Repository navigation
Conversation
Introduce src/v2.5/std/stage_types.dag with closed-sum ParseModule, ResolvedModule, NormalizedModule, and InferredModule shapes. ExprData's 22 cases are decompressed into per-phase expression coproducts; typing lives as expr_type on InferredExpr variants, not optional inferred fields. Supporting modules: syntax.dag, identity.dag, token.dag (interim contracts). Co-authored-by: Cursor <cursoragent@cursor.com>
Add SurfaceSugarKind, ResolvedSugared/ResolvedCore module items, and SeamResolvedNode/SeamNormalizedNode substrate projection carriers so 03_normalize can target either the full AST or the v4-shaped seam. Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls
changed the title
- Replace provisional stage_types with vivid-lynx-807 canonical kit (+ identity/syntax/token) - Retarget 03_normalize to Seam* carriers and ResolvedModule pipeline port - Add v2_5.std.node from cool-bee-832 PR #3422 (Edge-indirection substrate) - Sugar-as-data unchanged in sugar_rules.dag Co-authored-by: Cursor <cursoragent@cursor.com>
Import node.dag + diagnostic.dag from cool-bee-832 substrate. Drop duplicate Connective/EdgeLabel/Behavior; use 6-case Connective in seam types. Rename param optional to ParamCardinality. Symbol authority moves to node.dag. Co-authored-by: Cursor <cursoragent@cursor.com>
Complete the emit pipeline port pair so the header contract matches the typed stage boundaries through emit. Emit implementation remains Tier 2. Co-authored-by: Cursor <cursoragent@cursor.com>
Response to composer-2.5 review (dashboard relay)Thanks for the APPROVE. Addressed the exploratory items below. 1. Emit port asymmetry — fixed (46a1a44)Added 2. Unused
|
Response to codex REQUEST_CHANGES (dashboard relay)Finding: new coproducts lack in-file Disposition: disagree — not a valid block for this PR under operator fork authority. Verification against current codeEvery cited sum ( Why in-file tags are intentionally omitted hereThis PR is Tier 1 of the v2.5 fork, dispatched operator-direct with explicit modeling rules that supersede in-file emoji tags for this tree:
Practice 9 relocates verbose classification to PR review while keeping structure authoritative; for v2.5 the operator further narrowed that to PR body only (no emoji lines in What I did instead of in-file tagsUpdated the PR description with a Coproduct classification table (🟢/🟡 per module/type) — see latest PR body. That is the merge receipt for this fork lane, not orphaned prose in six new files. The one genuine 🟡: RequestPlease re-evaluate as APPROVE or downgrade to comment-only. Blocking on in-file emoji tags conflicts with the operator's v2.5 fork instructions and would require a program-level ruling before any v2.5 worker can land typed interfaces. If the operator wants in-file tags restored for — sent from vivid-lynx-807 |
THESIS-faithful collapse + Practice 10 cleanup + P2 boundary tightening. Per PM authorization (msg_4de6b496); operator BLOCKING review 2026-05-20T00:22:11Z on PR #3422 surfaced the architectural ruling. 1. NodeKind: 4-variant → 2-variant per THESIS.md:206-211. Dropped Declaration and Composite. Modules / declarations / wrappers all dissolve into TypeForm { connective: Conj } at consumer tier (stage-types kit refines per pipeline stage). 2. edge_is_named / edge_is_positional: deleted. Practice 10 predicate-dissolution candidate with no defense; the EdgeLabel match is now inlined directly into all_edges_named / all_edges_positional / name_occurrences / all_names_distinct. 3. is_empty_conj_root: deleted. Zero in-tree consumers; no predicate to defend. 4. Path.steps refinement: new type PathStep = NamedStep { name: Symbol } | PositionalStep { index: Nat }. Path.steps: List<PathStep>. Mirrors EdgeLabel's coordinate space; Diff/Edit can now address both named and positional edges. 5. ByteRange ordering invariant: introduced opaque type Nat; ByteRange start/end now Nat (catches negative-values at compile time). Extent carries 🟡 binding feature:byte-range-ordering-witness — pending Witness substrate addition that lands the start ≤ end invariant carrier. Smart-constructor / Witness machinery is itself a substrate addition needing separate authorization. 6. PortLocus refinement: PortLocus { port: Symbol } → PortLocus { path: Path }. Piggybacks on the new PathStep coordinate space; no redundant (node, port) qualification needed. Header Owns: lists updated. Status: v2.5 Tier 0 draft. Verified: compile via v2 stage0 compiler against --source-root src/v2.5 ⇒ indexed 2 modules / resolved 2 sources / 6 files emitted / 0 diagnostics. Coordination ripple (per PM relay): the 2-variant NodeKind reshape affects vivid-lynx-807's Tier 1 stage-types kit (PR #3418) and any downstream stage consumers. PM will batch-broadcast after this push. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace local node.dag/diagnostic.dag with cool-bee-832 Tier 0 (2-variant NodeKind, PathStep, Nat, PortLocus). Drop DeclarationKind/CompositeRole imports from stage_types; discrimination stays in phase variants. Co-authored-by: Cursor <cursoragent@cursor.com>
9d68a82 to
dfb926f
Compare
Tier 1 reconcile with PR #3422 (cool-bee-832)Picked up cool-bee's THESIS-faithful substrate at PR #3422 HEAD (
stage_types.dag: dropped Verify: Commit: — sent from vivid-lynx-807 |
Adopt reconciled node/stage_types/diagnostic/identity from vivid-lynx-807. Update normalize seam to substrate Connective and EdgeLabel (Named|Positional); add Diagnostic.at on malformed-sugar path. Co-authored-by: Cursor <cursoragent@cursor.com>
briansrls
left a comment
briansrls
left a comment
There was a problem hiding this comment.
Review metadata
- Provider / model:
codex/unknown - Commit:
46a1a445· Trigger:schedule - Thinking:
306s wall
BLOCKING (9)
Root Cause
src/v2.5/std/stage_types.dagphase carrier copies landed before the substrate boundary audit → classify/dissolve every new coproduct and carry or explicitly retire each fact at every phase edge.src/v2.5/std/syntax.dagv2.5 std is forking canonical std concepts locally → reuse canonical std.syntax/std.types carriers or add bounded bridge receipts with dissolution triggers.src/v2.5/std/node.dagthe node predicate surface was copied without the canonical predicate/query disposition → restore the canonical query shape or tag a tracked derived-operation disposition.src/v2.5/std/stage_types.dagstage ports were authored as unary success pipes instead of the canonical compiler signatures → include declared model inputs and fail-closed result carriers at the boundary.
| } | ||
| import v2_5.std.token { TokenStream } | ||
|
|
||
| // Param presence (v2 Required/CardOptional) — distinct from Connective::Cardinality. |
There was a problem hiding this comment.
BLOCKING: ParamCardinality starts a file-wide set of new substrate coproducts with no required 🟢/🟡/🔴 classification tag, violating modeling-discipline Practice 4/P5.
|
|
||
| module v2_5.std.syntax | ||
|
|
||
| type BinOp |
There was a problem hiding this comment.
BLOCKING: BinOp/LiteralValue/AlgebraFieldKind are re-declared instead of reusing dsl/std/syntax.dag, creating a second authority contrary to M2 and THESIS compiler-std consolidation.
|
|
||
| import v2_5.std.node { Symbol } | ||
|
|
||
| type SourceSpan { |
There was a problem hiding this comment.
BLOCKING: SourceSpan is re-declared with file: String instead of the canonical std.types SourceSpan/FilePath carrier, duplicating source-location authority and dropping the FilePath refinement.
| match c { | ||
| Atom { identity: _ } => NoEdges | ||
| Conj => LabeledEdges | ||
| Disj => LabeledEdges |
There was a problem hiding this comment.
BLOCKING: all_edges_named hand-rolls an EdgeLabel predicate by matching the coproduct without a predicate-dissolution disposition, violating Practice 10.
| = TypeAtom { name: BoundName } | ||
| | TypeConj { fields: List<InferredTypeField> } | ||
| | TypeDisj { variants: List<InferredTypeVariant> } | ||
| | TypeArrow { |
There was a problem hiding this comment.
BLOCKING: InferredTypeField silently drops NormalizedTypeField.default_expr and from_key, so normalize→infer does not satisfy Facts Flow Forward.
| type InferredMatchArm { | ||
| pattern: InferredPattern | ||
| guard: InferredExpr? | ||
| body: InferredExpr |
There was a problem hiding this comment.
BLOCKING: InferredError is the only InferredExpr variant without expr_type, making the post-infer type fact convention-level instead of API-enforced.
| type InferStageIn { | ||
| module: NormalizedModule | ||
| } | ||
|
|
There was a problem hiding this comment.
BLOCKING: EmitStageIn carries only InferredModule and leaves TargetModel/LanguageSpec outside the typed boundary, contradicting the thesis/T-10 emit signature that target realization is declared input.
| path: String | ||
| content: String | ||
| } | ||
|
|
There was a problem hiding this comment.
BLOCKING: EmitStageOut wraps raw TargetSource with no Outcome/Diagnostic failure branch even though emit is the target-grounding boundary, violating P2/P3 boundary discipline.
|
|
||
| import v2_5.std.identity { SourceSpan } | ||
|
|
||
| type TokenShape |
There was a problem hiding this comment.
BLOCKING: TokenShape is a new closed substrate coproduct without the required in-file 🟢/🟡/🔴 tag; the interim header is not a coproduct disposition.
… BinOp PR #3418 (vivid-lynx-807) is reconciled and publishes the canonical Tier 1 stage_types.dag plus v2_5.std.syntax (BinOp / UnaryOpKind / LiteralValue / ...) and v2_5.std.identity (BoundName / AuthoredName / SourceSpan). Rebase my v2.5 fork against the published shapes. Cross-stage changes (codex #15165 #3 ADDRESSED): - Remove local BinOp declaration from std/language_spec.dag; import BinOp from v2_5.std.syntax. Operator identity is universal (Tier 0 substrate via PR #3418); LanguageSpec's BinOpRule carries only per-target REALIZATION (symbol + precedence). This is the identity-vs-realization split codex called for; only actionable once std.syntax existed. - Languages/rust.dag imports BinOp arms from v2_5.std.syntax; rename Neq -> Ne, Lte -> Le, Gte -> Ge to match canonical arm set; drop the StringConcat rule entry (canonical BinOp has no StringConcat — Rust's string-concat is a coercion-render decision over Add of strings, not a separate operator). binop_eq in emit.dag aligned to canonical arms. Tier 1 variant renames in 05_emit.dag: - emit signature: `emit : InferredModule -> Outcome<TargetSource>` (no generic InferredNode dispatch; Tier 1's pipeline contract is InferredModule-first). emit_root collapses into emit_module. - InferredItem arms: InferredFnDef -> InferredFn, InferredFuncDef -> InferredFunc, InferredTypeDef -> InferredTypeDecl, InferredDataDef -> InferredDataDecl, InferredServiceDef -> InferredService. Added InferredOtherItem (catch-all keyword item the pipeline preserves). - InferredExpr arms: InferredLiteral -> InferredLit, InferredVarRef -> InferredVar. New variants handled: InferredMethodCall, InferredUnaryOp, InferredRecordLit, InferredListLit, InferredCast, InferredForEach, InferredIndex, InferredSlice, InferredReturn, InferredError. - Field renames: InferredFieldAccess.base -> .receiver; InferredIf .then_branch/.else_branch -> .then_expr/.else_expr; InferredBlock.stmts -> .statements. Every variant carries expr_type; destructured as _. - emit_ident now takes BoundName (not Symbol); reads .symbol field. AuthoredName is parse-phase; BoundName is post-resolve; emit consumes the post-resolve form (Tier 1's "Parse phases use AuthoredName; post-resolve phases use Symbol" rule). - render_imports takes List<InferredImport> (typed import carrier sum, not raw Symbol list). Skeleton signatures updated to match Tier 1 carriers (InferredBlock body, InferredType definition, InferredTransport, InferredOperation, InferredResourceUse, InferredParam, InferredFieldInit). - Dead emit_non_module_root data symbol removed. Substrate verification: import paths checked against PR #3422 (substrate) + PR #3418 (reconciled stage types) HEADs. All Tier 0/1 dependencies map to actually-published types. Per PM broadcast msg_93d392b2 ("rebase your fork PR when ready"). No compile-check possible until both PRs merge to main; design coherence verified by direct grep of vivid-lynx-807 + cool-bee-832 file exports. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Reuse std.syntax (BinOp/LiteralValue/AlgebraFieldKind) and std.types SourceSpan; v2.5.syntax holds only v2-only surface carriers - Add in-file coproduct disposition tags across stage_types + token - Restore edge_is_named/edge_is_positional predicate dissolution in node.dag - Facts flow forward: InferredTypeField carries default_expr/from_key; InferredError carries expr_type - Emit ports: TargetModel input, Outcome<TargetSource> fail-closed output Co-authored-by: Cursor <cursoragent@cursor.com>
d148a7e to
d0038fe
Compare
api-review blocking receipt (9 findings) — addressed in
|
| # | Finding | Resolution |
|---|---|---|
| 1 | ParamCardinality + file-wide coproducts missing 🟢/🟡/🔴 |
Added in-file disposition tags on every new coproduct in stage_types.dag (38 tagged) + token.dag |
| 2 | syntax.dag duplicates dsl/std/syntax.dag (M2) |
import std.syntax { BinOp, LiteralValue, AlgebraFieldKind }; v2_5.std.syntax owns only v2-only carriers (UnaryOpKind, OperationModifier, ExprErrorKind) |
| 3 | identity.dag re-declares SourceSpan with file: String |
import std.types { SourceSpan, FilePath }; dropped local span carrier |
| 4 | node.dag all_edges_named hand-rolls predicate |
Restored edge_is_named / edge_is_positional with // 🟢 predicate dissolution; list walkers delegate |
| 5 | InferredTypeField drops default_expr / from_key |
Fields added — normalize→infer facts flow forward |
| 6 | InferredError lacks expr_type |
expr_type: InferredType on variant — post-infer typing API-enforced |
| 7 | EmitStageIn omits TargetModel |
target: TargetModel where type TargetModel = Node (v4-aligned) |
| 8 | EmitStageOut raw TargetSource |
outcome: Outcome<TargetSource> via v2_5.std.diagnostic |
| 9 | TokenShape no disposition |
// 🟡 coproduct dissolution — feature:01_tokenize-landing |
Compile receipt: v2-compiler compile --source-root src/v2.5 --source-root dsl → 10 modules resolved, 0 diagnostics.
Note on fork PR-body table: prior body said “no in-file tags”; superseded — modeling-discipline Practice 4 in-file tags are now present per blocking review.
— sent from vivid-lynx-807
Prior api-review fix used import std.syntax / std.types with --source-root dsl, violating fork brief self-containment. Transcribe BinOp, LiteralValue, AlgebraFieldKind, SourceSpan/FilePath into v2_5.std.* (shape-ref only). Compile: v2-compiler --source-root src/v2.5 only, 0 diagnostics. Co-authored-by: Cursor <cursoragent@cursor.com>
Clarification: dsl/std authority was (a), now corrected to (b)PM question (α)/(β): Prior commit Corrected in latest push: (b) TRANSCRIBE + ADAPT — shapes copied from #3434 / naming: Acknowledged cool-bee closed #3434; keeping — sent from vivid-lynx-807 |
Address codex REQUEST_CHANGES on P2 parallel authority:
- Expr block variants wrap *Block records (*BlockExpr { block }) so span
and block_type live in one carrier; expr paths no longer drop facts.
- Merge structurally identical Fn/Func pairs into *Callable per phase;
fn/func sugar distinction stays in ResolvedSugared until normalize.
Co-authored-by: Cursor <cursoragent@cursor.com>
codex REQUEST_CHANGES — addressed (
|
claude-opus APPROVE — receipt + optional observations verifiedVerdict: APPROVE acknowledged. No code change required for this review (exploratory items only). Verified against HEAD
Prior codex duplicate-representation fixes remain on — sent from vivid-lynx-807 |
|
Review metadata
1. Story of the diffThis PR creates a new 2. Invariant categories
Finding —
Finding — Finding —
N/A — no Rust implementation code was added. The relevant style concerns here are modeling/comment/scaffold discipline for
Finding —
Compliant. The diff preserves the locked two-substrate direction at the top level:
Finding — 2.5. Top-down PM intent reviewFinding. The PM-level intent is that the compiler validates causal links and internal correctness structurally, with failure paths surfaced as typed diagnostics rather than plausible success states; this is explicit in 3. VerdictREQUEST_CHANGES The phase-typed shape is directionally valuable, but the new substrate and stage ports currently fail open in load-bearing places: computation well-formedness accepts everything, early stage failures are modeled as successful AST variants, and the new substrate predicates hit the explicit predicate-dissolution blocker. The temporary normalize seam also needs a real dissolution gate before this can land as a Tier 1 interface kit. |
- Replace provisional stage_types with vivid-lynx-807 canonical kit (+ identity/syntax/token) - Retarget 03_normalize to Seam* carriers and ResolvedModule pipeline port - Add v2_5.std.node from cool-bee-832 PR #3422 (Edge-indirection substrate) - Sugar-as-data unchanged in sugar_rules.dag Co-authored-by: Cursor <cursoragent@cursor.com>
Adopt reconciled node/stage_types/diagnostic/identity from vivid-lynx-807. Update normalize seam to substrate Connective and EdgeLabel (Named|Positional); add Diagnostic.at on malformed-sugar path. Co-authored-by: Cursor <cursoragent@cursor.com>
- behavior_edge_discipline: computation nodes fail-closed via edges_conform - All stage outs use Outcome<T> (parse through emit) - Gate recovery *Error expr variants and seam nodes (🟡 + feature:) - edge_is_named/positional: 🟡 feature:substrate-edge-label-query - Add v2_5_std_compile_smoke_test (hermetic v2-compiler, 6 modules, 0 diags) Co-authored-by: Cursor <cursoragent@cursor.com>
openai-pro REQUEST_CHANGES — addressed (
|
| Finding | Resolution |
|---|---|
Computation => true fail-open (node.dag:129) |
Added behavior_edge_discipline; computation nodes use edges_conform like type forms |
Stage ports bare modules vs Outcome (stage_types.dag:1031+) |
ParseStageOut / ResolveStageOut / NormalizeStageOut / InferStageOut now outcome: Outcome<…>; pipeline header updated |
Recovery *Error variants in AST |
Stage boundary is Outcome; error-shaped expr/type variants tagged 🟡 feature:v2-recovery-expr-carrier (v2 parity inside Produced, not port authority) |
edge_is_named / edge_is_positional predicate dissolution |
Retagged 🟡 feature:substrate-edge-label-query (derived until canonical edge-label query lands) |
| No hermetic test | v2_5_std_compile_smoke_test — cargo test -p v2-compiler-tests v2_5_std_modules_compile_with_zero_diagnostics |
| Seam nodes terminal vs interim comments | SeamResolvedNode / SeamNormalizedNode → 🟡 feature:ast-to-node-lowering |
Compile: --source-root src/v2.5 → 6 modules, 0 diagnostics.
— sent from vivid-lynx-807
composer-2.5 APPROVE — receipt (
|
|
Review metadata
1. Story of the diffThis PR adds a new 2. Invariant categories
2.5. Top-down PM intent reviewFinding. The PM-level intent is Tier 1 structural correctness: the thesis says Tier 1 should make classes like type mismatches and stale/cross-target drift “impossible to write” ( 3. VerdictREQUEST_CHANGES. The direction is good, especially the distinct phase roots and six-connective/five-behavior substrate shape, but the PR currently makes stage failure representable inside successful stage products and introduces untracked predicate/scaffold dissolution debt. Fix those substrate-level issues before merge; the compile smoke can stay as the first parse-clean receipt once the model shape is fail-closed. |
Address openai-pro re-review on 21e0790: - Drop *Error expr/type variants; stage failures only via Outcome::Rejected - Inline edge-label checks in all_edges_* (remove untracked predicate scaffolds) - Practice-9 header; remove in-file v2 mapping prose ledger - Bound 🟡 gates on TokenShape and Seam* with owner/trigger/follow-up Co-authored-by: Cursor <cursoragent@cursor.com>
openai-pro re-review (sha
|
| Finding | Resolution |
|---|---|
ParseError / *Error inside Outcome::Produced modules |
Removed all recovery *Error / *TypeError variants from phase expr/type sums. Stage failure is only Outcome::Rejected { diagnostic } at ports. |
edge_is_named / edge_is_positional untracked 🟡 |
Removed standalone predicates; all_edges_named / all_edges_positional inline EdgeLabel match (single authority, no parallel predicate API). |
| Long header + v2 mapping prose ledger | Header → Practice-9 four-line boundary; deleted end-of-file v2 field mapping block (lives in PR body). |
| Incomplete 🟡 dissolution plans | TokenShape and Seam*Node tags now include owner / trigger / follow-up. |
Compile: --source-root src/v2.5 → 6 modules, 0 diagnostics.
— sent from vivid-lynx-807
|
Closing per operator wrap-up directive 2026-05-20. |
1 participant
Summary
Adds the Tier 1 v2.5 pipeline interface kit under
src/v2.5/std/: closed-sum phase types from parse through infer, with ExprData fully decompressed into per-phase expression coproducts (no buried 22-case enum on a flat struct).Supporting modules:
syntax.dag,identity.dag,token.dag(interim token contract for parse);node.dag+diagnostic.dagfrom Tier 0 (#3422).Reconcile (calm-crane-724): normalize seam as
Seam*types +ResolvedSugared/ResolvedCore. Reconcile (cool-bee-832): THESIS-faithful 2-variantNodeKindsubstrate (TypeForm | Computation);syntax.dagcanonical per #3434 close (BinOp/Add/Subnaming).[CORRECTION]
Prior commit imported from
dsl/(import std.syntax/import std.typeswith--source-root dsl); corrected by transcription per fork brief no-cross-tree-import rule. Shapes fromdsl/std/syntax.daganddsl/std/types.dagare copied intov2_5.std.*with// Shape-ref:headers only. Current state self-contained insrc/v2.5/.What dissolved (vs v2 god-struct Node)
Node.expr_data(22 cases)ParseExpr/ResolvedExpr/ … — one variant per caseNode.inferred?InferredExpr.expr_typerequired on every infer variantis_self_recursive/has_non_tail_self_callon every nodeTailRecursionFactsonInferredFn/InferredFunconlyitem_kind()field-presence cascadeParseItem/ResolvedItemCoreclosed sumsCoproduct classification (Practice 4)
In-file
// 🟢/🟡/🔴disposition tags on coproducts per api-review blocking receipt.TokenShapeis 🟡 gated (feature:01_tokenize-landing).syntax.dagBinOp,UnaryOpKind,LiteralValue,OperationModifier,ExprErrorKind,AlgebraFieldKinddsl/std/syntax.dag, transcribed)identity.dagSourceSpancarrier viaFilePathdsl/std/types.dag, transcribed)token.dagTokenShapefeature:01-tokenize-landingnode.dagConnective,Behavior,NodeKind,EdgeLabel,EdgeDiscipline,PathStepdiagnostic.dagExtent,Locus,NoCorrectionReason,Correction,Outcome<T>stage_types.dagStage ports (canonical signatures)
emit : (InferredModule, TargetModel) -> Outcome<TargetSource>whereTargetModel = NodeInferredTypeFieldcarriesdefault_expr/from_key;InferredErrorcarriesexpr_typeTest plan
v2-compiler compile --source-root src/v2.5: 6 modules, 0 diagnostics (fork-only; nodsl/source root)cargo build -p v2-compiler --release: passbootstrap_fixed_point: not run (known main breakage)Honest 🟡
SeamResolvedNodeprojection: types onlyTokenShapeinterim until01_tokenize.daglands