Issue #1179: lower-std: lowered/fold_bound.rs flips between rule: &str and rule: String across runs

Open and answer this issue on GitHub.

ready · opened 2026-11-02 · updated 2026-11-02 · 1 comment

Running nikaia lower-std again with fold_bound.nika unchanged writes lowered/fold_bound.rs differently. Between runs, the parameter type flips between rule: &str and rule: String. The result is that crates/nikaia/src/fold_bound.rs:61 stops compiling with E0308 (expected String, found &str).

This was found by the M14 emit lane. The binary from before that lane’s change shows the same flip, so the emit change did not cause it.

The lowering is not deterministic here. A likely cause is an iteration order, such as a map or a parallel pass, that decides whether the borrowed form or the owned form is chosen. The sysroot test the_tools_lower_the_same_on_one_thread_as_on_every_core may catch the flip only some of the time.

Probe

for i in 1 2 3 4 5; do target/fast/nikaia lower-std >/dev/null; md5sum crates/nikaia-std/src/tools/lowered/fold_bound.rs; done

The hashes differ between runs.

To build

1. Find the source of the nondeterminism in the param-type (borrow/own) decision and make it ordered.
2. Add a test that lowers twice and compares the output, either in tests/compiler or as an extension of the existing sysroot comparison.

Comments

keywan-ghadami, 2026-11-02

Not reproduced on a clean git archive HEAD copy (bug lane, 2026-11-02). Two suspects to check next:

1. build.rs compiler_fingerprint hashes lowered/*.rs while another agent runs lower-std in the shared tree, so the fingerprint changes during a build.
2. The cache key sees other files without their bodies.

A probe needs two runs racing a lower-std on the same tree.

---

Generated by Claude Code

Back to Open Issues