[repeat98]

Draft: waits for the owner's unit test (a dev twin's crashtest). Not mergeable before that.

What

A crash record today holds registers and the stack but no module (crash.c), and diagnostics:pull prints only pc 0x…. Modules are placed at run time, so an address alone names nothing.

• The lookup. After crash.s writes the record, it calls crash.c's modwerk_crash_owners on a stack of its own (the faulting one may be what broke). That looks up the PC and the eight stack longs, in both SDRAM aliases, with a new loader function, modwerk_runtime_owner_module. It reads only the base's own owner records, never the pool. It writes {"MWCO", count, PC's module, eight stack modules} right after the record at 0x100ff8b4 (44 bytes, below the stall record at 0x100ff900).
• If the lookup itself faults, crash.s comes back for the next count, finds the block marked busy and restarts without names. It never faults again.
• At the next start, diag.c puts the PC's module in the CRASH incident's screen field and the distinct stack modules in its 56-byte id run, as a set record does. So the names travel with the incidents the Worker has always taken.
• Printing. diagnostics:pull prints crash pc 0x… in Mini Verb, stack through SC Ringmod, and device.mjs crash prints the owners. An older base's crash prints as before.

Base rebuilt with --beta --write. Base-bound packages relinked; the write-lease entries did not move, so Analog BD's reviewed recipe stays. docs/OCTATRACK_DRIVING.md lists the new battery RAM range.

How I checked

• test_crash.c (new): owners in both aliases, for the record's count, and re-entry skipped.
• test_diag.c: the CRASH incident carries the ids, and stale owners from another count are ignored.
• host_test.c (loader): any byte of the extent matches, nothing past it, and nothing once the module is discarded.
• diagnostics-pull.test.mjs: names, and an older base's crash.
• npm run pr:ready: Ready.

Not tested

• Hardware: not tested. No unit was available. The plan: RAM-boot a --dev twin, run crashtest, then crash. It should show the record and modules: PC 00000000 (crashtest runs base code), and the restart must still work. The owner power-cycles after the crash restart.
• A crash inside a module can't be made on purpose today, so naming on hardware is proven by the first real one.

Base batch

#1320 (the module store) also regenerates the base files. Whichever merges second merges main, then rebuilds the base and the base-bound packages.

Guides

docs/OCTATRACK_DRIVING.md names the owners' range. Nothing else was missing.

🤖 Generated with Claude Code