markdown-enable-math interferes with flymake-goto-next-error in an eglot managed buffer

[ywwry66]

Hi there,

Thanks for working on this great project!

I recently found a wierd issue when using eglot with markdown-enable-math set to t. Specifically, the error messages in the echo area produced by flymake-goto-next-error only flash for a second and disappear.

The following example is reproducible on my local environment (macOS Sonoma, latest stable Emacs and R, languageserver and lintr R-packages, markdown-mode from melpa, built-in eglot and flymake):

1. emacs -Q
2. Evaluate

(require 'package)
(package-initialize)
(setq-default markdown-enable-math t)

3. Open R script containing

## test
A <- 1

4. Run eglot
5. Run flymake-goto-next-error

I don't know whether this issue is caused by eglot, flymake, or markdown-mode. But I think you are the most knowledgeable in this regard and thus raised my question here.

[joaotavora]

I don't have the R tool chain, sorry. Maybe you'll have luck with someone else here. Maybe you can add debug-on-entry to the message function to know who is calling it to clear the echo area. But it's hard to get it to not interfere with normal Emacs usage to much.

…

On Sun, Apr 28, 2024, 20:01 Ruiyang Wu ***@***.***> wrote: Hi there, Thanks for working on this great project! I recently found a wierd issue when using eglot with markdown-enable-math set to t. Specifically, the error messages in the echo area produced by flymake-goto-next-error only flash for a second and disappear. The following example is reproducible on my local environment (macOS Sonoma, latest stable Emacs and R, languageserver and lintr R-packages, markdown-mode from melpa, built-in eglot and flymake): 1. emacs -Q 2. Evaluate (require 'package) (package-initialize) (setq-default markdown-enable-math t) 3. Open R script containing ## testA <- 1 4. Run eglot 5. Run flymake-goto-next-error I don't know whether this issue is caused by eglot, flymake, or markdown-mode. But I think you are the most knowledgeable in this regard and thus raised my question here. — Reply to this email directly, view it on GitHub <#1394>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAC6PQZGE5XZFFPO7RFRSMLY7VBP7AVCNFSM6AAAAABG5GLDOOVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZWGU3TMNBUGI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ywwry66]

Thanks for the tips. I was able to figure out the reason:

Running flymake-goto-next-error produces the following call sequence eglot-hover-eldoc-function -> eglot--hover-info -> eglot--format-markup -> gfm-view-mode -> gfm-mode -> markdown-mode. If markdown-enable-math is set, markdown-toggle-math is called, which messages "markdown-mode math support enabled".

What do you think is the proper way of fixing this?

[joaotavora]

What do you think is the proper way of fixing this?

I don't know honestly, but something like having Eglot inhibit messages inside the callees of eglot--format-markup, which is possible at least in Emacs 29.1 (maybe let-binding set-message-functions, which see).

[joaotavora]

Another equally good way, perhaps even more idiomatic, is for markdown-toggle-math to not call message if it' snot being invoked interactively.

[ywwry66]

I don't know honestly, but something like having Eglot inhibit messages inside the callees of eglot--format-markup, which is possible at least in Emacs 29.1 (maybe let-binding set-message-functions, which see).

It seems the inhibit message mechamism alway clears the echo area so that does not help. But I was able to use set-message-functions to fixed problem. The idea is to record what was already on the echo area and always output that message thereafter.

  (defun existing-message-advice (fun &rest args)
    (let* ((existing-message (current-message))
	   (set-message-functions (list (lambda (msg) existing-message))))
      (apply fun args)))
  (advice-add 'eglot--format-markup :around #'existing-message-advice)

Of course this line needs to be removed:

eglot/eglot.el

Line 1812 in db91d58

(let ((inhibit-message t)

Another equally good way, perhaps even more idiomatic, is for markdown-toggle-math to not call message if it' snot being invoked interactively.

You are right this is actually better. I will open a pull request upstream.

[ywwry66]

Fixed by upstream.