Do not delay mapped keys with common prefix

Question

What I want to do is have two mappings. One mapped to <s-k>, and one mapped to <s-k><s-k>. What happens right now is that if I press <s-k> vim delays for timeoutlenms, and if I don't press anything it does the <s-k> action, and if I press another <s-k> in that time it does the <s-k><s-k> action.

However, what I want it to do is do BOTH, i.e. if I press <s-k> it will do the <s-k> action immediately, and if I press another <s-k> in the timeoutlen period, it will do the second command as well.

Answer 1

It's not possible to achieve this with regular mappings of K and KK. If the K mapping fires then Vim is necessarily no longer waiting to see if KK is typed, and there are no configuration options that can change this behaviour.

There are ways to workaround this, however. filbranden suggested one possible method in the comments which D. Ben Knoble has implemented.

Rather than rely on autocommands to reset the state, I'd probably check the time since we last fired the mapping directly in the mapping itself by using reltime():

function KFunc() abort
  if !exists("s:last_press") || reltimefloat(reltime(s:last_press)) > &timeoutlen / 1000.0
    " do one

    " remember when we fired the mapping
    let s:last_press = reltime()

  else
    " do two

    " reset the time
    unlet s:last_press
  endif
endfunction

nnoremap K :call KFunc()<CR>

Answer 2

I think something like this would do:

let s:pressed = 0

function KFunc() abort
  let s:pressed += 1
  if s:pressed == 1
    " do one
  else if s:pressed == 2
    " do two
    let s:pressed = 0
  endif
endfunction

augroup ResetPressed
  au!
  au CursorMoved,CursorMovedI,CursorHold,CursorHoldI * let s:pressed = 0
augroup END

nnoremap K :call KFunc()<CR>

I included the insert mode variants of cursor moved/hold, but you probably don't need them unless it's an insert-mode map.

As @Rich points out in the comments, CursorHold and friends fire after 'updatetime', not 'timeoutlen'. Changing 'updatetime' has implications for swap files, and fixing the code seems tedious. That said, I changed 'updatetime' to 250ms (down from the default 4s) without any adverse effects.