1. Most turns: finding out which yaml test failed. I ran the package suite three times: once with tail, which hid the failing test’s name, once with a grep, and once more to see the full output. Running a test file on its own (nikaia run tests/X.nika) is refused in a project, which cost one more turn. Next came deciding whether the shifted error positions were a defect or the right result. I spent a turn looking in the spec for failure-position rules and found nothing useful.
2. What the brief could have handed me:
   • The brief points to is_hex “near line 1086”, but Hex was its only user.
   • It doesn’t say that errors_carry_line_and_column checks error offsets of bad escapes, which a guard-to-built-in change can move.
   • It doesn’t say that nikaia.contracts, nikaia.derived and nikaia.lock under packages/yaml are tracked and change with the source.
3. Checks:
   • Useful: the package suite, fmt --check, fmt::every_nika_file, pre-push-check.sh.
   • Wasted: two package-suite runs, which one run written to a file and grepped would have replaced. builder-facts already says to do that, and I didn’t.
   • I never read builder-facts whole, so that wasn’t wasted, but I could have missed a trap that was in it.
4. Change that would save the most: the stop condition “escape tests change” was ambiguous. The brief should say whether error positions count as escape-test output. For example: “error offsets may move to the offending character; update the .stdout and say so in Body:”. Or the opposite, if the architect wants a stop. Either way I wouldn’t have had to choose between stopping and deviating.

This site is open source. Improve this page.