For large configuration files it would ease navigation if a code outline (document symbol map) could be provided.

Sorry for my fault.

single \n in Markdown won't lead to a new line.
ref: https://stackoverflow.com/a/58465541

So a Markdown string

"Set base image\n## Parameters\nos (str): The operating system (i.e. 'ubuntu20.04')\nlanguage (str): The programing language dependency (i.e. 'python3.8')\nimage (Optional[str]): Custom image (i.e. 'python:3.9-slim')"

will become:

without any new line.

We would use \\\n instead of \n. I will give a patch soon.

The right string is:

Value: "Set base image\n## Parameters\nos (str): The operating system (i.e. `ubuntu20.04`)\\\nlanguage (str): The programing language dependency (i.e. `python3.8`)\\\nimage (Optional[str]): Custom image (i.e. `python:3.9-slim`)",

Any the right info:

Hello, I am a developer from tensorchord. We use starlark-lsp as a denepndency to build our LSP Server.
Much thanks for this project and its contributors!

I have a feature recommendation for starlark-lsp. If you think it is acceptable, I would be happy to help on it. If it could be merged, I could upgrade our project to introduce it either.

When we use Python extension, it would display Parameters and Returns for functions.

Hovever, starlark-lsp could only show first line(document) for Hover, it would be better for user if Hover could output arguments types.

Related issue: tensorchord/envd-lsp#35

Hover Support for custom symbols would be great, e.g. in the following snippet I would like to hover main and say_hello and get the docstring and function definution, like for builtin symbols.

"""
say something
"""
def say_hello(what):
    print("hello", what)

"""
main func
"""
def main():
    say_hello("name")

main()

starlark-lsp use LUCI docstring parser. Wheile parse an arg line, it doesn't allow any space inside argument, though it might be useful for add some type defenitions.

Examples:

We could revised it into ^(\S.*?)\s*:\s*(.*)$, then could support more flexible argument defenition.

Related issue: tensorchord/envd#1468

If you think this feature is acceptable, I would be happy to work on it.

Autocomplete for dicts would be very good. This would ease the configuration with structured data.

Given three files:

# a.tiltfile
load('ext.tiltfile', 'flag')
load('b.tiltfile', 'hello')
hello()

# b.tiltfile
load('ext.tiltfile', 'flag')

def hello():
  print("in b: " + flag)

# ext.tiltfile
flag = True

The starlark-lsp server reports a circular load: ext.tiltfile diagnostic back to the LSP client when there should be none.