I took a look at the count implementation for #20403 (cc @cossio), and was surprised to discover that the current definition is nearly identical to sum (albeit with a type instability):

function count(pred, itr)
    n = 0
    for x in itr
        n += pred(x)
    end
    return n
end

For example, count(sqrt, [1,2,3,5]) returns 6.382332347441762.

We could disallow this case entirely, but one possibility would be to make count(f, x) equivalent to countnz for non-boolean data. Indeed, I'm not sure why we have both countnz and count — couldn't we merge the two functions into a single count function?