I'm largely writing this update just to keep everyone informed in lieu of
submitting a fix, because I've made no real progress. I now know a bunch
of things that *don't* fix it, but nothing that really does.
I have nailed down the symptoms better, but can't really make sense of
them. Valgrind shows me that the problem always begins with a "Syscall
param write(buf) points to uninitialised byte(s)" inside
/lib/ld-2.11.2.so, inside snd_seq_drain_output, inside startClocks, inside
startClocksApproved (delayed for Jack). Then comes a cascade of "Invalid
write of size 4", then we crash.
But we're not passing it that anything that can fail to be initialized!
We pass the fatal call two things:
* The same m_midiHandle we use for a driver's entire lifetime.
* A NULL event. Alsa docs say this is OK.
I looked at the other events - I test it on a file of 72 simultaneous
notes - and every note and controller we ever pass to Alsa really, really
looks to be initialized correctly.
The closest thing I have to a fix is what I tried about the race
condition. That makes it a lot rarer, but it still happens. I actually
wait for clocks after STARTING_TO_PLAY, not before, because the sequencer
wants to starts clocks after we give the driver its first mapped events.
I have a few more things to try:
* Some types like MappedEvent::Audio are passed thru like events but do
nothing. They might be uninitialized. Now that I understand Alsa
better, I see that they don't belong. I'll remove them and see what
happens.
* I have a desperate hypothesis that the race condition might result in
half-written events. I hope that's not it, because that's tough to fix.
Other than that, I can't see how to fix it.
Tom Breton (Tehom)
------------------------------------------------------------------------------
Virtualization & Cloud Management Using Capacity Planning
Cloud computing makes use of virtualization - but cloud computing
also focuses on allowing computing to be delivered as a service.
http://www.accelacomm.com/jaw/sfnl/114/51521223/
_______________________________________________
Rosegarden-devel mailing list
[email protected] - use the link below to unsubscribe
https://lists.sourceforge.net/lists/listinfo/rosegarden-devel
