[tyeth]

Draft, for CI and hardware UAT only. Do not merge.

This branch exists to run CI and to produce a single FunHouse firmware with all the in-flight work combined. Every commit here is already up for review individually — review those, not this.

What is on it

Base: cfe5624e47 (upstream main, past 11.0.0-alpha.2).

commit	reviewed in
espressif: add audiopwmio.PWMAudioOut for the ESP32-S2	#33
synthio: stop MidiTrack treating every 0xF. byte as End of Track	#34
synthio: support running status in MidiTrack	#35
synthio: treat a Note On of velocity 0 as a Note Off in MidiTrack	#36
synthio: keep MidiTrack's timing accurate for long delays	#37
synthio: follow the MIDI file's tempo map in MidiTrack	#38

Build

adafruit_funhouse builds clean: 1,477,440 bytes flash used, 1,406,144 free; 217,256 bytes internal SRAM.

Status

Not yet working on hardware. The audiopwmio port stalled on a real FunHouse — playing stayed true with no audio and playback never completed. The cause was found by inspection (the sample refill was queued only from the ISR's buffer-exhaustion path, so a tick that found no ready buffer queued nothing and wedged permanently; and the queueing call is taskENTER_CRITICAL(), illegal from an ISR). That is fixed here, along with return-checking on all five gptimer_* calls so a timer that will not configure raises RuntimeError from play() instead of hanging. This has not been confirmed on hardware.

The synthio changes are verified against an independent MIDI parser on the host and build clean on arm and xtensa, but have also not been run on a device.

🤖 Generated with Claude Code

[tyeth]

Rebuilt on the internal-RAM fix (1f6a9ddbb3, #33).

The first hardware run failed at gptimer_register_event_callbacks with ESP_ERR_INVALID_ARG (0x102). CONFIG_GPTIMER_ISR_CACHE_SAFE is enabled in this build, which requires the callback to be in IRAM and user_data to be in internal RAM; the driver passed self, a MicroPython heap object, and port_malloc() allocates that heap from PSRAM on a 2MB board. All ISR-reachable state now lives in an internal-RAM control block, following the common-hal/frequencyio pattern.

Branch is now 7 commits on cfe5624e47. Builds clean for adafruit_funhouse: 1,477,824 bytes flash used, 1,405,760 free; 217,224 bytes internal SRAM, 77,688 free.

Still unverified on hardware.

[tyeth]

UAT passed. The combined build works on an Adafruit FunHouse: 440 Hz test tone, then a MIDI track played through audiopwmio on the onboard piezo to completion, clean decode (error_location: None).

Both halves of this branch are therefore exercised together on real hardware. Still a draft and still not for merge — the individual PRs #33–#38 are the reviewable units.

[tyeth]

UF2 for funhouse, with halloween midi file played through synthio using an audiopwmio output, loading the file after fixing a bit of midi from_file stuff.

funhouse_halloween_audiopwmio.zip

Code

# SPDX-FileCopyrightText: Copyright (c) 2026 Tyeth Gundry for Adafruit Industries
# SPDX-License-Identifier: MIT
"""FunHouse: 440 Hz test, then a polyphonic MIDI file via audiopwmio + synthio.

REQUIRES the custom firmware that adds audiopwmio.PWMAudioOut to the ESP32-S2.
Stock CircuitPython has no audiopwmio on espressif and this will raise
ImportError -- use the pwmio/adafruit_midi_parser version instead.

Unlike the piezo/pwmio version this is POLYPHONIC: synthio renders up to 12
voices (CIRCUITPY_SYNTHIO_MAX_CHANNELS on esp32s2) into a sample stream, which
audiopwmio pushes out of GPIO42 on a 78 kHz carrier.

The .mid must be prepared for synthio.from_file(), which is strict:
  * format 0, exactly one track
  * no running status, and no meta events at all (a tempo event is read as
    end-of-track and silently stops playback)
  * tempo is fixed at 2*division ticks/sec, so the real tempo must be baked
    into the delta times
Use make_synthio_midi.py to convert; a normal .mid will not work.
"""

import array
import math
import time

import audiocore
import audiopwmio
import board
import synthio
import time
time.sleep(3)

MIDI_FILE = "/melody.mid"
SAMPLE_RATE = 11025

dac = audiopwmio.PWMAudioOut(board.SPEAKER)

# --- 1. 440 Hz sine --------------------------------------------------------
length = 8000 // 440
sine_wave = array.array("H", [0] * length)
for i in range(length):
    sine_wave[i] = int(math.sin(math.pi * 2 * i / length) * (2**15) + 2**15)

print("440 Hz test tone")
dac.play(audiocore.RawSample(sine_wave, sample_rate=8000), loop=True)
time.sleep(1)
dac.stop()
time.sleep(0.5)

# --- 2. MIDI through synthio ----------------------------------------------
# Single-cycle waveform for the voices. Must be signed 16-bit ('h').
# Delete the waveform= argument for synthio's default 50% square, which is
# closer to a chip-organ and arguably more fitting for a toccata.
WAVE_LENGTH = 128
waveform = array.array("h", [0] * WAVE_LENGTH)
for i in range(WAVE_LENGTH):
    waveform[i] = int(math.sin(math.pi * 2 * i / WAVE_LENGTH) * (2**15 - 1))

envelope = synthio.Envelope(
    attack_time=0.03,
    decay_time=0.05,
    release_time=0.2,
    attack_level=1.0,
    sustain_level=0.8,
)

try:
    midi_data = open(MIDI_FILE, "rb")
except OSError:
    print("no", MIDI_FILE, "- nothing to play")
else:
    midi = synthio.from_file(
        midi_data,
        sample_rate=SAMPLE_RATE,
        waveform=waveform,
        envelope=envelope,
    )
    print("playing", MIDI_FILE, "at", midi.tempo, "ticks/sec")
    dac.play(midi)
    while dac.playing:
        pass
    # -1 means the track decoded cleanly; anything else is the byte offset
    # where MidiTrack gave up (usually running status or a meta event).
    print("error_location:", midi.error_location)
    midi.deinit()

print("stopped")
dac.deinit()

[tyeth-ai-assisted]

The zephyr-cp board failures here weren't caused by this PR. zephyr-cp's CI dependency list doesn't fetch lib/mp3 or extmod/ulab, and upstream normally gets them from a submodule cache hit, which this run missed. Tracked in adafruit#11655 and fixed by adafruit#11656. That fix is cherry-picked here (c8554c8), so this run should build those boards.