SurrealQL Language Server

A Language Server Protocol (LSP) implementation for SurrealQL, the query language of SurrealDB.

Features

• Syntax diagnostics via tree-sitter
• Semantic analysis with schema inference from DDL and query flow
• Hover with type info, permission posture, function signatures, and language badges (SurrealQL vs JavaScript)
• Contextual completions for record<table> types, field names, builtin functions, and statement keywords
• Go-to definition and references for tables, fields, functions, and params
• Safe rename of local function definitions
• Code actions for missing PERMISSIONS clauses
• Signature help for builtin and user-defined functions
• Call hierarchy with inbound/outbound function call tracking
• Document symbols outlining tables, fields, events, indexes, and functions
• function() { ... } bodies parse cleanly with no false diagnostics; DEFINE FUNCTION bodies containing scripting functions are detected and labelled as JavaScript

Requirements

The language server compiles against a tree-sitter SurrealQL grammar that must be checked out as a sibling directory:

parent/
├── surrealql-language-server/   ← this repo
└── surrealql-tree-sitter/       ← grammar (sibling checkout)

Run the setup script to clone or update the grammar:

bash scripts/setup-grammar.sh

Or set TREE_SITTER_SURREALQL_DIR to point to an existing checkout:

TREE_SITTER_SURREALQL_DIR=/path/to/surrealql-tree-sitter cargo build

Building

Native binary

cargo build --release
# binary at: target/release/surrealql-language-server

Browser WASM package

Build the wasm-bindgen npm package (outputs to pkg/):

bash scripts/build-wasm.sh

Requirements:

• wasm-bindgen CLI (cargo install wasm-bindgen-cli --version 0.2.108)
• wasm-opt (cargo install wasm-opt)
• On macOS, a wasm-capable clang (e.g. brew install llvm; the script auto-detects Homebrew LLVM)

Browser hosts (Surrealist, etc.)

Initialize the module with fetch + arrayBuffer, then pass the bytes to the default export — the same pattern used by @surrealdb/wasm. Avoid passing a URL string directly to init() when the build pipeline pre-gzips .wasm assets in place: browsers only gunzip automatically when the response carries Content-Encoding: gzip (S3 production uploads do; many static preview servers do not).

import init, { WasmLanguageServer } from "@surrealdb/surrealql-language-server";
import wasmUrl from "@surrealdb/surrealql-language-server/surrealql_language_server_bg.wasm?url";

const wasmCode = await fetch(wasmUrl).then((response) => response.arrayBuffer());
await init({ module_or_path: wasmCode });

const server = new WasmLanguageServer({ /* callbacks */ });

The ./surrealql_language_server_bg.wasm export is declared in pkg/package.json for bundlers that resolve deep imports.

Testing

cargo test

Repository Layout

.
├── src/
│   ├── main.rs               # LSP stdio entry point
│   ├── backend.rs            # tower-lsp request handlers
│   ├── config.rs             # workspace settings
│   ├── grammar.rs            # tree-sitter language binding
│   ├── providers/            # completion, hover, rename, etc.
│   └── semantic/
│       ├── analyzer.rs       # document analysis (parse + extract)
│       ├── model.rs          # merged workspace model, code actions
│       ├── types.rs          # DocumentAnalysis, TableDef, FunctionDef, ...
│       ├── type_expr.rs      # SurrealQL type expression parser
│       └── text.rs           # LSP range utilities
├── tests/
│   └── lsp.rs                # integration tests
├── build.rs                  # compiles tree-sitter grammar (C)
└── scripts/
    └── setup-grammar.sh      # clones/updates the grammar sibling repo

Editor Integration

The server communicates over stdio and works with any LSP-compatible editor.

Grammar Development

The tree-sitter grammar lives in the sibling surrealql-tree-sitter repo. After editing grammar.js:

cd ../surrealql-tree-sitter
npx tree-sitter generate
npx tree-sitter test

The src/parser.c is auto-generated and should not be edited directly. JavaScript scripting function bodies (function() { ... }) are handled by an external C scanner at src/scanner.c which tracks brace depth, strings, template literals, and comments.

CI

GitHub Actions runs cargo fmt --check and cargo test on every push and pull request. The grammar sibling repo is cloned automatically during CI.

Releases

Native binaries and crates.io

Push a v* tag (e.g. v0.1.5). CI builds platform binaries, uploads them to the GitHub Release, and publishes the Rust crate to crates.io.

Browser WASM npm package

The scoped package @surrealdb/surrealql-language-server is built with scripts/build-wasm.sh (cargo → wasm-bindgen → wasm-opt) and published to npm on the same v* tag. A .tgz is also attached to the GitHub Release.

Release checklist:

1. Bump version in Cargo.toml and pkg/package.json.
2. Push the tag: git tag vX.Y.Z && git push origin vX.Y.Z.
3. Confirm the wasm CI job succeeds and the package appears on npm.

npm publishing uses Trusted Publishing (OIDC from GitHub Actions). Before the first publish, an @surrealdb org admin must configure a trusted publisher on the package's npm Access page (@surrealdb/surrealql-language-server) with:

• Repository owner: surrealdb
• Repository name: surrealql-language-server
• Workflow filename: ci.yml (exact match, case-sensitive)

The CI workflow intentionally does not set registry-url on actions/setup-node — that option writes an .npmrc which forces token auth and breaks OIDC (npm/cli#8730). Do not add a NODE_AUTH_TOKEN secret for this job.

If the first CI publish still fails with a misleading 404, an org admin can bootstrap the package once locally (bash scripts/build-wasm.sh && npm publish --access public from pkg/), then configure the trusted publisher for subsequent tag releases.