ci(ra4m1): drop upstream PR #16409 checkout, carry USB LDO VDCEN as patch 0002

[cumin777]

Problem (ref #28)

The workflow deleted the lib/micropython submodule, cloned micropython fresh, checked out upstream PR #16409 (WEACT_RA4M1_CORE board profile — open, unmerged since 2024-12-13), and patched main.c/mpconfigport.h with two line-addressed seds. Depending on a 2-year-old open PR is the fragile pattern #28 set out to remove.

What the board actually needs from #16409 — only 2 hunks

PR #16409 = 23 WEACT board-profile files + a USB LDO VDCEN enable:

• Board profile: not needed. This repo already supplies the XIAO board via boards/seeed/xiao_ra4m1 (full FSP set: ra_cfg/, ra_gen/, linker, pins), and the renesas-ra Makefile natively supports BOARD_DIR.
• USB LDO VDCEN: needed. boards/seeed/xiao_ra4m1/ra_gen/bsp_clock_cfg.h sets USB_CFG_LDO_REGULATOR (USB_CFG_ENABLE) — USB is powered from the internal LDO, so USBMC.VDCEN must be set for enumeration. Carried as micropython/patches/v1.27.0/0002-renesas-ra-usb-ldo-vdcen.patch, verified with git apply --check on the v1.27.0 tree after 0001 (anchors identical to the PR's: main.c:366, mphalport.h:64).

Everything else is upstream at v1.27.0: RA4M1 family (EK_RA4M1, RA4M1_CLICKER, ra4m1_af.csv), BOARD_DIR support. Both old seds are obsolete: main.c now ships mp_cstack_init_with_top() as-is; MICROPY_FATFS_ENABLE_LFN is (2) (enabled, dynamic buffer — better than the forced (1)).

Cross-impact of growing the shared patch dir

0002 touches only ports/renesas-ra/* (disjoint from 0001's ports/zephyr/*), is guarded by #if USB_CFG_LDO_REGULATOR == USB_CFG_ENABLE, and the README now documents the disjoint-file-set contract. Jobs that don't compile renesas-ra apply it inertly (and vice versa for 0001 on this job).

Scope

• removed: fresh clone + gh pr checkout 16409 + both seds
• added: ./tools/apply_micropython_patches.sh, patch 0002, README section
• unchanged: make BOARD_DIR=../../../../boards/seeed/xiao_ra4m1, toolchain, board files

⚠️ Same caveat as #37: compiled, not yet hardware-tested — the VDCEN behavior (USB enumeration) specifically deserves a smoke test on a real XIAO RA4M1.

🤖 Generated with Claude Code

[cumin777]

Environment change notes (pre-merge summary)

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

The workflow deleted the lib/micropython submodule and rebuilt the source from scratch every run:

cd lib && rm -r micropython || true
git clone https://github.com/micropython/micropython.git
cd micropython && git submodule update --init --recursive && gh pr checkout 16409
sed -i '301,302c\    mp_cstack_init_with_top(...)' main.c
sed -i '170s/#define MICROPY_FATFS_ENABLE_LFN.*/...(1)/' mpconfigport.h
make BOARD_DIR=../../../../boards/seeed/xiao_ra4m1

• PR #16409 (WEACT_RA4M1_CORE board profile) has been open and unmerged since 2024-12-13 — a two-year dependency on someone else's pending PR, the exact fragile pattern issue Unify the MicroPython build environment; drop personal-fork / open-PR dependencies (CI red for 4 boards) #28 set out to remove.
• Only 2 hunks of that PR are actually load-bearing for us: the USB LDO VDCEN enable (the XIAO RA4M1 powers USB from the internal LDO — USB_CFG_LDO_REGULATOR = ENABLE in its FSP config — so USBMC.VDCEN must be set for enumeration). The 23-file WEACT board profile was only used as a path crutch.
• Hidden landmine: the XIAO board dir was copied from that PR and still referenced boards/WEACT_RA4M1_CORE/... paths (LD_FILES, FROZEN_MANIFEST) that resolved only because the PR checkout put the WEACT board in-tree. Removing the checkout exposed it (first CI run failed at freeze time) — fixed in the second commit by pointing at the board's own files via $(BOARD_DIR).
• The two line-addressed seds patched a then-current micropython master; both changes are upstream at v1.27.0 (mp_cstack_init_with_top() as-is; MICROPY_FATFS_ENABLE_LFN (2)).

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. New patch 0002-renesas-ra-usb-ldo-vdcen.patch carries the PR's 2 genuinely-needed hunks (verified with git apply --check on the v1.27.0 tree, after 0001). The RA4M1 chip itself is fully upstream at v1.27.0 (EK_RA4M1, RA4M1_CLICKER, ra4m1_af.csv, native BOARD_DIR support). make BOARD_DIR=... and the board files are unchanged (except the $(BOARD_DIR) path fix).

Test status / caveat:

✅ CI compiles green (artifact: firmware.hex), and the shared patch set (0001+0002) was confirmed side-effect-free on the nRF54LM20A/B jobs in this run.

⚠️ No on-hardware validation yet. The VDCEN change specifically affects USB enumeration — flash a real XIAO RA4M1 and confirm the USB CDC REPL comes up before shipping a release.