Stored registers are removed on multi cursor creation

[duianto]

Observed behavior

Stored registers are removed when multiple cursors are created.

Expected behavior

That stored registers remained after cursor creation.

Steps to reproduce

• Start Emacs
• Open the scratch buffer
• Type: abc
• Store abc in register a: "ayiw
• Look at the registers: :reg RET, there an entry for:

a          abc

• Switch back to the scratch buffer: C-w w
• Create a second cursor: grh
• Look at the registers again, the entry for register a is gone.

Notes

Any registers that were stored while multiple cursors were active. Remain after pausing, resuming or undoing the cursors.

System information

Evil-mc Version: 0.0.4 (20180921.1727)
Evil version evil-git-9865e7731
GNU Emacs 26.1 (build 1, x86_64-w64-mingw32) of 2018-05-30
Windows 10 Version 1803

[dchrzanowski]

Same spotted on Linux Emacs 26.1.

[dchrzanowski]

If someone has a good elisp fu, then perhaps an easy solution would be to make a copy of all text registers before the multiple cursors are made. Then when the cursors are killed the registers are restored...food for thought.

EDIT:
For the time being I am removing register-alist here and then byte-recompiling the file. Works just fine, it's not that I need named registers while multiple cursors are active anyway.

[duianto]

Update

No, the "fix" below only works when creating additional cursors with (evil-mc-make-cursor-here).
The named registers are still removed if one creates a cursor with for example:
(evil-mc-make-and-goto-next-match).

A to limited solution

I might have found a possible solution that doesn't remove the named registers.
By wrapping the call to (evil-mc-cursors-before) in a temporary register-alist variable.

(defun evil-mc-run-cursors-before ()
  "Runs `evil-mc-cursors-before' if there are no cursors created yet."
  (when (not (evil-mc-has-cursors-p))
    (let (register-alist)
      (evil-mc-cursors-before))))

The optimal solution would be to be able to paste from a register at every cursor, but with this "fix" it only pastes from a register at the main cursor.

But the major issue of not removing the named registers is probably enough for now.

[sooqua]

If someone has a good elisp fu, then perhaps an easy solution would be to make a copy of all text registers before the multiple cursors are made. Then when the cursors are killed the registers are restored...food for thought.

EDIT:
For the time being I am removing register-alist here and then byte-recompiling the file. Works just fine, it's not that I need named registers while multiple cursors are active anyway.

(setq ~+multiple-cursors--evil-mc-registers-copy '())

(defun ~+multiple-cursors*evil-mc-run-cursors-before (&rest r)
  (cl-loop for reg from 97 to 122 do
           (setq ~+multiple-cursors--evil-mc-registers-copy (nconc ~+multiple-cursors--evil-mc-registers-copy (cons (get-register reg) nil)))))

(defun ~+multiple-cursors*evil-mc-run-cursors-after (&rest r)
  (cl-loop for reg in ~+multiple-cursors--evil-mc-registers-copy
           for i from 97 do
           (when reg
             (set-register i reg)))
  (setq ~+multiple-cursors--evil-mc-registers-copy nil))

(advice-add #'evil-mc-run-cursors-before :before #'~+multiple-cursors*evil-mc-run-cursors-before)
(advice-add #'evil-mc-run-cursors-after :after #'~+multiple-cursors*evil-mc-run-cursors-after)

[duianto]

@sooqua
That only seems to work with:
evil-mc-make-and-goto-next-match and evil-mc-make-and-goto-prev-match

It doesn't work with:
evil-mc-make-cursor-here or evil-mc-make-all-cursors

And there might be an off by one issue, because while I was storing text in the a, b and c registers and creating evil-mc cursors, then it would sometimes show that the same text was stored in the } and | registers.

[duianto]

The author of this Spacemacs issue:
Problem persisting registers syl20bnr/spacemacs#12606

Says:

On exiting spacemacs the register list seems to persist alphabetic and numeric registers differently, as regards the buffer name.

In the - rather longer than is probably necessary - line below from .emacs.desktop the numeric registers seems to be stored along with the full path name of the buffer. Whereas a lot of registers seem to refer only to the file name.

That might be why the alphabetic registers are erased when evil-mc cursors are created.

[sooqua]

Quick fix:

(defun ~+multiple-cursors*evil-mc-write-cursor-state (state)
  "Write the state of the real cursor with values from STATE."
  (let ((names (evil-mc-get-cursor-variables)))
    (dolist (name names)
      (when (boundp name)
        (let ((p (evil-mc-get-cursor-property state name)))
          (when (not (and (string= name "register-alist") (null p)))
            (set name (evil-mc-get-cursor-property state name))))))))
(advice-add #'evil-mc-write-cursor-state :override #'~+multiple-cursors*evil-mc-write-cursor-state)

[duianto]

That seems to work for all evil-mc cursor creation commands 👍

It seems to also keep the 0 register from being deleted.
There's an open issue about it:
Nils out evil-was-yanked-without-register #70

[duianto]

Could that solution or something similar be implemented to also prevent the position markers from being removed when evil-mc cursors are created?

Position markers (I might not be the correct name), can be store with ma, mb, etc.
And jumped to with:

• `a (backtick and a) to jump to the character where the mark was set.
• 'a (single quote and a) to jump to the beginning of that markers line

[sooqua]

Yeah. I'm not sure why (null p) check doesn't work with evil-markers-alist, it seems like it doesn't write null in there but something breaks anyway.

(defun ~+multiple-cursors-evil-mc-write-cursor-state-a (state)
  "Write the state of the real cursor with values from STATE."
  (let ((names (evil-mc-get-cursor-variables)))
    (dolist (name names)
      (when (boundp name)
        (let ((p (evil-mc-get-cursor-property state name)))
          (when (not
                 (or
                  (eq name 'register-alist)
                  (eq name 'evil-markers-alist)))
            (set name p)))))))
(advice-add #'evil-mc-write-cursor-state :override #'~+multiple-cursors-evil-mc-write-cursor-state-a)

[sooqua]

You could probably just remove register-alist and evil-markers-alist from evil-mc-cursor-variables at this point instead of overriding evil-mc-write-cursor-state, for some negligible performance benefit, but I don't know what these variables are used for anyway.

(when (featurep! :editor multiple-cursors)
  (after! evil-mc
    (setq evil-mc-cursor-variables (mapcar
                                    (lambda (s)
                                      (remove 'register-alist (remove 'evil-markers-alist s)))
                                    evil-mc-cursor-variables))))

I'm using Doom Emacs, you will have to modify everything before and including after! to run this code only after evil-mc is loaded.

[duianto]

This seems to work to prevent the removal of:

• alphabetic registers
• position markers
• the 0 register

(with-eval-after-load 'evil-mc
  (setq evil-mc-cursor-variables
        (mapcar
          (lambda (s)
            (remove 'register-alist
                    (remove 'evil-markers-alist
                            (remove evil-was-yanked-without-register s))))
          evil-mc-cursor-variables)))

Thanks