mermaid graph input

[matt-beanland]

construct an artefact from a mermaid graph input, if valid
test full round trip from artefact new, mermaid out - construct new - mermaid out - compare.

[matt-beanland]

Design notes from yarning (2026-05-13)

Scope

Parse Mermaid legacy graph format into %Artefact{}. Two input profiles:

• Round-trip: source produced by Artefact.Mermaid.export/2 — nodes as id(("name<br/>Label1 Label2"))
• Import: hand-authored Mermaid from tools like Confluence/GitHub — nodes as id[text], id(text), etc.

Node content convention

Display text encodes name and labels. Parser splits on:

• <br/> → our export format: name before, space-separated labels after
• · (middle dot) → yarn convention: LABEL · name
• otherwise → content is the name, no labels

Node descriptions

click nodeId "tooltip text" carries the node description property — present in source, visible on hover, parseable. export/2 should be enhanced to emit these for nodes that have a description property.

UUID stability

Deterministic UUID from the Mermaid node id via Artefact.UUID.from_name/1 (SHA-256, forced version=7 and variant bits). Same Mermaid id always produces the same UUID — no external service needed for MVP. An Artefactory registry is the right answer at scale across authors.

What is not parsed

• Edges without a relationship label (no valid Artefact type)
• Node properties beyond name and description
• Subgraphs (ignored, not mapped to sections)

Exporter gap

export/2 currently emits accDescr: for the artefact description but nothing for node-level description properties. This should emit click id "description" lines.

Test plan

• Parse our own export output and compare node/rel counts and identity
• Round-trip: new! → export → from_mmd! → export — both exports equal
• Parse simple hand-authored Mermaid: bare ids, bracket nodes, edge labels
• click tooltip lands in node description property
• Same Mermaid id in two diagrams → same UUID (combine! finds the binding)