add thread safety to queue access in GPIOQueue - fix #975

[chaddcw]

No description provided.

[bennuttall]

Thanks! @waveform80 will you take a look?

[jogi-k]

Suddenly - after about 10 month working without any problems 24/7 - this issue occurred reproducible in my application, it didn't work any longer. I applied the pull-request on my installation. It works perfectly.

👍

[lurch]

Hmmm, if the value is in self.ignore, then I guess self.queue won't get appended to, and therefore len(self.queue) won't change, and therefore there's no chance of self.full getting set either?
I.e. perhaps this function could change to:

    def fill(self):
        try:
            while not self.stopping.wait(self.sample_wait):
                value = self.parent._read()
                if value not in self.ignore:
                    with self.lock:
                        self.queue.append(value)
                        if not self.full.is_set() and len(self.queue) >= self.queue.maxlen:
                            self.full.set()
                    if (self.partial or self.full.is_set()) and isinstance(self.parent, EventsMixin):
                        self.parent._fire_events(self.parent.pin_factory.ticks(), self.parent.is_active)
        except ReferenceError:
            # Parent is dead; time to die!
            pass

?

[chaddcw]

Moving the middle if in the function inside the first lock (and eliminating the second with self.lock:) looks like a good change.

I'm not sure about that last if in the function being protected by if value not in self.ignore:

The existing code allows self.parent._fire_events() to be activated when value is in self.ignore and self.partial is set. The suggested change won't allow that to happen.

Thoughts?