change. Keep a collection of all of the floating events, and only
reindex these events. It scales linearly with # of floating events, but
n may be small enough here. In theory we could partition away long past
events if necessary. Another improvement: change effectiveStartTime from
 being a calculated property to being a persisted attribute, which
avoids use of a CompareIndex, which avoids loading items when reindexing.
http://lists.osafoundation.org/pipermail/chandler-dev/2006-August/006534.html

Meetings
--------
Apps team meeting:
http://wiki.osafoundation.org/bin/view/Journal/AppsMeeting20060803

Chandler meeting:
http://wiki.osafoundation.org/bin/view/Journal/EngineeringMeetingNotes20060803


_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Open Source Applications Foundation "chandler-dev" mailing list
http://lists.osafoundation.org/mailman/listinfo/chandler-dev

Reply via email to