1. Biggest cost: I started the gate in the background with a 600 s limit, which is the foreground default, not the background maximum. It was killed partway through tests/compiler, so I had to re-run tests/compiler and clippy. That wasted about 10 minutes of machine time and two turns. After that, three quick lower-std rounds on slips I made myself:
   • ?? continue used as a for header, which the parser refused;
   • ref String from iterating a local Vec, which needed .clone();
   • .cloned() on a Seq, which doesn’t exist (I used .map(fn(p) { p.clone() })).
2. What I looked up: the brief’s locations were accurate. I still had to find:
   • keeps_handed’s signature, and that it doesn’t take holes_of;
   • seen_in’s signature in foreign.nika;
   • keep_viewed_name, and the type of walk.views;
   • the Seen variants.

   One line saying “keep_argument/keep_method_arguments lack holes_of; seen_in is in foreign.nika:67” would have saved two reads.

3. Checks:
   • Useful: the new tests on the old binary (confirmed the failure), the build cycle (one build, lowered output stable), the it guard set plus keeps:: (57 passed), and tests/compiler at both settings plus clippy (the second run).
   • Wasted: the killed first tests/compiler run. Also my glob a_lambda_*.nika ran three existing tests along with mine, a small waste.
   • Inaccurate: my report says Disk freed: 0 although I deleted my copy of the binary. The real figure is a few hundred MB, not reported.
4. Change that would have saved the most: builder.md should say plainly: “a background run of tests/compiler --both-settings needs timeout: 7200000; the default 600000 kills it.” It could also give the gate as one ready-made script with that timeout.

This site is open source. Improve this page.