1. What cost the most: about 6 turns on step 1, the rs2nika.names rows. Re-converting with them changed nothing, so I read the converter’s loader and ran a probe. It turned out the binary applies no [forms] rows at all, not even the global {x}.span.at() row. The second cost was waiting about 5 minutes for the it run and about 6 minutes for tests/compiler. I ran them in parallel, so that time was mostly unavoidable.
2. What I looked up that the brief could have given me:
   • Whether print/println lower inside a tools file. They do, as plain macros (one probe).
   • Whether retain, tuple-pattern closures or ?? continue exist in tools/*.nika. Tuple closures appear nowhere in tools, but they lower.
   • That SignatureOps::arguments() skips a leading self. The brief says it “is params”, which is not quite true.
   • The Rust signatures of check_against, foreign::check and render_*, and that diagnostics is diagnostics/mod.rs.
   • The plan row format, and where the clippy:: allow list sits in nikaia-std/src/lib.rs.
3. Checks run: cargo check --release, one lower-std → build → lower-std cycle, the it filters plus the guard set, --lib project::, tests/compiler --both-settings, clippy twice, pre-push-check and --score.
   • Wasted: the first clippy run, which failed on map_clone in the lowered code. Running the converter twice (--file and the tree engine) only confirmed the dead [forms] rows.
   • Too broad: the check filter matches most of it (765 tests). That was fine this time, but it is wider than this change needed.
4. One change that would save the most: builder.md should state that .map(fn(x) { x.clone() }) on a struct Vec trips clippy::map_clone, and that the fix is an entry in the tools allow list. Also, a brief whose step 1 depends on [forms] should first confirm that the current converter binary applies them; the driver could check with a one-line probe. That would have saved the clippy round and the converter investigation.

This site is open source. Improve this page.