ivy.el (ivy--filter): allow filter to be interruptible with input.

[kiennq]

This is an attempt to fix the issue in ba37ec22, while still allow ivy--filter to be interruptible by keyboard.

Test with and the issue reported in ba37ec221ce7a0fce6f881d83895b8f971636cc05 doesn't happens

(let ((ivy-expr '(counsel-describe-symbol))
      ivy-test-inhibit-message)
  (dotimes (i 50)
    (message "%s" i)
    (execute-kbd-macro (vconcat (kbd "C-c e") (kbd "RET")))
    (sleep-for 0.1)))

Note: I've test with both ivy-switch-buffer and counsel-describe-symbol

[abo-abo]

Why not just change ivy--filter and ivy--recompute-index to have the "interruptible" behavior by default?

[kiennq]

Why not just change ivy--filter and ivy--recompute-index to have the "interruptible" behavior by default?

I think ivy--filter and ivy--recompute-index may be used for other purpose than filtering items on minibuffer, to minimize the risk, I only allow ivy--filter which is called from ivy--update-minibuffer to be interruptible.

[abo-abo]

Thanks for the update. I think this PR complicates the code too much.

Ideally, I would like to have a single while-no-input in the whole code, located probably in ivy--exhibit.

Then we need to investigate the cause of the ivy-switch-buffer bug that I discovered before.

I think it might be because in C-x b RET, the RET has interrupted the initial filter. So maybe we can check that in ivy-done and perform the filter if the initial one wasn't finished.

[kiennq]

I think it might be because in C-x b RET, the RET has interrupted the initial filter. So maybe we can check that in ivy-done and perform the filter if the initial one wasn't finished.

The problem is the RET is even interrupt the collection creation. And special case handling in ivy-done, seems more like a hack to me :sad:
My PR although while it makes the code more complicate, it's more straight forward and no hacking, special checking anywhere. It just allow a potentially long running function ivy--filter the ability to be interrupted by keyboard, not at all time but under safe time, when all the related variable has been set to correct value.

[kiennq]

@abo-abo Friendly ping

[abo-abo]

@kiennq Thanks for the code. But I feel it complicates things too much.

I think we should instead update/simplify ivy--update-minibuffer to extend the existing while-no-input expression to envelop the ivy--filter expression.

[kiennq]

Let me see if I can do something.

[kiennq]

@abo-abo Sorry for late update. I've push a new update with much smaller size that's using non-essential variable as a cue to interrupt calculation when there's keyboard input or not.
Can you see if this is ok?

[kiennq]

@abo-abo Friendly ping

[basil-conto]

Sorry, I can't usefully review this, since I'm not and don't intend to become more familiar with Ivy's low-level architecture. And given @abo-abo's concerns, I especially can't merge this.

BTW: what if the BODY in ivy--while-no-input returns t? It seems like this isn't currently supported, so it might be useful to point this out in the docstring, and make sure that none of the callers can end up returning t.