Telvax for VS Code

Language support for Telvax, powered by the compiler itself:

• Syntax highlighting for .tvx files (keywords, @endpoint, f-strings with interpolation, the |> operator, tensor and table types).
• Diagnostics from the real parser and type checker, including tensor-shape mismatches.
• Hover with inferred types and function signatures.
• Inlay hints showing the inferred type of let bindings and loop variables.
• Completion for types, fields, table columns (also inside t[".. and t.sql("..), methods of built-in and your own types, enum variants, std. paths, functions and variables, ordered by the type the code expects.
• Go to definition, find references, rename and document symbols for variables, functions, globals, types, enum variants, fields and methods, across the files of the project.
• Debugger for .tvx files: breakpoints, step over / into / out, the calls in progress, variables that expand, and hovering a name or a field (p.x) to see its value. Press F5 on a Telvax file, use Telvax: Debug Current File, or add a telvax launch configuration. It runs the program in the interpreter through telvax dap.
• Formatting with the same rules as telvax fmt.
• Snippets for fn, main, type, enum, match, for, test, endpoint and trait.

Setup

Install the telvax executable so that telvax lsp is on your PATH:

curl -fsSL https://raw.githubusercontent.com/telvax/telvax/main/scripts/install.sh | sh        # Linux, macOS
irm https://raw.githubusercontent.com/telvax/telvax/main/scripts/install.ps1 | iex             # Windows (PowerShell)

or point the telvax.server.path setting at the executable. Each GitHub release also carries the .vsix, which installs with code --install-extension telvax-<version>.vsix.

Commands: Telvax: Restart Language Server, Telvax: Debug Current File and Telvax: Run Current File Without Debugging.

Debugging

F5 on a .tvx file starts a session without a launch.json. To set it up, add a configuration:

{
  "type": "telvax",
  "request": "launch",
  "name": "Debug the current Telvax file",
  "program": "${file}",
  "function": "main",
  "args": [],
  "stopOnEntry": false
}

args are the words you would give telvax run after the function name; allow takes the same list as --allow (fs.read, net, none, ...). The debugger stops before a statement, so a breakpoint on a line that starts no statement moves to the next one that does. telvax.server.path also selects the executable that is started as telvax dap. There are no conditional breakpoints, the debug console evaluates only names and their fields, and programs built with telvax build cannot be debugged.

Publishing

A release publishes the extension by itself (the vscode job of release.yml): it packages the .vsix, attaches it to the release and, when the secrets exist, publishes it.

1. Visual Studio Marketplace: create the publisher telvax at https://marketplace.visualstudio.com/manage (it needs an Azure DevOps organization), then create a Personal Access Token there with the scope Marketplace > Manage and All accessible organizations. Save it as the repository secret VSCE_PAT.
2. Open VSX (VSCodium, Cursor and others read from it): log in at https://open-vsx.org with a GitHub account linked to an Eclipse account, sign the publisher agreement, create a token under Settings > Access Tokens, and run npx ovsx create-namespace telvax --pat <token> once. Save the token as the repository secret OVSX_PAT.
3. Cut a release. A step is skipped while its secret is missing, so the two can be set up one at a time.

To publish by hand: npx vsce publish --packagePath telvax-<version>.vsix --pat <token> and npx ovsx publish telvax-<version>.vsix --pat <token>.

Development

npm install
npm run compile      # TypeScript -> out/
npm test             # tokenises sample code with the real TextMate engine
npx vsce package     # builds telvax-<version>.vsix