Support one-way aliases

[mitchellwrosen]

Hi there, I'd like to propose a feature similar to equivalence_classes, but one-way.

As a motivating example, suppose I want to make the . character equivalent to any special character, because it's easier to type than most:

equivalence_classes = { ".`~!@#$%^&*()-_=+[{]}\\|;:'\",<>/?" }

This works okay, but it does prevent me from ever jumping to (say) the comma character in this sequence:

","

because it's equivalent to an overlapping sequence of three dots:

...

If I could ask leap to treat . as any operator character, but not the other way around, then I could jump to the middle comma in the above example with s,"<label>, as usual.

Thanks!

[ggandor]

I should probably create a FAQ entry, this is an obvious request, but - just like smartcase - impossible by design: #89 (comment), #28.

[mitchellwrosen]

Thanks, I trust your assessment, but I'm not sure I understand.

Think about it: when targeting k+, after pressing k, the plugin cannot know whether you will type + or * (the wildcard character), so it cannot decide which label to show (the labels are different in the two cases, since the wildcard would match a lot more targets).

Why must the plugin know whether you will type + or * after k in this example? I should think that after a single k, we'll get a label after every k<c> pair, and pressing * would simply drop no labels (because all k<c> match).

Or, in my example, I don't exactly have a wildcard, but I would think that after (say) j., we could leave a label after every j!, j@, j#, etc. in the buffer.

[ggandor]

I should think that after a single k, we'll get a label after every k pair, and pressing * would simply drop no labels (because all k match).

The problem is that in phase 1 there are conflicting labels, you cannot just keep all of them. Each sublist of targets gets labeled by the same label list, e.g. kas get l1, l2, l3..., and kbs also get l1, l2, l3... The ambiguity is only resolved in phase 2, once you enter the second character, and only the chosen sublist remains.
Because of this (ahead-of-time labeling), any kind of aliasing/mapping can only work in a "symmetric" way in Leap. We could say that . will match everything, but then everything has to match everything (= equivalence classes); after pressing k, you either label a given ka match as part of the k. joker sublist, or the shorter ka sublist, we can't have the cake and eat it too.

[mitchellwrosen]

Oh, of course. I hadn't ever noticed that labels conflict but that does make sense!