Add a ReaderLock to Route::process_output_buffers().
But process_output_buffers() is always called with processor-lock
held. To avoid deadlocks, a processor WriterLock must always imply
a process-lock (IFF reconfigure-I/O is called with _processor_lock).
Otherwise: e.g.
* add_processor() -> takes processor-lock. set up and activate processor.
* simult. audio-engine process, process-lock -> call process_output_buffers() -> wait for processor-lock
* add_processor() continues -> calls reconfigure-io -> take process-lock -> deadlock.
fixes possible crash if a processor modifies port-count
1. a processor is inserted and activated with processor-lock held
2. only after that the process_lock() is taken, configure_processors() is called which reconfigures-IO
BUT if the processor that is inserted changes the channel count AND audio is processed before IOs are reconfigured
-> possible crash (invalid port-buffers)
To reproduce: Bus1 (2in, 3out), Bus2 (2in, 3out)
- add a send from Bus1 to Bus2,
- then add a processor to Bus1, just before the send which
increases the channel-count to 4 -> occasional crash or assert.
* clean up source (whitespace)
* fix speaker 3x3 matrix iteration
* update math to go along with Ardour Cartesian -- fixes rounding errors
* fix division by zero in cross_prod()
* disable old debug output
(NB PBD::spherical_to_cartesian() returns
3.7494e-33, 6.12323e-17, 1 for azimuth 90 elevation 90 distance 1
while it should return
0.000000, 0.000000, 1 for azimuth 90 elevation 90 distance 1
IOW cos(90.0 * 2.0 * M_PI / 360.0) != 0
Cause unknown. This is currently worked around check in vec_length()
)
Add a ReaderLock to Route::process_output_buffers().
But process_output_buffers() is always called with processor-lock
held. To avoid deadlocks, a processor WriterLock must always imply
a process-lock (IFF reconfigure-I/O is called with _processor_lock).
Otherwise: e.g.
* add_processor() -> takes processor-lock. set up and activate processor.
* simult. audio-engine process, process-lock -> call process_output_buffers() -> wait for processor-lock
* add_processor() continues -> calls reconfigure-io -> take process-lock -> deadlock.
fixes possible crash if a processor modifies port-count
1. a processor is inserted and activated with processor-lock held
2. only after that the process_lock() is taken, configure_processors() is called which reconfigures-IO
BUT if the processor that is inserted changes the channel count AND audio is processed before IOs are reconfigured
-> possible crash (invalid port-buffers)
To reproduce: Bus1 (2in, 3out), Bus2 (2in, 3out)
- add a send from Bus1 to Bus2,
- then add a processor to Bus1, just before the send which
increases the channel-count to 4 -> occasional crash or assert.
