fix: Make the GPIO chip, IRQ backend, and enable pins configurable (#85)

[cwill747]

Fixes #85.

The bug

The 40-pin header sits on /dev/gpiochip0 for CM4, but on /dev/gpiochip15 for CM5 and Pi 5 kernels, where it hangs off the RP1. We hardcoded chip 0 in three places and never passed openhop_core's gpio_chip kwarg, so affected users had no supported way to point at the right chip — the only workaround was a udev symlink.

Changes

Three hardware settings, each with an env override, a Settings > Hardware control, and a passthrough in create_radio():

Setting	Env var	Effect
gpio_chip	MESHCORE_GPIO_CHIP	which /dev/gpiochipN to open
use_gpiod_backend	MESHCORE_USE_GPIOD_BACKEND	polled IRQ edges instead of edge interrupts
en_pins	MESHCORE_EN_PINS	GPIO pins driven HIGH at init to power the radio

These are deliberately not in HARDWARE_PRESETS: the chip number and IRQ backend are properties of the SoC and kernel, not of the radio board, so picking a board preset must not reset them. There's a test pinning that.

Diagnostics

So the next reporter doesn't have to guess:

• doctor reports the configured chip and lists what the host actually has, instead of asserting /dev/gpiochip0:

  [FAIL] gpiochip: Configured GPIO chip /dev/gpiochip0 not found — available: 11, 12, 13, 14, 15. Set it in Settings > Hardware or via MESHCORE_GPIO_CHIP (CM5/Pi 5 kernels usually put the 40-pin header on 15, not 0).
  

• Pre-flight gains a GPIO_CHIP conflict naming the available chips. A missing chip short-circuits the pin probes, which would otherwise emit one redundant conflict per pin for the same root cause.
• Pin probes use the configured chip, and now include the configured enable pins.

Note on use_gpiod_backend

It does not actually select libgpiod while python-periphery is installed. In GPIOPinManager.__init__ the GpiodGPIO class and the globals()["GPIO"] = GpiodGPIO swap are nested inside the if backend == "auto" branch, so passing backend="gpiod" explicitly never installs the wrapper — and the [hardware] extra always pulls in periphery.

What the flag really changes is edge detection, from kernel edge interrupts to a polling thread. That's a workaround for kernels that reject the edge request and leave the radio connected but deaf, which is plausibly the more useful half here. It's labelled "Poll IRQ" in the UI and documented accordingly. We do not depend on openhop_core's gpiod extra: it would never be imported, and it needs libgpiod headers to build on the Pi.

Dependency migration (first commit)

Upstream renamed both the repo (rightup/pyMC_core → openhop-dev/openhop_core) and the PyPI distribution (pymc-core → openhop-core). pymc-core is frozen at 1.0.12; further releases land under the new name, starting at 1.1.1.

Comparing the wheels, the rename is near-mechanical: the module tree is identical apart from three new files, SX1262Radio.__init__ is unchanged, and protocol/identity.py is byte-identical. Internal names renamed alongside the imports (PyMCCoreSession → OpenHopCoreSession, etc.). CHANGELOG entries are left referring to pymc-core, as they were accurate when written.

Also fixes a dead entry in _RADIO_LOGGER_SUBSTRINGS: the "pyMC" substring never matched anything, since the library's module loggers were named pymc_core.* (lowercase) and no logger was literally named pyMC.

Testing

• 136 passed / 1 skipped (17 new), ruff check + format clean, mypy unchanged at its 85-error baseline.
• Settings panel rendered under Xvfb: the new row displays and round-trips gpio_chip=15, en_pins=16,17, use_gpiod_backend=True; natural width stays 58px, so no runaway-resize regression.
• Not yet verified on affected hardware — neither reporter has confirmed a fix. Worth holding for that.

Follow-ups (not in scope here)

• doctor's spidev check is still hardcoded to /dev/spidev1.0 despite now having bus_id/cs_id in hand — wrong for the waveshare preset, but doesn't affect the Change gpiochip0 to gpiochip15 #85 reporters.
• GPIOPinManager warnings (including "Could not setup EN pin") still don't reach the radio-error path. Adding it to _RADIO_LOGGER_SUBSTRINGS would surface exactly the failures reporters hit blind, but broadens what becomes a toast.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2bd420e0b0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Honor the GPIO environment override in doctor

When MESHCORE_GPIO_CHIP=15 meshcore-console doctor is run on a host with an accessible settings database, SettingsStore.load() normally succeeds and this path returns the persisted/default chip (usually 0), so the environment fallback is never reached. Doctor therefore checks /dev/gpiochip0 and may report a failure even though the radio CLI uses load_runtime_config() and will use chip 15; apply the environment override before returning the persisted configuration.

AGENTS.md reference: AGENTS.md:L160-L162

Useful? React with 👍 / 👎.