Process indicator as part of the lighter

[minad]

Since we've got this lighter, would it make sense to add a process indicator there too, which tells if jit-spell is active or not? The indicator should be implemented such that it doesn't introduce movement of the mode line or flickering. See for example the indicator in consult-ripgrep which switches from : to * and back, such that the prompt doesn't move.

[astoff]

Ah, cool idea. That one I agree will require some caching.

[astoff]

I've experimented with this, but at least in normal circumstances the busy state is so short that it just causes a little flicker in the mode line. Only if I set the wrong language in order to get lots of misspellings I see some fun stuff going on. So it looks like this is not a practical feature to provide.

[minad]

Did you try stealth mode?

[astoff]

I've created a branch with the mode line counter: https://github.com/astoff/jit-spell/tree/modeline-counter. I maintain my impression this is not a useful thing for the user, but it's interesting for debugging.

I noticed that without stealth mode, jit-lock does what you expect: fontify the area near the window and then stop. On the other hand, enabling stealth mode causes even a 1 character change in a comment to trigger many checks. I suppose you're getting the entire buffer refontified eventually.

[minad]

I noticed that without stealth mode, jit-lock does what you expect: fontify the area near the window and then stop.

Maybe I should just give up on stealth mode. I believe it is nice for font locking only. But still, I think #9 should be fixed. jit-spell has an efficiency problem, which I find hard to justify for a mode which I turn on everywhere.

[astoff]

Out of curiosity, why did you turn stealth mode on in the first place?

[minad]

Out of curiosity, why did you turn stealth mode on in the first place?

I've seen it recommended somewhere as a performance optimization (maybe Doom Emacs, could be just cargo culting). So far I never observed bad side effects of it. But maybe it is a useless optimization, since I find it also hard to observe positive effects. :-P

But even if I disable stealth locking, jit-spell is still loading my system too heavily. Aspell takes around 3% and Emacs takes 90%. Without jit-spell this does not happen.

[astoff]

Can you reproduce on emacs -Q? I have no tweaks whatsoever to GC, pipe buffer size, etc. I'm not even using native compilation. When I see high CPU usage, it's on the spellchecker side.

[minad]

Yes, I see the high load of Emacs with the following snippet starting from emacs -Q. It suffices to open a file like simple.el or embark.el. Or go to an open buffer of simple.el and turn jit-spell-mode off and on again. I have Emacs open and top next to it. As soon as I toggle jit-spell-mode, the load produced by Emacs increases dramatically.

(package-initialize)

(add-hook 'prog-mode-hook 'jit-spell-mode)

;; Causes high load until the whole buffer is fontified.
(setq jit-lock-stealth-time 0.01 ;; aggressive, not my actual setting
      jit-lock-stealth-nice 0.01 ;; aggressive, not my actual setting
      jit-lock-stealth-load 100
      jit-lock-chunk-size 3000)

;; Reduce load of aspell ~1%. Emacs is high > 90%.
(setq ispell-extra-args '("--sug-mode=ultra"))

;; Doesn't seem to make a difference.
;; (setq read-process-output-max (* 64 1024))

EDIT: I can reproduce this pretty reliably if I open org.el and toggle jit-spell-mode. I use overly aggressive setting in the recipe, in comparison to my actual configuration. I notice that both Emacs and Xorg process load go up. This may be specific to my installation, however it supports the hypothesis that jit-spell causes many redisplays while it is processing the aspell output.

(package-initialize)

;; Causes high load until the whole buffer is fontified.
;; All of the settings are overly aggressive.
(setq jit-lock-stealth-time 0.01
      jit-lock-stealth-nice 0.01
      jit-lock-stealth-load 1000
      jit-lock-chunk-size 10000)

;; Reduce load of aspell ~1%. Emacs is high > 90%.
(setq ispell-extra-args '("--sug-mode=ultra"))

[astoff]

On a tangent, the mode line seems to ignore the text properties returned from an (:eval ...) construct. How do I get around this?