Re: Recentish C-s M-y change

[Juri Linkov]

> I understand, but the consistency is broken anyway: C-u has an effect on
> yank, and no effect on isearch-yank-kill.

There is a difference between no effect and a completely different effect.

> And mirroring the effect of C-u on yank (exchanging point and mark) in
> isearch-yank-kill is not possible.

It's still possible to give C-s C-u C-y the meaning similar to C-u C-y
where point at the beginning of the search string means to replace the string,
not to append to the end as it does now by default.

> What you probably meant is that C-s <something> C-y M-y ... or C-s
> <something> M-y M-y ... cycle through the kill-ring entries and appends
> each of them in turn to the <something>, and that typing C-u M-y discards
> the <something> and enters the interactive selection of a kill-ring element
> in the minibuffer.

Actually the same problem exists for both cases:

C-s <something> C-y
C-s <something> C-u M-y

Both cases (either directly by C-y and M-y, or the interactive selection
by C-u M-y) append the string to the end of the search string.

> It could indeed make sense to use C-u C-u M-y (for example) to enter an
> interactive selection of a kill-ring element prepended with the
> <something>, but I'm not convinced it would be very useful in practice.

I agree, it makes no sense to prepend the old search string to the
initial input of the interactive selection.  This is why this would be
more convenient:

C-s <something> C-u C-y
C-s <something> C-u M-y

Both these cases will replace the search string, similar to what their
global versions do in regard to switching mark and point.

Then we need to find a new prefix instead of C-u M-y to keep
the old behavior of appending the yanked string to the end.
Maybe C-s <something> C-0 M-y.