Off-by-one in counsel-yank-pop selection

[ambihelical]

If I understand the recent changes correctly, the current kill-ring-yank-pointer position should be preselected. This seems to always be one off. If I have three lines which I've killed in order:

one
two
three

After killing "one", the counsel-yank-pop shows "one" and it is preselected.
After killing "two":

    two <- should be selected
    one <- selected

After killing "three":

    three  <- should be selected
    two  <- selected
    one

When I use the "rotate" command to change the current kill-ring-yank-pointer, it never sets the selection to the item selected, but is always one off. If I select "three", and use rotate (M-o r) to set the pointer, the next counsel-yank-pop shows "two" selected. If I select "two" and rotate, the next counsel-yank-pop shows "one" selected. If I select "one", the next invocation shows "three" selected.

For this issue I was using commit 5c096a2.

[abo-abo]

@basil-conto Can you have a look? Off-by-one results from the change in interactive spec - arg is usually 1 when it's expected to be zero.

[basil-conto]

TL;DR

Preselecting the next-to-last kill is intended.

Rationale

The recent changes to counsel-yank-pop try to mirror the default yank, yank-pop, rotate-yank-pointer, etc. functionality as close as possible, whilst providing convenient realtime visualisation/completion and custom actions via the Ivy/Counsel interface.

As (emacs) Yanking explains in further detail, the general idea is that yank retrieves the last kill, whereas yank-pop is for accessing earlier kills. This is also summarised in the docstring of yank-pop:

yank-pop is an interactive compiled Lisp function in ‘simple.el’.

It is bound to M-y.

(yank-pop &optional ARG)

Replace just-yanked stretch of killed text with a different stretch.
This command is allowed only immediately after a ‘yank’ or a ‘yank-pop’.
At such a time, the region contains a stretch of reinserted
previously-killed text.  ‘yank-pop’ deletes that text and inserts in its
place a different stretch of killed text.

With no argument, the previous kill is inserted.
With argument N, insert the Nth previous kill.
If N is negative, this is a more recent kill.

The sequence of kills wraps around, so that after the oldest one
comes the newest one.

This command honors the ‘yank-handled-properties’ and
‘yank-excluded-properties’ variables, and the ‘yank-handler’ text
property, in the way that ‘yank’ does.

The optional prefix argument defaults to 1, meaning yank the next-to-last kill. If you want to yank the last kill, you should pass it a prefix argument of zero.

The current implementation of counsel-yank-pop mirrors this by preselecting what vanilla yank-pop would have yanked. So it doesn't "select" the "wrong" candidate, it "preselects" the "default" candidate.

[abo-abo]

@basil-conto Thanks for the explanation. It makes sense in a way.

Yet in another way it doesn't. Using M-y RET instead of C-y seems like a nice use case. Of course, if you did M-w seconds ago, you know what's on the top of the kill ring and you can do C-y directly. But if you're just fairly sure that the thing you want is on the top of the kill ring, M-y RET gives you a preview, an option to back out with C-g, and an option to select something else.

Currently, the above use case was moved to the M-y C-p RET binding. I'm not sure it's worth the consistency with the defaults.

[basil-conto]

It makes sense in a way. Yet in another way it doesn't.

If you'll pardon the facetious retort, it doesn't make sense to discuss whether this makes sense. :)

Here's my understanding of why that is. Please to confirm/comment.

• Emacs provides intricate, central and well-documented idioms and interfaces around killing and yanking, with defaults and conventions which support a particular workflow.
• Counsel can extend the user-friendliness and flexibility of many such systems via Ivy filtering, actions, etc.
• The command counsel-yank-pop previously did no preselecting, meaning the first completion candidate was preselected by default.
• The command's recent reimplementation was intended to reuse the existing Emacs system but make it more Counsel-y. Part of using the existing system is respecting the yank-pop idiom of doing what's in the name and popping the kill-ring. This was not just to satisfy the author's compulsive pedantry, but also to satisfy a user request (#1190).
• Now, users (and owner :) seem (because no-one has stated it clearly) to be requesting that Counsel should return to and/or also support preselecting the last kill.

My conclusion is that, as is always the case, different people have different preferences, so we should simply do the Emacs-y and Counsel-y thing and add another user toggle which defaults to the old, non-Emacs behaviour, right?

[dieggsy]

@basil-conto A user toggle/custom var seems like a good solution here

[abo-abo]

@basil-conto You're right. counsel-yank-pop does say

"Ivy replacement for `yank-pop'.

Could you add a defcustom for this, with the current behavior as the default?

[basil-conto]

@abo-abo

Could you add a defcustom for this

I can't promise I can, but I can promise to try. :)

with the current behavior as the default

Are you sure you don't prefer the old behaviour as default?

[abo-abo]

Are you sure you don't prefer the old behaviour as default?

I was planning to customize it in my config, in the name of counsel-yank-pop staying true to its name.

We can change it later if needed.

[articuluxe]

@basil-conto I just wanted to say, thanks for this change. It makes this feature usable for me and it is a big improvement IMO to make the functionality consistent with standard emacs usage.

[basil-conto]

@articuluxe Glad it's useful for you; thanks are due to others, though. :)