@mildastudio/milda-lsp

A Language Server Protocol server for the Milda design-system DSL (.milda files).

It reuses the @mildastudio/milda parser directly, so the language server and the Studio speak the exact same grammar - there is no second parser to keep in sync. One server backs every LSP-capable editor: JetBrains IDEs, VS Code, Neovim, Zed.

Status

M1 - diagnostics. On open and on every edit, the server parses with the error-recovering parseMildaDocument and publishes one diagnostic per syntactic issue, each widened from the parser's point anchor to the token underneath it.

M2 - spans, semantics, structure. The parser now emits a source-span side-table (ParseResult.spans) keyed by the same ids the AST uses, without adding positions to any AST type. On top of it:

• Semantic diagnostics from engine/validate (unknown archetype, alias-by an undeclared context group, facet/layout references to missing tokens, dangling param→prop bindings, unknown states), each placed on the offending entity's span.
• Document symbols - a hierarchical outline (component → parts / props / events, context group → members).
• Folding ranges for every multi-line declaration block.

M3 - completion, hover, navigation. Context comes from the token before the cursor plus the enclosing declaration span; vocabulary is the live prelude, so it never drifts from the language.

• Completion - archetype names after archetype, UI-state names after state, declared token refs after a facet, facet names + keywords inside a part, and top-level / component-body keywords elsewhere.
• Hover - Markdown docs for the identifier under the cursor: document declarations (token type/value, component archetype + description, context-group members) and prelude vocabulary (archetype contract + composes, facet covers, state from).
• Go-to-definition - from a reference (token ref, component / type name) to the declaration's name range.

M4 - semantic highlighting + JetBrains plugin. The server now answers textDocument/semanticTokens/full, so syntax colouring travels with the language server (keywords, types, strings, numbers, comments, dotted refs, prop refs) instead of a per-editor grammar. A thin JetBrains plugin that launches this server lives in packages/milda-jetbrains.

Run

npm run build -w @mildastudio/milda-lsp   # emits a self-contained dist/index.cjs

The milda-lsp binary speaks LSP over stdio. Point any editor's generic LSP client at it for the milda language / .milda extension. For local development without a build step, editors that accept a command can run:

npx tsx packages/milda-lsp/src/index.ts

JetBrains

Install the plugin in packages/milda-jetbrains (paid JetBrains IDEs only — the LSP API is not in Community editions). It launches this server and wires up every feature above, including semantic highlighting. See that package's README to build and install.

Architecture

Each feature is a pure, transport-free, unit-tested core; server.ts binds them to the LSP stdio transport. All Milda-specific logic lives in the cores.

• src/diagnostics.ts - syntactic diagnostics (text → Diagnostic[]).
• src/semantic.ts - semantic diagnostics via engine/validate, placed by span.
• src/symbols.ts - hierarchical document symbols from the span table.
• src/folding.ts - folding ranges from the span table.
• src/completion.ts - context-aware completion items.
• src/hover.ts - hover docs for the identifier under the cursor.
• src/definition.ts - go-to-definition for references.
• src/semantic-tokens.ts - lexical classifier → LSP semantic tokens (highlighting).
• src/analysis.ts - shared position helpers (word-at-cursor, enclosing span, reference resolution) used by completion / hover / definition.
• src/range.ts - the single 1-based (Milda) → 0-based (LSP) position bridge.
• src/server.ts - transport wiring + advertised capabilities.
• src/index.ts - the milda-lsp executable; just starts the server.