1. Cost. The biggest cost was the test program. I ran nikaia run FILE inside tests/compiler and was refused, because a test file isn’t an entry point. Then I got the closure’s return type wrong twice: first 1, then Some(1). Only null in an if expression works. The lowering cycle (lower-std, build, lower-std, build again for the fixpoint) was the other main cost, but it was necessary.

2. Lookups the brief could have given.
   • How to run a single test program: put it in a scratch project with its own nikaia.toml and src/main.nika, then nikaia run.
   • nikaia test has no name filter.
   • A filter_map closure returns T? and the empty case is null. project_check.nika:442 shows this, but I only found it by grep.
   • The exact fix location was given. I would still have liked to know whether the file was meant to be formatted or kept as the long line.
3. Checks.
   • Ran: the fmt:: filter (40 tests, including the scan), the guard set, clippy beside the tests, the new program run by hand, and fmt idempotence on copies of both files. All were useful.
   • Wasted: the first nikaia run calls, and the attempt to filter nikaia test.
   • Skipped: running the new test against the unfixed formatter (needs the binary rebuilt without the fix) and nikaia test --both-settings --project tests/compiler. I also did not look at pre-push-check.sh’s exit status.
4. Brief and builder.md change. Add one line to builder-facts.md: “To try a test program in seconds, copy it to <scratch>/src/main.nika with a minimal nikaia.toml and nikaia run; nikaia test takes no filter.” Also state how to get a failing-before result without a rebuild, for example keep the pre-fix binary as $S/nikaia.old and run the new test with it.

This site is open source. Improve this page.