- Most turns: working out that step 2 contradicts the specification. About 8 turns went on probing b37’s probe and its views-3 spellings, searching for an “existing escape error” that does not exist, then reading Part I 6.6, NK2303 and NK2304 to confirm it. Finding that step 1 was already built took about 4 turns of probes.
- What the brief could have handed me:
- The error code. The code for a view flowing where an own value is needed is NK1106 for a field and NK1102 for a
push.
- No escape refusal exists. NK2303 is not a refusal, and 6.6 anchors a struct handed back.
- #1152’s probe. It is already
tests/compiler/tests/ref_field_from_local.nika.
- Ledgers of test programs. They are not recorded, so “ledger lists the view” can only be checked by hand with
nikaia lower.
- Checks:
- About 15 probes on the old binary; all were useful.
- One
nikaia test --bless in a scratch project, fmt --check, pre-push, and a by-hand ledger check with nikaia lower.
- No build, clippy,
lower-std or it run, since no code changed. Nothing was wasted except the grep for an escape-error message.
- One change: a pre-flight line in the brief, written before it is issued: “probe each step’s tests on the current binary; if they already pass, the step is tests only”. The supervisor should also check every refusal the brief orders against Part I/III: an order that 6.6 contradicts needs a decision before the brief goes out, not after a builder run.
This site is open source.
Improve this page.
