ci(mg24): drop dead-fork PR override, apply repo v1.27.0 patches for USER_C_MODULES

[cumin777]

Problem (ref #28)

Same dead-fork failure as #37: the workflow overrode lib/micropython with git reset --hard origin/master + gh pr checkout 18030, and upstream micropython PR #18030 was closed unmerged with its source fork since deleted → upload-pack: not our ref, CI red.

But unlike nRF54L15/nRF52840 in #37, this board genuinely uses the feature: it passes -DUSER_C_MODULES=src/cmodules/modadc;src/cmodules/modrtc, so the override can't just be deleted.

Fix

Build at the repo-wide pin (v1.27.0 submodule) and run tools/apply_micropython_patches.sh. The repo's shared 0001-zephyr-xiao-nrf54lm20b-runtime.patch already carries PR #18030's entire one-file diff:

include(${MICROPY_DIR}/py/usermod.cmake)     # in ports/zephyr/CMakeLists.txt
... ${MICROPY_SOURCE_USERMOD}                # into MICROPY_SOURCE_QSTR
... ${MICROPY_INC_USERMOD}                   # into include dirs

And py/usermod.cmake at v1.27.0 appends /micropython.cmake to directory entries (IS_DIRECTORY branch), so the existing -DUSER_C_MODULES="dir;dir" keeps working unchanged — our cmodules are already in native format (add_library INTERFACE + target_link_libraries(usermod ...)).

Also drops the sed '116s/4096)/4096 * 2)/' hack: v1.27.0 moved flash storage to zephyr_storage.c, which reads erase_block_size from the devicetree (4096 is only the DT_PROP_OR fallback) — the hardcoded doubling is superseded.

Cross-impact of the shared patch set on this job (checked)

• BT advertiser-option shim in 0001: version-guarded #ifdef (new/old Zephyr names) → fine on Zephyr v4.2.0
• xiao_nrf54lm20b_imu_power.c added to zephyr port sources: #if defined(CONFIG_BOARD_XIAO_NRF54LM20B_...) guard → empty translation unit on xiao_mg24
• modmachine/modtime/modzsensor/mpconfigport tweaks: port-wide, daily-verified on LM20A/B; MG24 × Zephyr v4.2.0 is the new combination CI will adjudicate

Scope

• removed: override block + main.c sed
• added: ./tools/apply_micropython_patches.sh
• unchanged: Zephyr v4.2.0, xiao_mg24 board, overlays/conf, USER_C_MODULES list

🤖 Generated with Claude Code

[cumin777]

Environment change notes (pre-merge summary)

What this board was building on before (the "old environment"):

The workflow overrode the lib/micropython submodule before compiling:

git reset --hard origin/master
gh pr checkout 18030     # upstream micropython PR, "zephyr: supports adding external user modules"
sed -i '116s/4096)/4096 * 2)/' ./main.c

• PR #18030 was closed unmerged (author-closed 2025-11-20) and its source fork has since been deleted, so gh pr checkout fails with upload-pack: not our ref — CI red since ~2026-07.
• Unlike nRF54L15/nRF52840 (ci(nrf54l15,nrf52840): drop dead-fork PR override, build at MicroPython v1.27.0 pin #37), this board genuinely used the PR's feature: -DUSER_C_MODULES=src/cmodules/modadc;src/cmodules/modrtc.
• Effective old baseline: fork-base master snapshot 9939565 (2025-08-26) + the PR's 3-line usermod wiring + a line-addressed sed doubling the hardcoded flash block size (4096 → 4096*2) in main.c.

What it builds on now:

The repo-wide pin — lib/micropython at v1.27.0 (78ff170) — plus tools/apply_micropython_patches.sh, same model as the green nRF54LM20A/B jobs. The shared 0001 patch already carries PR #18030's entire one-file diff, and v1.27.0's py/usermod.cmake appends /micropython.cmake to directory entries, so the existing -DUSER_C_MODULES="dir;dir" works unchanged. The sed hack is dropped: v1.27.0 moved flash storage to zephyr_storage.c, which reads erase_block_size from the devicetree (4096 is only the fallback). Zephyr v4.2.0, board and overlays unchanged.

Test status / caveat:

✅ CI compiles green (artifacts: zephyr.elf / zephyr.hex), including the usermod_adc / usermod_rtc modules being discovered and linked.

⚠️ No on-hardware validation yet. Recommend flashing and smoke-testing on a real XIAO MG24 (REPL over USB, import adc / import rtc, flash /flash filesystem with the new devicetree-derived block size) before shipping a release.