chore(release): merge main (0.9.1 + Windows hotfixes) back into develop - #754
Merged
Merged
Conversation
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
… into release/0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> # Conflicts: # client/rust/conformance/sysml.descriptor.binpb
… into release/0.9.1
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com> # Conflicts: # changes/unreleased/authoring-transitions.added.md
…ement' into release/0.9.1
…into release/0.9.1
…s' into release/0.9.1
…nto release/0.9.1
…into release/0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> # Conflicts: # internal/workspace/libs/stdlib.snapshot # internal/workspace/libs/stdlib/OpenSysML Libraries/MigrationMetadata.sysml
… into release/0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com>
…release/0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> # Conflicts: # internal/semantic/semantics/annotations.go # internal/workspace/libs/stdlib.snapshot
…o release/0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> # Conflicts: # docs/project/pilot-differential-baseline.json
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
…t pom Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
…narrating them twice Co-Authored-By: jason.han <hanhuijun@gmail.com>
* fix(runtime): use effective parameter ranges for writes Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(runtime): clarify action parameter write ranges Co-Authored-By: jason.han <hanhuijun@gmail.com> * feat(convert): convert a model of several documents as one graph (#653) -convert took one file and Convert refused a model_hash of several documents; a file converted alone wrote a reference into another as an unresolved name. The documents of a model are now analyzed together, as a workspace analyzes them, and each is encoded over that one analysis: a reference from one document to an element another declares links the subject that document writes, by the name and identity its own encoder gives it (a positional name included). Each document's root elements carry sysx:sourceDocument. - export.ModelToRDFWith, convert.ConvertModel - sysml a.sysml b.sysml -convert ttl|api-json - Convert of a multi-document model_hash to ttl or api-json; notation stays a one-document operation (FAILED_PRECONDITION, as before) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(export): a dependency's prefix metadata is an owned Annotation (#652) Dependency takes PrefixMetadataAnnotation, not PrefixMetadataMember: the metadata usage is the annotatingElement of an Annotation the dependency owns. The reader takes that Annotation as the ownership edge. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(check): distinguish shorthand succession multiplicity Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(examples): require single RocketEquation inputs Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): a flow's payload is a PayloadFeature (#652) FlowDeclaration takes 'of' FlowPayloadFeatureMember: a FeatureMembership owning a PayloadFeature. The declared payload is now written as that feature (m.cmd reaches it), 'of T' as one typed by T, and the reader rebuilds the 'of' clause from it instead of the sysx:payload expression. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(corpus): source positions move with a dropped blank line; lint Seven KerML Annex A examples begin with a blank line the rendering drops, so hop 2's positions sit a line higher: whitespace-only, not stable. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(export): keep reading a legacy sysx:payload; refuse stray payload features Review: a graph from an earlier release states a flow's payload as sysx:payload beside standard ends, and it lost its 'of' clause. It is read again; a flow stating both shapes is refused. Only the one payload feature a flow's head writes is kept out of its body; any other is refused, not dropped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(export): prefix annotations in sync; refuse one annotating another element Review: sync minted the dependency prefix's Annotation an id of its own, and minted the prefix metadata usage one it cannot declare, so the next export moved both back. The Annotation is a derived satellite that follows its metadata usage (_an, as _om does), and a '#' prefix is not minted. An owned Annotation whose annotatedElement is not its owner is refused rather than written as the owner's prefix. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs(pilot-diff): refresh baseline figures and narrative Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): a source range ends at the declaration's last token Review: a span runs on to the next token, so 'part a; /* note */' ended a's range after the note. The range now ends at the last token that is not trivia; a comment that is a declaration's body (doc /* ... */) stays in it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): keep add_member positional parameters stable Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(edit): accept other-language keywords in import targets lexesAsQualifiedName rejected every token the lexer marks Keyword, so a target naming an element after a word that is reserved only in the other language (type in SysML sources, part in KerML sources) was refused as not a qualified name. Read the token stream the way the parser does: a keyword of the file's own grammar is refused with a message telling the modeler to quote it, a keyword of the other grammar reads as a name, and KindUnknown reads as SysML. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): a model of several documents is checked as one Review: - identity scopes are counted across the documents, so one declared id in two projects is two subjects, not one; - a subject two documents declare is refused, naming both; - Convert of a modelHash refuses a document with syntax errors, with its diagnostics, as a single document is; - several files to a flexo:// branch is refused rather than written as a path. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(edit): keep visibility when replacing documentation Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(repl): read syntax-backed scopes for recorded files Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(doc): split a metadata column path outside quoted names Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(ci): give the CircleCI WebAssembly gate a 30-minute test budget The "Go gates and binaries" job ran go test ./tests/wasm with the default ten-minute budget. TestWasmBuilds alone takes 555-585 s on this executor, so TestWasmRuns hit the deadline while linking the js target ("panic: test timed out after 10m0s" at TestWasmRuns/js/version) on every develop run since the gate reached the tests. The gate now runs with -timeout 30m, matching the race job's budget for the same package, and the step's no-output timeout is raised alongside it. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(parser): parse rooted assert references Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(verify): preserve values in solver questions Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(verify): document satisfaction bindings and evaluation stamping Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(semantics): a parameter subsetting several features takes the intersection of their ranges A subsetting feature by default shares the subsetted feature's properties and may only restrict its multiplicity (KerML 1.0 §7.3.4.4), so several subsetted features bound the parameter to what satisfies all of them — the intersection of their ranges; a redefinition bounds it likewise (§7.3.4.5). EffectiveParameterRangeAlong replaces the first-found walk, which made the answer order-dependent, with a recursive intersect over explicit RelSubsets/RelRedefines targets and the implicit chain successor; a visited set ends cycles, and a target that states nothing, has no implicit default, and whose own targets give nothing contributes no bound. Ranges intersect by Range.Intersect, the same rule the runtime's inheritedMultiplicity uses. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(migrate): name the expression argument of a usage-owned document column Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(edit): cover plain documentation replacement and note the visibility fix Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): compare a legacy transition trigger by the element it names A graph carrying both a transition's trigger structure and an earlier mapping's sysx:trigger was refused when the text qualified the payload type differently ("T::Sig" beside accept Sig): an early check compared the text against the type's short name. Drop that check and compare sysx:trigger and sysml:via against the structure in transitionTrigger by resolving each differing name from the transition through its enclosing namespaces. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(edit): place metadata prefixes after declaration modifiers Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): give same-named toolkit roots positional identities Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): keep the name of a reference to its own scope Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(pilot-diff): regenerate baseline after merging develop develop@07cbe053e produces the same result without this branch because develop's optional bare in parameter change removes the warning at examples/runtime-showcase/delta-v-budget.sysml:93, and the committed baseline predates that change. The control also lacks the prior warning at Simple Tests/Behaviors.kerml:14; this additional movement is recorded in the refreshed counts. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(source): read a lone carriage return in a comment as a line break KerML 1.1 §8.2.2.1 makes \r, \n and \r\n line terminators, so Comment::body processing (§8.2.3.3.2) and documentation prose split on all three. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(migrate): key documentation sorts on the documentation the query reads docKey read the exact Comment::body of the doc comment it writes, which keeps the trailing space or line break single-line and multiline comments differ in, while the OrderBy it generates reads DocumentationOf's prose. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(parser): parse source multiplicities before then in nested action bodies Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(authoring): resolve sequence references to earlier declarations in nested bodies Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(authoring): refuse sequencing after a member that cannot be a succession source Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(diff): keep the parent's differential counts, restating the examples digest Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): count a calculation send by its result type, as the runtime sends it A send invoking a calculation sends the calculation's value, so the undeclared-signal lint now selects the invoked declaration with SelectCall and decides with the new Model.CallsCalc, which the runtime's invokesCalc now also uses. A calculation send records its result's declared type; any other invocation still records the invoked name. The when-trigger adjudication test now expects the lint's warning on an injected signal nothing sends. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(authoring): anchor sequence placement on declarations, not first references Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(runtime): a performance occurrence holds a parameter at its effective multiplicity Context.materialize builds the occurrence's feature shape through featureMultiplicity, which defaulted an unstated multiplicity to [1..1]: when an earlier parameter default evaluated this, the occurrence materialized, and a later bare parameter's empty binding was then written through SetFeatureValue and rejected with a multiplicity violation. featureMultiplicity now returns the parameter's §7.6.3 effective range, so a bare parameter stores its binding as [0..*], and statedMultiplicity answers the same for a parameter stating nothing. The parameter special cases in admitsNoValue and pinsOf fold into the shared rule, alongside the write-target path write_conformance already takes. The bindCalcParameters override stays: Decl.Target.mult comes from extractMultiplicity, the declared bound, not featureMultiplicity. Migrating SysML v1: direction parameters that wrote no multiplicity (in, out, inout and inout bindings, port and context members, Monte Carlo and activity pin declarations) now emit an explicit [1], preserving the v1 reading of exactly one value under §7.6.3 instead of silently widening to [0..*]; reduce-lambda parameters keep their bare form, which the grammar requires. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(semantics): tighten the effective parameter range comments Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): exempt only interfaces port-conjugation judges from the port lint A connector's concrete ends were skipped whenever its type declared port-typed ends, but port-conjugation judges only a two-ended interface, so a connection definition pairing unrelated ports hid the mismatch from both checks. The exemption now requires an interface with two port-typed ends. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): warn on non-unit source multiplicities in nested action bodies Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(runtime): drop a duplicated succession from the this-default action fixtures Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci: re-run checks after a transient Go module proxy error Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(referee): spell a parameter's multiplicity, one being unstated no longer A bare parameter now takes [0..*], so the fUML translation spells the multiplicity it means: [1] on in and inout parameters, pins and the context, and an out parameter's upper bound with a 0 lower and an empty default, since a performance's output holds nothing until a token arrives. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(python): restore develop's add_member positional order Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(solve): refuse failed defaults whose cause is a missing dependency Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): resolve quoted, global and library names in a legacy trigger sameNames split the clause with strings.Fields, cutting a quoted name with a space in two; it resolved $::T::Sig as a scoped name; and it looked names up only among the graph's elements, so a standard library type refused its short spelling. Split outside quotes, resolve a global name from the root, read the structure's first-pass names as fully qualified, look library names up in the catalog, and let a name only an import brings into scope agree when it ends the qualified name the other reaches. Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(ci): re-run checks Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): keep a looked-up trigger name in its written form declaredName returned the canonical NUL-joined key, which nameSuffix then re-read as a ::-delimited name. Return the element's qualified name as written and canonicalize only for the equality test. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(runtime): a write through a chain step holding one value reaches that object Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(fmi): spell a scalar FMU parameter's multiplicity, one being unstated no longer The FMU-to-notation converter wrote every scalar in, out and return parameter bare, which now reads [0..*], so a scalar output held a sequence and the tool:fmi engine rejected its own result. An array parameter keeps its spelled [n] nonunique. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs: regenerate documentation counts after merging the parent branch Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(diff): restate the examples digest after merging the parent branch Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): count a literal send's scalar type as the signal it sends The runtime types a sent literal by its scalar type; the lint now names it through the same classification, shared from semantics. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(semantics): judge recorded and redefined ports for port-type-mismatch A port's type is read from its record as from its declaration, conjugation included, and a port redefining another takes that one's type. A recorded interface definition is recognized by its definition kind. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): count computed scalar sends in undeclared-signal Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(check): count the signal type a send's message carries at runtime The undeclared-signal gatherer now reads the written message before the body payload, counts a feature's value type rather than the feature's name, and follows aliases to the definition they reach. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): order toolkit roots by the root namespace and count root relationships Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): let a partially indexed root namespace keep its member order Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): list the root namespace's members in decoded root order Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(migrate): keep a v1 parameter's declared multiplicity when redeclaring, binding or ordering it Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(migrate): keep a parameter's multiplicity and collection modifiers where it is redeclared or bound An operation's context body and a reception's parameter bindings wrote every parameter as [1], so an optional or plural v1 parameter became required and single in v2, and a reception binding a plural signal attribute was refused even though bindingMismatch accepted it. Both now write the parameter's own multiplicity through parameterShape, ordered/nonunique included, as do behavior parameters and activity pins. The implicit-singleton fallback for a v1 parameter that writes no multiplicity now applies before the collection modifiers: an ordered singleton wrote a bare ordered, which v2 reads as [0..*], and now writes [1] ordered. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): read a message's ends from standalone parameter memberships messageEnds only followed the flow's own sysx:ownedMembership and sysx:ownedRelationship links, so a graph whose ParameterMemberships name the flow from their side alone, through membershipOwningNamespace or owningRelatedElement, gave the message no ends and it could not be written back. The decoder now indexes those memberships by the namespace they claim and includes their event-occurrence members, in subject order, after the flow's own links. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): order a message's ends by memberIndex or source and target, refuse when neither can A message's ends come from two passes — the ParameterMemberships the flow links and the standalone memberships naming it back — so when a graph links only some ends from the flow, the standalone ends followed the linked ones and the message read back reversed. The ends now order by sysx:memberIndex or the flow's sysml:sourceFeature/sysml:targetFeature; a partly linked message neither can order is refused, as is one whose two orderings disagree. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): order a message's ends by memberIndex only when every index is well-formed and distinct A present sysx:memberIndex parsed through a lenient reader counted as an index even when malformed, and ends tied at one index kept discovery order — flow-linked first — so a partly linked message read back reversed. byIndex now orders the ends only when every membership, or else its end, states a well-formed and distinct integer; otherwise the source/target fallback and the partly linked refusal apply as before. Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(export): state the malformed memberIndex alone on its membership Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(runtime): pin this in a part-owned action to the owning part Co-Authored-By: jason.han <hanhuijun@gmail.com> * Fix formatting in then_after_members.golden.ttl * Fix missing newline at end of rdf_out.go Add newline at the end of rdf_out.go file * Fix missing newline at end of export_test.go Add missing newline at the end of the file. * Fix missing newline at end of names.go Add newline at the end of names.go file * Fix missing newline at end of rdf_out.go Add a newline at the end of the file. * Fix formatting in payload_declaration_bodies.golden.ttl * fix(export): an end-free flow writes its payload; refuse a payload with a body Review: 'message m of T;' states no ends, so its head never wrote the payload the body no longer holds, and it came back as 'flow m;'. A flow with no ends writes 'of <payload>' after its declaration (FlowDeclaration, MessageDeclaration). A PayloadFeature with members or a body is refused, since the 'of' clause has no place for one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Update shadowing_parameters.golden.ttl with source info Added source line and column information for shadowing parameters. * Add source line and column metadata to state machine * fix(export): a source range ends at the body the parser attached Review: a body comment was told from trivia by the token before it, so 'doc // note' with its body on the next line ended the range at 'doc'. The encoder records each body span the parser attached (Documentation, Comment, TextualRepresentation) and treats every other /* */ comment as trivia. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(solve): model short-circuit reachability of unreadable reads in holds/satisfiable The translator tracks the evaluator's left-to-right, short-circuit evaluation path (and/&/implies reach the right operand where the left holds, or/| where it fails, xor and not always, a conditional its selected branch, an element's conditions in order) as Var.Reached. UnfixedRead refuses a failed default read on every path, guards one reached only under a condition out of the query's models (Query.Unreadable, left out of Query.Free), and Query.Reached asks whether any assignment reaches it, so holds/unsatisfiable are claimed only where the evaluator answers on every assignment. Resolves #719 Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(runtime): memoize the parameter range, chain-host and chain-declaring checks a derivation repeats EffectiveParameterRange walked the redefinition chain on every calc parameter binding; shapeOf re-derived isChainHost from the relationships of every feature value on every shared-default check; and pendingNestedRedefinitions built its name and seen maps for objects whose types declare no chain at all. Each is memoized where its answer is fixed by the model, and takeShared/shareDerived intern the shape once. Co-Authored-By: jason.han <hanhuijun@gmail.com> * Bring the architecture self-model up to date with develop Model the Julia and MATLAB clients, the Cameo and SysON integrations, the HTML document backend, the SysML v1 migrator and calc code generator, the PSSM, fUML, validation-census and conformance-suite oracles, the sweep-run budget, the language server's full capability set and every source-edit operation; point the gates at the test packages that hold them; extend the self-model test to check the new language-server and edit-operation claims; and correct the README's stale figures. Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(grpc): lift the held-objects bound for the warm Instantiate benchmark The benchmark holds every object it makes, so with the default bound it fails past 10 000 objects and never reports a figure. Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(runtime): skip aliasing redefined feature values for a type with no redefinition group The two name maps were built on every materialization; a type whose features share no name has nothing to alias. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(examples): refresh the example walkthroughs and transcripts for 0.9.1 Replay every example command against the current binary and bring the transcripts back in line: the standing lines every analysis prints, the probability column of -schedule explore, the sysml: property names of OSLC query rows, the current RDF experimental note and byte counts, the %features form of event occurrences and library features, and the %check assignment listing. Fix the two commands that no longer ran as written: the disposal-robot state machine is exhibited by RobotController, so -state names the controller instance; and disposal-team's usableReach and teamEndurance returned duplicate values into unique collections, so their parameters are declared nonunique. Correct the semantic-layer README, whose feature names, section line numbers and REPL invocation had drifted from demo.sysml, and drop its dead PR link. Record that the Apollo 11 model now validates with two warnings, not four, since a parameter written without a multiplicity takes its effective [0..*] and is not reported unbound. List the semantic-layer walkthrough and the loose guide models in examples/README. Co-Authored-By: jason.han <hanhuijun@gmail.com> * Resolve SonarCloud code smells outside cognitive complexity Constants and small helpers for repeated literals, renamed shadowed predeclared identifiers, parameter structs for wide signatures, dropped unnecessary if-scoped declarations, split composite assertions and hoisted the tested calls out of assertThrows/pytest.raises lambdas. Pins setup-julia to a commit, adds the default case of race-shard.sh, and names the sponsor link by its visible label. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(examples): give the disposal-team verification case its verdict CheckPayload declared its verdict and never assigned it, so -requirement TeamCases::payloadHolds -satisfy reported the verification as an error with no value. The verdict is now PassIf over the payload comparison the requirement states. The README's validate transcript shows the two conforming-types warnings the sum over MassValue raises, as the runtime showcase does for the same pattern, and the calculation and satisfaction transcripts carry their standing lines. Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(ci): rerun checks after the static job hit its time limit Co-Authored-By: jason.han <hanhuijun@gmail.com> * Keep the sponsor pill's height with the icon wrapped Co-Authored-By: jason.han <hanhuijun@gmail.com> * Re-record the pilot differential baseline's examples digest The self-model files under examples/ changed, so the provenance digest of the examples root moved. Every verdict is unchanged: a fresh run agrees with the committed baseline entry for entry (380 files, 345 fully agreeing, 38 agreed, 41 only ours, 1614 only the pilot's). Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor(export): split toolkit graph normalization into per-pass methods Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor(export): split the remaining high-complexity decoder and encoder functions Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(diff): restate the examples digest for the corrected disposal-team model Only the provenance digest of examples/ moves; the differential itself is unchanged (380 files, 345 fully agreeing, 38 agreed, 41 only ours). Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor(migrate): split high-complexity view, context, deferral and classification functions Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor: split symbol stream reader, FMI notation writer and fUML closure walks Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): payload identity, ownership, ordering; legacy-end payloads Review: - a flow with legacy relatedFeature ends and a separate sysx:payload kept its ends and lost its payload; both shapes are read, disagreeing ones refused; - a payload's declared id is annotated 'about' it in the flow's body, as a cross feature's is, so it survives the graph alone; - only a payload owned through a FeatureMembership is written after 'of'; - 'ordered' and 'nonunique' on a declared payload are parsed (they belong to its multiplicity part), exported and written back; an unnamed payload stating one is refused. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * refactor(analysis): split record shape settling, CSV/regex reply readers and dry-run preview Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): a transition's trigger action carries its source range Review: the accept action a trigger declares is written without going through head, so it had no position. It is placed where the trigger is written; the goldens gain only those properties. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * refactor(runtime): split pin binding, invocation inputs, nested redefinitions, weights and tool calls Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor(edit): split add-member checks, trivia scan, layout bindings and statement formatting Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor: split remaining complex Go functions across passes, semantics, IR, REPL, parser and CLI Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(solve): guard a chain's declared value that does not evaluate instead of leaving it free ChainPins skipped every chain read failing with ErrNoValue, so a nested default reading a feature with no value (inner.dependent = input + 1.0) stayed free and the solver could witness a value for it; any other failure refused the question even where the evaluator never reaches the read. It now tells a feature holding nothing (still free) from a declared value that does not evaluate — the runtime marks the latter ErrFeatureValueMaterialization — and returns the latter as Unfixed naming the exact chain variable, which UnfixedRead refuses or guards by Var.Reached as it does a direct read. translatedQuestion applies the chain reads' guards, and reapplies every guard when chain pins require translating again. The differential harness classifies a chain's failed declared value as unreadable and its randomized generator declares one on a nested part read through a chain. Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor: split edit-request building in the Python and Java clients Co-Authored-By: jason.han <hanhuijun@gmail.com> * Match release tags on ASCII digits only Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(runtime): bind a calc's this hook once per occurrence and statement engine Every statement evaluation and binding env bound host.materializeOccurrence afresh, an allocation per step of a loop body; the bound hook now lives on the engine, the occurrence and the pooled invocation frame it belongs to. Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(semantics): memoize PortFeatures by port type The port-type-mismatch lint asks for the features of the same port types once per connection end; the answer is now journaled on the model as conjugatedSupertypes already is. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(solve): constrain Query.Reached only by the guards the evaluator checks before the read A definedness guard hoisted from an operation the evaluator performs only after reaching a failed default — a later required condition's divisor — could make Query.Reached unsat and certify a proof on a question that is unsat for its own reasons, though assignments reaching the read exist and the evaluator gives no verdict there. Each conditional read now records the guards asserted before its first reference (Var.Preceding); Reached requires them with the read's condition and asserts no other definedness guard. Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(runtime): memoize the chain-host check on the model, not the effective feature The self-model counts runtime.EffectiveFeature's fields; a private cache field does not belong in that shape. Key the isChainHost memo by feature symbol on the runtime model beside declaresChains instead. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci: rerun after a Go module proxy stream error in the static checks Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(export): untyped named payloads, flags before the typing, duplicate names Review: - a named payload with no typing was written 'of cmd', which reads back as a payload typed by cmd; a named payload needs its one typing and is refused otherwise; - 'of p[1] ordered : T' failed the declaration lookahead, which now skips the multiplicity part's ordered/nonunique before the colon; - a payload named like a body member is identified by its position, as any member whose name a sibling took first is. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs(project): record the 0.9.1 candidate's performance against 0.9.0 Every package benchmark on both revisions, whole-binary scaling on generated models and Apollo 11, and each regression fixed or priced with the commit that introduced it. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(changelog): performance entry for the 0.9.1 fixes Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci: rerun after the rest race shard stalled on the runner Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(performance): re-time the Apollo 11 validation on the 0.9.1 candidate Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(cli): refuse -o naming one of a model's several files The graph of a model of several files could be written over one of them; each file, and a link to one, is now refused as the output path as the migration's paths are. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(changelog): own fragment for the -o overwrite refusal Co-Authored-By: jason.han <hanhuijun@gmail.com> * feat(python): wrap ParseSources to parse several documents as one model Connection.parse_sources and opensysml.parse_sources take paths, (name, text) pairs and SourceDocuments, require the parse_sources capability (and inline_language/strict_conformance when asked for), and return the Model that load returns, with one root per document in Model.roots and their names in Model.documents. Model.find, get and the name walk search every root. The response's error field raises ModelError, as a strict load does. Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(export): the payload features carry source ranges develop's source ranges (#654) place every declared element, the payload features this branch adds included; the golden gains only those lines. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(client-node): rename the package to @openmbee/opensysml Derives the published name and the platform-package prefix from a single src/core/package.ts, so binary.ts, capabilities.ts, the platform-package generator and the tests all read the same source. Version follows client/python/opensysml/_version.py in lockstep. Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(release): check the Node client's version in lockstep Adds check_version.py --node, which verifies client/node/package.json against _version.py and the tag, and a ci-changed-areas rule so a PR touching only that file still runs the Python suite that gates it. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): publish the Node client to npm from the core tag publish-npm now runs in the release workflow on the v* tag, at the same version and carrying build-release's binaries — replacing the client-node-v* path, which never ran. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): tidy the npm publish job and its verification snippet Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(edit): resolve an ApplyEdits batch's references against the whole batch A succession reference or metadata prefix an operation writes is now checked once against the model the batch leaves, so it may name a member or metadata definition a later operation of the same batch declares. Names are still taken in operation order, and a `first x` label no longer takes or makes visible the name it borrows. Co-Authored-By: jason.han <hanhuijun@gmail.com> * feat(parser): accept a target-end multiplicity on action-body then successions The SysML v2 grammar lets the target a `then` references carry a crossing multiplicity (SysML.xtext TargetSuccession -> ConnectorEndMember): `then [m] x;`, `then [m] x { ... }`, `then [m] done;` and `[m] then [n] x;`. The pilot accepts them; the parser rejected them with "a multiplicity in an action body must precede then". The AST records the target multiplicity beside the source one, semantics carry it on the target end, RDF export writes it on the target connector end and reads every form back without source text, and the end-feature-multiplicity warning stays limited to source ends. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(python): refuse a bare document and non-str fields in parse_sources A string passed where the sequence of documents goes was iterated as its characters, so the complaint was about two documents named 's'; it is now refused as one document outside a sequence. SourceDocument fields of a type other than str raise TypeError up front instead of the protobuf's opaque one. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(skills): probes for parse_sources across documents Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(cli): check -o against every model file before reading any Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(python): generate a typed module from every document of a model generate_source collected definitions from model.root alone, so a model parsed with parse_sources produced a module of its first document only, and a type or redefinition crossing documents could not resolve. collect_definitions now takes every root and builds one facts map. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(python): say the typed-class command line takes one source file A model of several documents is generated from Python with generate_source; the command line still reads one file. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(parser): reference a declared node-word member after a bracketed then multiplicity namesEdgeEnd read the tokens at fixed offsets from `then`, so `then [m] fork;` with `action fork;` declared was taken as a new fork node with a source multiplicity. The lookahead is now relative to the offset after the multiplicity and prefix metadata, and a declared name followed by a body is an end as well. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(parser): keep a node keyword with a body a node declaration after then A control-node keyword followed by `{` declares an anonymous node with that body (SysML.xtext ForkNode, MergeNode) regardless of a member sharing the keyword's name; only `then <kw>;` and `then <kw> <name>;` are read as references to a declared member. Co-Authored-By: jason.han <hanhuijun@gmail.com> * feat(client-java): lock the version to the core and auto-publish to Central The pom and its consumers follow _version.py in lockstep, checked by check_version.py --java and a pytest gate that also runs on a manifest-only change; the release profile publishes the validated deployment itself and waits until it is on Central. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): publish the Java client to Maven Central from the core tag publish-maven runs in the release workflow on the v* tag, beside publish-pypi and publish-npm, signing and uploading at the core version from the restricted maven-central context. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(release): document the Maven Central publish Co-Authored-By: jason.han <hanhuijun@gmail.com> * test(release): import ElementTree at module level; tidy the context wording Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): read the Maven Central credentials from project variables Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(edit): follow a batch's deferred references through later renames A succession end or metadata prefix an operation writes is now anchored to the byte it starts at in the edited document and moved past the later operations' splices, so the deferred check reads the name the batch leaves there rather than the one the operation was given: a later rename of the metadata definition, of the prefixed declaration or of the sequenced node no longer refuses a valid prefix or skips the duplicate-prefix check. ActionNodeOfBody and FeatureSymbolInScope look past a `first g;` label to the inherited member of that name, so the label no longer hides it. Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(edit): judge a deferred succession end under its `$::` root The end read back from the final AST is passed to the visibility check, and reported, with the global root it was written with, so a `$::`-rooted end resolves through the index rather than the edited document's scopes. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): warn against debug output on the Maven publish step Co-Authored-By: jason.han <hanhuijun@gmail.com> * feat(client-rust): lock the crate version to the core Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): publish the Rust client to crates.io from the core tag Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(release): document the crates.io publish Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(changed-areas): run the lockstep tests when an editor's client reference changes Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(release): drop --offline from the Cargo.lock refresh Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(client-rust): accept a literal-string crate version Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(parser): `then fork F;` declares the node beside a member named fork namesEdgeEnd read a control-node keyword followed by a name as a two-ended succession to the member sharing the keyword's name. The grammar gives the keyword its own declaration (ForkNode, JoinNode, MergeNode, DecisionNode: 'fork' UsageDeclaration? ActionBody), so only the bare `then <kw>;` beside such a member is an edge end. Co-Authored-By: jason.han <hanhuijun@gmail.com> * refactor: resolve the SonarCloud findings left on develop after the cleanup Co-Authored-By: jason.han <hanhuijun@gmail.com> * perf(runtime): attribute and fix the 0.9.1 satisfaction, gRPC, migration and lint regressions Shared-default and shared-verdict tracing recorded every path of a nested shared value per read, deduplicated reads through a key string built per read, spelled a dotted path at every ancestor to ask whether a binding governs it, and every journal mark cloned the clock's waiter list. The trace now records a nested value once, compares paths in place, prefilters the binding question by the features a type's bindings start at (memoized on Model.bindingRoots), and the clock replaces its waiter list instead of editing it so a mark keeps the slice it saw. The migration writer reuses the buffers of closed blocks and indents from a table. The nested-usage subsetting memoizes declaredSubsettedNames on the runtime model and makes its deduplication and reachability sets on first use. The undeclared-signal lint walks the document once per analysis through kit.ScopedNodes instead of twice. The performance record's findings 5–8 carry the attribution, the six-run benchstat tables against v0.9.0 and the tree before this change, the profile frames, and what remains as the price of a rule. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): read the registry credentials from the org contexts Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): read the Maven Central credentials from its context Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): add a no-publish release rehearsal One pipeline parameter, release_rehearsal, runs the release workflow on a branch with every check live and the irreversible commands - cosign signing, ghr, twine upload, npm publish, mvn deploy, cargo publish - skipped. A rehearsal exports a stand-in CIRCLE_TAG translated from _version.py (PEP 440 to SemVer), and rehearsals probe the GitHub, npm and Central tokens and build and sign the Maven artifacts. The tag path is unchanged: the jobs drop only the branches-ignore filter, which the workflow-level when now replaces. Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(release): state the rehearsal's limits without overclaiming Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(editors): lock the editor versions to the core version The editors carry the core version the clients already follow: 0.9.0 in the two package.json files (locks regenerated, only the version lines move), the Cameo and SysON poms, and the child poms' <parent><version>. Nothing publishes the editors, but check_version.py --editors now reads every editor manifest - package.json, package-lock.json (both version copies must agree), pom and child <parent><version> - through the shared lockstep check, so a release tag fails early when one disagrees. The python changed-area trigger covers every editor manifest, build-python-package runs --editors before anything is built, and releasing.md's bump step lists the files. Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release): don't print Central's probe response in the rehearsal Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(release): set the client and editor versions to 0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(release): fold the late fixes into the 0.9.1 changelog Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(release): point the Java and editor READMEs at 0.9.1 Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(java): keep the parent's SCM URLs verbatim in the published client pom Co-Authored-By: jason.han <hanhuijun@gmail.com> * chore(java): publish the Java client as org.openmbee:opensysml Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(ci): accept a base64-encoded GPG signing key in publish-maven Co-Authored-By: jason.han <hanhuijun@gmail.com> * docs(readme): point the Apollo 11 defects at the showcase instead of narrating them twice Co-Authored-By: jason.han <hanhuijun@gmail.com> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <hanhuijun@gmail.com> Co-authored-by: Somesh Kashyap <someshsandbox@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Jason Han <jason.han@jpl.nasa.gov>
…uch paths Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
…ackage Co-Authored-By: jason.han <hanhuijun@gmail.com>
…looked Co-Authored-By: jason.han <hanhuijun@gmail.com>
* fix(docrender): rename the Windows-illegal image fixture and reject such paths Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release-windows): let the workflow re-run against an existing tag Co-Authored-By: jason.han <hanhuijun@gmail.com> * ci(release-windows): sign tagged dispatches like a tag push Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(client-node): lock the published sysml-grpc optional dependencies Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(client-node): let the binary-resolution tests hide the platform package Co-Authored-By: jason.han <hanhuijun@gmail.com> * fix(client-node): leave a skipped platform package out of the places looked Co-Authored-By: jason.han <hanhuijun@gmail.com> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <hanhuijun@gmail.com>
Co-Authored-By: jason.han <hanhuijun@gmail.com>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <hanhuijun@gmail.com>
Contributor
Author
|
I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".
|
HuiJun
marked this pull request as ready for review
September 30, 2026 20:37
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Co-Authored-By: jason.han <hanhuijun@gmail.com>
4 of 6 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What and why
Back-merge of
mainafter the 0.9.1 release (#614) and its two hotfixes (#752, #753), sodevelopcarries:0.9.1CHANGELOG.mdsection and all versions at 0.9.1 (the fragments folded into it are removed);tests/hygienetracked-path check;release-windows.ymltag dispatch and no-checkout MSI build, with thereleasing.mdrecovery notes.Plain
git merge origin/mainintoorigin/develop, no conflicts, no other changes. Only the two hotfix changelog fragments remain unreleased.How it was verified
python3 scripts/changelog.py check,gofmt -l .(empty),go build ./...,go test ./tests/hygiene/... ./internal/doc/docrender/...locally; full suite in CI.Checklist
make testandmake lintpass locally (targeted; full in CI)changes/unreleased/<slug>.<section>.md, not as an edit toCHANGELOG.md(release fold carried from main)make docs-countsrun if a gate count moved (n/a)F4,K5) in the body, docs, or changelogLink to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/e0c5a204e0984415814ed38767578b41
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/e0c5a204e0984415814ed38767578b41?variant=devin
Requested by: @HuiJun