Inconsistent state after some misterious crash. Had to use eglot--managed-mode

[tromey]

I was using eglot with Rust and "something" happened -- I don't know exactly what but I assume that RLS crashed.

After this, eglot ended up in a state where I could do very little. For example, I could not M-x revert-buffer, because some eglot hook function errored (due IIRC the process being dead). I couldn't kill eglot-managed buffers, either. M-x eglot-shutdown did not help.

In the end I skimmed through the eglot source and did M-x eglot--managed-mode.

I think eglot should be more careful about process failures and avoid allowing exceptions to unwind past the boundaries of hook functions.

[joaotavora]

I think eglot should be more careful about process failures and avoid allowing exceptions to unwind past the boundaries of hook functions.

I agree to the first part, though I don't fully understand the second. Let's keep this open and mark the eglot--managed-mode a workaround for such situations for now.

[joaotavora]

By the way, you can try killing RLS or otherwise making it crash on purpose. Eglot should recover automatically, if it doesn't please report here.

[tromey]

I don't fully understand the second

For example, in my case I think I found that the eglot function that is attached to kill-buffer-hook was throwing an exception. This was propagating out of the function and so it was terminating the kill-buffer. But I think it would be better not to stop kill-buffer from completing -- so the hook function would have to use condition-case (or ignore-errors, which is simpler) to avoid this.

If I reproduce this again I will be sure to attach more information to this bug.

[joaotavora]

(or ignore-errors, which is simpler) to avoid this.

Or with-demoted-errors so, you can debug the thing when it happens (if you happen to have debug-on-error turned on).