LSP v2

The poly-bench-lsp-v2 crate implements the Language Server Protocol. It provides syntax highlighting, diagnostics, hover, formatting, and embedded language support for .bench files. This page covers the server architecture, Tree-sitter integration, and embedded code handling.

---

Overview

LSP v2 is a rewrite that improves on v1 with:

• Error-tolerant parsing — Always produces a syntax tree, even with errors
• Incremental parsing — Fast re-parsing on edits
• Full semantic token coverage — All syntax elements highlighted
• Incremental formatting — Small edits instead of whole-document replacement
• Better diagnostics — Syntax errors from Tree-sitter + semantic validation
• Embedded language hover — Full hover support for Go, TypeScript, Rust, and Python code

---

Server Architecture

The server (poly-bench-lsp-v2/src/server.rs) uses tower-lsp and implements:

• PolyBenchLanguageServer — Main server struct with document store, embedded config, virtual file managers
• Document store — DashMap of open documents by URI
• Workspace root — Detected from workspace folders or initialized document

Handlers:

• textDocument/didOpen, didChange, didSave — Track document state, publish diagnostics
• textDocument/hover — Hover for DSL and embedded code
• textDocument/semanticTokens/full — Semantic highlighting
• textDocument/formatting — Format entire document
• textDocument/rangeFormatting — Format selection
• textDocument/completion — Code completion (if implemented)

---

Tree-sitter Integration

The LSP uses poly-bench-syntax (Tree-sitter-based) for parsing:

• Error-tolerant — Always returns a tree; parse errors become diagnostics
• Incremental — Reuses unchanged subtrees on edit
• Queries — queries/highlights.scm for semantic tokens

This differs from poly-bench-dsl (used by CLI), which uses a different parser. The LSP prioritizes responsiveness and partial results over strict correctness.

---

Diagnostics

Diagnostics (poly-bench-lsp-v2/src/diagnostics.rs):

• Syntax errors — From Tree-sitter parse failures
• Semantic validation — Fixture references, setup completeness
• Embedded code — Type-checked via virtual files (see below)

On save, the LSP runs embedded code checks (Go, TypeScript, Rust, Python) and merges those diagnostics.

---

Embedded Language Support

Embedded code blocks (e.g. setup go { ... }) are supported via:

• Virtual files — Extracted code is written to temp files and passed to language-specific tools
• VirtualFileManagers — Per-language managers (gopls, typescript-language-server, rust-analyzer, pyright)
• Embedded diagnostics — check_embedded_code runs type-checkers and maps errors back to source spans
• Embedded hover — get_hover resolves hover for embedded code by delegating to the language server

---

Semantic Tokens

Semantic tokens (poly-bench-lsp-v2/src/semantic_tokens.rs):

• Legend — Defines token types for the DSL (suite, bench, fixture, setup, etc.)
• Tree-sitter queries — captures in .scm files map nodes to token types
• Embedded code — Delegates to language servers for embedded blocks

---

Formatter

The formatter (poly-bench-lsp-v2/src/formatter.rs):

• Format document — Full document or range
• Incremental — Can format only changed regions

---

Further Reading

• LSP / Editor — User-facing setup and configuration
• Architecture — Pipeline overview
• DSL — Parser and AST