feat(codegen): emit typed WIT list from Array data types (#345)

[avrabe]

Summary

Closes the #345 gap: spar codegen --format wit mapped every AADL data type with Data_Representation => Array to an opaque type x = list<u8>, dropping the element type and dimension — so an inter-core WIT contract could only say list, not the real payload (the jess NavState 16 × f32).

This teaches codegen to emit a typed WIT list from a Data Modeling Annex Array data type.

What changed

• spar-hir-def (resolver.rs) — new DataShape::Array { element, len }. resolve_data_shape recognizes Data_Representation => Array, resolves the Base_Type classifier to the element shape (recursively, reusing the existing scalar/record resolution), and reads the first Dimension as the length. Both the maintainer's list-wrapped property syntax (Base_Type => (classifier(Pkg::T)), Dimension => (16)) and the bare forms are accepted.
• spar-codegen (wit_gen.rs) — a fixed Dimension emits a bounded fixed-size list<T, N> (no cabi_realloc, consistent with the RECORDS‑002 no_alloc discipline where only unbounded list/string force dynamic allocation); no Dimension emits list<T>. The element resolves through a shared shape_wit_type helper, so an array-of-record and an array-typed record field both work.
• spar-codegen (rust_gen.rs) — the Rust port type mirrors wit-bindgen's lowering: list<T, N> → [T; N], list<T> → Vec<T>.

Honest scoping / no regression

An Array whose Base_Type is absent or resolves to an unsized/unresolvable element (e.g. the standard Base_Types::Float, which carries no Data_Size) is not fabricated: array_of returns None, so a top-level array port gracefully keeps the legacy list<u8> (no regression, no hard error), and an array-typed record field takes the record no_alloc hard-error path — exactly as an opaque field would.

Not claimed here: list byte-layout ↔ canonical-ABI equivalence (that's #327's ordeal work); multi-dimensional arrays use only the first Dimension; no compiled-crate oracle for array ports on the Rust side.

Oracles (executed unit tests, spar-codegen/src/wit_gen.rs)

• codegen --format wit emits opaque list<u8> for every AADL data type — how to get a structured WIT record/list<f32> for an inter-core message? #345 NavState → type vehicle-state = list<f32, 16>;
• unbounded array → type samples = list<f32>;
• no Base_Type → type mystery = list<u8>; (negative control)
• unsized element → type blurry = list<u8>; (graceful fallback, no error)
• array-typed record field → nav: list<f32, 16>

Each asserts the emitted WIT parses + resolves under wit_parser (a real fixed-size-list bind, not a string match).

cargo test -p spar-codegen --lib → 56 passed. cargo test -p spar-hir-def --lib → 471 passed. cargo fmt --check clean; cargo clippy on both crates clean. An independent clean-room agent re-derived each claim against its own fixtures.

Tracked as REQ-CODEGEN-WIT-ARRAY-001.

Note: case (b) of #345 (a data implementation with subcomponents) already works today when the port references the implementation classifier (VehicleState.nav) rather than the bare data type — details in the issue reply.

🤖 Generated with Claude Code

---

Generated by Claude Code

[avrabe]

CI blocked by self-hosted runner disk exhaustion (infra, not this diff)

Three required checks failed — Proptest (extended), Fuzz smoke, Bench compile smoke — all with the same root cause: the self-hosted runners are out of disk.

• Bench compile smoke (runner5): error: failed to write to '.../target/release/deps/rmetaNfJLue/full.rmeta': No space left on device (os error 28)
• Proptest (extended) (runner11): ld terminated with signal 7 [Bus error], core dumped while linking spar-mcp (SIGBUS = the linker's mmap'd output hit a full disk)
• Both post-job hooks report disk: 100% used (threshold 70%), and the cleanup step could not reclaim space (cleanup done: disk 100% → 100%).

This is not related to this PR's change: it's isolated to spar-mcp linking / lsp-types codegen (crates this diff doesn't touch), and spar-codegen / spar-hir-def (the crates I changed) compiled cleanly earlier in the same logs. Locally: spar-codegen 56/56, spar-hir-def 471/471, fmt + clippy clean.

Two runners (runner5, runner11) are both at 100% — this looks fleet-wide, which likely also explains why the main v0.32.0 CI run has been stuck queued for hours.

Action needed (runner admin): reclaim disk on the self-hosted runner fleet (prune target/, sccache, and ~/.cargo/registry/cache — the current post-job hook isn't freeing enough). The required checks can't go green until then. I don't have permission to re-run failed jobs (rerun-failed-jobs → 403), and a re-trigger won't help while the runners are full. I'll re-verify and drive this to merge once the disks are back under threshold — nothing to change on the branch itself.

---

Generated by Claude Code

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}

[codecov]

Codecov Report

❌ Patch coverage is 94.57995% with 20 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/spar-codegen/src/wit_gen.rs	95.41%	11 Missing ⚠️
crates/spar-hir-def/src/resolver.rs	85.24%	9 Missing ⚠️

📢 Thoughts on this report? Let us know!