Phase 1 / Etch / Test-runner + test identity

[guysenpai]

Milestone M1.0.15 — Etch test-runner + test identity. Brief: briefs/M1.0.15-test-runner.md.

test "X" { } blocks now execute end-to-end (parse + validate + register existed since M1.0.8; zero execution surface until now), closing the M1.0.15 line of the C1.6 gap plan. Delivered gate-by-gate E1→E5.

Closing notes

What worked

• The E1 recon's five points held exactly — compile() is the full type-registration point, so "fresh World + Interpreter per test" fell out cleanly; no gap in bindToWorld.
• Reusing existing primitives paid off: spawn_with shares the observer-firing spawn path (spawnWithObservers factored from the deferred .spawn flush); emit method shares enqueueEvent; assert string comparison reuses eventValueEql; E0101 reused for duplicate tests (no new code).
• Adversarial-review workflows earned their cost: caught 5 real E4 defects (incl. a dangerous assert_neq false-PASS on strings) before they shipped.
• Break-tests (neuter a fix/assertion, confirm the failure, revert) proved both the UAF fix and that the new inline/driver tests actually execute — after the §13 wire-in gap surfaced.

What deviated from the original spec / brief plan

• Two mechanisms not in the brief's file-level plan, both minimal + confined and ack'd by STOP fixes: suppress_body_store_resets (E3 UAF) and in_test_body runtime flag (E4 test-scoped-builtin gate).
• A Tier-0 touch (observers.zig/world.zig spawnWithObservers) required by the brief's own "reuse the same application primitive" trap — one spawn path, no flush behavior change.
• test-etch runs the shim over the green fixtures in ONE invocation (clean un-interleaved ✓ output) rather than one Run step per fixture.

What to flag explicitly in review

• Event-clear semantics for emit; tick(1): stepOnce clears the event store at its head, so tick/tick_until set a one-shot suppress_event_clear to let a pre-tick event reach the first tick (§32). Non-test paths never set it.
• assert_eq/assert_neq are intentionally MORE capable than == (which rejects strings at type-check): they compare strings by bytes and reject aggregates — a test-only Eq surface, so assert_neq cannot silently false-PASS.
• The hunt:semantics adversarial agent stalled (never returned); the other three lenses + inline/driver tests cover measure/tick_until/Duration/asserts.

Final measurements — no perf gate this milestone. zig build test green debug + ReleaseSafe; zig build test-etch prints total: 10 passed / 0 failed / 2 skipped (exit 0); measure around a 5000-iter interp loop reports ~5 ms wall-clock. New diagnostic: E0910 (sole new code).

Residual risks / tech debt (intentional)

• Memory during a long tick(n)/tick_until inside a test grows (per-body store resets suppressed) — bounded by the test body's end (freed en bloc); test-scale, acked at the E3 STOP.
• test_world() is mono-world (spec'd v0.6 restriction, not debt); multi-world, async test bodies, the weld CLI proper, and filtering flags are out per §32.

Validation (Step 4)

• All "Scope" deliverables present (E1 identity + annotations + body type-check; E2 runner library; E3 world surface; E4 assertions + measure + tick_until; E5 shim + corpus + CI).
• No drift into "Out of scope".
• All "Acceptance criteria › Tests" pass in debug and ReleaseSafe (driver_test.zig + inline suites).
• "Observable behavior" demonstrable — zig build test-etch prints ✓/skip lines + aggregate (exit 0); etch_test failing.etch prints the ✗ line + file:line + exit 1.
• zig build, zig build test, zig build test-etch, zig fmt --check, zig build lint all green.
• CLAUDE.md updated (§3.4 current-state table + Tags row + open-decisions bullet) — commit docs(claude-md): update for M1.0.15.
• Language audit (§3.6.1) — no FR function-words in the branch diff or brief (the lone .le match is the Zig <= enum tag).
• Local drift audit (§3.6.1) — stale kw_test comment refreshed; no residual SymbolKind.test_ (removed E2); .test_ matches are AnnotTarget.test_.
• Brief "Closing notes" filled; Status: CLOSED, Closed: 2026-07-05; final commit docs(brief): close M1.0.15.

Merge + tag (v0.10.15-test-runner) + close decision are Guy's.

🤖 Generated with Claude Code