I had a quick look at commit logs, from what I can see the last time the u235 object file was changed was 16th of March 2015, and the most recent changes to the code that calls the audio code around February 2015. Time files, does it not?!
I do remember having all sorts of weirdness with sound fx missing. I remember sh3 telling me that he used to fire each sample twice so as to be sure it'd actually play. So that's certainly an option. Your sndplay/sndfreq might be the same thing as sh3 does - basically the audio code needs to be prodded twice due to some timing issues? Also if you check nyandodge project I also do some utter weird things with audio (just don't look at the mod playing, that was the best I could come up with to switch between modules).
Finally the only thing I could think of that could probably possibly might affect sample playing was my turning on gcc optimisations so more optimal basic code would be generated. If you would like to test this out, you just need to change -O2 to -O0 in build.bat in the two instances around this point: https://github.com/g.../build.bat#L183
That's pretty much all I can do with a closed source library, sorry. I couldn't get any answers back then and in any case it's too complex of an issue to lay it to other people as there are many points of failure here, i.e. basic->c->assembly, interface with library running on Jerry which I'd have to disassemble. How do you even start to debug this?
Edited by ggn, Tue Oct 23, 2018 11:29 AM.