Null language id makes rust-analyzer crash

[roife]

I have found that rust-analyzer often crashes when used together with eglot. Upon investigation, I discovered that this is because the languageId parameter in the didOpen request of eglot is set to null, causing rust-analyzer to panic during unwrapping.

Further investigation revealed that this is due to the request issued by eglot--maybe-activate-editing-mode in the after-change-major-mode-hook. The function normal-mode in lisp/files.el indicates that after-change-major-mode-hook is executed before set-auto-mode, causing the current major mode to be fundamental-mode at this point. As eglot cannot find the languageId for fundamental mode, it sends null as the languageId. However, according to the LSP spec, the languageId parameter must be a String, not String?, so we cannot send null.

In any case, eglot may need to set a default value for languageId, such as "Unknown".

[joaotavora]

Analysis is nice, but reproductions is better. I've reproduced this in one specific situation, using extend-to-xref. Would you mind sharing what you are trying to do following https://joaotavora.github.io/eglot/#Troubleshooting-Eglot?

[roife]

Well, here's the event buffer: https://pastebin.mozilla.org/YYbMbQmf.

Every time I use go-to-definition and this operation jumps to a file that was not previously open, I encounter this issue. Currently, I am using the following solution to override an existing function to address the problem:

(cl-defun eglot--languageId (&optional (server (eglot--current-server-or-lose)))
    "Compute LSP \\='languageId\\=' string for current buffer.
Doubles as a predicate telling if SERVER can manage the current
buffer."
    (or (cl-loop for (mode . languageid) in
                        (eglot--languages server)
                        when (provided-mode-derived-p major-mode mode)
                        return languageid)
	"Unknown"))

While this temporarily resolves the issue, the language server reports a duplicate DidOpenTextDocument error. I believe this is because the buffer is sending two didOpen requests when set to fundamental-mode and rust-mode.

A more detailed error report would require some time for me to write.

[joaotavora]

A more detailed error report would require some time for me to write.

When you don't spend that time, I will have to spend it in guesswork. So it's well invested. Let's hope I made the right guess in emacs.git af1fe69f05d803a6958f9d8a045d1013e2ce785c, master branch. I probably have since according to your event log, your M-. to a file outside your project, which was my conjecture, too. The fix is available from GNU-devel ELPA, if you want to fetch Eglot from there. There is no duplicate DidOpen request in that fix.

[roife]

Oh, you are right, I have made some attempts, but there were a few times when I couldn't reproduce it completely, so I am also investigating this issue. I never thought that the reason was that I was jumping into files (libraries) outside of the project. Thank you for your efforts 🙏.