OLPC News 2007-11-24

[Walter Bender]

1. Testing: The latest wireless firmware and associated patches to fix
the WLAN hang bug (Ticket #4470) were the focus of much of our testing
efforts this week. We are also very busy testing the candidate for
Ship.2, using a 1-hour "smoke test" with some additional concentration
on the browser.

2. Schedules: We have slipped Update1, but only for stabilization
purposes, not for new features.

3. Sugar: Marco Pesenti Gritti fixed a few regressions: a memory leak
when switching between activities (2 MB per switch), screen rotation,
compatibility of the Sugar shell with "raw" X applications and a few
other smaller bugs. He spent time reviewing patches and triaging
tickets. Tomeu Vizoso also worked on bug fixing in Sugar, the Journal
and the DataStore. Simon Schampijer spent the week testing and
detecting regressions, mainly with the clipboard (Ticket #5089). Simon
also worked on a schema for activities that process data that
originates
in another activity (Ticket #4909).

Reinier Herres landed numerous bug fixes in Joyride. These are mostly
bugs related to the palette (Tickets #4515 and #4887) and the search
field in the Mesh View and Journal (Ticket #4311). The "unfull-screen"
button  that was implemented for the Browse activity is moving to
sugar.graphics.Window, making this a standard behavior for all
activities. Reinier made a temporary patch to remove activities from
the donut again if they fail to launch (Ticket #4612); he is working
on improving the feedback to the user while unmounting volumes (Ticket
#2630). Reinier implemented support for rational numbers in Calculate.

4. Browse: Thanks to all the people who helped fix a bug in the
browser that prevented access to T-Mobile hotspots (Ticket #4280):
Marco Presenti Gritti, Simon Schampijer, Chris Ball, Jim Gettys, and
the Firefox developer team at Mozilla, including Chris Blizzard, Mike
Schroepfer, Kai Engert, Mike Beltzner, Johnathan Nightingale, and Dave
Camp. The Mozilla guys have been awesome; consequently we have an easy
workaround for the issue.

5. Translation: Sayamindu Dasgupta and Xavi Alvarez have Pootle is
fully open for translation and they have git integration almost
working (translators will be able to directly commit their work to the
master repository at http://dev.laptop.org). We have seen some intense
activity for quite a few languages; for example, the Hindi team has
completed nearly 100% of XO Core, Terminology, and Packaging in the
past week. We also saw a number of new languages getting added,
including Bengali, Dzongkha, and Persian.

Sayamindu fixed a bug in Sugar that caused PO files to be merged when
"setup.py genpot" is called (Ticket #4995). Separating the updating of
the POT files from PO files will presumably cause less confusion among
translators. Sayamindu helped setup support for localization for the
Watch and Listen activity. Reinier tokenized internationalization of
the help descriptions in Calculate.

6. Updater: C. Scott Ananian continued work on making olpc-update more
robust and user-friendly—the Libertas wireless improvements
integrating have helped in regard to robustness. A new wrapper around
olpc-update, called olpc-update-query, asks a central server what
build a laptop should be running, and runs an update if necessary
(olpc-update-query is being run in a cron job in joyride builds; stay
tuned to devel@ for instructions on how to subscribe to one of the
streams at https://activation.laptop.org/streams/).

7. Scott also wrote a patch to olpc-utils to make requesting a
developer key easy (Ticket #4271). This patch still requires some help
from to make it into the builds in a useful fashion (Please see
http://dev.laptop.org/git?p=users/cscott/olpc-utils-tmp;a=commit;h=5377149e8a326d1120f36669697561ac47ef83b9).

Marco wrote some scripts to automatically build git snapshots for
Joyride (using mock). Tomeu Vizoso wrote test cases so we detect more
easily when we break compatibility with the current release.

8. X Window System: Bernardo Innocenti tried to improve upon his fix
for the the "double mouse click" bug, but eventually gave up and
rolled an updated kernel containing only the one liner workaround. He
will investigate this further after the frenzy around Ship.2. Bernie
made numerous improvements to olpc-utils, including support for
determining the XKB settings from the new mfg-data tags. He migrated
the xkeyboard-config RPM in Koji; Dennis Gilmore helped with the
merging some more X11 packages. As a result, the number of rogue
packages has been reduced considerably. Bernie iterated over his bug
list and closed a number of internationalization- and keyboard-related
tickets that he had been deferring because the higher priority bugs
were always taking precedence.

9. Presence service: Guillaume Desmottes fixed a regression in the
Presence Service (PS) involving a timeout waiting for the roster to be
sent over the network (Ticket #4896); he fixed a bug in Salut
regarding bad handles (Ticket #5015); he modified the PS to accept
subscriptions coming only from trusted servers (Ticket #4993); and he
checked the status of PS patches in Update.1, enabling him to close
out a number of tickets. He implemented Activity.ListChannels in the
PS and wrote a Sugar patch using this new API (Ticket #5079).
Guillaume also began chasing down activities that still are using the
old sharing API and that don't catch TypeError when calling
get_preferred_connection; he is filling in more "could you use new
sugar sharing API" tickets. Finally, he began an investigation into
ejabberd's external component system.

Sjoerd Simons performed simulations and tests with Salut and worked on
some fixes for problems that those turned up there.

Morgan Collett is testing sharing within Rainbow; he chased down
issues with Buddy Handle tracking in the PS (Ticket #4920). He worked
with Simon and Guillaume on Chat's handling of URLs (within Rainbow we
cannot launch Browse directly—the workaround is to use the clipboard
(Ticket #5080).

10. Power management: Chris Ball continued to improve our power
manager: it now checks to see whether the Linux tty console is active
before deciding to suspend; it turns off wake-on-wireless when in
"sleep" mode; and has better alarm handling. Next up is inhibiting
suspend-on-idle when the CPU is extremely busy.

-walter
-- 
Walter Bender
One Laptop per Child
http://laptop.org