Motion tracking was storing a strong reference to the HRTF DSP node,
which was resulting in the DSP not shutting down or restarting properly,
leading to playback hanging.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This also fixes code signing for VGMStream bundle. All frameworks or
bundles which nest other frameworks or libraries must themselves be
signed first in building, not just on embedding.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Motion control, which requires macOS 14.0, was broken completely, ever
since the changes made to the code so it could compile on older Xcode
versions.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
And a bunch of potential memory leaks, and some misbehavior that could
occur due to not checking for errors properly.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Promote or demote all projects to Xcode 12.0, remove signing from any
libraries or frameworks that still have it, and clear identities in the
project files, so that they will derive from the out of repo local
settings file, and will stop inserting my ID into projects every time
they are modified.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Add setup error checking and failure states, and also add a failure
state for format changes while the device is already running, in case it
returns a nil format structure.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Add NSDictionary initWithOptionalObjects:forKeys:count: helper, to work
around several places that may have been setting nil dictionary valies
on placeholder dictionaries. Hopefully this fixes the crash someone
logged recently.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This code, based on some other vDSP information I found, performs the
DFT in a different fashion, and also pre-converts the spectrum data into
decibel levels, so the DeaDBeeF analyzer calculation code doesn't need
to convert the values itself any more.
Maybe this will also get rid of the use after free problem somewhere in
the audio code? Who knows?
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
A race on the source queue setter could cause a crash. Prevent that by
not attempting to set a nil input node.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Track fades could hang if fired in rapid succession. Now each fade will
be guaranteed to resume input feeding upon removing the current input to
the background fade-out queue.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This end of track restart notification may come from a different thread
than the processing thread. Wait on the thread to complete.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
A few places could have used nil checks, but there shouldn't be anything
passing nil URLs to these. Well, just in case, at least it won't lead to
crashes or something...
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This should fix an exception being thrown because the observer wasn't
registered, or known to be registered. Only register it when it will be
used, and only unregister it if it was registered.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Apparently, this doesn't work too well with real HDCD tracks, and causes
all sorts of weird volume issues. Just leave the volume alone, and let
HDCD decoding make tracks louder, possibly, rather than try to halve the
volume automatically on a detector that isn't terribly good.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
And disabled by default, at that. I can't actually hear the difference
of Peak Extension in the Rock track I have that claims to use it. And
Low Level Range Extension is more trouble than it's worth on tracks that
use it by mistake, or maliciously, if the case may be. I may add track
tag level control in the future.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Now using Accelerate methods to calculate and multiply ramps per channel
and add the remainder if necessary.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This code did not check the number of samples in a packet before adding
it to the output buffer, which apparently had the potential to cause the
output code to emit up to 512 samples of silence between tracks. This,
as one can guess, is a bad thing, and causes noticeable gapping between
tracks.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Output logging, a debugging feature that is only enabled at build time
if I need to chase down some audio mixing or output bug, was not logging
anything at all. Change to use Cocoa file writing methods, and actually
implement the output writer function again.
This code is left disabled 99% of the time anyway, and especially in
release builds. Like the node logging code elsewhere, it has the
potential to be very noisy and consume massive amounts of disk space.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This should be included, for safety purposes, in case the rounding up to
the nearest multiple of 256 samples doesn't bump the buffer size enough.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
The double members should be ordered first so they are aligned to an 8
byte boundary. The rest are fine as-is.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
The Visualization Manager PCM/FFT copy function was already observing
these parameters for null input and only returning output to the ones
which were not null. This just makes it clear that they are both
optional parameters. This is useful for future visualization adventures,
allowing PCM copy without invoking the FFT processing, or requesting FFT
without also having to keep the PCM.
This is mostly only a compile time change, and has no noticeable effect
on the current runtime behavior, as the only consumers of the Visualizer
data currently request both PCM and FFT.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
This affects User Defaults, but only has any effect on ChunkLists which
are being used for conversion, and only if they're processing DSD source
material. Thus, the observer should only be added on the one stream that
is converting DSD, and should definitely be removed when the object is
deallocated.
This fixes a serious crash bug that mostly appears to only affect Intel
Macs, and has no major side effects on Apple Silicon that I can tell.
It's a good thing I still own an Intel Mac or two to test on, even if
they are both trapped on older releases of macOS.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Apparently I somehow didn't notice this situation because I still had
Rubber Band enabled, and existing users kept it enabled ever since I
introduced it.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Move this commonly used string decoding helper to the CogAudio framework
and import it in every plugin that uses it.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Apparently, stringWithUTF8String: just returns nil when the encoding is
not UTF-8, rather than throwing an exception.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
The converter doesn't just require an output format call, it also
requires this input format change callback to actually signal it to
reopen the converter process with a new format setup.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Output format mostly requires stopping and restarting the output device,
and this also prevents us from using the latency function properly,
which apparently always returns 0 for output devices anyway. These
changes also prevent the output callback from hanging when resets occur.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Fix default output device logging, and also the preferences if no
default device happens to be set.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Fixes visualization latency under virtual machines, at least. Not sure
which local or native systems would be reporting high latency here, but
this should fix them as well.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
