Epic A: Stop depending on patched and external config #153

Activity

1. Strycher commented on Sep 7, 2026

   

   Strycher

   on Sep 7, 2026

   OwnerAuthor

   More actions

   _{@Strycher — 2026-08-09}

   Epic A complete — human sign-off received

   Ben, 2026-08-09: integration test #160 "functionally complete and tollgate passed", followed by explicit authorisation to complete the epic.

   Children

   Issue	Outcome
   #157	Rotation patch replaced by setRotation(3) in our own source. TFT_eSPI is stock — the vendor-source fork is gone
   #158	lv_conf.h vendored to include/, with scripts/sync-lv-conf.py as a documented bridge and drift check
   #159	Library + toolchain manifest recorded from the compiler's own resolution
   #160	Integration test — verified on hardware

   What the epic actually delivered

   The goal was to stop the build depending on files outside git. The blocking item was TFT_eSPI/TFT_Drivers/ST7796_Rotation.h, a patched vendor source file that existed on exactly one machine and that any library upgrade would silently revert.

   That is resolved without forking the library. The patch turned out to be a pure permutation — its case 1 emitted the identical MADCTL to stock case 3 — so one line of our own source replaced a maintained fork. Confirmed empirically: restoring the stock driver without the source change flipped the display exactly 180 degrees, precisely the MX|MY delta the analysis predicted.

   What it uncovered

   The epic was scoped as "vendor three files." Four unrelated defects were in the way:

   • build: main no longer links — DRAM overflowed by 4296 bytes #164 — main did not link at all. dram0_0_seg overflowed by 4,296 bytes. Not a code regression: the source was untouched since 2026-03-02 and the environment drifted underneath it. Measured to a single symbol, LVGL's 160KB static heap in internal DRAM, and fixed by routing LVGL's allocator to PSRAM. DRAM went from over-by-4,296 to 103,096 bytes (31%).
   • bug: SPI bus down — TFT, SD and FRAM all dead; every other bus healthy #166 — the entire SPI bus was dead. TFT, SD and FRAM all down while every I2C device, the UART and WiFi were healthy. Root cause: a wire in the wrong pin on the board (SCK). Diagnosed from the device's own /diags endpoint by reading failures by bus rather than by device.
   • bug: missing SD silently skips settings, mag cal, weather history and brightness at boot #168 — a missing SD card silently skips settings, mag calibration, weather history and saved brightness at boot, with no error surfaced. Open, P2.
   • docs: web diagnostics suite is undocumented — add docs/DIAGNOSTICS.md #169 — the whole web diagnostics suite was undocumented. Now docs/DIAGNOSTICS.md, referenced from CLAUDE.md.

   Known gap, carried forward deliberately

   lv_conf.h is still copied into the sketchbook by the sync bridge, and TFT_eSPI's User_Setup_Select.h and Setup_Field_Compass_ESP32S3_ST7789.h still live inside the library folder. arduino-cli cannot serve either from the repo without a machine-specific absolute path.

   Epic #154 retires both via build_flags = -I include -D LV_CONF_INCLUDE_SIMPLE (task B2). Signed off with the gap explicit rather than claimed as clean.

   Unblocks

   Epic #154 — PlatformIO / pioarduino migration.

   Two notes for whoever starts it:

   1. Port pio-flash device-identity guard from meshcore-firmware — prevent wrong-device flashing #172 (pio-flash device-identity guard) should land before B7: Epic B integration test — PlatformIO build on hardware #161. Epic B's integration test is itself a hardware flash, and this machine enumerates six serial devices.
   2. B6's parity baseline now exists. It was specified against "the known-good arduino-cli binary", which did not exist when the plan was written. It does now — the image currently running: flash 1,815,046 (62%), DRAM 103,096 (31%). Capture it before Epic B changes the toolchain.

   Agent: IndigoForge (session b5521fc0)
2. Strycher

   closed this as completedon Sep 7, 2026
3. Strycher

   added

   board:doneBoard sync: Done columnBoard sync: Done column

   priority:P0Priority: P0 - CriticalPriority: P0 - Critical

   type:epicParent issue with sub-issuesParent issue with sub-issues

   on Sep 7, 2026
4. Strycher

   added a parent issue on Sep 7, 2026

   • Feature: PlatformIO migration + hardware abstraction layer #151