>>if there are encodings that
discard information, why are they of interest?
You've never worked on a mix stage, have you? 8-) All Dolby
Stereo movie mixes -- probably even the print of Star Wars you
watched way back when -- use such encodings. That's reality.
it may well be reality. but its also true that in reality there is
very little audio hardware that uses floating point representation
internally. despite this, we decided (?) to adopt floating point
internally because it makes sense, not because we feel obliged to
reflect "reality".
clearly there is a need for *an* encoding of multichannel streams. but
such an "encoding" doesn't need to be something like dolby stereo or
5.1 or whatever - it could just be a set of N channels of mono data.
by supporting multiple encodings, we massively increase the complexity
involved in using plugins together. hosts no longer have to just count
channels in and out, but are required to track encoding-handling
capabilities, splice in converters where necessary, etc.
there seems to be little reason to have any encoding floating around
inside a GMPI graph, and every reason to make it possible to feed any
encoding in and out of any leaf node. instead, the conversion gets done
once in each direction (per leaf node using the encoding). less work
for the host (and host author(s)), less work for plugin authors, less
hassle, less areas for subtle bugs to emerge.
i'm not saying we should handle the encodings, but that somewhat like
MIDI, the encodings should be present only at leaf nodes, never in the
core.
