[HANG] DeviceReadBuffer leaks threads on error untill backend hangs

[Tony Lill <ajlill@…>]

If an error causes the DeviceReadBuffer? thread to exit on it's own, the Stop function doesn't reap the thread, and the backend eventually runs out and locks up. The attached patch uses the thread variable itself to determine if a thead_join needs to be done, rather than the running variable, which seems to have slightly different semantics.

[Tony Lill <ajlill@…>]

The correct patch

[danielk]

Thanks for catching this!

Can you rewrite this without using the thread variable value to determine if a join is needed? I don't believe that is portable.

[Github]

Refs #9927. Make sure we stop the DRB thread if we somehow reach the dtor with the thread running & also issue a wake if Start is called when the thread is already running.

Branch: master Changeset: a5d9c063e82ae4a45364ed57f8b5320b78857030

[Tony Lill <ajlill@…>]

Patch using new boolean instead of thread.

[Tony Lill <ajlill@…>]

Ok, I updated the patch to use a boolean instead of the thread variable.

I also looked at the code in master. It still suffers from the problem that IsRunning? tells you if the thread is active, not if it's been created and requires reaping to free up the resources. I don't know if the other changes to the code make that distinction irrelevant for the issue addressed in this bug, but I think it's still a problem.

I also think that there is potentially another problem in the while loop at the end of the Start function. If the ::run starts and exits before that loop can see the IsRunning? return true, it will be stuck there forever.

[Github]

Refs #9927. In DeviceReadBuffer?, make sure we check dorun when waiting for the run() to start in case it ran to completion (due to error) before we started waiting on it.

This also adds a QWaitCondition to avoid a tight loop waiting for the thread to start up.

Branch: master Changeset: c74387ced6ae2c0657e25278382283398cc141d9

[Github]

Refs #9927. Make sure DeviceReadBuffer::videodevice is never QString::null, this will cause the logging to segfault.

Branch: master Changeset: 9e5a094a0a3c1accfca1131b74f2c9173ba8cdf0

[danielk]

This was applied 0.24-fixes in [760c8db330134fbd4b084473bace157ea778aa27] but the ticket closing hook didn't fire :(