All model matching will be done simulator independent in
CContextSimulator. The simulator specific part is kept in the model
set.
This also caused the signal modelMatchingCompleted to be renamed to
aircraftRenderingChanged, since the name wasn't accurate anymore.
Both getInstalledModels(), getInstalledModelsCount() and iconForModel()
were removed from the ISimulator interface.
refs #765
The reason for moving the implementation out from CSoundGenerator into
its own class is, because CSoundGenerator was a very complex and
obscure class. It mixed many tasks in one place. CSelcalPlayer
is designed to play SELCALs only.
The following design changes have been made, compared to CSoundGenerator:
* Use pull mode instead of push mode. QBuffer is used as the QIODevice and
is a wrapper around QByteArray. Therefore it is not necessary to
implement our own QIODevice.
* Internally it uses a CThreadedSelcalPlayer to relieve the load of the
main thread. CThreadedSelcalPlayer inherits CContinuousWorker, no
low level QThread implementation was necessary.
* Push mode was not implemented.
* It is important that the QAudioOutput is allocated in the worker thread.
QAudioOutput allocates internal objects, which cannot be moved to
the worker thread.
* Data caching. The generated seclal audio data is cached.
refs #736
When a plugin is loaded, existing aircrafts should not be added
immediatly, but as soon as the simulator is simulating.
CContextSimulator also handles now the case that no simulator is
attached and all aircraft slots are now doing nothing instead of
raising an assert.
refs #549
CPluginManagerSimulator::createListener returns nullptr if no
listener could be created, e.g. because the plugin could not be loaded
due to missing dependencies. Handle this scenario properly instead of
raising an assert.
refs #669
Change MetarSet to MetarList:
CMetarSet was implemented as a collection. This only makes sense for values which have one member variable only or hardly ever change. METARs often change and therefore a collection didn’t make much sense.
Additional profiling showed that a sequence has better algorithmic performance.
