Skip to content

v0.58.0

Latest

Choose a tag to compare

@github-actions github-actions released this 18 Nov 17:34
· 5 commits to master since this release
966b147

[0.58.0] - 2026-11-18

Bug Fixes

  • Lesbarer Spurabstand fuer verschachtelte Binding-Lanes (Nacharbeit 84) (par)- MSI rename step no longer copies the file onto itself (release)

Documentation

  • Layout block keywords (sameLayer, first, last) per diagram family

Features

  • UML class diagram draws the general class above its specialisations (Nacharbeit 85) (layout)- BDD hub edges drawn as an orthogonal fan (layout)

Other

  • Ktlint formatting of the patchDesktopDeb task (desktop)- Note tone colours overridable per theme via KumlColors tokens

Notes (blueprint/annotation) used hard-coded tone colours in NoteTones, so
themes could not restyle them and dark/custom themes got unreadable notes.
KumlColors now carries optional note-tone tokens that NoteTones resolves
before falling back to the built-in defaults, keeping existing output
unchanged for themes that do not set them.

Review and security audit of the feature branch surfaced and fixed the
issues found there; coverage added: NoteToneTokensTest, a NoteGeometry
regression test and a theme serialization round-trip test. Docs
(themes, blueprint-notes) and CHANGELOG updated.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Steerable note placement (placement hint, W004/W005 diagnostics)- Nacharbeit 91: fewer files per Maven Central release

Maven Central caps a project at 1,000 files per release; a release had
2,884 files (limit request to Sonatype pending). Every multiplatform
module published a root component plus -jvm, -js and -wasm-js components,
each with jar, sources, javadoc, pom, module and signatures.

The js and wasmJs publications (and their signing tasks) of all 16
multiplatform modules are disabled in the root build.gradle.kts; the
targets still build. kuml-wasm-playground joins nonPublishedModules
(kuml.dev builds it from source). The release probe in release.yml checks
only the jvm leg; docs and the readiness note are adjusted. New
scripts/verify-central-file-count.sh publishes into a throw-away repo and
fails if a -js/-wasm-js artifact is published again.

Verification: publishToMavenLocal went from 715 files / 140 artifacts to
555 / 108; a JVM-only consumer resolves and compiles against the result.

Breaking for JS/Wasm consumers: dev.kuml:*-js and *-wasm-js coordinates
are no longer published. kuml-workspace and kuml-workspace-pdf stay
published because kuml-workspace-latex depends on them.

Not fixed: about 2,200 files remain (target was about 1,400); further
options need a decision. Whether Central accepts the root .module
metadata still listing js/wasm variants only shows at the real upload.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 92: anchor lines of notes in Service Blueprint and SEQ

Remaining list PAL wave 3, section A, N3: anchor lines of notes that run
through foreign shapes. Measured first on neutral scratch examples with the
freshly built CLI, then fixed only where the numbers showed a defect.

Why: in a Service Blueprint, once the four tracks below a lane's cards were
taken, an anchor line used the outer channel and ran straight up through
every card below its target (and two lines into one column overlapped).
In sequence diagrams the note placer returned the first margin column that
had any admissible position, even with crossings, so a note on a left-half
lifeline took the right margin and crossed all lifelines to its right.

Blueprint (BlueprintNoteRouter): such a line now leaves the outer channel
through the gap beside the card and ends on the card's side edge; lines in
one gap keep 6 px. Straight entry and card-less calls are unchanged.
SEQ (SeqNotePlacer.choose): all columns are searched, a later column wins
only with strictly fewer crossings; ties keep the old column order.

Results: Blueprint K1 (segments through a foreign card) 3 -> 0; SEQ
crossings reduced per note, vault examples 19/33/59/65 measure 0. New tests
BlueprintNoteAnchorLinesN3Test and SeqNoteAnchorCrossingTest fail without
the change and pass with it. Review/gate found only ktlint style findings
in the new files (fixed) and an unrelated detekt heap error in
kuml-runtime-tokenflow (repeated green). Known limits (shared gutter
overlap, labels as obstacles, multi-anchor SEQ notes, placement ignored)
are documented in CHANGELOG.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 93: test keeps layout.adoc effect table in sync with LayoutEffectMatrix

The layout-effect table in the handbook (layout.adoc) is maintained by hand
and had already drifted from LayoutEffectMatrix, the code-side source of
truth. Without a guard, every new layout option or changed effect silently
makes the documentation lie again.

Add LayoutEffectMatrixDocSyncTest, which parses the table from layout.adoc
and compares it against LayoutEffectMatrix, so any drift fails the build.
A small adjustment in LayoutEffectMatrix.kt makes the matrix readable for
the test. Review and security audit of the change were completed and their
findings fixed before this merge. CHANGELOG updated.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 94: cover wrapColumns on BDD hub fans with tests and docs

wrapColumns on BDD hub fans had no dedicated test coverage and was not
documented, so a regression in the wrapped-fan layout would have gone
unnoticed and users could not discover the option. Adds
BddHubFanWrapColumnsTest (bridge-level layout checks), a layout reference
entry in the handbook and a CHANGELOG note.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 95: probe BDD layout direction Up (hub fan, bus, allocation docking)

Why: Up is the natural direction for inheritance in a block definition
diagram (general block below its specialisations). Nacharbeit 86/94 only
covered Down/wrapColumns, so the generalisation bus, the hub fan (hub as
source and as target) and the allocation docking were unverified with
direction = Up (BottomToTop).

Finding: all probes passed; the existing passes already handle Up. No
production code was changed. Added BddDirectionUpTest (geometry only,
font independent, 9 tests: no foreign-box crossing, edge ends at hub top
and block bottoms, 6 px separation, bus trunk, wrapColumns rows,
determinism, docking no-op for Up/Down) plus a CHANGELOG entry and one
handbook sentence. Module check, detekt and ktlint green (866 tests, 0
failed). Not fixed, documented: allocation docking is a no-op for Up/Down
by design; rotated aspect candidate for Up untested; allocation probe
uses a hand-built graph.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 96: report wrapColumns and sameLayer inside C4 container boundaries with W001

Task: layout options such as wrapColumns and sameLayer were silently ignored for
nodes placed inside C4 container boundaries, so authors saw no effect and no
explanation. The layout runner now detects this case and reports it through the
existing W001 "option has no effect" diagnostic, backed by an entry in
LayoutEffectMatrix and surfaced via the CLI ValidationService.

Findings from review and security audit were fixed on the feature branch before
this squash. Tests cover the new C4 container cases (C4ContainerLayoutOptionsTest)
and the extended effect matrix. CHANGELOG and the layout reference page in the
handbook are updated.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 97: report wrapColumns and sameLayer as ineffective inside compound groups (UML, SysML 2 package)

Measured with the real ELK engine: both options are dead for members of compound
groups in UML class/package/component/deployment/state and SysML 2 package diagrams.
LayoutEffectMatrix.ignoredInsideCompound now lists them per kind, so KUML-LAYOUT-W001
is reported (render and kuml validate). No layout change.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- C4 notes honour the placement hint (Nacharbeit 98)

Task: C4 diagrams ignored the placement hint on notes, while UML notes
already honoured it via NotePlacer, and the CLI validator wrongly flagged
C4 with KUML-NOTE-W005 ("hint not supported"). The result was a silent
inconsistency between diagram families: a hint was accepted by the DSL but
had no effect on C4 output.

Fix: C4NotePlacement now passes the placement hint to NotePlacer and
reports misses (hint cannot be satisfied) as KUML-NOTE-W004, the same
diagnostic UML uses. NoteChecks no longer emits W005 for C4. Handbook
pages (c4-dsl, uml-dsl, validate) and CHANGELOG are updated to match.

Verification: new C4NotePlacementHintTest (hint honoured, miss reported,
default behaviour unchanged) and an extended NotePlacementHintCliTest;
review and security audit of the change found no blocking issues.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- UML notes honour the placement hint (Nacharbeit 99)

Task: UML diagrams ignored the note placement hint that other diagram
types already honoured, so notes landed wherever the layout put them.
Added UmlNotePlacement in the layout bridge and wired it into the SVG
and LaTeX renderers, plus the CLI note checks, so the hint now takes
effect consistently across output formats.

Review and security audit surfaced real findings (hint handling edge
cases in the renderer/CLI paths); these were fixed on the feature
branch and are covered by new UmlNotePlacementHintTest and extended
NotePlacementHintCliTest cases. CHANGELOG and handbook (uml-dsl,
validate) updated accordingly.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 100: ERM notes honour the placement hint

Aufgabe: ERM-Diagramme ignorierten den Platzierungshinweis (placement hint)
von Notizen, den UML-Diagramme bereits auswerten. Notizen landeten dadurch
unabhaengig von der Angabe im Skript an einer festen Position, und die
CLI-Pruefung (NoteChecks) meldete den Hinweis fuer ERM faelschlich als
wirkungslos.

Aenderung: neue ErmNotePlacement-Logik in der Layout-Bridge, vom
KumlSvgRenderer fuer ERM-Notizen verwendet; NoteChecks, Handbuch
(erm-dsl, uml-dsl, validate) und CHANGELOG entsprechend angepasst.

Tests: ErmNotePlacementHintTest, ErmNotesSvgTest und
NotePlacementHintCliTest decken Happy Path und Fehlerfaelle ab. Die in
Review und Security-Audit gefundenen Befunde wurden vor dem Merge auf dem
Feature-Branch behoben.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 101: Service Blueprint notes honour the placement hint (ABOVE/BELOW in the note column)

Blueprint notes accepted placement but ignored it and reported KUML-NOTE-W005.
BlueprintGeometry now evaluates the hint inside the note column against the first
anchor card: ABOVE and BELOW are honoured, RIGHT and NEAR keep the default, LEFT can
never be honoured (reported as KUML-NOTE-W004 with the source line). NoteChecks no
longer reports W005 for Service Blueprint. Output without a side hint is unchanged.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 102: sequence diagram notes honour the placement hint

Notes in sequence diagrams ignored the note placement hint (left/right/
above/below) that class and state diagrams already respected, so authors
could not steer where a note lands next to lifelines and messages. The
SeqNotePlacer now evaluates the hint, keeps notes clear of lifelines and
message arrows, and falls back safely when the hinted slot is occupied.

Review and security audit found real defects in the first draft, which
were fixed before this merge: hint handling edge cases in the placer and
the CLI note checks (NoteChecks), plus matching test gaps. New tests
cover the placer (SeqNoteHintTest) and the CLI validation path
(NotePlacementHintCliTest). CHANGELOG and handbook pages (uml-dsl,
sysml2, validate) are updated to describe the new behaviour.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 103: BPMN text annotations honour the placement hint

Task: the placement hint for notes (introduced for UML/SysML 2) was ignored
for BPMN text annotations; the CLI warned KUML-NOTE-W005 ("hint not
supported") instead. Annotations in process and collaboration diagrams now
honour the hint, so authors get the same predictable note placement across
all diagram families and the misleading warning disappears.

Implementation: new BpmnNotePlacement in the layout bridge places hinted
annotations with strict lane rules (no crossing of lane/pool boundaries, no
overlap with flow nodes), wired into KumlSvgRenderer; Sysml2NotePlacement and
NoteChecks/KumlNoteCodes adjusted accordingly (W005 no longer emitted for
BPMN). Handbook (bpmn-dsl, uml-dsl, validate) and CHANGELOG updated; new
BpmnNotePlacementHintTest and an extended NotePlacementHintCliTest cover the
happy paths and lane-rule edge cases.

Review and security audit of the wave ran through the standard pipeline and
their real findings were fixed on the feature branch before this squash.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 104: LaTeX Service Blueprint renderer honours the note placement hint

The SVG renderers already respected the note placement hint (Nacharbeiten
84-103), but the LaTeX/TikZ Service Blueprint renderer ignored it, so notes
anchored to a column landed in a different spot in the PDF than in the SVG.
BlueprintLatexRenderer and BlueprintNoteTikz now read the hint through the
shared ColumnNoteHint helper in kuml-layout-bridge, and BlueprintGeometry in
the SVG module uses the same helper so both back ends agree on placement.

Review and security audit of the change found and fixed real issues
(hint handling edge cases and unescaped/unsafe note input in the TikZ
output); the final state was approved in both loops. New
NotePlacementHintLatexTest covers happy path and fallback cases;
BlueprintNoteTikzTest adjusted. CHANGELOG and blueprint-notes reference
page updated.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 105: LaTeX BPMN renderer honours the note placement hint

The LaTeX renderer ignored the note placement hint for BPMN diagrams, so
notes were always drawn at a fixed default position while the SVG renderer
respected the hint. That made LaTeX output diverge from the SVG output and
could place notes over flow elements.

BPMN notes in the LaTeX backend now resolve their position through a new
BpmnNoteLayoutBridge and the reworked BpmnNoteTikz, using the same hint
semantics as the other diagram types. Handbook (BPMN and UML DSL reference)
and CHANGELOG are updated.

Tests: unit tests for the layout bridge and the TikZ output, a CLI-level
test for the hint, and a PDF build test that compiles the generated TikZ.
Review and security audit findings were fixed on the branch (named
arguments to satisfy detekt among them) before this squash.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 106: per-group layout direction, pilot C4 boundaryDirection

A group (for example the C4 system boundary) can now lay out its content in
a different direction than the whole diagram instead of inheriting the
global layout direction.

  • LayoutGroup.direction; mapped in ELK via SEPARATE_CHILDREN for
    self-contained compounds (HintsMapper/ElkLayoutEngine).
  • DSL: layout { boundaryDirection } for C4; LayoutSpecRunner and
    LayoutEffectMatrix adapted.
  • New warning KUML-LAYOUT-W009 when a relationship crosses the boundary:
    ELK cannot keep the separate direction then, so the effect would
    otherwise be silently void.
  • Tests (GroupDirectionTest, C4BoundaryDirectionTest, spike test), CHANGELOG
    and handbook (layout, c4-dsl) updated.

Squashed from the spike and implementation commits of the feature branch.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Nacharbeit 107: layer hints inside self-contained packages (UML package diagram)

Why: wrapColumns and sameLayer had no effect on the members of a package
and were reported as KUML-LAYOUT-W001, because ELK ignores layer hints
below an INCLUDE_CHILDREN root. A spike (GroupLayerHintsSpikeTest)
showed that ELK 0.12.0 honours partitions and layer constraints in a
compound with HIERARCHY_HANDLING = SEPARATE_CHILDREN, as long as no
edge crosses the compound border.

What: LayoutGroup.memberLayerHints (default false, output unchanged),
HintsMapper.applyCompoundLayerStrategy for flagged, self-contained
groups without nested groups, bridge flagging from the user's spec only
(so the automatic aspect iteration never changes default output),
LayoutEffectMatrix.honouredInSelfContainedCompound for UML_PACKAGE, and
CompoundLayerScope so validate / kuml validate / kuml render report W001
only where the hint really has no effect.

Findings fixed in review: wrapColumns rows were counted over the whole
diagram, so a class declared before the package shifted the rows; now
counted per package. A sameLayer group with an unknown id wrongly
reported the mixed-scope W001. Layering is applied only for flagged
packages in withSpec.

Not fixed: only UML package diagrams; other diagram types keep W001;
nested groups are not honoured.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Release 0.58.0: changelog highlights and version bump

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Release 0.58.0: correct the changelog highlights after review

The review found a false bullet (layer hints in C4/UML compounds only report
W001), a misleading one (class diagram items were documentation, not fixes),
and missing limits and the breaking JS/Wasm change.

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com- Release 0.58.0: changelog mentions the Windows MSI and its upgrade note

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

Tests

  • UML-Sichtbarkeits-Default und Enum-Attributtyp-Kante absichern