- Most cost: a wrong error location. The NK1125 (a
ref Ty? may be missing) pointed at the start of the statement, not at result.clone(), and I didn’t know Ty::Fn.result is a Ty?. That cost about 4 turns: a probe program and two lower-std cycles, one of them wasted because my sed edit addressed the wrong line number. A second full build cycle went on a fourth probe (a literal match) that was outside the brief: the number_asked attempt didn’t help, and I reverted it.
- Lookups the brief could have handed me:
- The brief named the wrong site.
check_core_lambda_leave’s outer is the enclosing function’s expected type. The lambda’s declared type is declared in check_core_ops_closure (check_core_ops.nika about line 1052). Tracing the callers took 3 turns.
- The field shapes
Ty::Fn { params, result: Ty?, .. }.
- That
emit_expr_unmixed reads some_tails and wrapped on any node. This would have settled “verify the emitter” in one line.
- Checks run:
- Useful: four probes on the old binary before the first build (they confirmed the cause and gave me the unfixed baseline); one fixpoint check by md5 after
lower-std; one background gate (clippy, the guard set, tests/compiler at both settings), all green; pre-push-check.
- Wasted: the
NIKAIA_LOWER_CACHE_SALT rerun (it wasn’t the cache), the second build with number_asked, and two no-op “waiting” tool calls.
- The one change: a brief’s Locations line should say where the value the fix needs actually lives, checked by the architect with a grep of the callers. Here that is “the lambda’s declared type is
declared in check_core_ops_closure”. It should also give the field types the edit touches (result: Ty?). Separately, builder-facts.md could warn that NK1125 points at the start of the statement, not at the method call it refuses.
This site is open source.
Improve this page.
