MachineShape construction wall: compile-gated single-constructor discipline by gunbai-bot[bot] · Pull Request #6698 · gunb-ai/gunbc

gunbai-bot · 2026-07-16T00:51:01Z

Summary

Compile-gated MachineShape single-constructor discipline — only sanctioned modules (std.machine_shape, extdeps.gpu.machine_shape catalog fill, test/claim witnesses) may materialize MachineShape record literals.

This PR's diff (3 files, stacked on #6701):

src/v2/lens/machine_shape.dag — module-context fold_node_topdown gate; Cons/Empty fail-closed match (determinism precedent, not HeadAbsent absorb)
src/v2/compiler/00_compile.dag — enrolled in always_required_root_lenses
dag/test/claim/machine_shape_construction_wall_test.dag — synthetic RED/GREEN fixtures

Stacked on #6701 (not bundled)

Base branch: session/vivid-owl-803 (#6701). The interim main-unbreak (#6686 hash-grounding revert + algebra import hygiene) lives exclusively in #6701 — this PR does not widen the fnv1a64 fork. Merge #6701 first, then rebase this onto main.

Also stacked on #6687 for std.machine_shape / extdeps.gpu.machine_shape subjects.

Test plan

git diff --stat origin/session/vivid-owl-803 shows only the 3 files above
Construction-wall witnesses green once std.machine_shape: allocation-as-contract Phase 1 landing #6687 subjects exist

gunbai-bot · 2026-07-16T01:00:39Z

Addressed: std→extdeps layering violation (da58cdb5dd)

The finding was valid. dag/std/machine_shape.dag no longer imports extdeps.*.

Catalog fill moved to extdeps: shape_from_catalog, ShapeFromCatalogInput, and GPU derivation helpers now live in dag/extdeps/gpu/machine_shape_catalog.dag (extdeps.gpu.machine_shape_catalog), which correctly imports std.machine_shape + extdeps.gpu.types.
PcieLink scaffold: Until the planned PcieLink de-fork (§3 item 2), InterconnectLink is LinkAttach { lane_count, generation_ordinal } in std — validation already ignores link payload; witness RED arm updated accordingly.
Construction gate: extdeps.gpu.machine_shape_catalog added to sanctioned fill modules alongside std.machine_shape and test/claim witnesses.

machine_shape_witnesses claim-run: green. dag/std/** has zero extdeps imports again.

— sent from proud-fox-374

…port x #6663 field wall) Partial revert of 4a60303: src/v2/std/node.dag, materialize.dag, materialize_witness_test.dag, inert_carrier.dag, and the DESIGN Sec-3-thread note pair (carrier + generated doc together). Keeps #6686's orthogonal cli_run.rs nfr-roster backfill (reverting it would re-red the falsifier lane). Root cause (proven by execution): #6663's new MissingField presence check resolves a literal's expected type by unqualified name through the flat module type env; #6686's 'import std.types { ContentHash }' in v2.std.node pulls dag-root std.algebra into the typecheck closure of every v2 module importing v2.std.node (704 files), so dag-root Monoid/Semigroup/BooleanAlgebra/ OrderedRing/CommutativeMonoid/CommutativeSemiring/AbelianGroup shadow the v2.std.algebra declarations at 28 literal sites in src/v2/std/{diagnostic,logic,nat,integer}.dag -> ci_floor_plan resolve fails -> every main push red. Neither PR alone was red: zero file overlap, merged 31 minutes apart, and the squash union tree was never CI'd (the PR's only CI run began 8 minutes before #6686 landed). Receipt: on the 60286e0 tree with the same seed binary, reverting exactly this .dag surface takes the plan resolve from 28 missing-required-field errors to green with the floor executing witnesses (control: #6678/#6685 left in tree). Re-land trigger (dissolution): the wall's presence leg consults the resolver's binding (or fails closed on ambiguous names) instead of the flat-env name lookup; the follow-up re-lands these files with that fix, using this exact 28-error tree as the RED control. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

gunbai-bot · 2026-07-16T01:42:03Z

CI fix pushed (5d99afcac6) — dashboard 623e382 was stale (auto-commit had stripped compile enrollment); branch restored earlier to cf6a57d, now rebased.

Root cause: pre-existing main-red from #6663 field wall × #6686 ContentHash cross-root import (not construction-wall diff). Fix: cherry-picked coordinated interim unbreak from vivid-owl-803 (76563774e3, 7fc39a33ad — reverts #6686 hash-grounding surface that poisoned v2 typecheck closure) + our 3-file construction-wall deliverable on current origin/main.

Floor job is running (resolve phase no longer failing at ~2m like prior runs). Proper wall fix (scope-correct binding lookup) remains operator-coordinated for main re-land.

— sent from proud-fox-374

gunbai-bot · 2026-07-16T01:45:26Z

Review artifact /api/reviews/38585 (composer-2.5 APPROVE): verified — no action required.

Checked at 5d99afcac6:

machine_shape_compile_gate uses module-context fold_node_topdown + callee Atom ^MachineShape recognition (determinism precedent)
Enrolled in always_required_root_lenses via 00_compile.dag
construction_justification: WallAfterGrounding present in src/v2/lens/machine_shape.dag
RED/GREEN synthetic fixtures in dag/test/claim/machine_shape_construction_wall_test.dag
#6686 revert commits are intentional interim unbreak (documented dissolution trigger); LocalAlias roster line orthogonal

No code changes from this review.

— sent from proud-fox-374

gunbai-bot · 2026-07-16T01:46:03Z

Addressed /api/reviews/38588 (opus REQUEST_CHANGES) — 3f29af2f1d

All three findings were valid. The #6686 revert commits (d82d7b52, 7fc39a33a) were smuggled scope from the interim main-unbreak lane and have been removed from this branch.

Finding	Resolution
§3 `Hash`/`content_hash` regression + DESIGN revert in MachineShape PR	Fixed — branch is again 3 files only vs `origin/main`; `v2/std/node.dag`, `materialize.dag`, `DESIGN.md` untouched
Deleted `node_content_hash_is_realization_plan_target` witness	Fixed by removal — that deletion lived in the reverted commits, not in the construction-wall deliverable
PR scope / commit-message honesty	Fixed — title/body updated; diff is exclusively lens + compile enrollment + claim witnesses

The interim main-unbreak belongs on a separate PR (operator/vivid-owl-803 coordination), not piggybacked here. Floor CI may re-red until that lands — acceptable per scope discipline.

— sent from proud-fox-374

…immune (rustfmt::skip) Batch-4 generated_artifact_drift_gate_passes red on this PR is the first run of that gate since main broke at 60286e0 -- #6677 landed its derived crate-layout artifact during the red window, so no tree ever executed batch 4 against it. Two writers disagreed on one artifact: the wet leg (stage0_crate_layout_emit.dag) emits plain multi-line arrays, while the committed bytes were rustfmt'd (trailing comma, collapsed one-liner). Committed bytes: fmt-clean, drift-red; emitter bytes: drift-clean, fmt-red -- no tree could satisfy both gates. Fix: the emitter stamps #[rustfmt::skip] on both consts, so its output is the fixed point of cargo fmt; artifact regenerated via main_wet. Receipts: wet regen leaves the tree clean, cargo fmt --check -p v1-compiler green, regen_stage0 --verify divergence 0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

gunbai-bot · 2026-07-16T02:21:24Z

Addressed /api/reviews/38607 (opus REQUEST_CHANGES) — 002a20b5c8

Finding	Verdict	Action
§3 `Hash`/`ContentHash` regression bundled with MachineShape wall	Valid	Split: PR base is now `session/vivid-owl-803` (#6701). Diff vs base is 3 files only — no `node.dag`, `materialize.dag`, `DESIGN.md`, or witness changes in this PR. Hash ungrounding lands exclusively via #6701 with explicit bootstrap justification on that carrier.
`HeadAbsent => outcome_accepted` absorbing fallback in `machine_shape_compile_gate`	Valid	Fixed in `002a20b5c8`: replaced `is_empty` + `list_head`/`HeadAbsent` with `Cons`/`Empty` match + `Rejected { NonEmptyDiagnostics }`, matching `determinism_compile_gate` precedent — no silent accept arm on invariant violation.

Merge order: #6701 → rebase this onto main → merge #6698.

— sent from proud-fox-374

Rebuilt from origin/main per operator recovery: three-file scope only (lens, compile enrollment, claim witnesses). Stacked on #6687 for std.machine_shape / extdeps.gpu.machine_shape subjects — no Phase 1 content copied. Removed unused std.algebra import that poisoned CI resolve. Co-authored-by: Cursor <cursoragent@cursor.com>

Replace is_empty + list_head/HeadAbsent absorb arm with Cons/Empty fold match — HeadAbsent silently accepted on invariant violation (§5). Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot · 2026-07-16T02:38:27Z

Review artifact 38618 (composer-2.5 APPROVE): verified — no code changes.

Checked at 49e1b5d64d (rebased onto latest session/vivid-owl-803 / #6696 algebra fix):

Rooted fold_node_topdown gate + ^MachineShape callee recognition
Cons/Empty fail-closed match (no HeadAbsent absorb arm)
Enrolled in always_required_root_lenses
ConstructionJustification { WallAfterGrounding { dissolves_to: SingleAuthority } }
RED/GREEN witnesses in machine_shape_construction_wall_test.dag
Diff vs base remains 3 files; ContentHash grounding preserved on base (not reverted in this PR)

— sent from proud-fox-374

gunbai-bot · 2026-07-16T02:38:28Z

Review artifact 38619 (opus APPROVE): verified — no code changes.

Checked at 49e1b5d64d:

Construction wall, compile enrollment, and synthetic witnesses match the approval read
Cons/Empty gate (prior REQUEST_CHANGES item) already landed in 002a20b5 / carried through rebase
Minor observation on machine_shape_walk_step propagating child_r.module_qn upward: acknowledged — findings aggregate via concat(acc.findings, child_r.findings) and child contexts derive from ctx in child_context, not acc; no correctness impact. Will clean up if this walker pattern is copied elsewhere.

— sent from proud-fox-374

gunbai-bot · 2026-07-16T02:55:01Z

Review artifact 38634 (composer-2.5 APPROVE): verified — no action required.

Checked at 49e1b5d64d:

Rooted fold_node_topdown + fail-closed Cons/Empty → Rejected { NonEmptyDiagnostics }
Enrolled in always_required_root_lenses (00_compile.dag)
ConstructionJustification { WallAfterGrounding { dissolves_to: SingleAuthority } }
RED/GREEN synthetic witnesses in machine_shape_construction_wall_test.dag
Diff vs session/vivid-owl-803 base: 3 files only; no hand-Rust, no absorbing fallbacks

— sent from proud-fox-374

gunbai-bot · 2026-07-16T02:55:02Z

Review artifact 38636 (opus APPROVE): verified — observation acknowledged, no code change.

Checked at 49e1b5d64d:

Recognition at machine_shape.dag:117 is exactly Transform call + callee Atom ^MachineShape (synthetic fixture at machine_shape_construction_wall_test.dag:37-53 exercises this path)
std.machine_shape / extdeps.gpu.machine_shape not in tree yet — confirmed (git ls-files '*machine_shape*' → lens + test only); gate is WallAfterGrounding scaffolding stacked on std.machine_shape: allocation-as-contract Phase 1 landing #6687

On the non-blocking flag: agreed — when MachineShape lands (#6687), we must verify production construction sites surface as Transform/Atom-callee nodes (as machine_shape_construction_wall_note already documents) rather than bare TypeNode { Conj } record literals. If the inhabited type's literal form differs, recognition extends before WallAfterGrounding dissolves — tracked as part of #6687 landing, not a merge blocker for this scaffold.

— sent from proud-fox-374

…-machine-shape-construction-walls

Main (#6698) added docs/plans/emitted-crate-partition-design.md without an inbound link; doc_graph_has_no_orphan_docs reded after merge. Co-authored-by: Cursor <cursoragent@cursor.com>

Main (#6698) added docs/plans/emitted-crate-partition-design.md without an inbound link. Add open-thread row in design_document.dag authority (not hand-edited DESIGN.md) so doc_graph and generated_artifact drift gates both pass. Co-authored-by: Cursor <cursoragent@cursor.com>

* Tier-2 homonym gate: split census ambiguity from reachability leg Census ABSENT must refuse the ambiguity leg (§5), not collapse to non-ambiguous; reachability invariants carry the gate until census coverage widens on main. Co-authored-by: Cursor <cursoragent@cursor.com> * Tier-2 census refusal: emit counted diagnostics and assert ABSENT RefusedAbsent arm now prints typed located refusals and reports 3/3 count; snapshot test asserts ABSENT as dissolve trigger when Co-authored-by: Cursor <cursoragent@cursor.com> #6640 widens census coverage. * Fix CI: link emitted-crate-partition-design into doc graph Main (#6698) added docs/plans/emitted-crate-partition-design.md without an inbound link. Add open-thread row in design_document.dag authority (not hand-edited DESIGN.md) so doc_graph and generated_artifact drift gates both pass. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Cursor <cursoragent@cursor.com>

gunbai-bot Bot marked this pull request as ready for review July 16, 2026 00:56

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch from da58cdb to cf6a57d Compare July 16, 2026 01:07

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch 2 times, most recently from cf6a57d to 5d99afc Compare July 16, 2026 01:29

Merge remote-tracking branch 'origin/main' into session/vivid-owl-803

1dcfb93

Brian Searls added 2 commits July 16, 2026 01:42

WIP: field-wall env precedence fix (scratch)

8876483

Merge commit '1dcfb936c6' into HEAD

ad8496a

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch from 5d99afc to 3f29af2 Compare July 16, 2026 01:46

gunbai-bot Bot changed the base branch from main to session/vivid-owl-803 July 16, 2026 01:58

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch from 3f29af2 to 6bca004 Compare July 16, 2026 01:58

gunbai-bot Bot changed the base branch from session/vivid-owl-803 to main July 16, 2026 02:01

Brian Searls and others added 3 commits July 16, 2026 02:01

WIP2: kernel/container guard + final regen

da36337

Merge commit '4cf86d3bc7' into HEAD

0680751

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch from 867545b to 002a20b Compare July 16, 2026 02:21

gunbai-bot Bot changed the base branch from main to session/vivid-owl-803 July 16, 2026 02:21

Brian Searls and others added 3 commits July 16, 2026 02:22

Merge remote-tracking branch 'origin/main' into HEAD

ada3d2a

machine_shape gate: fail-closed Cons/Empty match (determinism precedent)

49e1b5d

Replace is_empty + list_head/HeadAbsent absorb arm with Cons/Empty fold match — HeadAbsent silently accepted on invariant violation (§5). Co-authored-by: Cursor <cursoragent@cursor.com>

briansrls force-pushed the session/proud-fox-374-machine-shape-construction-walls branch from 002a20b to 49e1b5d Compare July 16, 2026 02:38

Base automatically changed from session/vivid-owl-803 to main July 16, 2026 03:16

Merge remote-tracking branch 'origin/main' into session/proud-fox-374…

2bd544d

…-machine-shape-construction-walls

briansrls merged commit 45808d7 into main Jul 16, 2026
3 checks passed

briansrls deleted the session/proud-fox-374-machine-shape-construction-walls branch July 16, 2026 04:45

gunbai-bot Bot mentioned this pull request Jul 16, 2026

Tier-2 namespace homonym triage: fail-closed reachability gates #6712

Merged

7 tasks done

gunbai-bot Bot mentioned this pull request Jul 16, 2026

CI PR latency: scope compile-clean by import closure (unblocks the #6239-gated path); scope regen to its input closure #6726

Merged

Navigation Menu

Repository navigation

MachineShape construction wall: compile-gated single-constructor discipline - #6698

MachineShape construction wall: compile-gated single-constructor discipline#6698
briansrls merged 12 commits into
mainfrom
session/proud-fox-374-machine-shape-construction-walls

gunbai-bot Bot commented Jul 16, 2026 •

edited

Loading

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

gunbai-bot Bot commented Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Stacked on #6701 (not bundled)

Test plan

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

gunbai-bot Bot commented Jul 16, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

gunbai-bot Bot commented Jul 16, 2026 •

edited

Loading