In module-bluetooth-device.c in filter_cb() there's this piece of code:

    if (acquire)
        if (bt_transport_acquire(u, FALSE) >= 0) {
            if (u->source)
                pa_source_suspend(u->source, FALSE, PA_SUSPEND_IDLE|PA_SUSPEND_USER);

            if (u->sink)
                pa_sink_suspend(u->sink, FALSE, PA_SUSPEND_IDLE|PA_SUSPEND_USER);
        }

The acquire variable is set to true if the bluetooth audio state changed to "playing". This can happen without pulseaudio requesting it, so if we want to have strict equivalence "sink/source suspended" == "audio state is 'playing'" (and I think we want that), then it makes sense that module-bluetooth-device unsuspends the sink/source. The problem here is that there's no clean "force unsuspend" mechanism in pulseaudio. pa_sink_suspend(u->sink, FALSE, PA_SUSPEND_IDLE|PA_SUSPEND_USER) may not actually unsuspend the sink, because there are also other suspend reasons than just IDLE and USER. So, depending on the situation, the sink may or may not get unsuspended.

module-bluetooth-device could list all possible reasons in the pa_sink_suspend() call, but that would be ugly. When adding new suspend reasons, it would be very easy to forget to update the pa_sink_suspend() call in module-bluetooth-device, and besides, if those suspend reasons are originally set by some other code, the original reasons to suspend don't really disappear when module-bluetooth-device unsuspends the sink, the old reasons are just overridden.

I think there should be a function for forcing unsuspending. The semantics would be such that the sink gets unsuspended regardless of any standing suspend reasons, but all standing suspend reasons would be remembered, and once a new suspend request would be made, the forced unsuspend would cease to have effect, the sink would suspend and things would return to normal.

A related thing is that I would like to replace the single suspend reason bitfield in pa_sink with an individual suspend request object for each suspend request. The single suspend reason bitfield is prone to conflicts if multiple places use the same reason (for example, module-bluetooth-device uses the USER reason which is also used for other purposes).