CallHierarchy support

[jbellis]

Hi there and thanks for creating multilspy!

I'm using it to create a language-agnostic-ish call graph to solve LLM code assistant context. [Article.]

Currently this is two passes:

1. Forward pass through request_document_symbols to collect method regions
2. For each symbol, call request_references and add edges

(1) is about 0.1s per file which is not great but I can live with it. But (2) is about 2.5s per source file.

Multithreading helps a little but not much (about a factor of 2x), because Python.

I think I could do (2) in request-per-file instead of request-per-symbol if I had CallHierarchy with outgoingCalls support.

WDYT?

[LakshyAAAgrawal]

Thanks a lot!

Do you mean this request (https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#textDocument_prepareCallHierarchy)?

[jbellis]

Yes

[jbellis]

the other nice thing about structuring it around outgoingCalls instead of request_references is you can easily rebuild just a single node when a method gets edited

[LakshyAAAgrawal]

Jonathan Ellis (@jbellis) could you please see if #47 helps?

[LakshyAAAgrawal]

Further, one feature of multilspy that I personally like is that it is very easy to access the underlying language server connection handler. This means, it is very easy to use a lsp feature that hasn't been implemented as a first class feature in multilspy.

Please refer to detailed steps I shared about it in another related thread (#12 (comment)). This isn't to discourage new feature requests, but so that multilspy users can get started with using the features they want, till a detailed implementation of it can be integrated first-class into multilspy.

Please let me know if you have any questions!

[jbellis]

Good timing with #47 -- it will be easy to add outgoing calls on top of that.