A tap now moves to the next multiple of VOLUME_STEP rather than adding
it, so 23 goes to 25 and 25 goes to 30 and the levels stay round. The
panel counts taps and lets the speakers do the rounding from whatever
level they are actually at, since the phone's copy can be seconds old;
the same rule is mirrored in JS so the optimistic number never has to
correct itself when the reply lands.
The input picker also asks the AVR which sources are still switched on
(SSSOD ?) and leaves out the ones deleted in its setup menu. Sources it
does not mention are kept, so a model that ignores the command shows its
whole list rather than nothing; deleted sources also keep their names, in
case the AVR is sitting on one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`hidden` is only a UA stylesheet rule, so .sheet-wrap's display:flex beat
it and the sheet was never hidden: a full-screen scrim sat over the app
from the moment it loaded, blurring it and swallowing every tap, with an
empty option list because openSheet() had never run to fill it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Raspberry Pi OS refuses pip install into the system Python, so the venv
step is the difference between the README working and not. The systemd
unit points at that interpreter for the same reason.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The HEOS app is unpleasant to use for the four things that actually get
done in this house, so this is those four things on one screen: volume
for the Home 400 and the Living Room pair, joining either to the AVR,
splitting them off again, and picking the AVR's input.
The bridge's logic moves in mostly intact, split into a HEOS client, an
AVR client and a controller, with two grouping bugs fixed on the way:
* Merging a room into the AVR sent only the pair's leader pid, which
left the second Home 200 behind. Group targets now expand to every
member pid, and unmerging re-forms the pair rather than leaving two
lone speakers. The members are learned while the pair is visible and
remembered in members.json so a restart can still rebuild it.
* Volume for the pair used its gid, which stops existing the moment it
joins the AVR's group. It now falls back to the member players.
Both sockets are kept open rather than reconnecting per command, which
is what made every button press feel slow; the AVR connection doubles as
a listener, so input changes made with the physical remote show up too.
The original query-string endpoints still answer, so anything already
pointed at the bridge keeps working.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
franzzandClaude Opus 5