1. Most time: proving the new test fails on the unfixed compiler. No older binary existed (no target/release, no target/debug), so I put back HEAD’s fmt.nika and fmt.rs by name, rebuilt, ran the test, restored my files and rebuilt again: two extra builds. The tests/compiler --both-settings run was the longest wait, but the gate requires it.
2. Looked up:
   • What a grammar body may hold (grammar_rule, frame_attr in parser/mod.rs), needed to tell a rule’s first line from its continuation.
   • That a here grammar needs a Bytes input (NK1187), and how to write one (Bytes([])).
   • That grammar is a reserved word and cannot name a struct field; this cost one cycle. The brief could have given all three in one line each.
3. Checks:
   • Useful: the probes on the old binary, the fmt:: tests, the guard set with clippy alongside, pre-push-check (it caught the badge), the full tests/compiler run, and the yaml-draft check.
   • Wasted: the first build cycle. lower-std refused the reserved field name, but the chained cargo build still ran on the old lowered code.
4. Biggest saving: builder.md could say “keep a copy of target/fast/nikaia before your first build (cp to the scratchpad)”. The test against the unfixed compiler then takes seconds, not two rebuilds. Second, add to the builder-facts syntax gaps: “grammar, rule, entry are reserved words; never use them as names in tools/*.nika.”

This site is open source. Improve this page.