Opus 5.5 crashed an ESP32, then debugged its own mistake

Stocko 🦾@_Stocko_

so fun story.. at work today opus 5.5 accidentally crashed some esp32 firmware into a tight boot loop by disabling the JTAG we don't currently use to free up enough internal RAM to enable the BLE stack

then it rebuilt the firmware to debug
but it couldn't flash because of the tight loop

then flashed it with blinker code (smaller binary gets in before the system reboots) and got the coredump
but lost the stack symbols because of the debug build it had made

then it found the thread by decompiling the binary pulled from the device. so it added debugging around that and flashed it to find the crash
but instead it got a corrupted stack in a completely different thread

then it noticed that stack was adjacent to the idle task stack and that disabling JTAG required 28 more bytes than the idle stack had during an interrupt and it would corrupt the _mesg field in the adjacent stack

it bumped the idle stack back up to 1.5K and JTAG could be disabled with no issue. then it ran git blame and found that someone 5 months ago had reduced the idle stack size from 1.5K to 1K to save internal memory.

then it turned off JTAG and moved the BLE controller blob into PSRAM with another config switch, saving 96K of internal memory

then i came back from dinner and was like..

"nice, remove the tombstones from the comments"