feat(migrate): encode deferred triggers in standard SysML v2 under -strict

[devin-ai-integration]

What and why

Built on #625 (feature/strict-migration, now merged), targeting develop.

Under -strict a UML deferrableTrigger on a SignalEvent used to be dropped as unmapped; by default it is written as the OpenSysML-only defer Sig;, which the OMG pilot's parser rejects (and in a large model such as TMT one rejected line cascades into tens of thousands of parse errors). Strict output now carries the deferral in standard SysML v2:

state Off {
    @MigrationMetadata::DeferredEvent { ref :>> signal : Door; }
    item deferred : Door[*] ordered;
    do action buffer {
        first start then receive;
        action receive accept kept : Door;
        then action keep { assign deferred := SequenceFunctions::including(deferred, receive.kept); }
        then receive;
    }
    exit action flush {
        for kept in deferred { send kept to self; }
    }
}

• One buffer, accept loop and flush loop per deferred signal; with several, the loops (and the state's own do behavior) are forked inside the one do action; the state's own exit behavior runs first in the exit action, the flush after it.
• Every synthesized member goes through the SynthesizedName machinery and is chosen clear of the state's own members; a state's nested do/exit behavior keeps its name, with its qualified name (used by view exposes and diagram layouts) now routed through the generated action.
• Every route the signal reaches the object by gets its own accept loop into the shared buffer, resolved through the same portRoutes a transition trigger uses: the direct accept plus accept … via <port> for each connector-derived port, or only the ports the deferrable trigger names. The flush replays to self, which a port-less trigger accepts (noted).
• Precedence: a signal accepted by an unguarded (or statically true) transition out of the state itself is not kept (Approximated, note names the transition). A guarded transition, or a transition out of a substate, does not suppress the deferral: the transition takes the occurrence while its guard holds / its substate is active, the loop keeps it otherwise (noted as an approximation of UML's priority). A state left by a completion transition keeps nothing (Unmapped, with a not migrated: defer Sig; comment in the state body: the accept loop would never let the do action complete).
• A state's own do behavior joins the fork only when it is written as an action; one with no action form (a StateMachine doActivity) leaves the accept loop alone, with the existing comment.
• Known limitation, documented in the report note and reference: with several deferred signals the flush replays one signal at a time, each in arrival order, so interleaving between signal types is not preserved (UML leaves event-pool order open, so this is a permitted approximation).
• Both modes now write @MigrationMetadata::DeferredEvent { ref :>> signal : Sig; } (metadata def DeferredEvent { ref signal : Base::Anything[1]; } added to the bundled library; the ref :>> signal : Sig form is what the pilot accepts). Default mode still writes defer Sig;. Non-signal deferrable events stay unmapped as before.
• Runtime support so the encoding executes: internal/ir/lower treats a state's item usages as state data (like its attributes), and lowers an explicit send … to self as a send to the sending object (the same route as a send with no target).

Specification basis

UML 2.5.1 §14.2.3.9 (deferred events: kept while the state is active, dispatched on exit; a transition on the same event takes precedence). No row of docs/project/spec-compliance.md moves. Documented in docs/reference/sysml-v1-migration.md (deferrable-trigger rows and a new "Deferred signals under -strict" subsection, including that the composite state's do action covers its substates' activation) and docs/guide/11-migrating-from-sysml-v1.md.

How it was verified

• go build ./... && go vet ./... && gofmt -l . && go test ./... clean; python3 scripts/changelog.py check, scripts/check-doc-links.py, scripts/check-doc-ids.py clean.
• tests/migrate/behavior_more_test.go: the ovenMachine default and strict tests now assert the metadata, the strict encoding and the absence of defer; new TestStrictDeferredSignalsAreKeptAndReplayed covers two deferred signals in one state with its own do and exit behavior, a composite substate, outgoing-transition precedence, a completion-transition state, report targets through the generated action, and executes the migrated model in the REPL session: signals sent while in the state are kept and delivered after it exits, a transition-triggered signal is not buffered, and the state's own do/exit behavior still runs. New TestStrictDeferredSignalsAreKeptByEveryRoute covers a connector-delivered signal (via inbox), a trigger naming a port (via side) and a guarded matching transition whose guard is initially false, executing that the connector-delivered occurrence is kept, then replayed and taken by the transition once its guard holds. New TestStrictDeferralSurvivesInactiveSubstateTransition executes a composite state deferring a signal that one substate's transition accepts: sent while the other substate is active it is kept and replayed on exit. New TestStrictDeferralWithoutRunnableDoBehavior covers a doActivity that is a StateMachine.
• Pilot batch validator (validate-sysml-batch, pinned pilot) over the strict output of every tests/migrate/testdata/xmi fixture plus our libraries: oven.sysml (the deferred-trigger fixture) 0 errors; 0 errors mention defer, the buffer, or the flush. The remaining 472 errors in that run are unrelated buckets (unresolved this.x references in opaque behaviors, transition/fork/else parse breaks in empty_behaviors, plant_states, station_points, .kerml libraries not collected).
• TMT (Open-MBEE/TMT-SysML-Model, TMT.mdzip, 823 deferred triggers), bin/sysml TMT.mdzip -strict -convert sysml, pilot over the output plus our libraries — before (base branch) vs after:
  • parse errors (no viable alternative / mismatched input / missing EOF): 50 → 50, none attributable to defer (the 50 are the decision-else ordering and transition/first/accept/do breaks owned by other sessions).
  • total error: lines: 32,536 → 32,592. The +56 are all additional references to signal/item types the pilot already fails to resolve on the base branch (e.g. GCFrame, PEAS-TCS signals): each such deferral now names the type three more times (metadata, buffer, accept). Not fixed here.
• Not done: a dedicated internal/exec/runtime conformance fixture for item state data / send … to self; that behavior is covered by the migrate execution test only.

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/2a00d846dce845f1b62739e756690000
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/2a00d846dce845f1b62739e756690000?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

[devin-ai-integration]

Fixed in 17f0ed7: a deferral dropped for a completion transition is now kept as a /* not migrated: defer Sig; — … */ comment in the state's body (and the bare state Name; shortcut no longer applies when such a comment exists). TestStrictDeferredSignalsAreKeptAndReplayed asserts the comment on Prep, and the no-defer checks now match whole statements so the comment is not mistaken for one.

[devin-ai-integration]

Re "Interleaved deferred signals replay out of order" (thread PRRT_kwDOTp2CH86mXFkH): confirmed — with one item buffer per signal the flush replays A1, A2, B1 for arrivals A1, B1, A2. This follows from the per-signal-buffer design the task specified ("one buffer per deferred signal, or one buffer typed by the common supertype if the signals share one"); a shared arrival-ordered buffer would need a common supertype the pilot accepts as the accept/send payload type, or an ordering key on every kept occurrence merged at flush time. Leaving this open as a design decision for the maintainer rather than changing the encoding silently.

[devin-ai-integration]

Resolved as a documented limitation (maintainer decision): UML 2.5.1 §13.3.3.3 makes the dequeuing order of the event pool a semantic variation point, so replaying one signal at a time, each in arrival order, is a permitted approximation. 77be761 states this in the migration report note on every deferred trigger of a state with several deferred signals (asserted in TestStrictDeferredSignalsAreKeptAndReplayed) and in docs/reference/sysml-v1-migration.md.