Find definitions/references

[emacs-tree-sitter]

Find definitions/references

[shackra]

that seems to work only for the current buffer, i.e.: if a reference is defined on another file and that file is not visited by Emacs then the implementation will move you to the first reference of that symbol in the same file.

Let me ask you something: do we really want such feature? (if we don't want it, you can always write a package and publish it) I see you are part of emacs-lsp, which makes me wonder what is wrong with lsp-mode that having such feature with help of emacs-tree-sitter seems so attractive?

[leungbk]

Let me ask you something: do we really want such feature? (if we don't want it, you can always write a package and publish it) I see you are part of emacs-lsp, which makes me wonder what is wrong with lsp-mode that having such feature with help of emacs-tree-sitter seems so attractive?

There are a couple of things I like about tree-sitter here:

1. I would expect tree-sitter-based symbol-at-point-highlighting to be faster than what we do in lsp-mode, which is to make a textDocument/documentHighlight request.
2. I would also expect a tree-sitter-based function that goes to the next occurrence in the buffer of the symbol at point to be faster than lsp-ui-find-next-reference, which makes a new textDocument/references request each time we call it.

that seems to work only for the current buffer, i.e.: if a reference is defined on another file and that file is not visited by Emacs then the implementation will move you to the first reference of that symbol in the same file.

xref is expected to support falling back on different backends someday. So having tree-sitter-based definition-jumping should be worth it if/when that day comes.

[theHamsta]

Why should copying our implementation be a problem for GNU MELPA? Isn't the actual problem for you the implementation in Lua? Our implementation could be improved regarding speed (better caching and look-in by hash tables) for using it for semantic highlighting (not only highlighting of current symbol but resolving the definition for queries that ask for the definition of a symbol). It's nevertheless very fast. We also helped with the draft of sidekick https://github.com/ElPiloto/sidekick.nvim which might be worth to look at.

The feature is handy for smart rename for languages that don't have a language server.

[leungbk]

Why should copying our implementation be a problem for GNU MELPA?

I've asked the #emacs and #fsf IRC channels about whether code released under a GPL-compatible license such as nvim-tree-sitter's Apache 2.0 license can be copied and included as part of Emacs without further steps (I think ELPA qualifies as "part of Emacs"). I did not receive an answer from anyone with any legal training, but some of the other people there who had contributed to Emacs speculated that for the code to be included, the people who originally wrote the copied code would need to sign papers -- presumably authorizing the FSF to redistribute the code under GPL 3+ or something.

Thanks for the implementation hints. If I ever get an affirmative answer on the whole copyrighting issue, I'll study your guys' code.

[theHamsta]

@leungbk we know all the contributers and could ask them about dual licensing if this is an issue.