1. Most cost: the tests/compiler --both-settings run took over 10 minutes of wall time. In turns, the costliest part was finding that the brief’s cause was only half the fix. After the build, the checker’s arm alone turned the rustc error into NK1104/NK1106. It took a grep and two reads to find the second place, grammar_reference_kinds in text_tiers.nika, which decides whether a rule’s result is a view. That meant a second build cycle.
2. Looked up:
   • How a test builds Bytes from a literal (Bytes("ab".bytes().collect())). The brief’s probe implied Bytes("ab"), which is NK1102.
   • The Pattern::Literal(_) variant’s shape.
   • That text_tiers.nika mirrors pattern_type for result tiers.

   The brief could have said: “the text(..) capture is typed in two places, check_grammar.nika reference_type and text_tiers.nika grammar_reference_kinds; add the arm to both.”

3. Checks run:
   • The probe on the old binary: useful, and it doubles as the run against the unfixed compiler.
   • Two lower/build cycles: the first could have been saved.
   • The gate guards with the grammar it filter, clippy in parallel, and pre-push: all needed.
   • tests/compiler with --both-settings: required because check_ changed.

   Nothing else was wasted. A slip in backgrounding (&& before &) lost $S in the foreground shell, but it cost nothing.

4. One change: in builder-facts’ grammar trap line, add: “a label’s view/owned tier comes from text_tiers.nika grammar_pattern_kinds; a change to pattern_type needs the same arm there”. The brief’s file list should then include text_tiers.nika, so the change is no deviation. Also, builder-facts.md’s Red on develop now line about sequences is stale: it passed here.

This site is open source. Improve this page.