CHG-0010: A Look for One Album, and a Strip of Eighteen

Summary

Two things, deployed together because they touch the same three programs.

The Dark Side of the Moon is a black sleeve with a thin spectrum across it, and its cover’s colors are not what that album should light a room with. After this change nowplayd says that album has a look, spectrum, and lightd lights its cover as six colors from red to violet instead of the cover’s own.

The stand has room for 18 LEDs across its front, not the 60 the firmware assumed. The firmware and the console’s drawing both become 18.

Goal

• nowplayd publishes eds/nowplaying/current/look beside every cover.
• lightd logs that it is listening for looks, and lights The Dark Side of the Moon as spectrum: six colors, the effect in LIGHTD_SPECTRUM_EFFECT.
• Every other album is lit from its cover as before, including the one played straight after.
• The stand reports 18 LEDs at start, and the console draws 18.

Scope

In scope: one topic granted to two accounts; the three services rebuilt; the stand reflashed for the strip’s length and wider chase patches.

Out of scope: the scene contract, which is unchanged: a look reaches the stand as an ordinary scene with an effect it already has. Spotify, which has no cover published and so no look. The LEDs themselves, still not attached.

Risk and impact

Procedure

make -C infra/deevnet-tenant-eds plan      # two accounts updated in place
make -C firmware/lp-stand test
make -C firmware/lp-stand build flash PORT=/dev/cu.usbserial-0001
make -C deploy deploy WHERE=builder

Verification

• eds/nowplaying/current/look holds a retained message naming the current cover’s hash.
• lightd logs listening for looks.
• The stand’s boot log names 18 LEDs.
• The console’s strip has 18 segments.
• The Dark Side of the Moon played in iTunes: lightd publishes a scene of six colors with look=spectrum, and the stand applies it. The next album is lit from its cover.

Undo

Remove LIGHTD_LOOK_TOPIC from deploy/lightd.env and run make -C deploy units: every album is lit from its cover. The grants can stay. The strip’s length is LP_LED_COUNT and CONSOLE_STRIP_LEDS.

Outcome

Every check under Verification holds, from the logs and the stand’s own reports. No LEDs are attached, so nobody has seen the spectrum on a strip.

Departures from the plan

• The apply came after the deployment, not before. For two minutes nowplayd published looks the broker refused without saying so. Nothing was lit wrongly; the album was lit as it would have been yesterday.
• The tag on this copy has no leading “The”. The rule allows for it, and it matched.
• The console’s page went stale after sitting idle, and this was fixed here. With iTunes paused for three quarters of an hour, the page did not show the next track for half a minute, and a reload took as long. The services were not late: each track was on the bus and on the stand within a second of the agent reporting it. The page’s stream had died without the browser knowing, and the page had no way to notice, because the console’s keep-alive was a comment a page cannot see. The keep-alive is now an event every 15 seconds, and a page that hears nothing for 40 reconnects, as it does on coming back to the tab or the network. That the stream was dead is inferred from the timings, not observed.

Follow-ups

• Look at spectrum on real LEDs, as sweep and as morph (LIGHTD_SPECTRUM_EFFECT), and say which it should be. That settles ADR-0019.
• Leave the console open and idle for an hour, then change the track: the page should follow within a few seconds, with no reload.
• The console is told the strip’s length. The stand could report it.