Sync from development - #48
Merged
Merged
Conversation
The Python client starts the service itself, but the wheel carries no Go binary and the release published none, so a user without a Go toolchain could not use the default path. Releases now publish raw sysml-grpc-<os>-<arch> files with .sha256 sidecars, which is what download_binary already expected, and its 'latest' path resolves the tag instead of raising NotImplementedError. Instantiate over gRPC also read raw slots, so a derived attribute came back unmaterialized where the REPL evaluates it; it now reads through Instance.GetSlot. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…nd-next-steps The Python client starts the service itself, but the wheel carries no Go binary and the release published none, so a user without a Go toolchain could not use the default path. Releases now publish raw sysml-grpc-<os>-<arch> files with .sha256 sidecars, which is what download_binary already expected, and its 'latest' path resolves the tag instead of raising NotImplementedError. Instantiate over gRPC also read raw slots, so a derived attribute came back unmaterialized where the REPL evaluates it; it now reads through Instance.GetSlot. Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
The roadmap now lives with the rest of the docs and is linked from docs/README.md; its contents are also captured as repo-pinned knowledge notes. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
The roadmap now lives with the rest of the docs and is linked from docs/README.md; its contents are also captured as repo-pinned knowledge notes. Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
The graph carries the SysML vocabulary and element IRIs the Flexo MMS SysML v2 service uses, so a converted model loads into that triplestore. Element IRIs are the qualified name percent-escaped, which makes them deterministic across conversions and reversible. The Turtle layer is hand-written rather than a new dependency: the subset needed here is small, and the parser rejects blank nodes, collections and literal shorthands with a line number instead of accepting them and dropping data. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
`ref part a : V`, `composite part b : V` and `derived attribute c` parsed with their modifiers consumed but never transferred, so the declaration came back with isReference, isComposite and isDerived all false. parseBodyMember had no branch for a modifier followed by a kind keyword, so those forms fell through to the anonymous-feature path. Parse the declaration and apply the modifiers already consumed, without clearing a flag the declaration parsed for itself. The flow_payload_declaration golden had locked in the wrong behavior: its two `ref part` members recorded ref=false, and now record ref=true. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
Each element carries its metaclass as rdf:type and its declaration as SysML metamodel properties. The kind tables are declared once and the reverse maps derived from them, so the two directions cannot drift. Three properties live in a separate urn:systemica:sysml: namespace because the metamodel does not define them and RDF cannot express them: memberIndex (declaration order, which the notation is sensitive to and a triple set is not), hasBody, and sourceText. Saving to notation writes the original source through the formatter rather than re-printing the graph, so comments and notes survive; only the Turtle direction goes through the mapping. Unsupported input is an error naming the element or line rather than a partial conversion: a missing rdf:type, an owner absent from the graph, a metaclass outside the mapping, or a syntax error on either side. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
The format follows the file extension. An unrecognized extension is an error rather than a guess, and an empty session writes no file. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
-convert takes the input path as its flag value rather than a positional argument, because Go's flag package stops parsing at the first positional and `-convert in.sysml -o out.ttl` would otherwise leave the output flags unparsed. Formats come from the file extensions, with -from/-to to name them when an extension is missing or unrecognized. The command writes nothing and exits non-zero on any input it cannot convert faithfully. convert_test.go drives the built binary as a process, which cmd/ had no coverage for previously. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
RDF_INTEROP.md covers the vocabulary, the deterministic element IRIs, the round-trip guarantees and every construct the mapping does not represent. SPEC_COMPLIANCE.md gains the capability map with honest status flags, and ROADMAP.md records the three items this work deliberately left open. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
Written files get 0600, matching the only other non-test WriteFile in the repo (internal/core/libs/cache.go). Reading the input the user named on the command line carries a documented #nosec G304, as the equivalent reads in internal/grpc and internal/core/model already do. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
End-to-end testing found the one silent loss in the feature: `//` and `/* */` trivia belongs to no element, so it does not survive notation -> RDF -> notation, while a direct .sysml save keeps it. The `doc` and `comment` keywords are declarations rather than trivia and do convert both ways. TestCommentsThroughRDF pins both halves of that, so the behavior cannot change without the documentation changing with it. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
The synthetic parameter of `action X accept p : T` took its span from the start of the enclosing member, so its source text was a copy of its own parent declaration, minus the `;` that had not been consumed yet. Nothing read that span before, which is why it went unnoticed; the RDF conversion does, and re-emitted the parent nested inside itself. The span now covers just the parameter. No existing golden changed. accept_action_shorthand covers both the plain and `via` forms. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…strict Homebrew scans the version from the release tag in the URL, so the explicit version line made 'brew audit --strict --online' fail as redundant. Drop it from the template and the render script, and document the tap repository naming rule (brew tap x/y only resolves x/homebrew-y). Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
Two message-clarity issues found in end-to-end testing of the refusal paths, both misleading rather than wrong. - `%save x.ttl` reported the position as `x.ttl:1:59` — the output file, which does not exist yet and is not where the position points. The session buffer has no path, so it is now named `<session>`. - The `then` error read as though the keyword were at the position given, but the flag hangs off the member on one side of it. It now says which member it is beside, by name. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…ew 6.x Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(packaging): make the rendered Homebrew formula pass brew audit --strict Homebrew scans the version from the release tag in the URL, so the explicit version line made 'brew audit --strict --online' fail as redundant. Drop it from the template and the render script, and document the tap repository naming rule (brew tap x/y only resolves x/homebrew-y). Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * docs(packaging): note that hand-cloned taps need brew trust on Homebrew 6.x Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
… tap trust Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Seven bugs found in review of the save/convert feature: - an unnamed `comment /* x */` lost its keyword and came back as trivia - a long Turtle literal ending in `"` emitted four quotes in a row - `derived var feature x : T;` produced an empty membership and dropped a leading visibility keyword, which the bundled KerML.kerml relies on - graph reads rescanned every triple, making conversion quadratic - prefix selection depended on Go map iteration order - `chain` was dropped when writing a usage back from a graph - the `then` refusal was only reached for usages, so a succession on a definition or package member converted anyway Kind keyword synonyms are now carried as `sysx:declaredKeyword` rather than normalized to the canonical keyword, and a synonym in a shape the graph cannot rebuild is refused instead. The metaclass names this mapping invents move to the `sysx:` namespace, and notation-to-notation conversion checks syntax like every other direction. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* feat(rdf): add a triple model with a Turtle writer and parser The graph carries the SysML vocabulary and element IRIs the Flexo MMS SysML v2 service uses, so a converted model loads into that triplestore. Element IRIs are the qualified name percent-escaped, which makes them deterministic across conversions and reversible. The Turtle layer is hand-written rather than a new dependency: the subset needed here is small, and the parser rejects blank nodes, collections and literal shorthands with a line number instead of accepting them and dropping data. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(parser): keep feature modifiers written before a kind keyword `ref part a : V`, `composite part b : V` and `derived attribute c` parsed with their modifiers consumed but never transferred, so the declaration came back with isReference, isComposite and isDerived all false. parseBodyMember had no branch for a modifier followed by a kind keyword, so those forms fell through to the anonymous-feature path. Parse the declaration and apply the modifiers already consumed, without clearing a flag the declaration parsed for itself. The flow_payload_declaration golden had locked in the wrong behavior: its two `ref part` members recorded ref=false, and now record ref=true. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * feat(export): convert between SysML notation and RDF Turtle Each element carries its metaclass as rdf:type and its declaration as SysML metamodel properties. The kind tables are declared once and the reverse maps derived from them, so the two directions cannot drift. Three properties live in a separate urn:systemica:sysml: namespace because the metamodel does not define them and RDF cannot express them: memberIndex (declaration order, which the notation is sensitive to and a triple set is not), hasBody, and sourceText. Saving to notation writes the original source through the formatter rather than re-printing the graph, so comments and notes survive; only the Turtle direction goes through the mapping. Unsupported input is an error naming the element or line rather than a partial conversion: a missing rdf:type, an owner absent from the graph, a metaclass outside the mapping, or a syntax error on either side. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * feat(repl): add %save to write the session as notation or RDF The format follows the file extension. An unrecognized extension is an error rather than a guess, and an empty session writes no file. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * feat(cli): add -convert for SysML/RDF conversion -convert takes the input path as its flag value rather than a positional argument, because Go's flag package stops parsing at the first positional and `-convert in.sysml -o out.ttl` would otherwise leave the output flags unparsed. Formats come from the file extensions, with -from/-to to name them when an extension is missing or unrecognized. The command writes nothing and exits non-zero on any input it cannot convert faithfully. convert_test.go drives the built binary as a process, which cmd/ had no coverage for previously. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * docs: document saving, the RDF mapping and its limitations RDF_INTEROP.md covers the vocabulary, the deterministic element IRIs, the round-trip guarantees and every construct the mapping does not represent. SPEC_COMPLIANCE.md gains the capability map with honest status flags, and ROADMAP.md records the three items this work deliberately left open. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(export): satisfy gosec on the new save and convert paths Written files get 0600, matching the only other non-test WriteFile in the repo (internal/core/libs/cache.go). Reading the input the user named on the command line carries a documented #nosec G304, as the equivalent reads in internal/grpc and internal/core/model already do. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * docs(export): note that the RDF hop drops lexical comments End-to-end testing found the one silent loss in the feature: `//` and `/* */` trivia belongs to no element, so it does not survive notation -> RDF -> notation, while a direct .sysml save keeps it. The `doc` and `comment` keywords are declarations rather than trivia and do convert both ways. TestCommentsThroughRDF pins both halves of that, so the behavior cannot change without the documentation changing with it. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(parser): scope the accept parameter's span to the parameter The synthetic parameter of `action X accept p : T` took its span from the start of the enclosing member, so its source text was a copy of its own parent declaration, minus the `;` that had not been consumed yet. Nothing read that span before, which is why it went unnoticed; the RDF conversion does, and re-emitted the parent nested inside itself. The span now covers just the parameter. No existing golden changed. accept_action_shorthand covers both the plain and `via` forms. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(repl,export): make the refusal messages name the right place Two message-clarity issues found in end-to-end testing of the refusal paths, both misleading rather than wrong. - `%save x.ttl` reported the position as `x.ttl:1:59` — the output file, which does not exist yet and is not where the position points. The session buffer has no path, so it is now named `<session>`. - The `then` error read as though the keyword were at the position given, but the flag hangs off the member on one side of it. It now says which member it is beside, by name. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(export): address review findings on the RDF conversion Seven bugs found in review of the save/convert feature: - an unnamed `comment /* x */` lost its keyword and came back as trivia - a long Turtle literal ending in `"` emitted four quotes in a row - `derived var feature x : T;` produced an empty membership and dropped a leading visibility keyword, which the bundled KerML.kerml relies on - graph reads rescanned every triple, making conversion quadratic - prefix selection depended on Go map iteration order - `chain` was dropped when writing a usage back from a graph - the `then` refusal was only reached for usages, so a succession on a definition or package member converted anyway Kind keyword synonyms are now carried as `sysx:declaredKeyword` rather than normalized to the canonical keyword, and a synonym in a shape the graph cannot rebuild is refused instead. The metaclass names this mapping invents move to the `sysx:` namespace, and notation-to-notation conversion checks syntax like every other direction. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
…rce text A comment inside a declaration head is part of the span the head was read from, so scanning that text for the kind keyword let a word inside the comment win: `attribute // the flow rate` recorded `flow` and came back from RDF as a FlowUsage. The keyword is now threaded from the token that was consumed. Also: a usage whose head is kept verbatim no longer has its synonym keyword refused (`verify R;` converted with an error before), and Turtle `PREFIX`/`BASE` only count as keywords when a whole word, so a prefixed name such as `base:Thing` is read as a term rather than rejected. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
- %save to notation warns and writes when the session does not fully parse - indent continuation lines one level deeper than their statement - fold a leading comment into the declaration it documents - expand ~, name a missing parent directory, advise per surface - state an overwrite, and write atomically at 0644
…e name A SatisfyRequirementUsage's `by` operand names the subject of the satisfaction. Taking it as the usage's own name declared a member that shadowed the real subject in the enclosing scope, so sibling references to that subject reported bogus subsets/redefines type errors. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…embers The inline (unbraced) action body loop kept consuming behavioral statements, so members following the node - the enclosing body's final node and its succession edges - became members of that node and the enclosing action lost its successors. The body is now one statement plus its 'then'-chained continuations only. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
while/loop-until/for and if/else parsed, resolved and typechecked, then were dropped at lowering, so a model that read correctly produced a wrong answer with no diagnostic. Lower them as body statements and execute them recursively, with a frame per body-local block and a step per iteration.
…ation The function library declares its numeric functions abstractly, with a signature and no body, so a call had nothing to invoke and `**` reached the runtime as an unsupported operator. A model could not compute a square root or a sine. Dispatch is by the declaration's qualified name, in the one calc invocation path, so tracing and diagnostics are unchanged and a model's own `calc sqrt` — which has a body — is evaluated from it. An unqualified call that resolves to nothing dispatches by local name, which is what makes `sysml -e "sqrt(2.0)"` work in a model that imports no part of the library. One shared `semantics.Pow` serves the constant folder and the runtime, so a folded and an evaluated exponentiation agree. Integer operands with a non-negative exponent give an Integer; anything else gives a Real. A domain failure, an overflow or a non-finite result is declined by the folder and reported by the runtime rather than becoming a NaN, an infinity or a wrapped integer. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…longs to the enclosing body Continuing the unbraced body on any 'then' still stole succession edges written directly after the node: 'then start bump;' names members of the enclosing body, so only the inline statement forms parseSuccessionEdge recognises continue the body. The enclosing mixed body also only consumes a 'then' as a chain marker when a keyword-introduced declaration or statement follows it, not a named edge. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
The section this PR added duplicated the parser A/B section already on main, and its copy of the corpus sweep reintroduced the word-splitting loop that compared 27 of 127 models and reported a false zero. Keep main's safe sweep and add only what is new: LSP hover/definition and short-name probes as sweeps 4-5, and the lowering/runtime naming probes as a subsection. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…name Deriving the name from the naming feature moved the single key a node is filed under rather than adding one, so a succession naming a short-labelled step by its label (first d then s;) stopped resolving while the derived name started to. The symbol layer registers both keys; node lookup and breakpoint matching now do too. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* Testing parser and naming changes end-to-end in the sysml REPL Co-authored-by: jason.han <jason.han@jpl.nasa.gov> * docs(skill): build baseline binaries with go build -C instead of cd * docs: correct claim that argv positional handles paths with spaces * docs(skill): fold the naming probes into the existing A/B section The section this PR added duplicated the parser A/B section already on main, and its copy of the corpus sweep reintroduced the word-splitting loop that compared 27 of 127 models and reported a false zero. Keep main's safe sweep and add only what is new: LSP hover/definition and short-name probes as sweeps 4-5, and the lowering/runtime naming probes as a subsection. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
…d negatives AGENTS.md 5.1 asks for a golden fixture and negative cases with any change to the accepted grammar. The golden shows each keyword member dumping as its symbol twin, differing only in the short name. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
A member redefining more than one feature derives no name (KerML 7.3.4.5), so a value on it reaches none of them and binds at best to a declared short name. It stays well-formed, so the constraint tier reports it as a warning next to the other redefinition rules. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…lationship-parity * fix(parser): read a relationship keyword after a short name as a relationship A declared short name is no declared name: KerML derives effectiveName from declaredName alone, so a usage stating only a short name still answers to the feature it references or redefines, and a relationship keyword after that short name states a relationship rather than spelling the name. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(lower): let a step answer to both its short name and its derived name Deriving the name from the naming feature moved the single key a node is filed under rather than adding one, so a succession naming a short-labelled step by its label (first d then s;) stopped resolving while the derived name started to. The symbol layer registers both keys; node lookup and breakpoint matching now do too. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * test(parser): lock the short-name relationship forms with a golden and negatives AGENTS.md 5.1 asks for a golden fixture and negative cases with any change to the accepted grammar. The golden shows each keyword member dumping as its symbol twin, differing only in the short name. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * feat(passes): warn when a value lands on a member no redefinition names A member redefining more than one feature derives no name (KerML 7.3.4.5), so a value on it reaches none of them and binds at best to a declared short name. It stays well-formed, so the constraint tier reports it as a warning next to the other redefinition rules. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
The negative fixture list gained cases on both sides; both are kept. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…reference A `then` before a keyword the body already declares a member with is an edge end (`then end;`, `then third end;`), not the kind of a member being declared, so the modifier and namespace cases no longer claim it first. A usage names a succession end by its effective name, the name lowering resolves it by, so `perform a; then action b;` sequences rather than warns. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
… printer Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
`perform b;` declares an occurrence usage, so a `then` between it and the named member before it is a succession rather than a statement chained into one node's body, where the keyword was dropped without a word. With no named member before it — a block's `assign …; then perform body;` — it still reads as that chain, which the lowered block already runs in order. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…half-named one `then b if x;` is a control flow edge, whose empty source the builder skipped, so lowering could not resolve it while the unguarded `then b;` worked; both node kinds now take the member before them. Reading a graph, a succession naming one end is reported the way the encoder refuses to write one, rather than printed as a bare `succession;` that declares no order. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* feat(parser): desugar a member-attached `then` into a succession edge
A `then` prefixing a body member sequences the members either side of it,
so it is desugared at parse time into the *ast.SuccessionEdge the `then a b;`
notation already built, and ast.Membership's HasSuccession, SuccessionTarget
and SuccessionGuard fields are gone with every site that set them.
Lowering and the RDF mapping already honour that node, so execution now
follows a member-attached `then` rather than member order, and the mapping
emits sysml:SuccessionAsUsage with source/target ends instead of refusing
the conversion.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser): take a `then` before an anonymous member rather than reading the kind keyword as an edge end
`then part;` and `then action { }` declare a member with no name, which an
edge end cannot reference. The keyword was handed to the edge parser, which
read the kind keyword as the target's name; it is now taken as the prefix it
is and warned about like any other unnamed end.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser): sequence from the member before the `then`, and name a short-form state
An unnamed member now clears the succession source, so a `then` after one warns
instead of sequencing from an older named member. A substate declared in the
short form (`state a;`) names its end, so `then state b;` builds a real edge the
state lowering turns into a completion transition.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser): read the succession edge form in every body that can hold one
A calc or requirement body reported the `then a b;` the RDF mapping writes back
as a syntax error, so a converted model no longer parsed; a state body read it as
an unnamed succession usage, which cleared the member a following `then`
sequences from and produced a different graph on a second conversion. A state
body's `then` now builds the same *ast.SuccessionEdge as every other body, a
region carries a member-attached `then` too, and a usage of an edge kind is not
taken as a succession's source.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser): read a keyword an edge end names, and a member named by reference
A `then` before a keyword the body already declares a member with is an edge
end (`then end;`, `then third end;`), not the kind of a member being
declared, so the modifier and namespace cases no longer claim it first. A
usage names a succession end by its effective name, the name lowering
resolves it by, so `perform a; then action b;` sequences rather than warns.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser): sequence a performed step after a named member
`perform b;` declares an occurrence usage, so a `then` between it and the
named member before it is a succession rather than a statement chained into
one node's body, where the keyword was dropped without a word. With no named
member before it — a block's `assign …; then perform body;` — it still reads
as that chain, which the lowered block already runs in order.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
* fix(parser,export): complete a guarded one-name succession, report a half-named one
`then b if x;` is a control flow edge, whose empty source the builder skipped,
so lowering could not resolve it while the unguarded `then b;` worked; both
node kinds now take the member before them. Reading a graph, a succession
naming one end is reported the way the encoder refuses to write one, rather
than printed as a bare `succession;` that declares no order.
Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
…laring scope An expression written inside an action or state machine body was evaluated with no scope at all, so it could see neither the imports nor the enclosing members of the document it was written in: a quantity attribute default reported "unresolved unit m" and an assignment naming a package member reported "unresolved feature". Lowering now carries the declaring scope on what it emits — the graph, each statement and block, each state and each transition — and the executors read it instead of re-deriving one from symbol.Decl. Frames are still pushed over that scope, so a token value or block-local binding keeps shadowing a same-named declaration. Unqualified evaluation resolves as a written reference does, and the declaration it finds is evaluated in its own declaring scope. Closes ROADMAP A3a: the residual failure there was the same defect in the REPL's %constraint/%requirement, which evaluated against the document root. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
A state written directly among a machine's members took the machine's scope, so a member the state declares was invisible to its own entry, do and exit behaviors, and every substate and region below it lost its scope too. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
The test compared two indexes built from the shared user cache, so a parallel test package populating it between the two builds made one index carry qualified names and the other local ones, failing over provenance rather than removal. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
…aluation * fix(runtime): evaluate action and state body expressions in their declaring scope An expression written inside an action or state machine body was evaluated with no scope at all, so it could see neither the imports nor the enclosing members of the document it was written in: a quantity attribute default reported "unresolved unit m" and an assignment naming a package member reported "unresolved feature". Lowering now carries the declaring scope on what it emits — the graph, each statement and block, each state and each transition — and the executors read it instead of re-deriving one from symbol.Decl. Frames are still pushed over that scope, so a token value or block-local binding keeps shadowing a same-named declaration. Unqualified evaluation resolves as a written reference does, and the declaration it finds is evaluated in its own declaring scope. Closes ROADMAP A3a: the residual failure there was the same defect in the REPL's %constraint/%requirement, which evaluated against the document root. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(lower): give a state declared in a machine body its own scope A state written directly among a machine's members took the machine's scope, so a member the state declares was invisible to its own entry, do and exit behaviors, and every substate and region below it lost its scope too. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * test(libs): give the removal test its own library cache The test compared two indexes built from the shared user cache, so a parallel test package populating it between the two builds made one index carry qualified names and the other local ones, failing over provenance rather than removal. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
…ery path A name in the unit position of a quantity expression resolves to the nearest declaration (KerML 8.2.3.5.3/8.2.3.5.4) and the position only checks that it conforms (8.2.3.5.1), so a sibling named like an imported unit shadows it with a diagnostic naming the declaration and the unit it hid. Conditions used to reach past the nearer declaration; they now evaluate in their own body scope, so all four evaluator paths answer one routine. The prompt evaluates in the namespace the session works in, so imported units resolve unqualified; %calc parses its arguments as expressions, so a quantity survives; and a quantity renders as written in a violation and by the Real convention in a result table. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…e it A whitespace-separated argument list is cut into arguments before the expression parser sees it, so `5 -3` is two arguments while `5 - 3` — an expression left unfinished across the space — stays one. Also name the shadowed unit when the shadowing declaration sits in the namespace that declares the import, by resolving the name again with that declaration hidden instead of searching its parent scope. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
An `=` nested in a call, a bracket or a string belongs to that expression, so it no longer refuses `add(x = 1, y = 2)` as an argument; only a bare identifier bound at the argument's top level is a named argument. A qualified name in unit position resolves to what it names, so it is reported without the shadowing explanation \u2014 whose suggestion would otherwise paste the whole written name onto the unit's namespace. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…g it %eval retained a lookup failure so an imported name could still be evaluated, which also swallowed the ambiguity error. A typed AmbiguousNameError separates the two: a name found nowhere falls through to the expression path, a name several declarations answer to is reported. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
… hidden unit A unit owned by no namespace has the same qualified name as its simple one, so the suggestion repeated the name that had just failed. The diagnostic now names the shadowed unit without advising a spelling in that case. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…-resolution * fix(semantics): resolve a unit name by ordinary name resolution on every path A name in the unit position of a quantity expression resolves to the nearest declaration (KerML 8.2.3.5.3/8.2.3.5.4) and the position only checks that it conforms (8.2.3.5.1), so a sibling named like an imported unit shadows it with a diagnostic naming the declaration and the unit it hid. Conditions used to reach past the nearer declaration; they now evaluate in their own body scope, so all four evaluator paths answer one routine. The prompt evaluates in the namespace the session works in, so imported units resolve unqualified; %calc parses its arguments as expressions, so a quantity survives; and a quantity renders as written in a violation and by the Real convention in a result table. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(repl): separate a signed %calc argument from the expression before it A whitespace-separated argument list is cut into arguments before the expression parser sees it, so `5 -3` is two arguments while `5 - 3` — an expression left unfinished across the space — stays one. Also name the shadowed unit when the shadowing declaration sits in the namespace that declares the import, by resolving the name again with that declaration hidden instead of searching its parent scope. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(repl): scope the named-argument check to the argument's own binding An `=` nested in a call, a bracket or a string belongs to that expression, so it no longer refuses `add(x = 1, y = 2)` as an argument; only a bare identifier bound at the argument's top level is a named argument. A qualified name in unit position resolves to what it names, so it is reported without the shadowing explanation \u2014 whose suggestion would otherwise paste the whole written name onto the unit's namespace. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(repl): report an ambiguous name at the prompt instead of resolving it %eval retained a lookup failure so an imported name could still be evaluated, which also swallowed the ambiguity error. A typed AmbiguousNameError separates the two: a name found nowhere falls through to the expression path, a name several declarations answer to is reported. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * fix(semantics): drop the remedy clause when no spelling qualifies the hidden unit A unit owned by no namespace has the same qualified name as its simple one, so the suggestion repeated the name that had just failed. The diagnostic now names the shadowed unit without advising a spelling in that case. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
Adds a `publish-pypi` job in a `release-python` workflow filtered to `pysysml-v*` tags, authenticating from the restricted `pypi-publish` context. The package's version is declared once, in `python/pysysml/_version.py`, read by the packaging metadata and by `pysysml.__version__` through the installed distribution; `python/setup.py` is gone and a test fails if a second version literal reappears. The job resolves the version from the tag, refuses a version the index already has, builds a wheel and an sdist, checks them with `twine check --strict`, installs the wheel into a clean virtualenv and imports it, and only then uploads - pre-release versions to TestPyPI. Also cuts the 0.0.5 changelog section (with the changes that landed after most of it was written, and an honest current limitation set verified on the built binary), corrects the README/QUICKSTART transcripts and the test counts repeated across the docs against a real run, and documents the PyPI release path and the upstream promotion the tag needs in docs/RELEASING.md. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…n prompts The interactive pass over the rewritten transcripts found three doc-side divergences from the binary: the first %save block omits the "(replaced the existing file)" suffix the REPL always prints there, the Runtime Features Engine/Car snippet needs "import ScalarValues::*;" for Real to resolve in a fresh session, and the behavioral blocks used an unindented continuation prompt. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov>
…and-0.0.5-prep * feat(release): publish pysysml to PyPI from CircleCI and prepare 0.0.5 Adds a `publish-pypi` job in a `release-python` workflow filtered to `pysysml-v*` tags, authenticating from the restricted `pypi-publish` context. The package's version is declared once, in `python/pysysml/_version.py`, read by the packaging metadata and by `pysysml.__version__` through the installed distribution; `python/setup.py` is gone and a test fails if a second version literal reappears. The job resolves the version from the tag, refuses a version the index already has, builds a wheel and an sdist, checks them with `twine check --strict`, installs the wheel into a clean virtualenv and imports it, and only then uploads - pre-release versions to TestPyPI. Also cuts the 0.0.5 changelog section (with the changes that landed after most of it was written, and an honest current limitation set verified on the built binary), corrects the README/QUICKSTART transcripts and the test counts repeated across the docs against a real run, and documents the PyPI release path and the upstream promotion the tag needs in docs/RELEASING.md. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> * docs(quickstart): correct save suffix, missing import and continuation prompts The interactive pass over the rewritten transcripts found three doc-side divergences from the binary: the first %save block omits the "(replaced the existing file)" suffix the REPL always prints there, the Runtime Features Engine/Car snippet needs "import ScalarValues::*;" for Real to resolve in a fresh session, and the behavioral blocks used an unindented continuation prompt. Co-Authored-By: jason.han <jason.han@jpl.nasa.gov> --------- Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: jason.han <jason.han@jpl.nasa.gov>
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.
No description provided.