Advice of `set-window-buffer` unconditionally enables `evil-local-mode`

[condy0919]

Issue type

• Bug report

Environment

Emacs version: Emacs 29.0.50
Operating System: archlinux
Evil version: 1.14.0
Evil installation type: melpa
Graphical/Terminal: GUI
Tested in a make emacs session (see CONTRIBUTING.md): No

Reproduction steps

• Start Emacs

Eval the following recipe.

(package-initialize)
(setq evil-want-minibuffer nil)
(require 'evil)
(evil-mode +1)

(defun test-setup ()
  (let* ((temp (generate-new-buffer "*temp*"))
         (win (with-minibuffer-selected-window (display-buffer temp))))
    (message "BEFORE %S" evil-local-mode)
    (set-window-buffer win (current-buffer))
    (message "AFTER %S" evil-local-mode)
    (kill-buffer temp)))

(defun test ()
  (interactive)
  (minibuffer-with-setup-hook #'test-setup
    (read-from-minibuffer "Prompt: ")))

M-x test

Expected behavior

ESC could exit the minibuffer.

Actual behavior

ESC is bound to evil-force-normal-state since evil-local-mode is enabled.

Further notes

It's originally reported at minad/vertico#171, minad has identified the bug:

evil/evil-core.el

Lines 362 to 373 in 652d726

	;; When a buffer is created in a low-level way, it is invisible to
	;; Evil (as well as other globalized minor modes) because no hooks are
	;; run. This is appropriate since many buffers are used for throwaway
	;; purposes. Passing the buffer to `set-window-buffer' indicates
	;; otherwise, though, so advise this function to initialize Evil.
	(defadvice set-window-buffer (before evil)
	"Initialize Evil in the displayed buffer."
	(when evil-mode
	(when (get-buffer (ad-get-arg 1))
	(with-current-buffer (ad-get-arg 1)
	(unless evil-local-mode
	(evil-local-mode 1))))))

The advice of set-window-buffer unconditionally enable evil-local-mode. IMHO, the buffer (ad-get-arg 1) shouldn't be evilfied if it satisfies (minibufferp) and evil-want-minibuffer is t.

See

evil/evil-core.el

Lines 167 to 178 in 652d726

	;; The function `evil-initialize' should only be used to initialize
	;; `evil-local-mode' from the globalized minor-mode `evil-mode'. It is
	;; called whenever evil is enabled in a buffer for the first time or
	;; when evil is active and the major-mode of the buffer changes. In
	;; addition to enabling `evil-local-mode' it also sets the initial
	;; evil-state according to the major-mode.
	(defun evil-initialize ()
	"Enable Evil in the current buffer, if appropriate.
	To enable Evil globally, do (evil-mode 1)."
	(unless (and (minibufferp) (not evil-want-minibuffer))
	(evil-local-mode 1)
	(evil-initialize-state)))

[Delapouite]

Hi

Sorry in advance if I hijack the thread but I have a feeling that what I observed is related to this topic.

I have many org-mode buffers with tables with about 100 lines and 5 columns (so not that huge).
Every time I press the Tab key in order to trigger the realignment of cells after some editing, the operation is verryy slow (about 15 secs) and my laptop fans start spinning.

Here's an extract of a profiler-report, just analyzing what's going on :

10436  98% - command-execute
10436  98%  - call-interactively
 9846  92%   - funcall-interactively
 9839  92%    - org-cycle
 9836  92%     - call-interactively
 9836  92%      - funcall-interactively
 9836  92%       - org-table-next-field
 9836  92%        - org-table-align
 5028  47%         - cl-mapcar
 5024  47%          - cl--mapcar-many
 5024  47%           - apply
 5014  47%            - org-table--align-field
 5011  47%             - org-string-width
 4749  44%              - set-window-buffer
 4749  44%               - apply
 4749  44%                - ad-Advice-set-window-buffer
 4708  44%                 - evil-local-mode
 1936  18%                  + run-hooks
 1563  14%                  + add-hook
 1185  11%                  + evil-initialize-state
    3   0%                  + evil-concat-lists
    3   0%                  + evil-disabled-buffer-p
    3   0%                    evil-initialize-local-keymaps
   41   0%                 + #<subr set-window-buffer>
  235   2%              + #<compiled -0x1c8ae47fe622b71b>
   13   0%              + version<
 4795  45%         - org-string-width
 4564  43%          - set-window-buffer
 4564  43%           - apply
 4564  43%            - ad-Advice-set-window-buffer
 4517  42%             - evil-local-mode
 1870  17%              + run-hooks
 1547  14%              + add-hook
 1084  10%              + evil-initialize-state
    4   0%              + evil-concat-lists

So my current understanding is that to perform string width calculation, a throw-away buffer is created, but as an extraneous evil-local-mode is triggered, it slows down everything.

I guess in this context, the advice should not enable evil-local-mode as well.

Thanks