Add runtime language configuration (#1794)

[zen3ger]

• Add set-language typable command to change the language of current buffer.
• Add completer for available language options.

Closes #1794

[kirawi]

I think creating a function like language_config_for_scope would be more homogenous with the code.

Edit: Ah, you're using fuzzy finding. Never mind.

[zen3ger]

Yes. 🙂 Though, now that you mentioned it, I could've added a language_config_for_id functions which could've been useful in the actual command implementation...

[archseer]

Rather than a custom implementation, set_language2 already takes a scope name and looks up the appropriate language. See the discussion on #1787

[kirawi]

Maybe a set_language3 that takes the name of a lang? doing :lang scope.rust isn't intuitive IMO.

[archseer]

set_language2(format!("scope.{}", language),...)

[kirawi]

If that's the approach taken, then the fuzzy finding for the languages should be changed to look through the scopes rather than the language names. There are a few langs where the names and scope endings aren't 1-to-1 (e.g. git-diff v. scope.diff)

[zen3ger]

I didn't use scope because the prefixes don't seem to follow the source.<lang> format all the time and the language id doesn't always match with suffix either.
Here's a list of those that don't match with language id:

• protobuf source.proto
• c-sharp source.csharp
• javascript source.js
• typescript source.ts
• html text.html.basic
• latex source.tex
• ocaml-interface source.ocaml.interface
• racket source.rkt
• comment scope.comment
• llvm-mir source.llvm_mir
• llvm-mir-yaml source.yaml
• markdown source.md
• git-commit git.commitmsg
• git-diff source.diff
• git-rebase source.gitrebase
• git-config source.gitconfig
• solidity source.sol

Edit: I agree that adding set_language3 that takes the language id would be nicer, also adding syntax::Loader::language_config_for_language_id() could be added to not poke into other structs too much.

[archseer]

Similar to #1787 you also need to call refresh_language_server to change language servers. The function also needs to be fixed: it the server changes we need to send a text document close to the old one.

[zen3ger]

The text document close seems to be already handled in Editor::launch_language_server(), more specifically from L464:

        if let Some(language_server) = language_server {
            // only spawn a new lang server if the servers aren't the same
            if Some(language_server.id()) != doc.language_server().map(|server| server.id()) {
                if let Some(language_server) = doc.language_server() {
                    tokio::spawn(language_server.text_document_did_close(doc.identifier()));
                }

                let language_id = doc.language_id().map(ToOwned::to_owned).unwrap_or_default();

                // TODO: this now races with on_init code if the init happens too quickly
                tokio::spawn(language_server.text_document_did_open(
                    doc.url().unwrap(),
                    doc.version(),
                    doc.text(),
                    language_id,
                ));

                doc.set_language_server(Some(language_server));
            }
        }

The only fixing that's needed for Editor::refresh_language_server() is to check if user already set a language or not, otherwise the detection will overwrite the user set value.

     pub fn refresh_language_server(&mut self, doc_id: DocumentId) -> Option<()> {
         let doc = self.documents.get_mut(&doc_id)?;
-        doc.detect_language(self.syn_loader.clone());
+        // try detection only if language is not already set for this document.
+        if doc.language.is_none() {
+            doc.detect_language(self.syn_loader.clone());
+        }
         Self::launch_language_server(&mut self.language_servers, doc)
     }

I'm not sure if this check should be part of Document::detect_language() or not.

[zen3ger]

I've played around with using set_language2 and the scope names, but as kirawi pointed out using the whole scope name is not user friendly, and looking at the list I've collected where the scope names don't follow the source.<lang> format, I think using the language_id is the least hackish way of doing this.

In the last commit I've added a set_language3 which does the look-up based on the id, and some other smaller helpers to keep things more aligned with what's already in the code base... If set_language2 is still the way to go, then I'm happy to rework this if someone can "enlighten" me how to go about the oddball scopes. 🙂

[archseer]

Just one minor comment

[archseer]

Nice work!