Succinctly expressing random number generation in a functional language

random-number-generation.elm

-- Original code to generate random numbers

let (nChars, seed2) = generate (int 3 5) seed1
    (vIndices, seed3) = generate (list nChars (int 0 1000)) seed2
    (cIndices, seed4) = generate (list nChars (int 0 1000)) seed3
    (length, seed5) = generate (int 0 2) seed4
    config = {vIndices = vIndices,
              cIndices = cIndices}

-- Dictionary that represents computation 
-- Same format found in http://dask.pydata.org/en/latest/spec.html

{'nChars':   (int,3,5),
 'vIndices': (list,'nChars',(int,0,1000)),
 'cIndices': (list,'nChars',(int,0,1000)),
 'length':   (int,0,2)}

The original code has a lot of boilerplate to pass around the seed and to execute the generator.
Compare them:

(vIndices, seed3) = generate (list nChars (int 0 1000)) seed2
'vIndices': (list,'nChars',(int,0,1000))

Every call into the original code had to:

1. Assign a new seed
2. Specify the input seed
3. Call the generate function

These can all be abstracted away.

The key insight from dask is that you can specify the graph of dependencies very simply with a dictionary of tuples.