11%

Section 6: Capabilities System

Goal: Effect tracking (moved earlier to unblock Section 8 cache and Section 11 FFI)

SPEC: spec/20-capabilities.md DESIGN: design/14-capabilities/index.md

Status: In-progress — Core evaluator working (6.1, 6.4 verified; 6.2, 6.3 WRONG TEST; 6.5 partial; 6.9 partial, was incorrectly marked not-started; 6.10 partial); composition (6.11), resolution (6.12), intrinsics (6.14), stateful handlers (6.16) pending. LLVM tests missing. Zero #compile_fail negative tests. Verified 2026-03-29.

6.1 Capability Declaration

  • Implement: uses clause [done] (verified 2026-03-29) WEAK TESTS

    • Rust Tests: ori_parse/src/tests/parser.rs — uses clause parsing (4 tests: single, multiple, with where, no uses) (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/declaration.ori (3 tests: pure function, with capability, generic with capability) (verified 2026-03-29)
    • Negative Tests: No #compile_fail tests for missing capabilities (S06-06)
    • LLVM Support: ARC lowering treats WithCapability as transparent passthrough; no dedicated LLVM/AOT tests verify this path
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • Implement: Multiple capabilities [done] (verified 2026-03-29) WEAK TESTS

    • Rust Tests: ori_parse/src/tests/parser.rs — test_uses_clause_multiple_capabilities (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/propagation.ori — fetch_and_log with uses Http, Logger (verified 2026-03-29)
    • LLVM Support: LLVM codegen for multiple capabilities in function signatures
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.1) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.1 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.1: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.2 Capability Traits

  • Implement: Capability traits WRONG TEST — reopened (was [done] 2026-02-10, failed verification 2026-03-29, S06-01)

    • Rust Tests: WRONG TEST — roadmap claimed “7 tests for capability trait validation” in ori_types/src/check/tests.rs; these tests DO NOT EXIST. module_checker_function_scope tests has_capability() but is not a capability trait validation test.
    • Ori Tests: WRONG TEST — tests/spec/capabilities/traits.ori is 126 lines ALL COMMENTED OUT (TODO: “Type checker needs capability support”). Zero active tests. Violates impl-hygiene.md: “No commented-out code ever.” (S06-09)
    • Fix: Delete or uncomment traits.ori — commented-out code is a hygiene violation
    • Fix: Write actual capability trait validation tests (Rust + Ori)
    • LLVM Support: LLVM codegen for capability trait dispatch
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.2) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.2 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.2: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.3 Suspend Capability

Note: Renamed from Async to Suspend per proposals/approved/rename-async-to-suspend-proposal.md

  • Implement: Explicit suspension declaration STALE TEST — reopened (was [done] 2026-02-10, failed verification 2026-03-29, S06-02)

    • Rust Tests: WRONG TEST — roadmap claimed “4 tests (marker trait, signature storage, combined capabilities, sync function)” in ori_types/src/check/tests.rs; ZERO such tests exist.
    • Ori Tests: STALE TEST — tests/spec/capabilities/async.ori is 7 lines with only a comment: “This file is intentionally empty - async is not a language feature.” Zero executable tests.
    • BUG: Prelude defines pub trait Async {} (line 250) but spec section 20.3 and approved rename-async-to-suspend-proposal.md say it should be Suspend. Rename not applied. (S06-04)
    • Fix: Rename Async to Suspend in prelude
    • Fix: Write actual suspension declaration tests (Rust + Ori)
    • LLVM Support: LLVM codegen for explicit suspension declaration
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • Implement: Sync vs suspending behavior STALE TEST — reopened (was [done] 2026-02-10, failed verification 2026-03-29)

    • Rust Tests: WRONG TEST — test_sync_function_no_suspend_capability does not exist in ori_types/src/check/tests.rs
    • Ori Tests: STALE TEST — tests/spec/capabilities/async.ori has zero executable tests

  • Implement: No async type modifier [done] (verified 2026-03-29)

    • Rust Tests: ori_parse/src/tests/parser.rs — 3 tests pass: test_no_async_type_modifier, test_async_as_identifier, test_uses_async_capability_parses (verified 2026-03-29)
    • Ori Tests: Design notes document this (verified 2026-03-29)
    • Note: Roadmap previously cited test_async_keyword_reserved which does not exist — actual test is test_async_as_identifier (opposite semantics: async is NOT reserved)

  • Implement: No await expression STALE TEST — reopened (was [done] 2026-02-10, failed verification 2026-03-29)

    • Rust Tests: WRONG TEST — test_await_syntax_not_supported does not exist in ori_types/src/check/tests.rs. Evaluator does have CanExpr::Await(_) => await_not_supported() but no test verifies this.
    • Fix: Write test for await rejection

  • Implement: Concurrency with parallel — spec/20-capabilities.md § Suspend Capability

    • Deferred to Section 8: parallel pattern evaluation
    • Ori Tests: tests/spec/patterns/parallel.ori (Section 8)
    • Note: Evaluator has a loud stub for parallel in can_eval.rs — replace when Suspend capability is implemented

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.3) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.3 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.3: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.4 Providing Capabilities

  • Implement: with...in expression [done] (verified 2026-03-29)

    • Rust Tests: ori_parse/src/tests/parser.rs — with expression parsing (3 tests: expression, struct provider, nested) (verified 2026-03-29)
    • Rust Tests: ori_eval/src/interpreter/mod.rs — CanExpr::WithCapability uses with_binding() at can_eval/mod.rs:350 (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/providing.ori — 17 tests, all pass (basic provision, struct provider, scoping, nested, shadowing, different types, conditionals, method calls, capability through function) (verified 2026-03-29)
    • Ori Tests: tests/spec/expressions/with_expr.ori — 14 tests (11 pass, 3 skipped) (verified 2026-03-29)
    • Negative Tests: No #compile_fail negative tests for with...in (S06-06)
    • LLVM Support: LLVM codegen for with...in capability binding
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • Implement: Scoping [done] (verified 2026-03-29)

    • Ori Tests: tests/spec/capabilities/providing.ori — scoping, shadowing, three-level nesting, closure interaction (verified 2026-03-29)
    • LLVM Support: LLVM codegen for capability scoping (push/pop)
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.4) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.4 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.4: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.5 Capability Propagation

  • Implement: Runtime capability propagation [partial] (verified 2026-03-29)

    • Changes: prepare_call_env() in interpreter/function_call.rs:149-156 passes capabilities from calling scope to called function (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/traits.ori — ALL COMMENTED OUT, zero active tests (S06-01, S06-09)
    • Ori Tests: tests/spec/capabilities/providing.ori::test_capability_through_function — passes via scope-based name lookup (verified 2026-03-29)
    • Implement: with...in capability provision propagates to called functions — with Cap = impl in callee() should make Cap available inside callee()
    • Ori Tests: tests/spec/expressions/with_expr.ori — 2 tests skipped: test_basic_with and test_multiple_capabilities (“capability provision to called functions not implemented”)
    • Investigate: Skip reason says “not implemented” but prepare_call_env DOES pass capabilities — skipped tests may pass now (issue may be trait method dispatch vs simple name lookup)
    • LLVM Support: LLVM codegen for runtime capability propagation through calls
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • Implement: Static transitive requirements WEAK TESTS — reopened (was [done] 2026-02-10, failed verification 2026-03-29, S06-03)

    • Rust Tests: WRONG TEST — roadmap claimed “7 tests for E2014 propagation errors” in ori_types/src/check/tests.rs; ZERO such tests exist. E2014 error code IS implemented and registered.
    • Ori Tests: tests/spec/capabilities/propagation.ori — 7 tests, all pass (valid propagation patterns) (verified 2026-03-29)
    • Negative Pin: No #compile_fail("E2014") test exists anywhere — need negative test verifying E2014 fires on missing capability
    • Fix: Write Rust tests and #compile_fail("E2014") negative pins

  • Implement: Providing vs requiring [done] (verified 2026-03-29)

    • Rust Tests: ori_types/src/infer/expr/calls/constraints.rs — check_capability_propagation function exists (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/propagation.ori — tests with…in providing capabilities (verified 2026-03-29)

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.5) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.5 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.5: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.6 Standard Capabilities

STATUS: 9 of 13 standard capability traits defined in library/std/prelude.ori (INCOMPLETE MATRIX, S06-07) Real implementations deferred to Section 7 (Stdlib).

  • Define: Trait interfaces [partial] (verified 2026-03-29) INCOMPLETE MATRIX

    • Location: library/std/prelude.ori — trait definitions (verified 2026-03-29)
    • Traits defined: Http, FileSystem, Cache, Clock, Random, Logger, Env, Async (should be Suspend), Unsafe — 9 traits in prelude (verified 2026-03-29)
    • Missing from prelude (spec section 20.8 lists 13): Crypto (spec 20.8.3 has full trait), Print (spec lists it), Intrinsics (spec 20.8.4 has detailed trait), FFI (spec lists it) (S06-07)
    • DRIFT: Cache signature uses (key: str) -> Option<str> but spec uses generics <K: Hashable + Eq, V: Clone> (S06-08)
    • DRIFT: Clock signature uses @now () -> int, @today () -> str but spec uses @now () -> Instant, @local_timezone () -> Timezone (S06-08)
    • BUG: Prelude has pub trait Async {} but should be pub trait Suspend {} per approved rename proposal (S06-04)

  • Implement (Section 7): Real capability implementations

    • std.net.http — Http capability impl
    • std.fs — FileSystem capability impl
    • std.time — Clock capability impl
    • std.math.rand — Random capability impl
    • std.cache — Cache capability impl (new module)
    • std.log — Logger capability impl
    • std.env — Env capability impl
    • std.crypto — Crypto capability impl (missing from prelude)
    • std.io — Print capability impl (missing from prelude)

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.6) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.6 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.6: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.7 Testing with Capabilities

STATUS: Complete — mocking works via trait implementations, demonstrated in propagation.ori

  • Implement: Mock implementations [done] (verified 2026-03-29) WEAK TESTS

    • Ori Tests: tests/spec/capabilities/propagation.ori — MockHttp and MockLogger demonstrate with...in mocking pattern (verified 2026-03-29)
    • Note: Uses string mocks, not trait-implementing structs — weak demonstration of real mocking
    • LLVM Support: LLVM codegen for mock capability implementations
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs (file does not exist)
    • AOT Tests: No AOT coverage yet

  • Implement: Test example [done] (verified 2026-03-29)

    • Ori Tests: tests/spec/capabilities/propagation.ori — shows test patterns with with...in (verified 2026-03-29)

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.7) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.7 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.7: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.8 Capability Constraints

STATUS: Partially implemented — E2014 code exists but NEEDS PIN (S06-03, S06-06)

  • Implement: Compile-time enforcement NEEDS PIN — reopened (was [done] 2026-02-10, failed verification 2026-03-29, S06-03)

    • Rust Tests: WRONG TEST — roadmap claimed “7 tests for E2014 propagation errors” in ori_types/src/check/tests.rs; ZERO such tests exist.
    • Implementation: check_capability_propagation exists in constraints.rs, emits E2014 (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/propagation.ori — positive tests pass (verified 2026-03-29)
    • Negative Pin: No #compile_fail("E2014") test exists anywhere — completely untested with negative pins (S06-06)
    • Fix: Write #compile_fail("E2014") test verifying error fires when capability is missing
    • GAP: Error codes E1200-E1207 from spec not implemented (only E2014 exists) (S06-10)

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.8) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.8 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.8: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.9 Unsafe Capability (FFI Prep)

Proposal: proposals/approved/unsafe-semantics-proposal.md

PREREQUISITE FOR: Section 11 (FFI) The Unsafe capability is required for FFI. Implement this before starting FFI work. STALE STATUS: Was marked “not-started” but basic unsafe { } IS implemented and tested (S06-05). Updated to in-progress.

  • Implement: Unsafe marker capability (compiler intrinsic, like Suspend)

    • Add Unsafe to standard capabilities list in type checker
    • Prelude: pub trait Unsafe {} defined (line 254) (verified 2026-03-29)
    • Unsafe not treated as marker capability (no E1203 for with Unsafe = ... in)
    • No UnsafeContext tracking in type checker
    • No E1250 diagnostic
    • No enforcement that unsafe operations require unsafe { }
    • Generalize E1203 to cover all marker capabilities (not just Suspend)
    • Ori Tests: tests/spec/capabilities/unsafe/ — basic tests, E1203 binding error
    • LLVM Support: LLVM codegen for unsafe { } blocks (transparent — same as inner expr)
    • LLVM Rust Tests: ori_llvm/tests/capability_tests.rs — unsafe block codegen
    • AOT Tests: No AOT coverage yet

  • Implement: unsafe { } block expression (basic parsing + eval) [partial] (verified 2026-03-29)

    • ExprKind::Unsafe(ExprId) in IR (ori_ir/src/ast/expr.rs:306) (verified 2026-03-29)
    • Parser: parse_unsafe_expr() in ori_parse/src/grammar/expr/primary/specials.rs:111 (verified 2026-03-29)
    • Type checker: transparent pass-through (ori_types/src/infer/expr/mod.rs:216) (verified 2026-03-29)
    • Evaluator: transparent pass-through (ori_eval/src/interpreter/can_eval/mod.rs:337) (verified 2026-03-29)
    • ARC lowering: transparent (ori_arc/src/lower/expr/mod.rs:298) (verified 2026-03-29)
    • Visitor: supported in ori_ir/src/visitor/walk_expr.rs (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/unsafe_block.ori — 6 passing tests: single expr, multi stmt, nested, in block, type, bool (verified 2026-03-29)
    • Not implemented: UnsafeContext tracking in type checker, E1250 diagnostic
    • Not implemented: Enforcement that unsafe operations require unsafe { }
    • LLVM Tests: No LLVM codegen tests

  • Implement: Unsafe capability requirements (deferred to Section 11)

    • Required for: raw pointer operations (future)
    • Required for: C variadic function calls (future)
    • Required for: transmute operations (future)
    • Tests added when FFI implemented

Note: Unsafe is a marker capability — it cannot be bound via with...in. There is no AllowUnsafe provider type. Unsafe code is tested by testing safe wrappers.

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)
  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.
  • Subsection close-out (6.9) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.9 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.9: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.
  • /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.

6.10 Default Implementations (def impl)

Proposal: proposals/approved/default-impl-proposal.md Status: Partially implemented — parser + evaluator registration work, type checking + name resolution + end-to-end NOT working (verified 2026-03-29)

Introduce def impl syntax to declare a default implementation for a trait. Importing a trait with a def impl automatically binds the default.

Implementation

  • Implement: Add def keyword to lexer — grammar.ebnf § DECLARATIONS (verified 2026-03-29)

    • Rust Tests: Lexer has TokenKind::Def (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/default-impl.ori — ENTIRELY COMMENTED OUT (S06-09, violates hygiene rules)

  • Implement: Parse def impl Trait { ... } — grammar.ebnf § DECLARATIONS (verified 2026-03-29)

    • Rust Tests: ori_parse/src/grammar/item/impl_def/ — 5 parser tests (basic, public, multiple methods, empty, multiple def impls) (verified 2026-03-29)
    • Ori Tests: tests/spec/capabilities/default-impl.ori — ENTIRELY COMMENTED OUT (S06-09)

  • Implement: IR representation for DefImpl (verified 2026-03-29)

    • DefImplDef type exists, tracked in Module.def_impls (verified 2026-03-29)

  • Implement: Type checking for def impl

    • Rust Tests: ori_types/src/check/registration/ — register_def_impls
    • Verify trait exists
    • Method signatures converted to ImplMethodDef
    • Methods are associated (no self parameter)
    • One def impl per trait per module (coherence check)

  • Implement: Evaluator registration for def impl (verified 2026-03-29)

    • Rust Tests: ori_eval/src/module_registration.rs — collect_def_impl_methods (2 tests) (verified 2026-03-29)
    • Methods registered under trait name for TraitName.method() calls — end-to-end dispatch NOT working

  • Implement: Module export with default — 12-modules.md

    • Mark exports as “has default” when def impl exists
    • Bind default when importing trait

  • Implement: Name resolution — 14-capabilities.md

    • Check with...in binding first
    • Check imported default second
    • Check module-local def impl third
    • Ori Tests: tests/spec/capabilities/default-impl.ori

  • Implement: Evaluator dispatch for def impl — route method calls through default impl resolution in ori_eval

    • Dispatch method calls to bound default
    • Override via with...in works
    • Ori Tests: tests/spec/capabilities/default-impl.ori

  • Fix: Delete or uncomment tests/spec/capabilities/default-impl.ori — entirely commented out, hygiene violation (S06-09)

  • Implement: LLVM backend support

    • LLVM Rust Tests: ori_llvm/tests/default_impl_tests.rs
    • AOT Tests: No AOT coverage yet
    • Codegen for def impl methods

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.10) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.10 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.10: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.11 Capability Composition

Proposal: proposals/approved/capability-composition-proposal.md

Specifies how capabilities compose: partial provision, nested binding semantics, capability variance, and resolution priority order.

Implementation

  • Implement: Multi-binding with syntax — grammar.ebnf § EXPRESSIONS

    • Parser: Extend with_expr to support comma-separated bindings
    • Rust Tests: ori_parse/src/lib.rs — multi-binding with expression parsing
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Partial provision — providing some capabilities while others use defaults

    • Rust Tests: ori_types/src/check/tests.rs — partial capability provision
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Nested with...in shadowing — inner bindings shadow outer

    • Rust Tests: ori_types/src/check/tests.rs — capability shadowing
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Capability variance — more caps can call fewer, not reverse

    • Rust Tests: ori_types/src/check/tests.rs — variance checking
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Resolution priority order — inner with → outer with → imported def impl → local def impl → error

    • Rust Tests: ori_types/src/check/tests.rs — resolution priority
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Suspend binding prohibition — with Suspend = ... is compile error

    • Rust Tests: ori_types/src/check/tests.rs — suspend prohibition (E1203)
    • Ori Tests: tests/spec/capabilities/composition.ori

  • Implement: Error codes E1200-E1203

    • E1200: missing capability
    • E1201: unbound capability
    • E1202: type doesn’t implement capability trait
    • E1203: Suspend cannot be explicitly bound

  • Implement: LLVM backend support

    • LLVM Rust Tests: ori_llvm/tests/capability_composition_tests.rs
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.11) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.11 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.11: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.12 Default Implementation Resolution

Proposal: proposals/approved/default-impl-resolution-proposal.md

Specifies resolution rules for def impl: conflict handling, without def import syntax, re-export behavior, and resolution order.

Implementation

  • Implement: without def import syntax — grammar.ebnf § IMPORTS

    • Parse use "module" { Trait without def } to import trait without its default
    • Rust Tests: ori_parse/src/lib.rs — without def import modifier parsing
    • Ori Tests: tests/spec/capabilities/def-impl-resolution.ori

  • Implement: Import conflict detection — one def impl per trait per scope

    • Error E1000: conflicting default implementations when two imports bring same trait’s def impl
    • Rust Tests: ori_types/src/check/tests.rs — import conflict detection
    • Ori Tests: tests/spec/capabilities/def-impl-resolution.ori

  • Implement: Duplicate def impl detection — one per trait per module

    • Error E1001: duplicate default implementation in same module
    • Rust Tests: ori_types/src/check/tests.rs — duplicate def impl detection
    • Ori Tests: tests/spec/capabilities/def-impl-resolution.ori

  • Implement: Resolution order — with…in > imported def > module-local def

    • Innermost with...in binding takes precedence
    • Imported def impl overrides module-local
    • Rust Tests: ori_types/src/check/tests.rs — resolution priority
    • Ori Tests: tests/spec/capabilities/def-impl-resolution.ori

  • Implement: Re-export with without def — permanently strips default from export path

    • pub use "module" { Trait without def } re-exports trait without default
    • Rust Tests: ori_eval/src/interpreter/module/import.rs — re-export stripping
    • Ori Tests: tests/spec/capabilities/def-impl-resolution.ori

  • Implement: Error messages with help text

    • E1000: “use Logger without def to import trait without default”
    • E1001: show location of first definition
    • E1002: “def impl methods cannot have self parameter”
    • Rust Tests: ori_diagnostic/src/ — error formatting tests

  • Implement: LLVM backend support

    • LLVM Rust Tests: ori_llvm/tests/def_impl_resolution_tests.rs
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.12) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.12 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.12: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.13 Named Capability Sets (capset)

Proposal: proposals/approved/capset-proposal.md

Transparent aliases for capability sets. Expanded during name resolution before type checking. Reduces signature noise and creates stable dependency surfaces.

Implementation

  • Implement: Add capset keyword to lexer — grammar.ebnf § DECLARATIONS

    • Rust Tests: ori_lexer/src/lib.rs — capset token recognition
    • Ori Tests: tests/spec/capabilities/capset.ori

  • Implement: Parse capset_decl — grammar.ebnf § DECLARATIONS

    • Rust Tests: ori_parse/src/grammar/item/capset.rs — CapsetDecl AST node parsing
    • Ori Tests: tests/spec/capabilities/capset.ori

  • Implement: Name resolution expansion

    • Expand capset names in uses clauses to constituent capabilities
    • Transitive expansion (capsets containing capsets)
    • Deduplication (set semantics)
    • Rust Tests: ori_types/src/check/ — capset expansion tests

  • Implement: Capset cycle detection — topological sort in ori_types during capset expansion

    • Topological sort of capset definitions
    • Error E1220 for cyclic definitions
    • Ori Tests: tests/spec/capabilities/capset-errors.ori

  • Implement: Capset validation rules — error reporting in ori_types/src/check/

    • Error E1221: empty capset
    • Error E1222: name collision with trait
    • Error E1223: member is not capability or capset
    • Warning W1220: redundant capability in uses
    • Ori Tests: tests/spec/capabilities/capset-errors.ori

  • Implement: Capset visibility checking — pub capset members must be accessible

    • pub capset must not reference non-accessible capabilities
    • Ori Tests: tests/spec/capabilities/capset-visibility.ori

  • Implement: Enhanced E1200 error messages

    • Show capset expansion context in “missing capability” errors
    • Rust Tests: ori_diagnostic/src/ — error formatting tests

  • Implement: LSP support

    • Show capset expansion on hover
    • Autocomplete capset names in uses clauses

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.13) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.13 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.13: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.14 Intrinsics Capability

Proposal: proposals/approved/intrinsics-capability-proposal.md (v1) Proposal: proposals/approved/intrinsics-v2-byte-simd-proposal.md (v2 — generic API, byte SIMD, Mask type)

Generic SIMD API, byte-level SIMD, Mask<$N> type, bit manipulation, and hardware feature detection. Atomics deferred to separate proposal.

Implementation

  • Implement: Generic SIMD type validation — spec/20-capabilities.md

    • Validate T x N combinations (byte: 16/32/64, int: 2/4/8, float: 2/4/8)
    • Validate operation availability by lane type (div/sqrt/abs float-only, shuffle byte-only)
    • Error E1063 for invalid T x N combinations
    • Error E1064 for operation not available for lane type
    • Rust Tests: ori_types/src/check/ — SIMD type validation tests
    • Ori Tests: tests/spec/capabilities/intrinsics-type-validation.ori

  • Implement: Mask<$N> type — spec/20-capabilities.md

    • Add Mask as compiler-known type with const-generic parameter
    • Methods: bits, any, all, count, first_set
    • Implement BitAnd, BitOr, BitNot for Mask<$N>
    • Valid N values: 2, 4, 8, 16, 32, 64
    • Rust Tests: ori_types/src/check/ — Mask type tests
    • Ori Tests: tests/spec/capabilities/intrinsics-mask.ori

  • Implement: Add Intrinsics trait to prelude — spec/20-capabilities.md

    • Generic SIMD operations (simd_add<T, $N>, etc.)
    • Comparison operations returning Mask<N>
    • Byte-specific: simd_shuffle<$N>
    • Aligned loads: simd_load_aligned<T, $N>
    • Select: simd_select<T, $N> (mask-driven)
    • Ori Tests: tests/spec/capabilities/intrinsics.ori

  • Implement: Generic SIMD operations — float (128/256/512-bit)

    • simd_add, simd_sub, simd_mul, simd_div, simd_sqrt, simd_abs
    • simd_cmpeq, simd_cmplt, simd_cmpgt (return Mask<N>)
    • simd_min, simd_max, simd_sum, simd_splat
    • Ori Tests: tests/spec/capabilities/intrinsics-simd-float.ori

  • Implement: Generic SIMD operations — int (128/256/512-bit)

    • simd_add, simd_sub, simd_mul
    • simd_cmpeq, simd_cmplt, simd_cmpgt (return Mask<N>)
    • simd_min, simd_max, simd_sum, simd_splat
    • simd_and, simd_or, simd_xor, simd_andnot
    • Ori Tests: tests/spec/capabilities/intrinsics-simd-int.ori

  • Implement: Generic SIMD operations — byte (128/256/512-bit)

    • simd_load, simd_load_aligned, simd_add, simd_sub, simd_mul
    • simd_cmpeq, simd_cmplt, simd_cmpgt (return Mask<N>)
    • simd_min, simd_max, simd_sum, simd_splat
    • simd_and, simd_or, simd_xor, simd_andnot
    • simd_shuffle (byte-only), simd_select
    • Ori Tests: tests/spec/capabilities/intrinsics-simd-byte.ori

  • Implement: V1 deprecated aliases

    • Map simd_add_f32x4 → simd_add<float, 2>, etc.
    • Emit deprecation warning for v1 names
    • Ori Tests: tests/spec/capabilities/intrinsics-v1-compat.ori

  • Implement: Bit manipulation operations

    • count_leading_zeros, count_trailing_zeros, count_ones
    • rotate_left, rotate_right
    • Ori Tests: tests/spec/capabilities/intrinsics-bits.ori

  • Implement: Hardware feature detection

    • cpu_has_feature with valid feature strings (add ssse3, avx512bw)
    • Error E1062 for unknown features
    • Ori Tests: tests/spec/capabilities/intrinsics-feature-detect.ori

  • Implement: def impl Intrinsics (NativeWithFallback)

    • Native SIMD when platform supports
    • Scalar emulation fallback
    • Ori Tests: tests/spec/capabilities/intrinsics-fallback.ori

  • Implement: EmulatedIntrinsics provider

    • Always uses scalar operations
    • For testing and portability
    • Ori Tests: tests/spec/capabilities/intrinsics-emulated.ori

  • Implement: Intrinsics error diagnostics

    • E1060: requires Intrinsics capability
    • E1062: unknown CPU feature
    • E1063: invalid SIMD type x width combination
    • E1064: operation not available for lane type
    • Rust Tests: ori_diagnostic/src/ — error formatting tests

  • Implement: LLVM backend SIMD codegen

    • Generic dispatch: monomorphize to platform vector intrinsics
    • Mask<$N> codegen: <N x i1> in LLVM IR
    • Byte SIMD: <16 x i8>, <32 x i8>, <64 x i8>
    • Float SIMD: <2 x double>, <4 x double>, <8 x double>
    • Int SIMD: <2 x i64>, <4 x i64>, <8 x i64>
    • count_ones → llvm.ctpop.i64
    • count_leading_zeros → llvm.ctlz.i64
    • Runtime CPUID for feature detection
    • LLVM Rust Tests: ori_llvm/tests/intrinsics_tests.rs
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.14) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.14 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.14: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.16 Stateful Handlers

Proposal: proposals/approved/stateful-mock-testing-proposal.md

Extend with...in to support stateful effect handlers. The handler(state: expr) { ... } construct threads local mutable state through handler operations, enabling stateful capability mocking while preserving value semantics.

Implementation

  • Implement: Add handler as context-sensitive keyword — grammar.ebnf § EXPRESSIONS

    • Rust Tests: ori_lexer/src/lib.rs — handler token recognition (context-sensitive)
    • Ori Tests: tests/spec/capabilities/stateful-handlers.ori

  • Implement: Parse handler(state: expr) { op: expr, ... } — grammar.ebnf § EXPRESSIONS

    • Rust Tests: ori_parse/src/grammar/expr/with_expr.rs — handler expression parsing
    • Ori Tests: tests/spec/capabilities/stateful-handlers.ori

  • Implement: IR representation for handler expressions

    • Add HandlerExpr to expression AST (state initializer, operation map)
    • Rust Tests: ori_ir/src/ast/expr/tests.rs — handler AST node

  • Implement: Type checker — verify handler operations match trait signature

    • State replaces self in operation signatures
    • Operations return (S, R) where S is state type, R is trait return type
    • State type inferred from initializer, consistent across all operations
    • All required trait methods must have handler operations
    • Default trait methods used if not overridden
    • Rust Tests: ori_types/src/check/tests.rs — handler type checking
    • Ori Tests: tests/spec/capabilities/stateful-handlers.ori

  • Implement: Error codes E1204-E1207

    • E1204: handler missing required operation
    • E1205: handler operation signature mismatch
    • E1206: handler state type inconsistency
    • E1207: handler operation for non-existent trait method
    • Rust Tests: ori_diagnostic/src/ — handler error formatting
    • Ori Tests: tests/spec/capabilities/stateful-handler-errors.ori

  • Implement: Evaluator — handler frame state threading

    • Create handler frame with initial state on with...in entry
    • Thread state through capability dispatch calls
    • Independent state per handler (nested handlers)
    • with...in returns body result only (state is internal)
    • Rust Tests: ori_eval/src/interpreter/with_expr.rs — handler evaluation
    • Ori Tests: tests/spec/capabilities/stateful-handlers.ori

  • Implement: LLVM codegen for stateful handlers

    • Handler frame state allocation
    • State threading through operation calls
    • LLVM Rust Tests: ori_llvm/tests/stateful_handler_tests.rs
    • AOT Tests: No AOT coverage yet

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.16) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.16 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.16: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.

6.17 Section Completion Checklist

  • 6.1 declaration — uses clause verified WEAK TESTS, needs negative pins and LLVM tests (reopened, verified 2026-03-29)
  • 6.2 traits — WRONG TEST, zero active tests, traits.ori all commented out (reopened, verified 2026-03-29)
  • 6.3 suspend — STALE TEST (explicit suspension, sync vs suspending, await rejection have zero tests), No async modifier verified (reopened, verified 2026-03-29)
  • 6.4 providing — with...in and scoping verified with 31+ passing tests (verified 2026-03-29)
  • 6.5 propagation — partial: providing vs requiring verified, static transitive WRONG TEST (needs negative pins), runtime partial (reopened, verified 2026-03-29)
  • 6.6 trait definitions — 9 of 13 defined, 4 missing (Crypto, Print, Intrinsics, FFI), signature drift vs spec (verified 2026-03-29)
  • 6.7 testing/mocking — mock pattern verified WEAK TESTS (verified 2026-03-29)
  • 6.8 compile-time enforcement — implementation exists but WRONG TEST (zero Rust tests) and no negative pins (reopened, verified 2026-03-29)
  • 6.9 Unsafe — basic unsafe { } block IS implemented (6 passing tests), but marker capability enforcement NOT implemented (was not-started, now in-progress, verified 2026-03-29)
  • 6.10 Default implementations (def impl) — parser + IR + eval registration work, type checking + name resolution + end-to-end NOT working, test files all commented out (verified 2026-03-29)
  • 6.11 Capability Composition — not started (verified 2026-03-29)
  • 6.12 Default Implementation Resolution — not started (verified 2026-03-29)
  • 6.13 Named Capability Sets (capset) — not started (verified 2026-03-29)
  • 6.14 Intrinsics Capability (Generic SIMD, Byte Ops, Mask Type) — not started (verified 2026-03-29)
  • 6.16 Stateful Handlers — not started (verified 2026-03-29)
  • LLVM codegen for capabilities — zero LLVM/AOT test files exist
  • #compile_fail negative pins — zero exist for any capability feature (S06-06)
  • Prelude Async -> Suspend rename (S06-04)
  • Fix commented-out test files: traits.ori, default-impl.ori (S06-09)
  • Implement error codes E1200-E1207 from spec (only E2014 exists) (S06-10)
  • Full test suite: ./test-all.sh
  • /tpr-review passed — independent Codex review found no critical or major issues (or all findings triaged)
  • /impl-hygiene-review passed — implementation hygiene review clean (phase boundaries, SSOT, algorithmic DRY, naming). MUST run AFTER /tpr-review is clean.
  • /improve-tooling retrospective completed — MANDATORY at section close, after both reviews are clean. Reflect on the section’s debugging journey (which diagnostics/ scripts you ran, which command sequences you repeated, where you added ad-hoc dbg!/tracing calls, where output was hard to interpret) and identify any tool/log/diagnostic improvement that would have made this section materially easier OR that would help the next section touching this area. Implement every accepted improvement NOW (zero deferral) and commit each via SEPARATE /commit-push. The retrospective is mandatory even when nothing felt painful — that is exactly when blind spots accumulate. See .claude/skills/improve-tooling/SKILL.md “Retrospective Mode” for the full protocol.

Exit Criteria: Effect tracking works per spec (6.1-6.8 evaluator complete, 6.9-6.14, 6.16 pending) Status: Verified 2026-03-29. Significant accuracy problems found: phantom Rust tests cited in 6.2/6.3/6.5/6.8, stale status on 6.9, commented-out test files.

Remaining for Section 7 (Stdlib):

  • Real capability implementations (Http, FileSystem, etc.)
  • Add missing standard capabilities to prelude (Crypto, Print, Intrinsics, FFI)
  • Fix signature drift (Cache generics, Clock types)
  • Integration with stdlib modules

Remaining for Section 11 (FFI):

  • Unsafe capability enforcement for extern functions

  • /tpr-review passed — independent review found no critical or major issues (or all findings triaged)

  • /impl-hygiene-review passed — hygiene review clean. MUST run AFTER /tpr-review is clean.

  • Subsection close-out (6.17) — MANDATORY before starting the next subsection. Run /improve-tooling retrospectively on THIS subsection’s debugging journey (per .claude/skills/improve-tooling/SKILL.md “Per-Subsection Workflow”): which diagnostics/ scripts you ran, where you added dbg!/tracing calls, where output was hard to interpret, where test failures gave unhelpful messages, where you ran the same command sequence repeatedly. Forward-look: what tool/log/diagnostic would shorten the next regression in this code path by 10 minutes? Implement improvements NOW (zero deferral) and commit each via SEPARATE /commit-push using a valid conventional-commit type (build(diagnostics): ... — surfaced by section-6.17 retrospective — build/test/chore/ci/docs are valid; tools(...) is rejected by the lefthook commit-msg hook). Mandatory even when nothing felt painful. If genuinely no gaps, document briefly: “Retrospective 6.17: no tooling gaps”. Update this subsection’s status in section frontmatter to complete.

  • /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.