Formerly it tried to pass 'va_list' as '...' to debug. Now it's first creating
a String containing the output via String::vformat and then passing it
to debug.
Found by buildbot (master-dc). The warning message was:
audio/softsynth/mt32.cpp:65: warning: cannot pass objects of non-POD type
`struct va_list' through `...'; call will abort at runtime
Formerly the frequency was at 10000Hz. This resulted in 3,4 or even only 1
sample to be generated between callbacks on my system. All the other MIDI
drivers use a much lower frequency here. The MidiDriver_Emulated subclasses
use a frequency of 250Hz (by default) and the MidiDriver_MPU401 subclasses
(which are for example the ALSA output) use 100Hz. With the new frequency
of 250Hz 128 samples are generated between callbacks. This will hopefully
reduce the overhead of the MT-32 emulator (the engine's code was run 10000
times a second too) a bit.
I talked with KingGuppy and it seems the value was increased in the past
(still with the very old MT-32 emulator code) because there were accuracy
issues. However, I gave the lower frequency a quick test with the MI1, MI2
and ITE intro and didn't spot any obvious differences. As a result, KingGuppy
and I agreed to lower it back to 250Hz. If there are any problems coming up
we can still slightly increase the frequency to 1000Hz for example.
Thanks to waltervn for noticing this. Thanks to KingGuppy for discussion.
This fixes warnings that would appear after a little while in
Kyrandia 2, which would happen because every time the music
changed it would add a new loop point. This was probably harmless
because once the list was full it would keep re-using the last
element, and I imagine all songs were set to loop forever. But
this should be more correct, as I understand it.
This fixes a Protracker module in the OS/2 version of
Hopkins FBI (bug #3612101). In row 0x30 of the first
pattern, the set channel offset effect in the fourth track
pushes the offset past the sample (repeat) length.
This is not error; the mixing function already handles this
case flawlessly. No assert() is necessary there.
A large part of cadavre.mod is broken. No new notes play, and only
the old background sounds remain. It's possible, perhaps even
probable, that the original MOD player didn't have this problem,
but all standalone players I've tried do, so I'm assuming it's the
file that's broken. We work around it by changing the length of
the song after it's been loaded to only include the good parts.
This is a manual merge of a slightly adapted pull request #296.
The changes made are:
- Each time the theme format changes, the version was increased
- default.inc has been regenerated in the same commit as the theme changes
To help people familiar with Qsynth (I'm not, but it seems to be
one of the more polished FluidSynth front ends), use the same
presentation and terminology for the FluidSynth settings.
More to follow.
I don't really understand what these parameters do, or what the
sensible values are, so for now the sliders are limited only by
the allowed (or, in one case, "safe") values.
The MUNT debug messages are called from an audio callback, which is not allowed to
update the screen, as per the OSystem documentation in common/system.h:401
