feat(migrate): a strict migration refers to no OpenSysML library; -portable inlines the ones a default migration does

[devin-ai-integration]

What and why

A SysML v1 migration under -strict is documented as output "an interchange partner could read", but it still referred to OpenSysML's own libraries: MigrationMetadata::SynthesizedName/StandIn/LibraryNameAvoided markers, #StateMachines::<kind> pseudostates, @Stochastic::Probability, DocumentQueries documents, @DiagramLayout geometry, Simulation::*, RandomFunctions::uniform and OpenSysMLMathFunctions::quotient/ceiling. A tool shipping only the standard library — the pilot implementation, or a modeling tool built on it such as CATIA Magic — fails on every one of them: on a 1124-element Cameo model the pilot reports thousands of unresolved-reference and knock-on type errors, none of them about the model.

Under -strict the migrator now writes nothing such a tool cannot resolve. Where the standard has a form, that is written and reported approximated where it loses something; where it has none, the construct is refused with a report line:

• the three MigrationMetadata name markers become comments (// names the migration made up: a, b); the report carries the same information
• choice/junction → plain state x; with the pseudostate's guarded transitions (Approximated); shallowHistory/deepHistory and the transitions into and out of them → Unmapped
• a decision node's branch probability → then b; // probability 0.25 (Approximated, noted on the edge whether or not its guard migrated)
• a duration constraint over an interval → a fixed wait of the midpoint when both bounds are literal numbers (Approximated), refused otherwise
• Java / over whole numbers → RealFunctions::floor((x - x % y) / y); Math.ceil → -RealFunctions::floor(-x) (both exact)
• «Document» and «DiagramTable»/«InstanceTable» → Unmapped with a /* not migrated: ... */ comment; a Monte Carlo analysis is written as the ordinary analysis without its Simulation::MonteCarlo generalization (Approximated); a simulation configuration's run settings → a comment in place of @Simulation::Configuration
• views carry no DiagramLayout geometry (noted on the view's report entry when a layout was available)

The one OpenSysML reference a strict migration keeps is the deferral encoding's @MigrationMetadata::DeferredEvent / #MigrationMetadata::DeferredKeeper: the encoding itself is standard notation, the runtime and lowering read those annotations to schedule the deferral, and a tool that does not know them reads the state as written. This is documented as the exception.

The default (non-strict) migration is byte-for-byte unchanged on the SEM model and every fixture.

-portable is the other route to a file such a tool loads, for the default migration: it leaves what is written unchanged and appends the OpenSysML library packages the output refers to — every package named as the first segment of a qualified name, and every one those refer to in turn (MOSA pulls in DocumentQueries) — after the model, under a comment saying so. A standard library package is written as a library package, which a file of one's own may declare (the pilot warns on the former); a KerML library (RandomFunctions, OpenSysMLMathFunctions) is written in its SysML spelling, its functions as calc defs, and one whose words are KerML's alone (StateActivity) is left referenced with a report line. The report's summary names the packages inlined (inlined 5 OpenSysML library package(s): DiagramLayout, DocumentQueries, MigrationMetadata, StateMachines, Stochastic) and the JSON report lists them under libraries. -strict -portable composes: whatever a strict migration still refers to (the deferral annotations) is inlined. The flag is a CLI/Go-API option (Options.Portable); the gRPC MigrateRequest does not carry it yet.

Specification basis

No new clause: -strict already promised notation of the pinned production. ModelingMetadata, RealFunctions::floor and % are KerML/SysML v2 standard library (SysML v2 1.0 Kernel Function Library, Semantic Library). No row in docs/project/spec-compliance.md moves.

How it was verified

• New TestStrictMigrationRefersToNoOpenSysMLLibrary in tests/migrate migrates every testdata/xmi/*.xmi fixture under Options{Strict: true} and fails on any line naming an OpenSysML library package (read from the library directory, so a new library is covered automatically), the deferral annotations excepted. TestStrictProbabilityNoteReachesAnEdgeWhoseGuardIsNotMigrated pins the report note on a weighted edge whose guard is kept as a comment.
• New TestPortableMigrationInlinesTheLibrariesItRefersTo migrates every fixture under Options{Portable: true} and checks that each OpenSysML package the model part names is appended exactly once as a library package (none marked standard), is accounted for in Report.Libraries and the summary, that nothing is appended when nothing is referenced, that the output still validates, and that a default migration accounts for no libraries; TestPortableMigrationSpellsAKerMLLibraryInSysML pins RandomFunctions arriving as calc def uniform; package-level tests pin the transitive case (MOSA → DocumentQueries), the refused KerML-only library and a package merely named like a library.
• Strict expectations updated where the output deliberately changed (TestStrictPropertyProbabilityUsesOnlyLiteralDefaults, the choice/junction/history state tests, the deferred-signal tests, TestDiagramNotesDoNotNameOmittedVertices); a shared wantMadeUp helper accepts either marker spelling.
• go test ./tests/migrate/ ./internal/translate/... ./cmd/..., go vet, gofmt -l, make man-check, scripts/changelog.py check, scripts/check-doc-ids.py all green.
• By hand: the SEM model migrated with -portable (views and all) validates in the pinned pilot batch validator on its own with 0 errors and 0 warnings, and under sysml -validate with no errors; so does -strict -portable.
• By hand and by an end-to-end run against a binary built from develop: the 1124-element Cameo SEM model migrated with -strict has zero extension references, loads under sysml -strict -validate with no errors, and validates in the pinned pilot batch validator with the standard library alone (0 errors, 0 warnings), where the develop binary's strict output fails; its default migration is byte-identical to develop's and still validates with the extension library; -render-all writes the same 33 views from either; deferral replay in the strict output still runs.

Checklist

• make test and make lint pass locally (the packages touched; staticcheck is not installed on this machine, CI runs it)
• 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 (none moved; packaging/man/man1/sysml.1 regenerated for the -strict and -portable usage text)
• 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/dbff7ccc1454416fb497412e51cf5513
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/dbff7ccc1454416fb497412e51cf5513?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]

Runtime verification (binary built at 9c22c3b, compared with one built from develop at 4e70d6a)

Strict output of a 1124-element Cameo model (SEM)

• -migrate sysml -strict: 954 mapped, 161 approximated, 9 unmapped; the notation names no OpenSysML library package.
• sysml -strict -validate: 0 errors (9 duplicate-import warnings, as on develop).
• Pilot batch validator with the standard library alone: passes. The develop binary's strict output fails the same validator (unresolved MigrationMetadata, DiagramLayout, Stochastic, StateMachines, DocumentQueries types and their knock-on errors); a deliberately broken file fails it too, so the pass is not vacuous.

develop strict output under the pilot	this branch's strict output

Regression controls

• Default (non-strict) SEM output is byte-identical between the two binaries; the pilot accepts it only with the extension library, as before.
• -render-all writes the same 33 views from strict and default output; strict views carry no canvas geometry.
• 12 fixtures (plant_states, documents, montecarlo, simconfig, ui_stimuli, meter, reactor, tables, weighted_decision, accept_via_context_port, deferral, mission) migrate under both binaries × both modes; default notation and reports are byte-identical, every strict output validates natively.
• Strict deferral still runs in the REPL: Alarm/Beep buffered in Busy, replayed through Alarmed → Done, buffers cleared, no stale replay on the second visit.
• Strict arithmetic rewrites execute correctly (Ceil(2.1) = 3, Ceil(-2.9) = -2; meter quotients and remainders unchanged); a [1, 8] interval wait becomes 4.5 with its midpoint note.

Found by this run and fixed in 3f39a59: a weighted edge whose guard is kept as a comment reported the guard only, not that its probability is written as a comment. Not exercised at runtime: strict choice pseudostates and non-literal interval waits (unit tests cover both).