1. What cost the most. The fix took one edit and one build, so the overhead was in testing:
   • The scratch-project recipe in Traps already met has no src/main.nika, so my first nikaia test --project run died with “cannot read …/src”. That was one wasted run.
   • I widened the related-test set by grepping for NK2202, NK1152 and sync. That pulled in two programs that need support files from tests/compiler, so they failed in the scratch project and cost an extra run on the old binary to clear them.
2. What the brief could have handed me:
   • That bare sleep is refused (NK1117) and the pausing call is use std::time with time::sleep(1).
   • That probe 2’s comptime W: String is refused by design (NK1167) and needs ref String.
   • That tests/compiler has no sync_* programs. The real related tests are the it modules sync_of_a_lambda, comptime* and build_time.
   • The brief’s location for the cause (a_named_call) was exact. I also had to look up callee_named in calls.nika to see how a bracket call on a non-function value (fs[0](x)) would be treated.
3. Checks.
   • Useful: the probes on the old and new binaries, the build cycle, the fixpoint cmp, the two new tests with --bless, the same tests on the old binary, the it filters, fmt, clippy 1.97 and pre-push-check.sh.
   • Partly wasted: the broad 24-program scratch run, whose two failures had nothing to do with my change, and the failed bless run.
   • Partial, and my report should have said so: I ran clippy 1.99.1 on nikaia-std only, not on the whole workspace.
4. The one change that would have saved the most. Add mkdir -p $D/src && printf 'fn main() {\n}\n' > $D/src/main.nika to the one-program recipe in builder-facts. Second choice: the brief names exactly which related tests to run (tests/compiler programs and it modules) instead of patterns that match nothing.

This site is open source. Improve this page.