Phase 1 / Etch / Graduate const, private, and test

[guysenpai]

M1.0.8 — Graduate const (top-level), private, and test

Brief: briefs/M1.0.8-const-private-test.md

Graduates the last three reserved non_s3_keywords parser-up (lexer → AST → parser → resolver), the same shape as import in M1.0.7. override stays reserved (waits for a Tier-1 overridable module). Also clears the two M1.0.7 deferred debts: cross-file const resolution and activation of the dormant E0107 ImportPrivateItem check.

Closing notes

• What worked: The M1.0.7 import graduation was a faithful template — lexer table move + AST slab + parse-fn + lockstep set + pass-1 registration + buildExports row. The E0107 path was genuinely wired-and-dormant in M1.0.7, so activation was a single change (buildExports reading Item.visibility). checkConstValue reused the field-default const surface verbatim (isConstEvaluable + synthExpr + literalTypeFits).
• What deviated from the original spec: Only the three recorded deviations (test placement / the one-vs-two enumeration / a stale comment). No scope or behavioural deviation. parseStmt was left untouched (block-level const is a parse error, per scope); override stays reserved.
• What to flag explicitly in review:
  1. Cross-file tests placed in tests/etch/import_resolve_test.zig, not inline in types.zig — forced by the tier dependency (types.zig cannot reach root.validateProject). The single-file tests (E0200, E1101, registration) are inline in types.zig as specified.
  2. src/etch/diagnostics.zig touched (outside the brief's file list) for a one-line §3.6.1 orphan-reference patch — the E0107 comment said "wired but dormant until M1.0.8", now stale post-activation.
  3. The lockstep had exactly ONE top-level error-message enumeration on the surface (parser.zig:690), not the two the brief states.
• Final measurements: N/A (parser/resolver graduation, no perf gate). Suite: 872 tests pass debug + ReleaseSafe; +17 new tests (token 1, lexer 1, parser 6, types 4, import_resolve 5).
• Residual risks / tech debt left intentionally: A string-named test "X" registers under the byte sequence X via registerSymbol, sharing the name namespace with identifier-named symbols (test "Foo" would collide with component Foo → E0101). Acceptable for M1.0.8 (the brief mandates registerSymbol); revisit when the M1.0.9 test-runner formalizes test identity. Out-of-scope-by-design: statement-level const (tri-doc drift, deferred to KB-audit), visibility inheritance + W0902, qualified m.Type, override.

Validation checklist

• All Scope deliverables present (lexer graduation, AST nodes, parser, resolver symbols + const validation, exports + E0107 activation, cleared M1.0.7 debts)
• No drift into Out-of-scope (parseStmt untouched, override reserved, no test execution, no m.Type, no W0902/visibility inheritance)
• All acceptance Tests pass in debug and ReleaseSafe
• Benchmarks: N/A
• Observable behavior demonstrable (two-file fixture: ROOM_CAP import resolves clean; Secret import → E0107)
• zig build (exit 0, no warnings), zig build test (debug + ReleaseSafe), zig fmt --check, zig build lint all green
• CLAUDE.md §3.4 updated and committed on the branch
• Brief "Closing notes" filled (4 rubrics + measurements + residual debt)
• Brief Status: → CLOSED, Closed: date set
• commit-msg hook green on every commit; pre-push gate green (build + test + test-release)

Merge + tag (v0.10.8-const-private-test) are left to Guy after review.

🤖 Generated with Claude Code