Verify that 'bits' really is one of 8, 12 or 16 before decoding
the data. It's probably always the case (unless the data files are
damaged) but if it isn't we'll either try to queue NULL to the
audio stream, or queue the same buffer more than once, or free the
buffer more than once. All of which are bad, though Coverity only
noticed the last of these cases.
1. Remove _sc55 bool. All Roland GS-capable devices have MT-32 sound/drum maps, so they should always be used when _enable_gs is set.
2. Always enable _native_mt32 if Roland GS mode is selected. I don't know why I never did this originally, since _enable_gs is automatically disabled for SCUMM v6+ games that use General MIDI tracks instead of MT-32 music.
3. Set master tune for GS devices to 442.0kHz. This is the master tune setting for the MT-32.
The initializeLocals() function assumes that it can copy 25
elements when being provided an array of values. But this array
was frequently a lot smaller than that. I've introduced a constant
for the number of locals (though VirtualMachineState has one more
for some reason), and fixed the array sizes in a number of places.
CID 1003951, 1003952, 1003953, 1003955, 1003956, 1003959, 1003960,
1003961, 1003963, 100394, 1003965
Added some comments to make it more readable. Switched usage of a temporary buffer and a double memcpy with a memmove.
Also fixed a potential out-of-bounds write, thanks to LordHoto.
Clang will produce plenty of warnings (most of them seem to be of
the "if the stars align the wrong way, this may happen" variety),
but I don't have the time or patience to look at all of them.
Unlike the EGA DOS version, it doesn't seem to be the scripts
that keep track of double-clicks. Instead, the engine sets a
variable to indicate them. Unlike the DOS version, we don't check
that the second click happens close enough on the screen to the
first one. We could, but it seemed needlessly complicated.
After listening to the original music in a Mac emulator (which
unfortunately doesn't handle the music very well), I can only
conclude that note value 1 means the note should continue playing.
At first I thought maybe it was supposed to fade the current note,
or perhaps change its volume, but I can't hear any traces of
either. So I'm going to assume it just means "hold the current
note", though for the life of me I cannot think of any valid
reason for such a command. So it may be wrong, but it sounds
closer to the emulator than it did before.
I completely forgot to delete the dummy iMUSE object after using it
to skip over the old music save information. Thanks to Lordhoto for
pointing this out.
Initialise _channel[] even when the instruments aren't available.
Otherwise, ScummVM will crash in a number of places including,
but not limited to, when loading savegames.
After some discussion on #scummvm, the player now locks the sound
resource while the music is playing. This prevents the resource
manager from expiring the resource, which at best could cause
music to restart where it shouldn't.. At worst, I guess it could
have crashed, but I never saw that happen.
In looped music, prevent the music channels from drifting out of
sync over time. This was noticeable after a few minutes in the
SCUMM Bar. We do this by extending the last note (which is just
zeroes, so we didn't even use to play it) so that it has the
exact number of samples needed to make all channels the exact
same length. (This is calculated when the music is loaded, so it
does not need any extra data in the save games, thankfully.)
As a result, the getNextNote() is now responsible for converting
the duration to number of samples (out of necessity) and for
converting the note to a pitch modifier (out of symmetry). I made
several false starts before I realized how much easier it would
be this way.
At least on my computer, when the note ended abruptly there would
be an annoying "pop" at the end. This was particularly noticeable
at the end of the distaff notes in Loom. To get around this, fade
out the last 100 samples. There's nothing magical about 100 in
particular, but it's a nice even number and it should be short
enough that it's never a noticeable part of the note, even at low
sample rates.
Mac Loom's drafts appear to be stored from variable 55 and upwards.
I'm working under the assumption that there's either only one
version of Loom for the Mac, or that they all behave the same. I
could be wrong about that.
