1. Most cost: the background waits for the two clippy runs and the 24-test selection, not the code. Of the turns, two lower-std rounds were spent on Nikaia text rules: NK1214 refused cutting a text at len() - 6, and NK1185 refused ?? stem because the two sides of ?? differed (ref String against String). I also spent one turn learning that nikaia run of the test program fails, because its target/fast/nikaia fallback is not found under run.
2. Looked up that the brief could have handed me:
   • The brief says a stray .bounds is an error “as a stray .stderr is”, but output_tests silently ignores a lone .stderr. I had to choose where the error lives: a Rust-side check in project::program_tests, chosen to avoid the Nikaia throw-type question.
   • With a dependency, check --bounds prints a --- path block for every unit, so the entry’s report has to be cut out of the output.
   • Failed program tests print through Option<&OutputTest> = None, so the explanation of a difference needed its own hook.
   • Text cutting in tools/*.nika takes only positions from find, and strip_suffix(..) ?? "" is the idiom to use.
3. Checks run:
   • Build and lower-std, which ran twice: before the build and as the fixpoint.
   • cargo check, clippy 1.97 and 1.99.1, the project:: unit tests, pre-push-check.
   • The scratch project: bless, re-run, and the old binary against it.
   • The 24 selected tests, with --both-settings.

   Wasted: running check --bounds from inside tests/compiler/tests (it read the whole project and printed parse errors from other files), and the nikaia run attempt on the test program.

4. One change: add to builder-facts: “to run some tests/compiler tests by name, copy nikaia.toml, src/ and the selected tests/NAME.* into a scratch project and run nikaia test --project there; never check a file inside tests/compiler.” Also have the brief state that check --bounds with dependencies prints a block per unit.

This site is open source. Improve this page.