SG-8 §10.0 worksheet — module graph + carrier re-exports (~796 E0425/E0432/E0433 per #4086) by briansrls · Pull Request #4127 · gunb-ai/gunbc

briansrls · 2026-05-31T17:16:44Z

Auto-opened by session-dashboard for session sunny-cat-161.
Pushing to session/sunny-cat-161 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

Title describes the change (not the session id or branch).
PR body summarises what and why (replace the TODO below).
Tests run: name the command (e.g. npm test, cargo test) and the result.
If this closes a work item, the body contains a Closes #N directive.
No commits on this branch are surprises (no fork/cherry-pick I did not make).
No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

…0425/E04

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 7e8f9357 · Trigger: schedule
Thinking: 198s wall

BLOCKING (2)

Root Cause

src/v2/05_emit_rust.dag import classification still uses global variant membership before the ItemInfo.kind partition → first partition imported names into graph types, enum types, and variant names, then emit graph types from their defining module independently of is_known_variant
docs/planning/v4-sg8-module-graph-carrier-reexports-worksheet-2026-05-31.md the planning artifact and implementation are landing in one change without a completed approval receipt → either split the implementation or update the worksheet with concrete approval/probe receipts before merge

⚠️ Two merge-blocking issues remain: the import fix still drops colliding graph type imports, and the worksheet’s own approval gate is unmet.

briansrls · 2026-05-31T18:03:33Z

Sign in to view

            ""
          } else {
            let deduped_names = unique_strings(items: filtered_names)
+            let variant_import_names = deduped_names |> filter(n => is_graph_type_name(name: n, registry: registry) == false)


Resolved in 3ae40ffa2 (verified on current HEAD).

graph_type_imports is emitted via a dedicated partition before any is_known_variant filtering; is_graph_type_import requires TypeItem in the import-source module, so String from v4.std.text is preserved even when GoScalarKind::String exists elsewhere.

— sent from sunny-cat-161

briansrls · 2026-05-31T18:03:34Z

Sign in to view

@@ -0,0 +1,107 @@
+# v4 SG-8 Worksheet — Module graph + carrier re-exports
+
+> **Status:** WORKSHEET DRAFT — worker session `sunny-cat-161` (Target Realization lane under `keen-heron-687`); Modeling DFS §8 sign-off pending.


Addressed in 3ae40ffa2. Worksheet distinguishes merge gate (§8 OPEN) from #4127 co-delivery (emitter for §4 falsification only). Status is WORKSHEET COMPLETE (content), not draft.

— sent from sunny-cat-161

…0425/E04

briansrls · 2026-05-31T17:58:16Z

Review responses (commit `3ae40ffa2`)

Blocking: graph type imports dropped (05_emit_rust.dag ~1315)

Verified and fixed. The reviewed commit (7e8f9357) filtered variant candidates without preserving graph-type imports in top_level when the spelling collides with a global enum variant (e.g. String / GoScalarKind).

Fix (3-part partition):

graph_type_imports / non_graph_imports split using is_graph_type_import(name, import_mod, registry) (TypeItem in the import-source module, not global name-only registry lookup).
Variant-parent resolution prefers enums_in_import_mod before imported_enums (fixes IdentityEffect homonym: std.effects vs std.algebra).
Merge import names per target module before emitting pub use (fixes duplicate SizeBound / IdentityEffect lines from multiple import statements).
Parent enums are not pub-re-exported (top_with_parents = graph types + enum types only); variant parents only emit use Parent::{…} when parent_mod == import_mod.

v2-compiler builds after regen_stage0; v2_compiler_effect_derivation.rs no longer imports CollectionSizeEffect::IdentityEffect alongside IdempotencyEvidence::IdentityEffect.

Blocking: worksheet draft vs emitter in same PR (worksheet L3)

Addressed in worksheet prose: status is WORKSHEET COMPLETE (content) with §8 sign-off OPEN; mechanical dispatch rule now distinguishes merge gate (§8 required) from #4127 co-delivery (emitter present for §4 falsification probes only, same pattern as SG-RC #4100 + implementation PR collapsed for review).

Merge remains blocked on Modeling DFS §8 per checklist; co-delivery is intentional for falsification review under Target Realization dispatch.

…0425/E04

briansrls

Review metadata

Provider / model: codex / unknown
Commit: 84d876e3 · Trigger: schedule
Thinking: 252s wall

⚠️ No new distinct findings beyond the previously reported unresolved blockers.

…0425/E04

briansrls · 2026-05-31T18:47:03Z

Review metadata

Provider / model: openai-pro / gpt-5-5-pro
Commit: fabc5a19 · Trigger: manual
Comparison: main @ 60e3bfcb ... session/sunny-cat-161 @ fabc5a19
Conversation: View conversation

1. Story of the diff

This PR documents SG-8 as a module/export-surface failure rather than as a pile of unresolved Rust names, then patches the live v2 Rust emitter used for the v4 M1 probe. The worksheet identifies two root causes: import emission treats every imported spelling as a possible enum variant parent from the import-site module, and generic type aliases like List<T> are classified as type declarations but emitted as nothing. The code responds by importing TypeItem, partitioning graph type imports away from variant-parent expansion, routing pub use lines through ItemInfo.module_name, grouping specific imports by module, and emitting parametric aliases via pub type Foo<T> = ...; the stage0 Rust file mirrors those .dag emitter changes. chatgpt-review-0b393e5e-7367-4c…