[M13] Resolve and lower the accepted protect-ban OSTW semantic slice into HIR

[Teakowa]

Parent: #90
Depends on: #117

Goal

Implement the Wright-owned OSTW semantic resolution and HIR lowering required by the accepted protect-ban project entry-point reachable graph, without expanding semantic support merely because other .ostw/.del files exist in the repository/corpus.

Context

#115 established protect-ban as the first accepted reference project under pinned OSTW v3.4.0. #117 owns project loading and syntax. Pinned-reference inspection during #117 acceptance clarified an important boundary: ds.toml.entry_point is the main file for analysis/compilation, and upstream Importer.FromEntryPoint(...) recursively includes only files reached through imports.

Therefore the first semantic/compile support claim is defined by the ds.toml.entry_point reachable import graph. Other committed protect-ban sources remain licensed parser/corpus evidence and may justify later semantic expansion, but they do not automatically become compilation inputs.

The accepted reachable graph still provides the concrete semantics this issue must inventory and implement; any additional construct is added only when it is actually reachable from the pinned entry point or separately reference/corpus justified.

Scope

• Consume the corrected [M13] Establish native OSTW project and syntax frontend foundation #117 project graph rather than scanning every source file as a semantic root.
• Build OSTW scopes/name resolution for declarations and rule/function contexts exercised by the reachable protect-ban graph.
• Support the core types actually exercised by that graph, including primitive Workshop-facing types, arrays, enum types, union/null forms, and other evidenced type forms.
• Resolve typed globalvar / playervar, constants/define-like declarations, initializers, explicit variable IDs where present, and player-variable receiver access where reachable/evidenced.
• Resolve enums and enum members used by the accepted graph.
• Resolve function/macro declarations and calls, parameters/returns, and compile-time/value semantics required by the accepted graph.
• Lower accepted expressions such as literals, arrays, indexing, unary/binary operators, casts, ternaries, calls, named arguments, and member access where evidenced.
• Lower accepted control flow and rule structure such as rules/events/conditions, if/else, while, foreach, switch/fallthrough, break, continue, and return where evidenced.
• Resolve Workshop actions/values through Wright-owned canonical Workshop catalog/signature semantics rather than importing OSTW game-derived tables.
• Lower the accepted semantic surface into existing Wright HIR so downstream analyzer/IR services remain frontend-neutral.
• Preserve structured, source-located semantic diagnostics and cross-file provenance.
• Add focused pinned-reference probes/fixtures where observable behavior needs clarification, plus semantic regression coverage for the complete reachable graph.
• Keep a separate inventory of syntax-parsed but semantically unsupported/unreachable protect-ban constructs so future expansion is evidence-driven rather than accidental.

Non-goals

• Treating every source file under the project root as a semantic/compilation root.
• Classes, inheritance, heap semantics, generics, lambdas, pattern matching, or other advanced OSTW categories unless they are demonstrated by the accepted reachable graph or separately approved evidence.
• Making MOBAwatch compile successfully; the pinned reference itself currently rejects that project.
• OSTW-specific analyzer/IR forks.
• Workshop emission/output comparison; tracked separately in [M13] Compile the accepted OSTW corpus slice to Workshop with reference differential validation #119.
• Workshop -> OSTW reconstruction.
• Upstream implementation parity or copying source/data from the unlicensed OSTW compiler.

Acceptance criteria

• The complete ds.toml.entry_point reachable graph of pinned protect-ban resolves semantically through Wright's native OSTW frontend and lowers to valid Wright HIR under a documented support boundary.
• Unreachable project files do not affect semantic success/failure merely because they exist under the project root.
• Cross-file imports/names, variables, enums, functions/macros, rules, control flow, arrays, Workshop calls, and player-variable access actually exercised by the accepted graph resolve consistently with pinned reference behavior.
• Unsupported/invalid reachable semantic forms fail with deterministic structured source-located diagnostics rather than being deferred to Workshop emission.
• Workshop call/value semantics reuse Wright-owned canonical catalog/signature data; no OSTW-specific duplicate game-content table is introduced.
• HIR produced by the OSTW frontend is consumable by existing Wright semantic/analyzer services without language-specific branches in HIR/WIR.
• Parser robustness over additional committed protect-ban sources remains a separate test/evidence layer and does not silently expand the semantic support claim.
• Advanced categories remain explicitly unsupported unless new corpus/reference evidence is added.
• Current tests, compatibility gates, and CI remain green.

Relationships

• Implements the semantic phase of the first [M13] Source-language interoperability and OSTW compatibility #90 corpus-defined OSTW support slice.
• Depends on [M13] Establish native OSTW project and syntax frontend foundation #117 project-graph/syntax foundation.
• Enables OSTW -> Workshop compilation ([M13] Compile the accepted OSTW corpus slice to Workshop with reference differential validation #119) and tooling integration ([M13] Integrate the accepted OSTW semantic surface with Wright tooling services #120).

[Teakowa]

Semantic inventory and implementation boundary (pinned OSTW v3.4.0)

This replaces a whole-directory reading with the corrected ds.toml.entry_point contract. The evidence identity is OSTW v3.4.0 / commit 769ce7aab097178cfe905bf21f0326d8e0d12e6b / release SHA 1ae882…e3a88, and the pinned protect-ban revision is f8c2353ed8447f13038fbf6b9938031cced5796f. The existing oracle observation is accept, 79 elements, Workshop hash 75722f…03912 (compatibility/ostw/reference.json, results.json). It is S/N-reference evidence for that full project run, not semantic parity evidence.

1. Entry-point reachable compilation graph

Starting at ds.toml: entry_point="main.ostw", following active quoted imports only (cycles deduplicated):

main.ostw
├─ Credits.ostw
├─ interface/HeroSelect.del
│  ├─ main.ostw                     (cycle)
│  ├─ interface/HeroSelectConfig.del
│  │  ├─ interface/HeroSelect.del   (cycle)
│  │  └─ interface/MapData.del
│  ├─ interface/HeroSelectFunctions.del
│  │  ├─ HeroSelectConfig.del       (cycle)
│  │  └─ HeroSelect.del             (cycle)
│  └─ ../OSTWUtils/{OnScreenText,Cursor,StringSorting}.del (not committed)
└─ interface/miscSetup.del
   └─ main.ostw                     (cycle)

Thus the in-repository semantic root is exactly seven files: main.ostw, Credits.ostw, interface/HeroSelect.del, HeroSelectConfig.del, HeroSelectFunctions.del, MapData.del, and miscSetup.del. Three active, source-located imports leave the licensed corpus. They must remain a deterministic ostw-missing-import boundary until a pinned, redistributable source/probe establishes their reference resolution; they do not authorize Cursor/classes or OSTWUtils semantics. The existing #117 parser test already records these three as out-of-closure; its fourth missing import belongs only to unreachable protectBanFull.ostw.

Exercised construct	Syntax / name and type behavior	Required semantic lowering; reusable Wright representation	Catalog/reference status
Imports and cross-file names	Quoted relative imports, cycles, program-wide declarations referenced across files. ds.toml selects only main.ostw.	Build the deduplicated graph, collect declarations before bodies, retain FileId/spans. HIR already owns file registry and spans; no frontend AST leak.	Pinned project accepts, but entry-only vs all-open LSP behavior and the three missing edges need probe P1/P2 below.
Global/player variables and IDs	Typed globalvar and playervar, inferred/default initialization, cross-file references. Types used: Number, String, Boolean, Team, Hero, Player, Vector, Any, arrays; exact explicit global id is i 127. No reachable explicit player id.	Reuse HIR GlobalVar/PlayerVar (including index, initializer, spans), then WIR variable tables and synthesized initializer rules. Validate explicit IDs/collisions/allocation before WIR.	Existing pinned output proves the project accepts; exact allocation/collision policy and 127 placement require P3, rather than copying OSTW allocation rules.
User enum and enum values	enum Phase { Waiting, Protect, Ban, ExtraBan, Gameplay }; members are used as Phase.X. Built-in qualified enums include Team/Hero/Button/Color/etc.	Built-in enum values fit HIR/WIR Expr::Enum; resolve them through the Wright Workshop catalog. User enum declaration/member identity has no HIR node today, so add a frontend-neutral representation or explicitly block it; do not encode as a catalog enum.	Phase acceptance is evidenced by protect-ban; member value/type and coercion behavior need a minimal enum probe (P4).
Null, unions, arrays, indexing	`Team	Number, Hero	Hero[], Cursor
Typed declarations/functions	Expression-bodied typed declarations (Number TeamIndex(Team team): ..., computed names such as PhaseName), void functions with typed parameters, calls/returns, and cross-file calls.	Calls fit HIR Expr::Call / WIR value/action calls only after resolution. Current HIR Subroutine is parameterless and models Workshop subroutines, not typed value-returning functions; parameter scopes, return values and call expansion/lowering need a frontend-neutral HIR extension.	Accepted project is evidence of syntax/reference acceptance, not a contract for function evaluation order/defaults; P5 required.
Macros / define	Not exercised by the seven-file graph.	No semantic support claim. Existing HIR Constant/Macro is potential reuse only after separately evidenced OSTW define semantics.	Syntax-only evidence exists in unreachable utils/{AltFont,Math,Colors,ServerLoad}.del; later.
Rules, events, conditions	Active/disabled rule:, omitted event (global), Event.OngoingPlayer, ordered rule "Create about and credits" -1, and repeated rule if conditions.	Reuse HIR Rule/Event/conditions and WIR Global/EachPlayer event forms. Preserve disabled flag. Rule priority has no HIR/WIR field: add a neutral field or reject until proven—never drop it.	Pinned project accepts; exact omitted-event and priority semantics need P1/P5-style focused probe before lowering.
Control flow	if/else, while, C-style for (i = 0; stop; step), typed foreach, switch/cases/default with intentional fallthrough, break, return; no reachable continue.	HIR/WIR already represent if/while. C-for/foreach can only reuse WIR ForGlobalVariable after semantic equivalence is proven; local loop bindings, switch/fallthrough/break/return require an extension or documented desugaring.	Pinned acceptance only; fallthrough/return/loop timing are observable and require P5 before desugaring.
Expressions	Number/string/bool/null literals; formatted <"…", …> strings; vectors; unary !/negation; arithmetic, comparison, &&/`		, ternary, assignment and compound assignment; currentPhase` cast.
Named arguments	Named arguments occur heavily in Workshop calls (VisibleTo:, Header:, Reevaluation:, Player:, etc.), mixed with positional arguments.	Resolve labels/defaults/reordering against a Wright-owned signature table before constructing ordered HIR calls. Do not preserve OSTW positional rules by guesswork.	Current Workshop catalog has only 58 action/value/event IDs and only a small subset of this call surface (for example wait, createHudText, setGravity, teleport, allPlayers, countOf, mappedArray, vector). P6 must establish named/default behavior; catalog expansion is a Wright-catalog task with provenance, not OSTW data import.
Receiver/member semantics	Player-variable receiver reads/writes (EventPlayer().isReady, AllPlayers().isReady, LocalPlayer().selectionIndex); method calls on the unresolved Cursor (getX, getY, getAimSpeed, updateAimSpeed, close).	Reuse HIR/WIR PlayerVar { player, variable } for the player-variable subset, including write/modify actions. Do not treat arbitrary .member() as a Workshop call; Cursor members are blocked by P2/source availability.	Player-variable access is reachable and must have a targeted P3 probe. Cursor/class behavior is later, evidence-prioritized.
Workshop values/actions	The graph calls settings functions, array/value functions, HUD/world-text/effect/camera/movement/game-mode actions and enum arguments.	Every builtin resolves by canonical Wright catalog id plus owned signature/type metadata, then maps to existing HIR Call and WIR Call/Action. Add only the exact exercised catalog entries/signatures; no OSTW-specific game table.	Pinned output proves some accepted aggregate, not per-signature semantics. Current catalog is materially incomplete for this set, so each added signature needs pinned oracle probe evidence (P6).

2. Syntax-only, non-compilation evidence

The other nine committed sources are deliberately outside this semantic claim: coreDebug.ostw, protectBanFull.ostw, interface/{ClickArea,PlayerInterface}.del, and utils/{AltFont,Colors,Math,ScreenToWorld,ServerLoad}.del. They remain S-level parser robustness inputs only. In particular they introduce define macros/default parameters, classes/constructors/new, generic define locals, casts on class values, i++, and an additional missing customGameSettings.lobby import. They must not make classes, macros, generic/define typing, or project-settings imports part of #118.

3. Explicitly later / unsupported

Until separately evidenced, reject with source-located diagnostics: classes/inheritance/heap/constructors/delete; the missing OSTWUtils APIs; macros/define semantics; generics/lambdas/pattern matching; structs/ref/in semantics; extended collections; project settings other than entry_point; vanilla Workshop embedded syntax; optimizer/output parity; and Workshop-to-OSTW reconstruction. MOBAwatch remains outside this support claim.

Small pinned-reference probe set (do before committing to ambiguous lowering)

• P1 entry closure: entry -> a -> b plus an unrelated invalid .del; run the pinned server with only entry opened and with all files opened. Record whether only the transitive graph affects compile/diagnostics, omitted-event behavior, and priority output.
• P2 missing import: one active missing import, then the protect-ban-shaped missing-OSTWUtils edge. Record acceptance, diagnostic code/range, and whether an ambient/module source can satisfy it. This decides the exact boundary; no inferred external library lookup.
• P3 variables: automatic globals around globalvar Number i 127, duplicate IDs, and EventPlayer().p read/write. Snapshot emitted variable table and diagnostics.
• P4 types/expressions: user enum, null assigned/compared through a union, <Number> cast, ternary and short-circuit side-effect operands. Snapshot output/diagnostics rather than assume C#-like coercion.
• P5 functions/control: typed value-returning function and void function, parameter/default behavior, C-for/foreach, switch fallthrough/break/return. Use a behavior-sensitive or output-structural oracle observation before desugaring.
• P6 catalog/signatures: one representative reordered named call with defaults, plus the exact per-call additions needed by the reachable graph. Record canonical ordered Workshop output and diagnostics; populate Wright-owned catalog/signature data with that provenance.

PM/Engineer-ready boundary: implement only the seven-file graph, stopping at the three missing imports; resolve the listed base variable/enum/array/expression/rule surface through Wright-owned HIR/catalog contracts; add neutral HIR support only where the table identifies a real semantic gap; and run the probes before choosing behavior for the marked ambiguities. Parser success of the other nine files is neither a lowering requirement nor acceptance evidence for #118.

[Teakowa]

QA semantic probe evidence for the accepted inventory (pinned OSTW v3.4.0, contentCommit 769ce7aab097178cfe905bf21f0326d8e0d12e6b, asset SHA-256 1ae88289…e3a88), all via Deltinteger --langserver (docker linux-x64). Committed in compatibility/ostw/probes/ (9d2ce5d, Refs #118); re-run with python3 compatibility/ostw/run_oracle.py --probes (verified deterministic across runs). Probe sources are QA-authored originals. Each probe dir has result.<run>.json, workshop.<run>.txt, and findings.md separating observed facts from decision support. All six probes answered; no probe left unanswered.

Baseline correction (P1): the reference LSP compiles the LAST-OPENED document plus its transitive import closure; ds.toml entry_point is not the compile selector. The corpus baseline "accept, 79 elements, 75722f…03912" is the per-document compile of the last-opened utils/ServerLoad.del (outside the entry closure) — not the seven-file graph. Compiling the entry closure rejects on the three missing ../OSTWUtils/… imports and on Math (reachable only through non-entry PlayerInterface.del). Pinned identities are unchanged; this only reclassifies what the baseline hash evidences.

• P1 entry closure (probes/p1-entry-closure/): only the last-opened document's closure compiles; an unrelated invalid .del produces no diagnostics and no compile effect unless it is the compile root (scatter experiments E1/E2 in findings). Omitted event → Ongoing - Global; priority literals sort emitted rules (-1 first, +2 last, implicit 0 in source order); Workshop text carries no priority value (order only). Decision: entry-point-only compilation membership is Wright-owned and diverges from reference LSP behavior.
• P2 missing import (probes/p2a-simple-missing/, p2b-protectban-shape/, p2c-ambient-resolution/): hard error, severity 1, source-located on the quoted path, message names the normalized absolute path (e.g. <root>/OSTWUtils/OnScreenText.del). Resolution is strictly relative to the importing file; no ambient/module source satisfies an import, and opening the file does not register it. The three OSTWUtils edges are confirmed hard ostw-missing-import reject edges.
• P3 variables (probes/p3a…/, p3b…/, p3c…/): autos fill lowest-free slots and skip explicit IDs — table 0:a, 1:b, 127:i, 2:c, 128:d, 3:e, 0 diagnostics. Duplicate explicit IDs reject per collection: The id 127 is already reserved in the global collection. / …player collection. Player-variable receiver EventPlayer().p read/write/+= → Set/Modify/read Player Variable(Event Player, p, …).
• P4 types/expressions (probes/p4-types-expressions/, p4b-assignment-expression/): user enum members are implicit 0-based integers (Fruit.Banana → 1); null → Workshop Null; Number | Team union is a raw Number cell; <Number> cast is an emission pass-through; ternary → If-Then-Else value; &&/|| → And/Or conditions with side-effecting operands hoisted UNCONDITIONALLY before the If (no short-circuit ordering — if (selector == 0 || ping()) runs ping's action always). Assignment is statement-only: (selector += 1) and (selector = 2) in expression position are parse errors.
• P5 functions/control (probes/p5-functions-control/): value functions inline as expressions (foldable, no param storage); void functions inline as actions with per-call PLAYER-variable parameters (by, by_0, …) and defaults materialized at the call site; return in a rule → Abort; C-for over a global → For Global Variable; foreach → index For Player Variable + Value In Array; switch → Skip-array dispatch with sequential fallthrough (break = Skip over remaining bodies, default = last body).
• P6 catalog/signatures (probes/p6-catalog-signatures/): named args bind by name regardless of source order; omitted params take defaults; the canonical emitted argument order is the signature order (reordered WorkshopSettingToggle(SortOrder:5, Default:true, Category:…, Name:…) emits (Category, Name, Default, SortOrder)). Receiver defaults: All Players (BigMessage Header-only call), Event Player (IsButtonHeld/SetInvisible). User functions bind named/default args identically (userCall(C:9, A:1) → 1+2+9, folded).

Caveats: evidence is output-structural (workshopCode payloads + diagnostics), not runtime behavior; vector spellings (Vector(0,0,0) → Subtract(Left, Left), Vector(1,0,0) → Left) are reference output-syntax/folding artifacts, not semantics.

[Teakowa]

PM acceptance recheck — implementation landed, two blockers remain

7ba2fec establishes the intended native OSTW semantic phase and the frontend-neutral HIR boundary. The P1-P6 probe evidence is useful and the implementation correctly leaves the new HIR nodes for #119 rather than adding an OSTW-specific Workshop backend.

Two blockers prevent closing #118:

1. Current CI is not green. Rust quality (stable) fails on crates/wright-ostw/src/semantic.rs with clippy::collapsible_match under -D warnings. This is [M13] Resolve and lower the accepted protect-ban OSTW semantic slice into HIR #118-owned and should be fixed without semantic changes.

2. Workshop semantic/catalog ownership is duplicated. crates/wright-ostw/src/signature.rs currently stores Workshop builtin kind, canonical parameter order, en-US Workshop spelling, and enum-domain/member data. wright-workshop already owns the canonical Workshop catalog. [M13] Resolve and lower the accepted protect-ban OSTW semantic slice into HIR #118's own acceptance criterion requires OSTW Workshop calls/values to reuse canonical Wright Workshop catalog/signature data and forbids an OSTW-specific duplicate game-content table.

Before closure, move/extend the canonical Workshop-owned metadata needed by the exercised OSTW surface and make wright-ostw consume that contract. OSTW may retain source-language aliases/binding metadata only where genuinely OSTW-specific, but it must not own the canonical Workshop spelling/enum/content table.

Do not broaden the OSTW language surface, implement #119 lowering, or start a general content/plugin registry. Keep all P1-P6 evidence and current HIR semantics intact, rerun full CI, and leave a concise final evidence comment.

[Teakowa]

Implementation evidence — #118 acceptance: catalog ownership + stable gate restored

Commits: 359a055 (feat(ostw): resolve builtins and enum domains through the canonical Workshop catalog (#118), the substantive change) and 133922d (test(ostw): assert representative builtin and enum resolutions through the catalog (#118)); final main HEAD 133922d, on top of the concurrent oracle rebaseline 0aac7e7/7d54334.

Catalog ownership AC (satisfied)

• The exercised Workshop surface moves into the canonical Wright-owned catalog (wright-workshop/src/catalog): ~100 builtin entries (canonical params + en-US spellings), 26 enum domains (149 exercised members plus the OPY-side members of shared domains), and per-parameter expected enum domains (paramDomains, validated by the strict loader).
• wright-ostw/src/signature.rs is now OSTW-only source bindings: OSTW source name → (kind, canonical catalog id) for builtins and OSTW member name → canonical member id per enum domain. No Workshop spelling/param/enum tables remain (grep-audited).
• The semantic phase resolves builtins and enum domains through the catalog at its consume sites (bind_builtin_args uses catalog params; enum-member validation checks the catalog domain/member). No general registry/plugin system was added.
• Supporting moves: Catalog implements ExpectedDomain; the Workshop parse path chains the OPY manifest with the catalog context (ChainedExpectedDomain in wright-core); Debug/Print HUD emission realigns to the canonical 11-param Create HUD Text layout so its emission reparses deterministically.

Behavior preservation

• All P1–P6 probe-backed semantics and HIR behavior preserved: the existing wright-ostw semantic regressions pass unchanged (zero assertions modified; two tests added: an all-bindings-resolve catalog-ownership invariant and a representative action/value/enum resolution through the shipped semantic path), byte-stability holds, and compatibility/ostw/results.json + probes/ are unmodified.
• One documented consequence: the catalog now records genuinely-ambiguous Workshop spellings (Up in Vector/Rounding, shared Visible To* reevaluation members), so OPY-corpus parser/roundtrip/analyzer tests parse with the catalog context; overpy-cake's bare Up (OverPy folds the vector-up constant inside Add) stays an asserted, documented ambiguity ([Workshop semantics] Resolve ambiguous enum members using expected parameter domains #111 no-guessing).
• MinWait() binds to the canonical wait action (single "Wait" spelling per the strict loader); its 0-arg source form fills the two canonical slots with zero literals, refined at [M13] Compile the accepted OSTW corpus slice to Workshop with reference differential validation #119 emission.

Tests

• cargo test --workspace --all-targets --all-features: 50 suites, all pass.
• cargo fmt --all -- --check: clean; cargo clippy --workspace --all-targets --all-features -- -D warnings: zero warnings on stable (1.97) and 1.85.0.

CI

• Run 31886223409 on 133922d: Rust quality (stable) and Rust quality (1.85.0) success, OSTW reference baseline success, v1 gates / differential / adapter / oracle / dist (ubuntu + macos) success. The only failure is Distribution validation (windows-latest) on the "Run npm distribution smoke tests" step, which fails identically on the baseline commit 0aac7e7 (run 31885241269) and on the commit that introduced the step (631e556) — pre-existing and unrelated to this Rust-only change.

No #119 (Workshop emission) or #120 (tooling) work was started.

[Teakowa]

PM acceptance — accepted

#118 is accepted at 133922d (substantive catalog-ownership correction 359a055).

The two prior blockers are resolved:

• stable and Rust 1.85 quality gates pass with zero clippy warnings;
• canonical Workshop builtin parameter/spelling/enum metadata is owned by wright-workshop, while wright-ostw retains only OSTW-specific source bindings.

The original semantic/HIR work from 7ba2fec remains intact, P1–P6 probe-backed behavior is preserved, and #119/#120 responsibilities remain separate.

The only failing check on the accepted commit is the Windows npm package-mode smoke failure introduced by the distribution path; it predates these #118 corrections and is now tracked separately as #123. It is not evidence of an OSTW semantic regression.

Closing #118 as completed. #120 is now unblocked.