First using list comprehensions:

let codes =
    [("A",".-");   ("B","-..."); ("C","-.-."); ("D","-.."); ("E",".");
     ("F","..-."); ("G","--.");  ("H","...."); ("I","..");  ("J",".---");
     ("K","-.-");  ("L",".-.."); ("M","--");   ("N","-.");  ("O","---");
     ("P",".--."); ("Q","--.-"); ("R",".-.");  ("S","..."); ("T","-");
     ("U","..-");  ("V","...-"); ("W",".--");  ("X","-..-");("Y","-.--");
     ("Z","--..")]

let rec decode input =
    if input = "" then [""]
    else [ for c, code in codes when input.StartsWith(code)
           for rest in decode(input.Substring(String.length code))
           -> c + rest ]

As it can be seen the code is almost identical to the Python code. Incidentally, I could not find a function equivalent to Python’s startswith method in the O’Caml standard library (without using regular expressions). Fortunately F# came with one from the .NET library.

Much to my the surprise the compiled F# (running on Mono 1.2.4) is 4 times slower that the Python code. I then rewrote the program to use sequence comprehensions:

let rec decode input =
    if input = "" then { -> "" }
    else { for c, code in codes when input.StartsWith(code)
           for rest in decode(input.Substring(String.length code))
           -> c + rest }

This version runs faster, and uses only a constant amount of memory. Still the Python version is three and half times faster.

I then tried to run the programs on an other computer with Microsoft’s .NET implementation. This improved the F# running times a lot. However, they are still 40% to 80% slower than the Python version.

My current guess is that Python is much better at handling strings than .NET.

The actual numbers:

Update 2007-11-12 Added test files:

• Python code: morse.py
• F# code using list comprehensions: morse.fs
• F# code using sequence comprehensions: morse-seq.fs

Without further ado here is the code I ended up with:

#!/usr/bin/env python

import string

letters = [('A',".-"),   ('B',"-..."), ('C',"-.-."), ('D',"-.."), ('E',"."),
           ('F',"..-."), ('G',"--."),  ('H',"...."), ('I',".."),  ('J',".---"),
           ('K',"-.-"),  ('L',".-.."), ('M',"--"),   ('N',"-."),  ('O',"---"),
           ('P',".--."), ('Q',"--.-"), ('R',".-."),  ('S',"..."), ('T',"-"),
           ('U',"..-"),  ('V',"...-"), ('W',".--"),  ('X',"-..-"),('Y',"-.--"),
           ('Z',"--..")]

def decode(input):
    if input == "" :
        return [""]
    else:
        return [ letter + remaining
                 for letter, code in letters if input.startswith(code)
                 for remaining in decode(input[len(code):]) ]

# Some Testing code
def test(s, code):
    if s in decode(code):
        print code + " can be decoded as " + s
    else:
        print code + " can NOT be decoded as " + s

test("SOFIA", "...---..-....-")
test("SOPHIA", "...---..-....-")
test("EUGENIA", "...---..-....-")

Interesting, my solution is rather similar to Patrick Logan’s Erlang solution. And I find it simpler to understand than the Haskell solutions at the HaskellWiki.