Fix unstable audio channel order

[justjanne]

We used to order audio channels using enumerate(set(t.split("+"))). In Python, the iteration order for sets is random and changes with every run.

This commit uses a stable order (as specified in the config) instead

[justjanne]

Demonstration of the issue:

test.py

streams = [None] * 10
for i, channel in enumerate(set("0+1+2+3+4+5+6+7+8+9".split("+"))):
    streams[i] = channel
print(streams)

Results in:

$ python3 test.py
['3', '0', '2', '5', '6', '1', '7', '9', '8', '4']

Every run has a different result:

$ for i in {0..10}; do python3 test.py; done
['3', '5', '9', '0', '1', '2', '6', '8', '4', '7']
['5', '4', '8', '6', '0', '1', '2', '9', '3', '7']
['8', '5', '3', '6', '0', '7', '9', '2', '1', '4']
['3', '7', '2', '1', '8', '5', '9', '6', '0', '4']
['6', '8', '0', '2', '3', '9', '7', '4', '5', '1']
['1', '8', '4', '7', '0', '3', '6', '2', '5', '9']
['9', '3', '4', '0', '7', '6', '8', '5', '1', '2']
['5', '1', '7', '4', '3', '0', '6', '9', '8', '2']
['1', '4', '5', '7', '9', '6', '0', '8', '3', '2']
['9', '5', '1', '2', '8', '7', '0', '6', '4', '3']
['7', '4', '0', '1', '5', '2', '6', '8', '3', '9']

[justjanne]

The code that runs is this:

for i, channel in enumerate(set(t.split("+"))):
    name = source if use_source_as_name else r.group(1)
    if self.has_stream(name):
        log.error("input audio stream name '%s' can't be addressed a second time within source '%s'", name, source)
    else:
        audiostreams.append(AudioStream(source, i, name, channel))

Assume we've got this configuration:

audio.cam1 = "0+1"

Then it gets parsed by t.split("+") into ["0", "1"]

Now set() deduplicates duplicate entries, but the order of sets is different every time python is launched, so there's a 50/50 chance for either ["0", "1"] or ["1", "0"]

Afterwards enumerate turns that into either [(0, "0"), (1, "1")] or [(0, "1"), (1, "0")].

Which means we either execute

audiostreams.append(AudioStream(source="cam1", channel=0, name="cam1", source_channel="0"))
audiostreams.append(AudioStream(source="cam1", channel=1, name="cam1", source_channel="1"))

or

audiostreams.append(AudioStream(source="cam1", channel=1, name="cam1", source_channel="0"))
audiostreams.append(AudioStream(source="cam1", channel=0, name="cam1", source_channel="1"))

Which then gets used in the audio mix matrix, which now has a 50/50 chance to be either

1 0
0 1

or

0 1
1 0

[Kunsi]

Thanks!