ivy-switch-buffer: calling current-buffer soon afterward returns old buffer

[mehw]

Hello.

Could someone confirm the following, please?

Description

Switching buffer from A to B via ivy-switch-buffer changes the buffer to B as expected, but calling current-buffer soon afterward returns A not B!

Versions affected

• ivy MELPA 20180917.2237
• https://github.com/abo-abo/swiper/tree/71b60a72d30cf110ed906ea744cbe273869130d2

How to replicate

Register the command testme-ivy-switch-buffer below, then run it via M-x testme-ivy-switch-buffer. Then, try to switch to a different buffer and take note of the printed message.

(defun testme-ivy-switch-buffer ()
  (interactive)
  (ivy-switch-buffer)
  ;; NOTE: Delay `message' with `run-with-timer' to get proper results,
  ;; i.e. (run-with-timer 0.5 nil (lambda () (message ...))).  On my
  ;; system just 0.00005 seconds are enough to get proper results.
  (message "(current-buffer): %s | (window-buffer (selected-window)): %s"
           (current-buffer) (window-buffer (selected-window))))

Expected odd results:

• (current-buffer) returns the old buffer.
• (window-buffer (selected-window)) returns the buffer switched to.

Implementation notes

There's a sort of oddity in ivy-call:

(with-current-buffer (ivy-state-buffer ivy-last)
  (unwind-protect (funcall action x)
    (ivy-recursive-restore)))

When action is ivy--switch-buffer-action, it switches buffer to x, but the wrapping with-current-buffer set it back to ivy-last's registered :buffer... the old buffer.

Proposed solution

Excuse me for not opening a PR... my bad...

I pushed a commit on a new branch and there should be a referenced solution right here.

[basil-conto]

Ccing @raxod502 who had some insights on related code in #1607.

[basil-conto]

@mehw Can you please review #1767?

[basil-conto]

the wrapping with-current-buffer set it back to ivy-last's registered :buffer... the old buffer.

Which is why I think this with-current-buffer can/should be removed, though I'm not 100% sure.

[mehw]

Ccing @raxod502 who had some insights on related code in #1607.

Please, also see @ericdanan 717#issuecomment-321920963.

@mehw Can you please review #1767?

It seems to work. Thanks!

Which is why I think this with-current-buffer can/should be removed, though I'm not 100% sure.

Peer reviews will be helpful... Please, kindly review #1767.

[abo-abo]

Thanks, all.