Section 02: TagAccess Full Migration + Derive Fix

Context: 66 codegen consumer sites across 10+ files hardcode enum field indices (extract_value(..., 0) for tag, extract_value(..., 1) for payload). This includes an active miscompile: derive enum bodies (enum_eq.rs, enum_comparable.rs, enum_hashable.rs) assume {tag, payload} struct layout but TAGGED_PTR_CODEGEN_READY is already true — user-defined enums can qualify for tagged-pointer layout TODAY. When they do, derives produce wrong code (Eq returns false, Comparable returns Equal, Hashable panics). §02.1 fixes this urgently.

Approach: TagAccess already exists in compiler/ori_llvm/src/codegen/arc_emitter/tag_access/mod.rs with 22 consumer sites using it correctly. This section migrates the remaining 66 sites. TagAccess becomes a thin LLVM adapter that reads EnumLayoutInfo from ori_repr (§01) and translates queries into LLVM builder calls.

Phase ordering: §02.1 (derive fix) is URGENT and must be done first. §02.2 (builtins) and §02.3 (remaining) follow.

02.1 Derive Enum Bodies Fix (URGENT — Active Miscompile)

File(s): compiler/ori_llvm/src/codegen/derive_codegen/enum_bodies/enum_eq.rs, enum_comparable.rs, enum_hashable.rs

Why URGENT: TAGGED_PTR_CODEGEN_READY = true (verified at canonical/type_repr.rs:251). User-defined enums like type MaybeIter = Empty | Holds(it: Iterator<int>) qualify for tagged-pointer layout via can_use_tagged_pointer(). When derive Eq/Comparable/Hashable runs on such enums, the code does extract_value(self_val, 0, "eq.tag.self") — but the LLVM type for tagged-pointer enums is i64 (not {i64, payload}), so extract_value(..., 0) either extracts nothing or produces garbage. Eq returns false on the failed extraction; Comparable returns Equal; Hashable panics at enum_hashable.rs:86.

Write failing TDD matrix FIRST — AOT tests in compiler/ori_llvm/tests/aot/:
- enum_derive_tagged_ptr.rs: Test #derive(Eq) on a tagged-pointer-eligible enum → currently produces wrong result. Pin: after fix, equality compares correctly for both the unit and pointer variants.
- enum_derive_explicit_tag.rs: Regression pin — existing explicit-tag derives still work correctly after the migration.
- Matrix: Eq × Comparable × Hashable for Explicit, TaggedPtr, (Niche gated by NICHE_CODEGEN_READY=false)
Add TagAccess-aware dispatch to each derive file. Before the existing extract_value(..., 0) tag extraction, query EnumLayoutInfo to determine the tag encoding strategy. Dispatch:
- Explicit → existing code path (extract tag at tag_gep_index, payload at payload_gep_index)
- TaggedPtr → use tagged_ptr_decode_tag() and tagged_ptr_decode_ptr() from TagAccess
- Niche → extract the niche field and compare against the niche value (gated by NICHE_CODEGEN_READY)
- None → single variant, no tag comparison needed
Migrate enum_eq.rs (lines 34-35, 118-122, 195-200):
- Replace extract_value(self_val, 0, "eq.tag.self") with TagAccess-aware tag extraction
- Replace extract_value(self_val, 1, "eq.*.payload") and struct_gep(..., 1, ...) with payload access via TagAccess
- Replace i64_offset += field_bytes.div_ceil(8).max(1) with slot_count() from §01
Migrate enum_comparable.rs — same pattern as enum_eq.rs (lines 31-34, 102-107, 206-207)
Migrate enum_hashable.rs — same pattern (lines 36, 78-85, 165-166)
Verify ./test-all.sh passes in both debug and release. ORI_CHECK_LEAKS=1 on derive-related test programs.
Subsection close-out (02.1) — Run /improve-tooling retrospectively. Commit separately.
/sync-claude section-close doc sync — verify Claude artifacts across all section commits. Map changed crates to rules files, check CLAUDE.md, canon.md. Fix drift NOW.
Repo hygiene check — run diagnostics/repo-hygiene.sh --check and clean any detected temp files.

02.2 Option/Result Builtins Migration

File(s): compiler/ori_llvm/src/codegen/arc_emitter/builtins/option_result.rs (15 sites), option_result_helpers.rs (10 sites), compound_type_impls/option.rs (6 sites), compound_type_impls/result.rs (5 sites), debug_helpers.rs (3 sites), collections/list_builtins/helpers.rs (2 sites)

Note: Option and Result have a runtime ABI contract — ori_rt writes {i64 tag, T payload} via C functions. The EnumLayoutInfo for Option/Result will have is_runtime_abi: true, which means TagAccess must use the runtime-compatible encoding regardless of what niche analysis says. This contract is enforced by canonical_option() and canonical_result() using IntWidth::I64 tags while NICHE_CODEGEN_READY = false.

Migrate build_option_struct() and build_result_struct() in option_result_helpers.rs:
- Read tag_gep_index and payload_gep_index from EnumLayoutInfo instead of hardcoding 0 and 1
- Use tag_encoding.tag_width() instead of hardcoding i64_ty
- These 2 functions transitively fix 9 call sites that use them
Migrate resolve_type_for_option() and resolve_type_for_result() in option_result_monadic.rs:
- Read struct layout from EnumLayoutInfo instead of constructing {i64, T} inline
Migrate tag extraction sites in option_result.rs (15 sites), compound_type_impls/option.rs (6), result.rs (5), debug_helpers.rs (3), list_builtins/helpers.rs (2):
- Replace extract_value(receiver, 0, "opt.tag") with tag_access.extract_tag(receiver) or equivalent
- Replace extract_value(receiver, 1, "opt.payload") with tag_access.extract_payload(receiver, variant_idx)
AOT regression tests: Run existing Option/Result AOT tests. Add 2 new tests:
- option_result_abi_contract.rs: Verify Option and Result<str, Error> roundtrip correctly through runtime C functions (ori_list_first, ori_map_get)
- Verify the is_runtime_abi flag prevents niche optimization from changing Option/Result layout
Subsection close-out (02.2) — Run /improve-tooling retrospectively. Commit separately.
/sync-claude section-close doc sync — verify Claude artifacts across all section commits. Map changed crates to rules files, check CLAUDE.md, canon.md. Fix drift NOW.
Repo hygiene check — run diagnostics/repo-hygiene.sh --check and clean any detected temp files.

02.3 Remaining Consumer Migration + Packing Consolidation

File(s): instr_dispatch.rs (1 payload site), rc_helpers.rs (1 payload site), drop_enum.rs (1 packing site), abi/mod.rs (2 packing sites), type_info/enum_layout.rs (2 packing sites)

Migrate remaining hardcoded payload access:
- instr_dispatch.rs:140: extract_value(val, 1, "proj.payload") → TagAccess payload extraction
- rc_helpers.rs:89: extract_value(val, 1, "rc.opt_inner") → TagAccess payload extraction
Replace scattered packing computations with ori_repr::layout::slot_padded_size() and slot_count() from §01:
- drop_enum.rs:380: current += field_size.div_ceil(8).saturating_mul(8) → current += slot_padded_size(field_size)
- abi/mod.rs:178,268: size.div_ceil(8) * 8 → slot_padded_size(size)
- enum_layout.rs:100,117: size.div_ceil(8) * 8 and max_payload_bytes.div_ceil(8) → slot_padded_size() and slot_count()
Run grep audit to verify ZERO remaining scattered computations:
```
rg 'div_ceil\(8\)' compiler/ --glob '!**/ori_repr/**' --glob '!**/tests*'
```
Must return zero matches.
Run grep audit for hardcoded enum indices:
```
rg 'extract_value\(.*,\s*[01]\s*,' compiler/ori_llvm/ --glob '!**/tag_access/**' --glob '!**/tests*'
```
Review every match — must be struct field access (legitimate) not enum tag/payload access.
Subsection close-out (02.3) — Run /improve-tooling retrospectively. Commit separately.
/sync-claude section-close doc sync — verify Claude artifacts across all section commits. Map changed crates to rules files, check CLAUDE.md, canon.md. Fix drift NOW.
Repo hygiene check — run diagnostics/repo-hygiene.sh --check and clean any detected temp files.

02.4 Completion Checklist

02.R Third Party Review Findings

None.