Section 04: Verification & Testing

Context: TPR-07-006-codex confirmed that niche-helper tests in option_result_helpers/tests.rs use include_str! source-text matching instead of actual IR emission — they test the text of the source file, not the LLVM IR the helpers produce. This section replaces those tests with emitter-driven verification and adds a codegen gate audit to prevent future TagAccess bypass.

Note on finding 6 (niche stubs): The niche monadic stubs (emit_option_niche, emit_result_niche) still return None — this plan does NOT implement them. Niche stub implementation is deferred to repr-opt §07.2’s NICHE_CODEGEN_READY gate flip. This section fixes the TESTS (finding 9), not the stubs (finding 6). The emitter-driven tests directly invoke the niche helpers from a synthetic harness, bypassing the gate, so they exercise the dead-code path as a regression guard.

04.1 Emitter-Driven Niche Tests

File(s): compiler/ori_llvm/src/codegen/arc_emitter/builtins/option_result_helpers/tests.rs (rewrite)

The implementation plan from TPR-07-018 (already documented in plans/repr-opt/section-07-enum-repr.md lines 754-768) describes the exact test architecture. This subsection executes it.

Read the existing test file to understand what it currently does (source-text assertions via include_str!)
Build synthetic emitter harness following the pattern from compiler/ori_llvm/src/codegen/arc_emitter/tests.rs (drop_fn_trivial_generates_rc_free pattern):
1. Construct a Pool with pool.option(Idx::STR) for Option, pool.result(Idx::STR, Idx::STR) for Result<str, str>
2. Create LLVM Context, SimpleCx, IrBuilder
3. Declare runtime functions via declare_runtime_functions or similar
4. Create a host function with no params, position builder at entry
5. Construct synthetic TagEncoding::new(EnumTag::Niche { field_index: 0, niche_value: 0, niche_variant_idx: 1 }, 2) for Option
6. Allocate synthetic receiver via builder.const_zero_ty(opt_str_llvm_ty)
Write emitter-driven tests (replace ALL source-text assertions):
- emit_option_niche_unwrap_contains_panic_and_rc_inc: Call em.emit_option_niche("unwrap", ...), assert IR contains "ori_panic" AND "ori_str_rc_inc"
- emit_option_niche_expect_contains_panic_and_rc_inc: Same for expect
- emit_option_niche_unwrap_or_contains_conditional_rc_inc: unwrap_or has RC inc on the conditional merge path (no panic)
- emit_result_niche_unwrap_and_unwrap_err_differ: Assert IR for Result.unwrap and Result.unwrap_err are NOT identical (proves the collapsed-arm bug is fixed, not just textually present)
- emit_result_niche_methods: Test all 5 Result methods produce valid IR
Negative pin: Test that invoking a niche helper with a NON-niche encoding (explicit tag) returns None — the gate bypass is only for niche-encoded types
Remove include_str! assertions from the test file — they are replaced by the emitter-driven tests. The file keeps its location (option_result_helpers/tests.rs) but the content is entirely rewritten.
Subsection close-out (04.1) — 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.

04.2 Codegen Gate Audit

File(s): diagnostics/ (new script or extension to existing tool)

Create diagnostics/enum-layout-audit.sh — a script that detects TagAccess bypass:

#!/bin/bash
# Detect hardcoded enum field access that bypasses TagAccess.
# Exit 0 if clean, exit 1 if violations found.

# Check for hardcoded i64-slot packing outside ori_repr
PACKING=$(rg 'div_ceil\(8\)' compiler/ --glob '!**/ori_repr/**' --glob '!**/tests*' -c)

# Check for hardcoded enum indices outside tag_access
# (filter legitimate struct field access via heuristic: enum names in the label)
INDICES=$(rg 'extract_value\(.*,\s*[01]\s*,.*\b(tag|payload|opt\.|res\.)\b' \
  compiler/ori_llvm/ --glob '!**/tag_access/**' --glob '!**/tests*' -c)

This script becomes a gate for future development — run it in CI or as part of /impl-hygiene-review to prevent regression.

Verify the audit passes after §02 and §03 are complete — zero violations
Add to ./clippy-all.sh or document as a manual gate (depending on false-positive rate)
Subsection close-out (04.2) — 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.

04.3 Integration Verification

04.4 Completion Checklist

04.R Third Party Review Findings

None.