In March, b67c2d72d6 moved some MIDI
initialization from the main thread to the MIDI thread. This caused
MidiPlayer_Midi::sysEx() to run on the MIDI thread for the first time.
This is a problem because it calls SciEngine:sleep(), which polls
events, and that causes MacOS to throw an exception for calling
SDL_PollEvent() on a non-main thread.
While investigating, it also turns out that MidiPlayer_Midi::sysEx()
and MidiPlayer_Fb01::sysEx() were calling OSystem::updateScreen(),
and that also shouldn't be happening on a non-main thread.
Now SciEngine::sleep() is only called on the main thread, and
OSystem::delayMillis() is called on the MIDI timer thread.
Continuing to call sleep() on the main thread keeps the UI responsive
when loading patches, which can take several seconds.
The OSystem::updateScreen() calls had no effect and have been removed.
Fixes bug #12947
- The mutex was added to avoid the triggering of the assert in backends/midi/windows.cpp, line 95. Meanwhile, this issue has been addressed differently.
- SCI does not per se require a mutex for the sound drivers. The engine is mostly thread-safe by avoiding driver calls through the main thread.
Fix SCI0 (LATE/EARLY) variants of the driver which were broken (didn't play at all). This might be of my own doing, since I introduced the initTrack() method into the driver class and the fb01 driver didn't have one so far.
SCI1 didn't seem to require much fixing. I modified some things according to my findings in the driver disasms. QFG2 and JONES seem to be fine. I am not too happy with KQ5. It has volume issues, but they might be present in the original, too.
I also added an isOpen() check and a mutex to avoid threading issues. When aborting SCI (either quitting ScummVM or returning ot the launcher) while using the fb01 driver I frequently (more often than not) got the assert from backends/midi/windows.cpp, line 95. This fixes that.
I've done plenty of checks and experiments with the sound bank initialization. But I found no bugs or possible improvements there. Hard to tell whether the sound is right. That device seems to have a mind of its own...
Fix up SCI0_LATE variant of the driver so that it actually plays anything (and even correctly). SCI1 hasn't seen any testing from me so far. I don't know whether that version works. Same for SCI0_EARLY...
Many engines follow the advice in audio/midiparser.h and create a
"pseudo-MidiDriver" subclass. But MidiParser really only needs a tiny
subset of the MidiDriver capabilities, namely those found in
MidiDriver_BASE. So we try to subclass from that whenever possible; this
allows us to remove many stub methods, and enables further future
simplifications.
Donovan Watteau
Eugene Sandulenko
Orgad Shaneh
sluicebox
Filippos Karapetis
Bastien Bouclet
athrxx
Colin Snover
Johannes Schickel
strangerke
Ori Avtalion
Max Horn
Walter van Niftrik