Is alwaysBeen from the MTL module inefficient?

[ivanperez-keera]

I'm doing some tests, and the amount of code generated for an alwaysBeen is huge. Is there a way to make this shorter?

What would the ideal look like in C?

[ivanperez-keera]

I wonder if it's be more efficient to do the following. This is a simplification for the case where the bounds are from n steps ago to 0 steps ago:

alwaysBeen n s = constant n == numTruesInWindow
  where
     -- Number of true values from 0 samples ago to n samples ago
     numTruesInWindow :: Stream Word64
     numTruesInWindow = count s - count sNAgo

     -- Stream s, but n samples ago. We introduce Falses since we are counting
     -- the Trues
     sNAgo :: Stream Bool
     sNAgo = replicate n False ++ s

     count :: Stream Bool -> Stream Word64
     count = integrate . toInt

     integrate :: Stream Word64 -> Stream Word64
     integrate s = g
       where
         g = s + ([0] ++ g)

     toInt :: Stream Bool -> Stream Int
     toInt s = ifThenElse s 1 0