Comments on: Playing With F# http://www.netcrucible.com/blog/2007/12/30/playing-with-f/ The software industry from a rational perspective Thu, 20 Nov 2008 11:12:49 +0000 http://wordpress.org/?v=2.2.1 By: allenjs http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51656 allenjs Mon, 31 Dec 2007 04:39:10 +0000 http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51656 Janne: The range function isn't tail recursive, either, but I thought it didn't matter anymore. I actually tested before posting, and F# returns 10,000 deep almost instantly -- so I assumed the compiler was doing some magic to implement tail recursion under the covers (it's smart enough to infer all my types, so I thought "why not"), and 10,000 deep is an ungodly level of recursion. Anyway, I just tested with bigger numbers and was able to blow the F# stack at 100,000 deep on my machine. So you're right. I also just tested the same routine in C#, and found that it actually does 10,000 deep on my machine (sum only, not range) as well. I am amazed at this; I would expect it to blow up after a few hundred. C# is a lot slower than F# at this level of recursion, though; and I can blow up C# at just 15,000 -- where F# functions fine and near-instant at 15,000. It seems that F# perf doesn't degrade as quickly as C# before blowing up. For completeness sake, the range function is better implemented as: let rec range2 a b = [for i in a to b -> i] Janne: The range function isn’t tail recursive, either, but I thought it didn’t matter anymore. I actually tested before posting, and F# returns 10,000 deep almost instantly — so I assumed the compiler was doing some magic to implement tail recursion under the covers (it’s smart enough to infer all my types, so I thought “why not”), and 10,000 deep is an ungodly level of recursion.

Anyway, I just tested with bigger numbers and was able to blow the F# stack at 100,000 deep on my machine. So you’re right.

I also just tested the same routine in C#, and found that it actually does 10,000 deep on my machine (sum only, not range) as well. I am amazed at this; I would expect it to blow up after a few hundred. C# is a lot slower than F# at this level of recursion, though; and I can blow up C# at just 15,000 — where F# functions fine and near-instant at 15,000. It seems that F# perf doesn’t degrade as quickly as C# before blowing up.

For completeness sake, the range function is better implemented as:
let rec range2 a b = [for i in a to b -> i]

]]> By: Janne http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51620 Janne Sun, 30 Dec 2007 20:25:13 +0000 http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51620 whoops, the last function should read let sum = let rec loop acc ... whoops, the last function should read

let sum =
let rec loop acc …

]]> By: Janne http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51619 Janne Sun, 30 Dec 2007 20:18:24 +0000 http://www.netcrucible.com/blog/2007/12/30/playing-with-f/#comment-51619 Are you sure that your sum function will not blow up the stack? It's not tail recursive: let rec sum list = match list with | h::tail -> (sum tail) + h (* 0 something like List.fold_left (fun acc e -> acc+e) 0 list or let rec sum = let loop acc = function x::xs -> loop (x+acc) xs | [] -> acc in loop 0 would be. Are you sure that your sum function will not blow up the stack? It’s not tail recursive:

let rec sum list =
match list with
| h::tail -> (sum tail) + h (* 0

something like

List.fold_left (fun acc e -> acc+e) 0 list

or

let rec sum =
let loop acc = function
x::xs -> loop (x+acc) xs
| [] -> acc in
loop 0

would be.

]]>