[devin-ai-integration]

What and why

Step 4 of the execution-owned IR migration (docs/internals/execution-ir-design.md, "Migration as a stack of pull requests", row 4): origins replace declaration handles for the text a run prints, so the executors, the gRPC session and the document state query name things from program.Origin and the lowered records rather than from ast nodes. Expressions stay AST-based; the exec/runtime → syntax/ast edge is not removed here.

Stacked on #2144 (feat/ir-action-exec-ids), with #2145 (feat/ir-state-exec-ids) merged in --no-ff. Until #2145 merges, this diff shows #2145's changes as well as its own; the step-4 diff alone is git diff fa62971a6b..HEAD. Required merge order: #2137, then #2144 and #2145, then this PR (retargeted to develop after they land).

Kind words. program.NodeKind.Word() is the construct's SysML v2 name and Origin.Kind carries it for action nodes: initial node, final node, fork node, join node, merge node, decision node, action, perform action, accept action, send action, assignment action, terminate action, if action, while loop, block, statement. Vertices keep the lower.VertexKind words (state, pseudostate, terminate action, …); regions, transitions, entry transitions and bindings are region, transition, entry transition, binding. Origin.String() is node <name> when named, the <kind> otherwise, the node for the zero Origin. An unnamed node that a message once printed as the *ast.IfActionNode is now the if action.

Public surface:

• program: NodeKind.Word(); Origin.String() defines the zero value; Origin.Kind is a construct name, never a Go type name.
• lower: (*ActionGraph).OriginOf(node); EntryTransition.Origin and EntryTransition.SourceText (the lower.EndpointText of a TransitionMember's source, lowered once); Binding.Origin (no Path: a binding belongs to no set it is unique within).
• runtime: nodeDescription is gone; describeNode(graph, node) and NodeRef.Origin() read the graph's origins (guardHolds takes the graph). firedEntry, entryFromName, entryTransitionEffectHead and stateSourceOrigin read EntryTransition.Origin/SourceText and the state origin table instead of lower.DocOf/lower.EndpointText. Context.bindingOwners holds program.Origin, Model.declared is keyed by symbols.ElementKey (symbols.KeyOf), and NoValueError.Ref is removed: nothing read it.
• gRPC session.go: branchesOf/gateOf read Kinds, Origins and ActionEdge.Else instead of type-switching on ast.DecisionNode/ast.ControlFlowEdge.
• queryexec/states.go: state and region names come from Origins/RegionOrigins, the state symbol through Provenance.State(id).

Conformance expectations moved: none. No .expected.json, trace or gRPC golden carried a the *ast.X rendering, so every baseline is byte-identical; the step-1 pin test is rewritten as TestOriginDescriptions (node <name> / the <NodeKind.Word>, kind word, span and document over every conformance action) rather than deleted.

Merge of #2144 and #2145: no textual conflict. One semantic conflict — both sides removed the last ast use in snapshot.go, leaving an unused import that failed the build — fixed in its own commit.

Where the design met the code:

• Origin.Element and Origin.Docs stay empty. Element ids come from identity, which needs a semantics.Model: creating one per lowered graph is not an option (semantics.NewModel attaches itself to the shared resolver and clears its caches — tests/stressmodel TestEditsHoldNoStaleState caught the first attempt as 40 MB of state outliving edits), reading the model the resolver's owner attached works but importing identity from lower costs sysml-engine 32 KB gzipped (it never linked identity) and puts sysml-wasm 526 bytes over its budget. Element wants an id provider the model's owner registers (or a fill through Provenance in the frontends that already hold an identity table); that is a follow-up, not a budget raise. Docs need the source text (semantics.Model.DocumentationOf), and lowering has no channel for it: the graphs are lowered from a *resolve.Resolver threaded through some twenty functions, several in the calc-lowering files step 5 owns, and the index holds no text.
• describeNode is //go:noinline: inlined, its error-path callers each carried the origin lookup, 17 KB of wasm code; the combined budget holds at 10,498,449 gzipped (10,494,609 on the bare merge, budget 10,500,000; no budget moved).
• NoValueError.Ref was dropped rather than lowered: no caller read it.
• The machine state has an id past the origin table (step 3), so stateOrigin answers the zero Origin for it and stateSourceOrigin falls back to the machine's document as before.

How it was verified

go build ./..., go vet ./..., gofmt -l . (empty), make lint, go test ./..., go test -count=1 ./tests/corpus -run 'TestTrainingExamples|TestPilotCorpora' with both corpora downloaded, go test ./tests/hygiene, go test -C tools ./..., go test ./tests/wasm (budgets unchanged), python3 scripts/check-doc-ids.py, python3 scripts/changelog.py check; the comment-length check over the merge-base diff prints nothing. The same gate ran on the bare merge of #2144 and #2145 before any step-4 code (it found the snapshot.go import).

Tests that pin the step: TestExecutionConformance (error text), TestOriginDescriptions, TestNodeKindWord/TestOriginString, TestAssignIDs_ActionOriginsNameWhatMessagesName, TestGRPCConformance, conformance_docquery_test, tests/grpc and the runtime robustness suites.

Checklist

• make test and make lint pass locally
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/42ffb49bd028432a81e2981ea0a239bf
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/42ffb49bd028432a81e2981ea0a239bf?variant=devin
Requested by: @HuiJun

---

🟢 Devin Review found nothing to fix · View findings · Follow Devin's work

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

Devin Review found 1 finding. Devin automatically fixed it.

Fixed by Devin (1)

• Implicit binding cycles go unreported — EndValueBindings lowered connector-end value bindings without an Origin, so binding owners keyed by origin could not tell them apart; they now carry the end attribute's origin, and TestEndValueBindingsLowerEndBodyValues pins that the two differ.

View findings in Devin Review · Follow Devin's work