1. What cost the most: the search for the “existing not built yet path” the brief told me to use. I spent about 8 turns grepping the checker, the emitter and the Part III code table before I was sure it doesn’t exist. The longest wait was the whole it binary, about 25 minutes. builder-facts says a parser change needs the whole binary, while the brief named only four filters. For a change that only shrank the reserved-word list, the filtered run would probably have been enough.
2. What the brief could have handed me:
   • that async was already in RESERVED_C, which makes half of step 1 a no-op;
   • that the AST is declared in tools/ast.nika, not ast/mod.rs, so a new node touches about 36 files of exhaustive matches;
   • which NK code, or which helper, the “not-built path” means;
   • a sample overlap { } program to copy into a test.
3. Checks run:
   • Used: a --both-settings bless run on a scratch project, the fails-before check with the old binary, cargo fmt --check, clippy, pre-push, and the whole it binary (green).
   • Partly wasted: the whole it run, as above. I also ran clippy while it was going, which risked a relink; it turned out harmless.
   • Not wasted: lower-std, which I skipped because no .nika changed.
4. The change that would have saved the most: a brief pre-flight against the tree. Before ordering an “existing path”, the supervisor should grep for it and put the exact function and NK code in the brief, or the reserved NK block if none exists. It should also count the Expr match sites first, so a node-adding step is sized as its own brief, not tucked into “S=1, C~3”.

This site is open source. Improve this page.