Skip to main content

Module jit_array

shape_jit

Module jit_array 

Source
Expand description

JIT array — Route A close.

§Route A close (ADR-006 §2.7.14 / W11-jit-new-array)

Q15 resolved Route A: kinded per-element-kind Arc<TypedArrayData> monomorphization. The deleted UnifiedArray heap layout (#[repr(C)] with data / len / cap / typed_data / element_kind byte) is gone. Under Route A the JIT and VM share the shape_value::v2::typed_array::TypedArray<T> carrier (24-byte header: HeapHeader at offset 0, *mut T at offset 8, len: u32 at offset 16, cap: u32 at offset 20). Element kind is stamped at the producing call-site (jit_v2_array_new_<kind>) and not stored on the heap object — it lives on the HeapValue::TypedArray(arc) variant the HeapHeader.kind field discriminates (§2.7.6 / Q8 single-discriminator).

Cranelift offsets for the typed-array layout live inline at the consumer sites (mir_compiler/v2_array.rs’s DATA_PTR_OFFSET = 8 and LEN_OFFSET = 16). The five legacy offset constants (DATA_OFFSET, LEN_OFFSET, CAP_OFFSET, TYPED_DATA_OFFSET, ELEMENT_KIND_OFFSET) that the W10-cascade close marked #[allow(dead_code)] are deleted — under Route A there is no ELEMENT_KIND byte on the heap, and the remaining offsets are no longer the JIT’s canonical reference shape.

§Forbidden under any future expansion

  • UnifiedArray revival under any renamed shape (CLAUDE.md “Renames to refuse on sight” — broader-family regex).
  • tag_bits-based element decoder.
  • Bool-default fallback for unknown element kind.
  • Re-introducing an ELEMENT_KIND byte on the heap object — the discriminator lives on HeapValue::TypedArray(arc) per ADR-005 §1 single-discriminator.