diagrams.mbt

moonrockz/diagrams-bpmn

This pure adapter converts moonrockz/bpmn documents into the independent moonrockz/diagrams/bpmn visual model. to_diagram returns an optional Diagram::Bpmn and ordered diagnostics. render_svg reads, converts and calls the SVG renderer. Neither operation reads files, fetches imports or changes XML.

The converter covers process and collaboration graphical kinds, including typed tasks, events and their definitions, gateways, subprocesses, expanded process calls, pools, nested lanes, data, groups, annotations and connections. It dispatches on the effective schema type, including xsi:type. BPMN-looking elements inside opaque vendor payloads remain inert. Choreography and conversation documents retain their typed reader/writer/traversal support; selecting them returns bpmn/unsupported-view. They never become generic tasks.

Select a view

Use ViewSelector::Process(id), Collaboration(id) or Plane(id). A plane selector accepts its plane ID or its parent diagram ID. View(id) selects an identity first and diagnoses a deferred or invalid kind explicitly.

Automatic selection examines planes first, then collaborations, then processes. The first nonempty tier must contain exactly one candidate. One plane and its referenced process count as one view. A collaboration includes its participant processes, rather than competing with them. Multiple matching planes need an explicit plane selector, even when they refer to the same process.

///|
test "select a semantic process without requiring DI during conversion" {
  let source = "<b:definitions xmlns:b='http://www.omg.org/spec/BPMN/20100524/MODEL' targetNamespace='urn:example'><b:process id='P'><b:userTask id='T' name='Review'/></b:process><b:process id='Other'/></b:definitions>"
  let document = @bpmn.read(source).document.unwrap()
  let ambiguous = @adapter.to_diagram(document)
  assert_eq(ambiguous.diagnostics[0].code, "bpmn/ambiguous-view")
  let selected = @adapter.to_diagram(document, selector=Process("P"))
  guard selected.diagram is Some(Bpmn(view)) else { fail("missing process") }
  assert_eq(view.source.source_id, Some("P"))
  assert_eq(view.nodes[0].kind, UserTask)
  assert_eq(view.nodes[0].geometry, None)
}

Render supplied geometry

render_svg(source, geometry_policy=RequireDi) paints complete process and collaboration DI, including the standard graphical kinds and their markers. The default PreferDi uses all supplied geometry when complete; missing or partial DI gives bpmn/incomplete-di and computes the whole view. RequireDi returns no output when required bounds or waypoints are missing. Invalid supplied geometry is an error under either policy. Legibility warnings preserve authored positions rather than moving individual elements.

Automatic explicitly ignores DI during conversion and geometry generation, including DI endpoint overrides. It can render semantic input with broken DI. With an explicit process selector, it does not require choosing between that process's planes. Automatic layout measures labels, ranks each scope after cycle condensation, packs lane/group bands and routes orthogonal connections around shapes and labels. Bounded search failure returns a layout-limit diagnostic without partial output.

Static SVG stays script-free. Labels are escaped text; source IDs, resource keys, containment paths, visual endpoints and render properties are inert metadata. See moonrockz/diagrams-svg/bpmn-layout for the pure geometry and legibility APIs.

///|
test "render semantic input or require supplied DI explicitly" {
  let source="<b:definitions xmlns:b='http://www.omg.org/spec/BPMN/20100524/MODEL' targetNamespace='urn:example'><b:process id='P'><b:userTask id='T' name='Review'/></b:process></b:definitions>"
  let automatic=@adapter.render_svg(source,geometry_policy=Automatic)
  assert_true(automatic.output != None)
  assert_eq(automatic.diagnostics,[])
  let required=@adapter.render_svg(source,geometry_policy=RequireDi)
  assert_eq(required.output,None)
  assert_eq(required.diagnostics[0].code,"bpmn/incomplete-di")
}

Source identities and geometry

SourceIdentity keeps the caller's resource key, optional original XML ID and source-order containment path. Node id and flow id are unique render IDs. An uninstanced root element with an XML ID keeps that spelling. Imported elements, participant instances and anonymous elements use length-prefixed generated IDs. The original ID stays in source.source_id or identity.source_id. These IDs are never written into the document. Multiple input documents may reuse local IDs.

Nodes keep pool/lane/subprocess ownership and boundary attachment. Markers keep loop mode, compensation, event-subprocess/interrupting flags, collection status, called activity kind/identity and event definitions. Expressions and timer text remain descriptions. Group labels resolve category values; category membership is retained in Ownership.groups for automatic group bounds. Data references keep their referenced name and collection marker. Connections distinguish sequence, message, association and data-input/output kinds. A many-source data association has one visual connection per source with a shared original identity. A valid source-less association has has_source=false and no semantic source identity. No transition or execution outcome is invented.

Flow semantic_source and semantic_target retain actual document endpoints. Visual source/target may identify an owning activity, or a selected DI sourceElement/targetElement, for example the boundary of a black-box pool. A shared process can have distinct participant instances. A reference with several possible visual instances needs an unambiguous selected DI endpoint; the adapter returns an ambiguity diagnostic instead of choosing an arbitrary instance.

Selected DI retains exact finite decimal/exponent tokens in Number, shape and label bounds, waypoints, fonts, display flags and endpoint-shape identities. bounds/points remain the original integer convenience inputs. Fractional DI is retained in geometry, without rounding it into those convenience fields. Absence or partial geometry remains explicit for the renderer's whole-view policy. Multiple DI occurrences for an otherwise identical projected identity return bpmn/ambiguous-di, rather than assigning coordinates by source order.

The core model also accepts caller-supplied RenderProperties and Link values through immutable constructors and with_properties. Vendor extensions never supply active styles, markup or links. The model has no dependency on bpmn.

///|
test "conversion retains an anonymous node without editing the source" {
  let document = @bpmn.read("<b:definitions xmlns:b='http://www.omg.org/spec/BPMN/20100524/MODEL' targetNamespace='urn:example'><b:process id='P'><b:task/></b:process></b:definitions>").document.unwrap()
  let before = @bpmn.write(document)
  guard @adapter.to_diagram(document).diagram is Some(Bpmn(view)) else { fail("missing process") }
  assert_eq(view.nodes[0].source.source_id, None)
  assert_eq(view.nodes[0].source.path, [0, 0, 0])
  assert_eq(@bpmn.write(document), before)
}

References, findings and limits

Supply ResourceSet and optional root resource_id explicitly. Import location keys and target namespaces follow the standalone module's strict QName rules. Each imported process resolves references using its own import closure. Reused immutable indexes avoid reparsing XML. No local-ID fallback crosses namespaces.

Malformed or ambiguous selected references prevent a diagram. Findings outside the selected view remain warnings; an unrelated unresolved import does not block a self-contained process. map_diagnostic preserves diagnostic code/severity and the source's half-open UTF-16 offsets with one-based lines/columns. CRLF and non-BMP characters use the same positions as the source reader. Constructed values have no invented spans. Conversion-specific findings use the affected element's source location when available.

Limits bounds validation, source tokens, depth, diagnostics and conversion work. Conversion charges index construction, identity paths and indexed field lookups against nodes. Reference fields are indexed once instead of rescanning every reference for every node. Imported re-resolution charges the existing resource closure's record count before rebuilding its resolver. Exhaustion returns no partial diagram. bpmn/conversion-limit differs from unresolved or unsupported input.

Conversion adds these codes to the standalone module's validation codes:

  • bpmn/view-not-found, bpmn/missing-plane, bpmn/ambiguous-view, bpmn/ambiguous-plane, bpmn/unsupported-view and bpmn/unsupported-element.
  • bpmn/invalid-reference, bpmn/ambiguous-endpoint, bpmn/ambiguous-di, bpmn/duplicate-id, bpmn/invalid-geometry and bpmn/conversion-limit.

The checked graphical coverage manifest is tests/corpus/bpmn-model/graphical-coverage.json. Its embedded copy is checked against the converter inventory and a concrete assertion for each of the 38 standard graphical kinds. Additional assertions cover all ten event-definition symbols, ownership, markers, imports, exact DI and deferred views. Non-graphical implicit throw events remain in multi-instance descriptions, without invented visible event shapes. Base tDataAssociation has no standalone global element; an invalid xsi:type downcast is diagnosed, while its two concrete association declarations convert.

Match and constructor migration

Adding Bpmn to the public moonrockz/diagrams/render.Diagram enum changes source compatibility for exhaustive matches. Add a Bpmn(view) case or return an unsupported-kind renderer result. Existing sequence constructors and SVG bytes remain unchanged. In the provisional BPMN constructors, pass integer convenience geometry as bounds= or points=; use geometry= for exact selected DI. No version change, release or module publication is part of this change.

View source on GitHub