- Time and turns.
- The biggest cost was waiting on K4’s uncommitted
keeps.nika. It made lower-std and cargo check fail for the whole shared tree, so I lowered my file in a scratch NIKAIA_SYSROOT copy. Then I polled with a background cargo check loop until the tree compiled.
- Smaller losses: a wait loop I wrote badly on the first
lower-std output, and about six check cycles on small syntax refusals.
- What the brief could have handed me.
use std::i64 is refused (NK1156), and i64::from needs a use line (NK1117). Widening in tools/*.nika is x as i64. builder-facts says the opposite.i64::min(a, b) takes one argument, so it is a.min(b).grammar is a reserved word as a variable name.- A
comptime cannot hold a Vec of tuples or of structs (NK1167). It can hold ref Array[Struct] with ref String and ref Array[ref String] fields.
for x in v over an owned Vec lowers to a borrow, so it needs .drain().- A function that calls itself recursively lowers as
async unless declared sync.
text.bytes.len() trips clippy::needless_as_bytes, which needed an allow-list entry.nikaia check on a lone tools file reports false “no type” errors, because it needs the package.
- Checks.
- Useful: the scratch-sysroot
lower-std, the cargo check and clippy runs, the tests the brief named (8 and 36 passed), pre-push (it caught the stale self-hosted.svg), and the lowered.rs diff.
- Wasted:
nikaia check on the standalone file, and the first lower-std that failed on K4’s file.
- I did not run the new-binary fails-before test, since the brief ordered no new test. I could not compare the assembled package byte for byte against the old output. Equality rests on
lower-std leaving every lowered/*.rs unchanged.
- One change. The brief should list the lane’s dirty files. The builder should be told that a shared-tree refusal in another agent’s
.nika means running lower-std with NIKAIA_SYSROOT on a scratch copy at once. A pre-flight “tree compiles?” check would also have saved the polling.
This site is open source.
Improve this page.
