feat: sound module-qualified type & effect paths, dot separator (Refs #228)

[hyperpolymath]

Refs #228

STAGE-C peer of #225 (typed-wasm Http) and #160 (C-spine). Language
decision is human-gated per the repo ISSUE-CLOSURE rule, so this is
Refs #228, not Closes — do not auto-merge, do not close the issue.

What

The type/effect grammar had no module-qualified path production:
Externs.Res / prelude.Option / any Pkg.Type in type or effect
position failed parse error at the ., contradicting already-settled
ADR-011 (real modules with qualified paths). This implements full,
sound module-qualified type & effect paths with . as the canonical
separator (:: stays value/import per ADR-011). Recorded as ADR-014.

Resolution is by sound module-scoped member lookup, never flat
scope: load module [A;B] through the loader, resolve+typecheck it
exactly as an import would, then look the member up inside that
module's own resolved symbols, requiring Public/PubCrate and the
right kind (SKType for types, SKEffect for effects). An unknown or
wrong module, a missing member, a private member, or a wrong-kind
member is a resolution error at the use-site span — never a parse
error, never silently accepted. Importing a member flatly does not make
a bogus A.B.C resolve.

New grammar productions (verbatim)

path_seg:
  | s = lower_ident { s }
  | s = upper_ident { s }

module_prefix:
  | s = path_seg { [s] }
  | p = module_prefix DOT s = path_seg { p @ [s] }

qualified_type_path:
  | p = module_prefix DOT m = upper_ident
    { (p, m, $startpos, $endpos) }

type_expr_primary gains qualified bare / [args] / <args> forms
(TyCon/TyApp (mk_qualified_ident ...)); effect_term gains
qualified bare / [args] forms (EffVar/EffCon). Module-prefix
segments accept lower OR upper case because real stdlib modules are
prelude/option/string/result as well as Network/Http.

Files changed

File	Why
lib/ast.ml	ident gains modpath : string list; add mk_ident helper so synthetic idents set it in one place
lib/parser.mly	mk_qualified_ident; path_seg/module_prefix/qualified_type_path; qualified type & effect productions
lib/typecheck.ml	lower_type_expr/lower_effect_expr validate qualified paths via loader-backed qualified_member_check; Qualified_path_error → QualifiedPathError at the check_program boundary; validated qualified effect admitted as declared effect (#59)
lib/resolve.ml	make_qualified_member_check (sound module-scoped lookup) inside the resolve_and_typecheck_module recursive group
bin/main.ml	thread the validator into the 6 patterned check_program calls
lib/{trait,codegen_gc,borrow,verilog_codegen}.ml	route ~10 internal ident literals through Ast.mk_ident
docs/specs/SETTLED-DECISIONS.adoc, .machine_readable/6a2/META.a2ml	ADR-014 (same shape as ADR-011/012/013)
test/test_e2e.ml, test/dune, tests/conformance/qualified-paths/**	5 conformance fixtures wired into the gate
test/golden/*.expected	regenerated — adding modpath reflows every AST dump; span-normalised golden harness confirms no structural change

Parser-conflict delta: ZERO

menhir --explain summary is byte-identical to the pre-change
parser:

21 states have shift/reduce conflicts.
one state has reduce/reduce conflicts.
68 shift/reduce conflicts were arbitrarily resolved.
7 reduce/reduce conflicts were arbitrarily resolved.

The upper_ident vs upper_ident DOT … decision is the expected
benign DOT shift (Menhir shifts; disclosed per ADR-012). No new
unexplained conflict; just build masks the same count as before.

Test gate

	Before	After
dune runtest	258 PASS / 0 FAIL	263 PASS / 0 FAIL (+5 #228 conformance cases)

Zero regressions. The 12 golden tests pass with regenerated .expected
(the change is purely the new modpath field + ppx_deriving.show
reflow; the span-normalising golden harness verifies no structural
change).

Oracle re-audit (build is the new oracle)

28 repos / 1176 .affine files. Pre-fix baseline vs post-fix:

Class	Pre-fix	Post-fix	Δ
PASS	552	552	0
DRIFT-SYNTAX	498	491	-7
TYPE-ONLY	126	133	+7

7 files moved DRIFT-SYNTAX → TYPE-ONLY (accessibility-everywhere ×5,
standards ×2): their only parse blocker was a qualified type/effect
path; they now parse and resolve. Spot-check (before → after):

• accessibility-everywhere/tools/cli/src/Index.affine: 6:17 parse error → resolves (now UndefinedModule, sound)
• standards/a2ml/.../Json.affine: 15:29 parse error at A2ml.Inline → resolves
• standards/k9-svc/.../K9_Renderer.affine: 17:37 parse error at K9_Types.Pedigree → resolves

The PASS count is unchanged (not the "large jump" the spec hoped for)
because the remaining DRIFT-SYNTAX is unrelated, out-of-#228-scope
syntax — extern fn … = "console" "warn" FFI string-bindings,
pub extern fn, record-construction expression syntax — confirmed by
sampling the residual parse errors. And because the audit harness uses
a bare loader without each repo's local module search paths, a
qualified path to a repo's own Externs/Types module now correctly
reports UndefinedModule (sound — not silently accepted) rather than
masquerading as a parse error. This is the correct sound outcome per
the spec ("a wrong/unknown module … is an error").

A/B sanity probes (all hold)

• prelude.Option[Int] / prelude.Option<Int> → Type checking passed
• bare Option[Int] → still Type checking passed
• Foo.Bar[Int] → resolution error Unknown module 'Foo' … (ADR-014) (not parse)
• effects.IO → Member 'IO' of module 'effects' is private … (ADR-014) (sound visibility)
• effmod.Logging (a pub effect) → Type checking passed
• prelude.Option used as effect → 'prelude.Option' resolves to a type, not a effect (ADR-014)
• prelude.DefinitelyNotHere → Module 'prelude' has no member … (ADR-014)
• Int -{IO}-> Int → still parses (Type checking passed)

Deviations from spec (reality)

1. Resolution happens in typecheck.ml, not resolve.ml. The spec
   directed edits to Resolve.resolve_type_ident/resolve_effect_ident,
   but those functions are dead code — actual type/effect name
   binding is in Typecheck.lower_type_expr/lower_effect_expr.
   typecheck.ml cannot depend on resolve.ml (would be circular), so the
   sound module-scoped validator is built in resolve.ml (where the
   loader + resolve_and_typecheck_module live) and injected into
   the typecheck context via an optional qualified_member_check
   closure threaded through check_program. Same soundness, no
   circular dependency.
2. Module-prefix segments must accept lowercase. The spec said "a
   path of upper_ident segments", but real stdlib modules are
   lowercase (prelude, option). The grammar accepts lower-or-upper
   prefix segments (mirroring module_path/use), member is
   upper_ident. Conflict count still zero-delta.
3. Qualified effect admitted as declared effect. A qualified effect
   that passes the module-scoped gate is registered in
   declared_effects so issue-[Repo] Effect-row v1 name list: define IO/Async/Partial/Throws[E]/Mut early #59's effect-name validator accepts it
   (a sibling module's public effect is as legitimate as a local one).
   Soundness preserved — only after the gate passes.
4. Typecheck/codegen logic otherwise unchanged (as predicted): a
   qualified ident is treated by its resolved symbol exactly like an
   unqualified one.

🤖 Generated with Claude Code

Known limitation (disclosed, not a blocker)

The qualified-type positive conformance fixture is self-contained
(prelude.Option — a stdlib module). The qualified-effect positive
fixture (tests/conformance/qualified-paths/valid/qualified_effect.affine)
references a sibling support module effmod because no stdlib module
currently declares a public effect. Consequence:

• It passes the test gate (the harness adds the fixture dir to the module
  search-path) but fails a bare affinescript check with
  Unknown module 'effmod'.
• Bare-oracle audits — including the estate dialect-conformance
  .affine-audit harness — must treat that one fixture as harness-only
  and exclude it from drift counts. This is now stated in the fixture
  header and in ADR-014.
• Qualified-effect resolution itself is sound and independently proven
  by invalid/private_member.affine (loads stdlib effects, rejects the
  private member). The limitation is fixture self-containment, not the
  feature.

Also disclosed: the estate-payoff numbers from the re-audit are
understated by the bare-loader audit harness (repo-local modules
report UndefinedModule). A re-audit with each repo's own module graph
resolvable is pending and is the load-bearing review evidence.

[hyperpolymath]

#228 estate-payoff proof (controlled before/after)

Method: same .affine files, two binaries — before = affinescript main
@417b97c (untouched), after = stage-c/228-qualified-type-effect-path.
Per-file first-error position+class captured. The coarse PASS/DRIFT count
understates #228 (oracle stops at the FIRST parse error); this measures
whether that first error stops being a module-qualified-path parse error.
Harness: .affine-audit/payoff.sh, raw: .affine-audit/payoff.tsv.

Result — #228 works, scope is narrow (honest correction)

Seed: 13 repos, ~98 files. Only accessibility-everywhere had
qualified-path drift. The other 12 (airborne-submarine-squadron,
idaptik, idaptik-dlc-vm, stapeln, aerie, ambientops, echidna,
game-server-admin, git-scripts, invariant-path, ubicity, panll) showed
0 wall-cleared / 0 wall-advanced — their drift is ReScript-surface /
record-literal = #229 territory, NOT #228. This corrects the earlier
"single dominant fault estate-wide" framing: the qualified-path gap was
dominant in the diagnostic seed (accessibility-everywhere), not estate-wide.

accessibility-everywhere — controlled before/after (per-package cwd)

Bucket	n	BEFORE (main)	AFTER (#228)
qualified-path files	11	all parse error	5 parse-clean → resolution; 6 parse-advanced past the qualified wall; 0 regress
Externs.affine support mods	4	PASS	PASS (unchanged)

Sharpest proof: tools/monitoring-api/src/Server.affine — parse error
95:31 (the mut Externs.Res / -{Externs.Net}-> qualified wall) →
264:32: 169 lines of qualified-path walls removed, residual is a
later non-qualified construct (#229).

WALL-CLEARED (parse error → no longer a parse error): Cli, cli/Index,
github-action/Index, core/Index, scanner/Index.
WALL-ADVANCED (qualified wall gone, later non-qualified error): Server
95→264, Arangodb 103→127, Scanner 81→111, Action 105→114, Button
42:23→42:36, react/Index 17→30.

What this means for review

• LANG: type/effect grammar has no module-qualified path — Pkg.Type/Pkg.Effect unrepresentable (estate-wide port blocker; ADR-014) #228 is necessary, correct, and attributable — it removes a
  language-level wall that made an idiomatic construct unrepresentable,
  and demonstrably unblocks the one repo whose drift genuinely was that
  gap. No regressions.
• LANG: type/effect grammar has no module-qualified path — Pkg.Type/Pkg.Effect unrepresentable (estate-wide port blocker; ADR-014) #228 is not, by itself, an estate green-maker: most estate drift is
  non-qualified (ESTATE: total ReScript-surface elimination from all .affine (committed goal; blocked-on #228) #229), and full typecheck additionally needs each repo's
  coherent (often multi-package) module graph. Those are out of LANG: type/effect grammar has no module-qualified path — Pkg.Type/Pkg.Effect unrepresentable (estate-wide port blocker; ADR-014) #228's
  contract by design.
• Net: land LANG: type/effect grammar has no module-qualified path — Pkg.Type/Pkg.Effect unrepresentable (estate-wide port blocker; ADR-014) #228 for the language-correctness win (ADR-011 consistency,
  sound qualified paths); the estate sweep value is realised later via
  ESTATE: total ReScript-surface elimination from all .affine (committed goal; blocked-on #228) #229 + per-repo module coherence, measured with this same harness.

[hyperpolymath]

Independent gate re-confirmation (not relayed from the implementing agent)

Both #231 gates re-verified from scratch:

Test suite — dune runtest in the worktree: exit 0, alcotest reports
Test Successful — 263 tests run, 0 failures. The pre-existing suite
is a subset and is fully green; the +5 over baseline are exactly the new
tests/conformance/qualified-paths/** fixtures. No regressions.

Menhir conflict delta — standalone menhir 20260209 (not via dune) on
git show main:lib/parser.mly vs git show HEAD:lib/parser.mly, same
--table flag:

21 states have shift/reduce conflicts.
68 shift/reduce conflicts were arbitrarily resolved.
7  reduce/reduce conflicts were arbitrarily resolved.
one state has reduce/reduce conflicts.

Byte-identical on both grammar versions — diff empty. Confirms ADR-014's
"the masked-conflict count is unchanged … identical to the pre-change
parser". The new qualified-path productions add the expected benign DOT
shift and introduce zero new conflicts.

Net: implementation gates verified independently. #231 residual is now
only the human review + merge it explicitly requires (Refs #228, not
Closes; human-gated per ISSUE-CLOSURE).

[github-actions]

🔍 Hypatia Security Scan

Findings: 44 issues detected

Severity	Count
🔴 Critical	12
🟠 High	21
🟡 Medium	11

⚠️ Action Required: Critical security issues found!

View findings

[
  {
    "reason": "Stray AI.a2ml in root -- use 0-AI-MANIFEST.a2ml only",
    "type": "banned",
    "file": "AI.a2ml",
    "action": "delete",
    "rule_module": "root_hygiene",
    "severity": "high"
  },
  {
    "reason": "Superseded by 0-AI-MANIFEST.a2ml",
    "type": "banned",
    "file": "AI.djot",
    "action": "delete",
    "rule_module": "root_hygiene",
    "severity": "high"
  },
  {
    "reason": "Issue in quality.yml",
    "type": "missing_workflow",
    "file": "quality.yml",
    "action": "create",
    "rule_module": "workflow_audit",
    "severity": "high"
  },
  {
    "reason": "Issue in security-policy.yml",
    "type": "missing_workflow",
    "file": "security-policy.yml",
    "action": "create",
    "rule_module": "workflow_audit",
    "severity": "medium"
  },
  {
    "reason": "Action hyperpolymath/standards/.github/workflows/governance-reusable.yml@main needs attention",
    "type": "unpinned_action",
    "file": "governance.yml",
    "action": "pin_sha",
    "rule_module": "workflow_audit",
    "severity": "high"
  },
  {
    "reason": "TypeScript file detected -- banned language",
    "type": "banned_language_file",
    "file": "/home/runner/work/affinescript/affinescript/affinescript-deno-test/example/smoke_driver.ts",
    "action": "flag",
    "rule_module": "cicd_rules",
    "severity": "critical"
  },
  {
    "reason": "TypeScript file detected -- banned language",
    "type": "banned_language_file",
    "file": "/home/runner/work/affinescript/affinescript/affinescript-deno-test/cli.ts",
    "action": "flag",
    "rule_module": "cicd_rules",
    "severity": "critical"
  },
  {
    "reason": "TypeScript file detected -- banned language",
    "type": "banned_language_file",
    "file": "/home/runner/work/affinescript/affinescript/affinescript-deno-test/mod.ts",
    "action": "flag",
    "rule_module": "cicd_rules",
    "severity": "critical"
  },
  {
    "reason": "TypeScript file detected -- banned language",
    "type": "banned_language_file",
    "file": "/home/runner/work/affinescript/affinescript/affinescript-deno-test/lib/compile.ts",
    "action": "flag",
    "rule_module": "cicd_rules",
    "severity": "critical"
  },
  {
    "reason": "TypeScript file detected -- banned language",
    "type": "banned_language_file",
    "file": "/home/runner/work/affinescript/affinescript/affinescript-deno-test/lib/runner.ts",
    "action": "flag",
    "rule_module": "cicd_rules",
    "severity": "critical"
  }
]

Powered by Hypatia Neurosymbolic CI/CD Intelligence

[hyperpolymath]

Superseded by #241 (merged): ADR-014 module-qualified type/effect paths landed with zero Menhir conflict delta, 270/270 gate, ADR recorded in META.a2ml + SETTLED-DECISIONS.adoc. #231 was the grammar-only DRAFT; #241 is the complete, gated implementation.