EpgContainer: Fix Deadlock when calling Observer with lock

[fritsch]

Quick fix especially for v13. Longterm should use proper async handling.

Edit: This is also relevant for master of course.

[fritsch]

jenkins build this please

[fritsch]

Backtrace: http://paste.ubuntu.com/7479831/

[t-nelson]

Maybe do this against Gotham instead? Otherwise we probably pull our normal shit and not do the right thing later for master since there's no incentive.

[fritsch]

It's a bug in master, too. So a non deadlocking master is highly appreciated until @opdenkamp is less busy.

Edit: s/segfault/deadlock/

[jmarshallnz]

Ah, the famous "quick fix", huh? :)

What was the goal here? To get the notification outside the critical section? If so, there's actually a double critical section usage here (to be sure, to be sure?) See ln 673.

[fritsch]

Yes - you are perfectly right. That was in deed a quick fix and I simply overlooked that other section. The "fix" does not make any sense now. I will close that PR and leave it to the experts.

[jmarshallnz]

I think you can just drop the inside critical section (or the outside one if doing a premature update instead of correctly interpreting m_iNextEpgActiveTagCheck isn't an issue).

Then unlock the crit section prior to your current change.

@opdenkamp comments?

[FernetMenta]

I think you can safely move the other block too. the good thing about bReturn is that it always exits the methond at its end.

[fritsch]

jenkins build this please

[opdenkamp]

the diff is annoying to read like this on github, but looks okay :)

[t-nelson]

@opdenkamp fritsch@820b852?w=0

?w=0 is your friend :)

[opdenkamp]

oh wow, why isn't that enabled by default? thanks! :)

[jmarshallnz]

That is a neat trick :)