devices/liquidsfz/liquidsfz.cc: fix for LiquidSFZ race cond by Stefan…

… Westerfeld, #44

#44

swesterfeld commented Jun 18, 2024:
>
> tim-janik commented Jun 18, 2024:
> > all input events must always be processed in render
>
> I agree. At least the filename changes, midi events can safely be dropped while a file is being loaded.
>
> Your patch is not good, because it will introduce race conditions of both threads accessing the same data.
> Since we cannot use locks due to RT constraints, I guess the real fix will be to use a lock-free queue to send events to the loader thread.
> I'll look into it and come up with a real fix and submit a PR.
> However, here is a quick fix that just avoids the initial load problem and is still safe as in it doesn't introduce race conditions

Signed-off-by: Tim Janik <[email protected]>

devices/liquidsfz/liquidsfz.cc

@@ -73,7 +73,7 @@ class LiquidSFZLoader
   {
     if (state_.load() == STATE_IDLE)
       {
-        if (want_sfz_ == have_sfz_ && want_sample_rate_ == have_sample_rate_)
+        if (want_sfz_ == have_sfz_ && (want_sample_rate_ == have_sample_rate_ || want_sfz_ == ""))
           return true;
       }
     state_.store (STATE_LOAD);