Finding
Every crate in the workspace uses thiserror for its error enum, contradicting the standing non-negotiable shape for this fleet's Rust work: "snafu for errors, one error enum per crate, #[non_exhaustive]. Never thiserror, never anyhow in a library."
Evidence
grep -rl "derive(thiserror::Error" across crates/*/src/error.rs matches all 11 crates that have an error.rs: loader, cache, core, decode, embed, encoders, hipcore, praxis, taxis, kernels, tokenize, transformers. thiserror also appears as a direct dependency in every crate's Cargo.toml (loader/Cargo.toml:13, cache/Cargo.toml, taxis/Cargo.toml:14, etc. — 13 Cargo.toml hits total). No crate uses snafu.
Why this matters
This is repo-wide, not local drift in one file — a genuine migration, not a spot-fix. Filed separately rather than folded into #56/#60 (a loader-only audit-batch PR) because touching every crate's error type would explode that PR's scope far past its stated findings, and this box currently has other agents concurrently working sibling crates in the same shared .git — a workspace-wide error-handling rewrite here would conflict with all of them.
Desired correction
Either:
- Migrate every crate's error enum from
thiserror to snafu, crate by crate, each as its own reviewable PR; or
- If
thiserror is the deliberate, accepted choice for this repo (overriding the fleet-wide default), update the standard to say so explicitly for logismos, the same way kanon#3473 already documents the #[must_use]-on-Result contradiction between RUST.md and the enforced lint set.
Whichever direction is chosen, the current state — a written rule everyone violates uniformly — is the thing to resolve; leaving it unstated is what lets it get re-violated by every future PR.
Done when: either every crate's error.rs derives snafu::Snafu instead of thiserror::Error, or the standard has an explicit, dated exception recorded for this repo.
Finding
Every crate in the workspace uses
thiserrorfor its error enum, contradicting the standing non-negotiable shape for this fleet's Rust work: "snafu for errors, one error enum per crate,#[non_exhaustive]. Neverthiserror, neveranyhowin a library."Evidence
grep -rl "derive(thiserror::Error"acrosscrates/*/src/error.rsmatches all 11 crates that have anerror.rs:loader,cache,core,decode,embed,encoders,hipcore,praxis,taxis,kernels,tokenize,transformers.thiserroralso appears as a direct dependency in every crate'sCargo.toml(loader/Cargo.toml:13,cache/Cargo.toml,taxis/Cargo.toml:14, etc. — 13Cargo.tomlhits total). No crate usessnafu.Why this matters
This is repo-wide, not local drift in one file — a genuine migration, not a spot-fix. Filed separately rather than folded into #56/#60 (a loader-only audit-batch PR) because touching every crate's error type would explode that PR's scope far past its stated findings, and this box currently has other agents concurrently working sibling crates in the same shared
.git— a workspace-wide error-handling rewrite here would conflict with all of them.Desired correction
Either:
thiserrortosnafu, crate by crate, each as its own reviewable PR; orthiserroris the deliberate, accepted choice for this repo (overriding the fleet-wide default), update the standard to say so explicitly forlogismos, the same waykanon#3473already documents the#[must_use]-on-Resultcontradiction betweenRUST.mdand the enforced lint set.Whichever direction is chosen, the current state — a written rule everyone violates uniformly — is the thing to resolve; leaving it unstated is what lets it get re-violated by every future PR.
Done when: either every crate's
error.rsderivessnafu::Snafuinstead ofthiserror::Error, or the standard has an explicit, dated exception recorded for this repo.