bug#74437: 30.0.92; completion-preview-idle-delay is delayed by flyspell

[Eshel Yaron via "Bug reports for GNU Emacs, the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>]

Hi there,

Tomasz Hołubowicz writes:

> Hello,
>
> This is my configuration for completion-preview:
>
> (use-package completion-preview
>   :hook (after-init . global-completion-preview-mode)
>   :bind (:map completion-preview-active-mode-map
>               ("M-n" . completion-preview-next-candidate)
>               ("M-p" . completion-preview-prev-candidate))
>   :custom (completion-preview-idle-delay 0.2)
>   :config
>   ;; Org mode has a custom `self-insert-command'
>   (push 'org-self-insert-command completion-preview-commands)
>
>   ;; Hide `completion-preview' overlay when invoking `completion-in-region'
>   (with-eval-after-load 'consult
>     (advice-add 'consult-completion-in-region :before (lambda (&rest _) (completion-preview-hide)))))

Side note: it is better to use (completion-preview-active-mode -1)
instead of (completion-preview-hide) to dismiss the preview, since
disabling the minor mode performs additional cleanup.  Perhaps I should
mention that in the documentation somewhere... :)

> When I enter org-mode file and flyspell-mode is toggled on then the
> completion-preview-idle-delay is much longer in reality that 0.2s. After
> I disable flyspell-mode, then the delay is truly 0.2s. It
> seems that completion-preview is being delayed by flyspell checking
> currently written word. It makes sense if we consider 'idle' literally,
> but I'm not sure if this is desired.

Thank you for this report.  I can reproduce the extended delay here.
(In the future, please provide a minimal recipe to reproduce the issue
starting from emacs -Q, if possible.)

The root cause is a peculiarity of flyspell-mode: it calls sit-for and
blocks Emacs for 3 whole seconds (by default, see flyspell-delay) after
certain commands, including after (org-)self-insert-command.  This also
blocks the idle timer that Completion Preview mode uses, unfortunately.

A quick search shows that this behavior of flyspell affects other
features as well.  For example, IIUC, auto completion in Corfu switched
to using run-at-time instead of run-with-idle-timer due to this issue.

I think flyspell should be modified to use a timer instead of sit-for,
so as to avoid blocking idle timers.  I can come up with such a patch,
but it's not quite trivial, so I wonder what others think about this
issue and how it should be addressed.

As a stopgap, you can try setting flyspell-delay to 0.


Best,

Eshel