Caacupé war bullettin -- day 1

Bernie Innocenti bernie at
Thu Jul 1 02:44:55 EDT 2010

Today we started testing of F11-0.88 with one 6th grade classroom in
Caacupé. Within a couple of hours from updating the laptops, an
overwhelmed amount of new bugs popped up:

* Restore from XS does not make system reboot.
tch sent a fix today, I need to build new Sugar packages.

* Journal does not reindex after restore
Silbe and aa helped me diagnose this. aa is working on a fix.
Meanwhile, I temporarily dropped the sort-by-size patch series.

* Touchpad control panel does not show up with new kernel
Dsd filed dlo#10191, pgf figured out why. we need to issue a new
olpc-utils package

* Missing spanish translation of backup strings
tch updated the po files, but there is plenty of line numbers
noise in the diffs. How do we usually handle po updates?

* Icon of backup is a laptop, should be a school
garymartin made a bunch of icons, need to select one ASAP

* Date not being updated
One laptop booted with clock set to the Epoch. Oddly, the lease
was accepted anyway. We need to figure out why the clock is not
being updated from the network. Why aren't we using ntp?

* When registration to XS fails, you get no error messages
Ditto. I'm not sure it happens in all cases, though...
(someone wants to pick this up?)

* Can't register to XS just after connecting to the AP for the first
time. Works after restarting Sugar. This may not be the same issue of
Python's urllib permanently caching the DNS lookup failure.

* Missing icons for the datastore sort feature
aa sent them later today, I need to integrate them

* New kernel still has "one-dot" boot hang bug
Did not see this yet, but we need to remember to apply the fixes from

* Missing spanish translation for the datastore sort feature
(Who could do this? aa?)

* When we uninstall an activity, we should remove its data dir as well.
This may sound controversial (think data loss), but if we don't laptops
become unusable after the user has tried out too many activities. In one
case, a user had 200MB of invisible junk in

* Another typical offender of free space is Browse, with >50MB of junk
in its data dir. We should consider reducing the cache to 5-10MB.

* Volumes don't disappear from journal volumes toolbar.
(tch fixed it today, need to rebuild Sugar packages).

* Plenty of datastore corruption issues.
These have been observed with journals created by 0.84, but the bugs are
probably still there in 0.88.

== Reflections ==

* We need to dedicate more resources to testing. Many of the above bugs
could have been avoided with better testing in the lab.

* The current test plan sucks (contains only trivial tests, it is not
representative of real-world usage)

* Many of these bugs are not regressions (i.e.: Sugar 0.84 and 0.82 were
also affected).

* We need a public test schoolserver. We could provide a VM on treehouse
and Carlos could install it, but schoolserver security is inadequate for
public Internet access. Ideas?

* Even if we could find and fix all bugs leading to datastore
corruption, we must deal with upgrades. The datastore need to become
more robust when dealing with various cases of corrupted entries and

* I have a macabre collection of datastores mangled in various ways.
Since they potentially contain privacy-sensitive data, I can't publish
this horror show. Contact me if you would like to analyze them.

* Datastore corruption is still the #1 cause of user and teacher
THIS PROBLEM. It makes Sugar look worse than Vista.

After an initial round of discussion, I'll transcribe the remaining bugs
in the F11-0.88 page so we don't forget. Anyone is also welcome to file
them as tickets in the OLPC/SL trackers. Much more importantly, anyone
is *very* welcome to work on a fix for one of these, but hurry up: we
have only one month before general availability.

   // Bernie Innocenti -
 \X/  Sugar Labs       -

More information about the Devel mailing list