Read sharing can deadlock.

[mstone]

I olpc-update'd two B4s from 703 -> 704. Both have Read-44 installed and both are connected to a channel 1 XS Mesh with Salut running.

I resumed a previously created Read instance containing Phyla.pdf which I shared publicly from MS2 and joined from MS1. I switched away to do other work, then came back to the Read instances to find them frozen with dark grey screens containing only the activity title text entry box. I ran 'strace -p' on both Read pids and found that the two instances were deadlocked, both blocking in recv().

How might this have occurred?

[mstone]

I just olpc-updated from 706 -> 708 and registered both laptops. olpc-netstatus reports that both laptops are connected to schoolserver.laptop.org and that gabble is running happily.

I shared Read-44 on MS1 and joined from MS2. A few minutes later, with no visible progress, I straced both activities. This time, both are blocked in poll().

[mstone]

My terminology was a bit inaccurate; unlike the recv() deadlock, these read instances responded to 'Quit' events from Sugar. Presumably they were sleeping in poll() waiting such events. However, it seems that no data was being transfered. How could we have reached this state?

The activity logs are quite remarkable. :)

[gregorio]

Milestone Never Assigned deleted

[morgs]

Somehow the client got a 404 error page, which it saved as the document, and then evince didn't like text/html.

[morgs]

Fix pushed to git master. Will be considered for release post 8.2.0.

[morgs]

In unstable release 60.