- What cost the most. Finding out why a build-time value failed. NK1127 “couldn’t build the program that computes it” gives no reason. I had to find the generated
main.rs by hand. The brief points to ~/.cache/nikaia/, but for a run with --project it sits in <project>/target/nikaia/build-time/comptime/<hash>/. When lowering refused the program, no file was written at all. Step 3 needed three build cycles, one per layer:
- the dump of
T?;
- the Rust type spelling (
Option<&str>);
- the earlier
comptimes the sub-program carries (a Some as a bare literal, null with no type).
- What I had to look up. These five points were not in the brief:
- the sub-program re-checks earlier
comptimes as literals (comptime_unit_without_its_constants), so every new value shape must also survive as a carried literal;
rust_constant_type / declared_below skip Ty::Nullable;- a text
comptime is declared ref String, which is Named str view;
Ty::fits exists;- the decoder already reads
(v Option …).
- Checks run, and which were wasted.
- Run: build cycles with the
lower-std fixpoint, single-test runs with one.sh, and one batch of 54 tests with --both-settings, which was efficient.
- Also run: clippy twice (stable and 1.99.1),
fmt, pre-push-check, and nikaia-std --lib for the three modules.
- Wasted: the
nikaia-std --lib runs, which found 0 tests (the modules are lowered .nika), and clippy on local stable once 1.99.1 was installed.
- Missing: the proof that the new tests fail on the unfixed compiler is old probes, because I kept no copy of the old binary.
- One change that would have saved the most. In builder.md: “Copy
target/fast/nikaia to your scratchpad before your first build, for the fails-unfixed run.” Plus one line in the brief: “a new comptime value shape must also work as the literal the sub-program carries (comptime_literal_of → re-check); the generated main.rs is in <project>/target/nikaia/build-time/comptime/, and none is written when lowering refuses.”
This site is open source.
Improve this page.
