feat(migrate): faithful states, events and opaque behaviors in the v1 transformation census

[devin-ai-integration]

What and why

This closes the census group for states, events and opaque behaviours in the SysML v1 → v2 transformation census. Each row is now either faithful to OMG's SysMLv1Tov2 mapping (or a valid v2 equivalent of it) or deliberate, with the measured reason recorded.

Migrator

• Opaque bodies. An OpaqueAction, OpaqueBehavior or FunctionBehavior body becomes rep language "<lang>" /* body */, which is OMG's TextualRepresentation, pairing bodies and languages by index. A translated body keeps its executable statements and now also keeps its source text. An untranslated body gets the representation instead of the old "body not migrated" comment. A body with no language stays a comment. A representation keeps the raw body, leading indentation included. Bodyless elements become plain usages or definitions and are reported as mapped.
• Terminate pseudostates. They are written as action <name> terminate;, with incoming transitions targeting that action. Before this change they were written as a transition to done, which only ended the enclosing region.
• Single-region composite states. These are written as OMG's parallel wrapper: state S parallel { …S's actions, deferrals, metadata… state <region> { entry; then A; … } }. Orthogonal regions are unchanged. The single region of a state machine is still flattened (see Candidate errata).
• Time triggers. The existing accept after <duration> / accept at <instant> notation is the textual form of OMG's TimeTrigger structure, so those rows are recorded as faithful.

Lowering and runtime

• parallelOwnedMember accepts @Meta { … } (prefix metadata) in a parallel state's body, as it already accepted metadata …;. Wrapped states with deferrals need this.
• A composite state's region with no entry no longer fails to instantiate. On default entry, such a region counts as complete if it is left with no active child: either its stand-in sub-state is entered and nothing inside it starts, or it has no stand-in and no start. A composite whose regions are all complete in this way completes at once, which is PSSM Entering 004: the flattened form already did this, and the wrapper now keeps it. A region that a later transition or fork branch makes active must reach its final state before its owner completes. A running do-action, on the owner or on a region's stand-in, still defers completion, and the owner completes when that action ends. A first entry through a history with no record uses the same default-entry rule. Top-level machine regions with no start are still refused, and so is a region whose stand-in is itself parallel.
• The former lowering refusals of a no-entry or fork-only region now assert the supported behaviour instead: tests renamed in lower/fork_plan_test.go, plus the PSSM referee classifier and docs.
• Overlap: open PRs fix(runtime): interleave executors due at one instant move by move #850, fix(runtime): = values in action, calc, block and state bodies hold at all times #842 and fix(runtime): write-granular change triggers, unordered composite entry, entry-action transition effects #833 also touch internal/ir/lower/state_graph.go and state execution. This PR changes parallelOwnedMember and recordRegionInitials there, plus region entry and completion in the runtime.

Census

Totals (faithful / approximate / not implemented / deliberate), measured on develop at bc3333c11: 500 / 50 / 42 / 191 → 529 / 17 / 42 / 195.

Rows	Before	After
OpaqueAction_Mapping, OpaqueBehavior_Mapping, OpaqueBehaviorMembership_Mapping, OpaqueBehaviorSpecification_Mapping	approximate	✅ faithful
all 22 TimeTrigger* rows	approximate	✅ faithful
CommonPseudostate_Mapping, PseudoState_Mapping	approximate	✅ faithful
Region_Mapping	approximate	✅ faithful
TimeEvent_Mapping, ChangeEvent_Mapping, ChangeEventReturnParameter_Mapping, ChangeEventReturnParameterMembership_Mapping	approximate	⛔ deliberate. OMG writes the event as a package-owned calculation, where expressions that name context features do not resolve. The expression stays inline at the trigger instead.

Candidate errata

• OpaqueBehaviorSpecification_Mapping pairs body with language by index, so a body with no language is dropped. The migrator keeps it as a comment.
• The pseudostate target is a plain StateUsage, which drops the kind: a choice becomes a waiting state, and a fork fires one transition. The migrator writes choice, junction and history as #StateMachines::<kind> state, fork and join as nodes, and terminate as action … terminate.
• For a state machine with a single region, OMG writes a non-parallel state definition whose only sub-state (the region) has no entry into it, so nothing ever enters it. The migrator keeps that region flattened.

Specification basis

• OMG SysML v1→v2 transformation (SysMLv1Tov2.xmi, 2025-02-01): Region, Pseudostate, OpaqueAction, OpaqueBehavior and TimeTrigger mapping classes.
• PSSM 1.0 (ptc/18-11-06), Entering 004: a composite state whose region has no initial pseudostate completes on default entry.
• UML 2.5.1 §14.2.3.4.5 / §14.2.3.5 (default entry, fork).
• docs/project/spec-compliance.md has rewritten rows for region default entry and completion, and for fork-omitted regions.

How it was verified

• Gates: go build ./..., go vet ./..., gofmt -l . (empty), OPENSYSML_REQUIRE_PSSM_SUITE=1 go test ./..., make lint, make docs-check, scripts/check-doc-ids.py, scripts/changelog.py check, census -check -require-xmi plus go test -C tools ./census/transformation, and TestPSSMSuiteMigration.
• New tests:
  • Migration tests under tests/migrate for opaque representations, terminate and the wrapper (entry, nested transition, exit, connection points, state actions, deferrals, orthogonal regions, no-entry regions).
  • Conformance fixtures with trace goldens: state_parallel_region_without_entry_completes (the Entering 004 shape), state_parallel_region_without_entry_inactive, state_fork_only_region_entered_by_default, state_nested_fork_starts_outer_region_inactive, state_history_restores_inactive_parallel_region, state_transition_cross_region_inactive_wrapper, state_parallel_deferred_event, state_history_first_entry_inactive_region, state_parallel_stand_in_do_action_defers_completion, and the single-region wrapper fixture.
  • Robustness tests: TestRuntimeRobustnessInactiveRegion and TestRuntimeRobustnessParallelWrapperLeavesInnerRegionInactive.

PSSM

• The referee is unchanged from develop bc3333c11 at 56 pass / 6 fail / 34 not-expressible / 7 differs-by-design. Entering 004 passes.
• The migration ratchet:

	mapped	approximated	unmapped	skipped	validation errors
develop bc3333c11	33,013	7,325	8,764	6	12
branch	33,086	7,329	8,759	6	12

• Movements:
  • 3 terminate pseudostates move from approximated to mapped.
  • 6 initial-transition-effect activities (Entry 002-D, History 002-D, Junction 001, Junction 002, Standalone 002, Transition 023) move from unmapped to mapped. They newly report their contents, one of which is unmapped: an object flow whose target has no v2 name.
  • Opaque bodies are now reported as mapped.
  • In total: +73 mapped, +4 approximated, −5 unmapped, every element accounted for.

TMT 2024x (-migrate sysml -layout, then -validate, -compare-results -seed 1, analysis_report.py and HTML rendering). Develop at bc3333c11 can't migrate TMT on its own: since df6dc8876 (from #1033), a dependency or comment on a nested requirement usage is written with a feature chain, so the notation doesn't parse. #1051 fixes that. Both sides below include #1051: develop bc3333c11 + #1051 → this branch + #1051.

	default: develop → branch	-strict: develop → branch
mapped / approximated / unmapped	33,904 / 11,234 / 4,063 → 33,933 / 11,205 / 4,063	33,103 / 11,003 / 5,465 → 33,132 / 10,974 / 5,465
validation errors	0 → 0	0 → 0
runs attempted / completed / stopped / refused	41 / 31 / 10 / 21 → same	41 / 31 / 10 / 21 → same
comparable observables, exact / differ	142 / 14 → same; the differing list is identical	142 / 14 → same; the differing list is identical
diagrams laid out	1,242 / 1,347 → same	1,242 / 1,347 → same
HTML documents	104 → the same 104 files, with the same 14 failures as develop	none declared → same

The only verdict movement in either mode is 29 elements from approximated to mapped: 14 FunctionBehavior and 15 OpaqueBehavior, whose bodies are now textual representations. No element becomes unmapped.

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/census-states-events.changed.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/7e80a1d5715b4e8b9bf7039b09d8d589
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/7e80a1d5715b4e8b9bf7039b09d8d589?variant=devin
Requested by: @HuiJun

[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