Incremental search

[fbergroth]

I think adding incremental search would be nice.

There are some issues with entering a search pattern via meow-visit if meow-visit-sanitize-completion=t.

• The pattern is always quoted, so searching for a regex will fail (but docstring says it'll do a regex search!)
• If you have a buffer containing meow and meow-variable, you can't easily search for the prefix meow, since symbol boundaries will be added via completions (typing meow results in \_<meow\_>)
• No incremental feedback

I'm experimenting with binding meow-isearch to /, which I quite like.

(defun meow-isearch ()
  (interactive)
  (if (meow--direction-backward-p) ; not sure how should work
      (isearch-backward t)
    (isearch-forward t))
  (when isearch-success
    (thread-first
      (meow--make-selection '(select . visit) isearch-other-end (point))
      (meow--select))))

Now, there is some overlap between this and meow-visit. I think it would be nice if v did meow-visit with completions being words (\<word\>), and V did meow-visit with completions being symbols.

What do you think @DogLooksGood @eshrh?

[eshrh]

I've encountered some of the problems you mentioned (not being able to easily search for a part of another symbol is a big one, missing regex is another), so personally, I actually have v bound to swiper. I think binding an isearch on / by default is a good idea, since that key is currently unoccupied.

To check if you're going backward I think you can check whether the variable current-prefix-arg is string= to "-". The - key in meow just modifies the prefix arg by simulating C-u.

As for meow-visit, I am totally in favor of binding v and V to word and symbol. This is in-line with other current keybinds and seems like a very useful addition that saves a lot of C-n presses when looking for just words.

[DogLooksGood]

I will not add incremental searching for meow-visit, because meow-visit use the builtin completing-read, it can be baked by a lot packages, like helm, ivy, vertico, selectrum, etc.

But I can definitely understand what you pointed out. I'd like to have an integration with incremental search, but not by providing a command. Because there are builtins. Additionally, there are some packages provides the enhance alternatives, like phi-search and ctrlf. I haven't do it because I have no idea how to get a general solution.

For isearch, I'd like to have something like this

(defun meow--post-isearch-function ()
  (unless isearch-mode-end-hook-quit
    (when isearch-success
      (let ((beg (car isearch-match-data))
	    (end (cadr isearch-match-data)))
	
	(thread-first
	  (meow--make-selection '(select . visit)
				beg
				(if isearch-forward end isearch-other-end))
	  (meow--select (not isearch-forward)))))))

(add-hook 'isearch-mode-end-hook 'meow--post-isearch-function)

It should work for both C-s, C-r, C-M-s, but it doesn't work for other packages, obviously.
I'll add incremental search support, whenever I found nice way to integrate.

[DogLooksGood]

I think a variable with function is not enough here. Implementing meow-search with isearch, phi-isearch and ctrlf are quite different.

[fbergroth]

It's not what I mean. If I do C-s some-pattern RET, I expect M-x meow-search to search forward using some-pattern. Currently it does not, but it works with meow--post-isearch-function.

[DogLooksGood]

meow-search will use regexp-search-ring.
Currently, meow-mark-word, meow-mark-symbol and meow-visit will push regexp into this variable.

You want provide a custom function to control which variable to be used?

[fbergroth]

Ah sorry. C-s some-pattern RET activated a region, and then I called meow-search, which pushed it into regexp-search-ring. I thought meow--post-isearch-function was inserting the pattern into the ring. 🤦‍♂️

[DogLooksGood]

It will be better if a region is activated after isearch, and cancel the existing region before isearch.

But changing the behavior of builtin commands is not good. So I'd like to make no changes.