[Haskell-cafe] isLetter vs. isAlpha

[wren ng thornton]

On 11/21/12 4:59 PM, Artyom Kazak wrote:
> I saw a question on StackOverflow about the difference between isAlpha
> and isLetter today. One of the answers stated that the two functions are
> interchangeable, even though they are implemented differently.
>
> I decided to find out whether the difference in implementation
> influences performance, and look what I found:
>
>> import Criterion.Main
>> import Data.Char
>> fTest name f list = bgroup name $ map (\(n,c) -> bench n $ whnf f c) list
>> tests = [("latin", 'e'), ("digit", '8'), ("symbol", '…'), ("greek", 'λ')]
>> main = defaultMain [fTest "isAlpha" isAlpha tests,
>> fTest "isLetter" isLetter tests]
>
> produces this table (times are in nanoseconds):
>
>                   latin digit symbol greek
>                   ----- ----- ------ -----
>         isAlpha  | 156   212   368    310
>         isLetter | 349   344   383    310
>
> isAlpha is twice as fast on latin inputs! Does it mean that isAlpha
> should be preferred? Why isn’t isLetter defined in terms of isAlpha in
> Data.Char?

FWIW, testing on an arbitrary snippit of Japanese yields:

     benchmarking nf (map isAlpha)
     mean: 26.21897 us, lb 26.17674 us, ub 26.27707 us, ci 0.950
     std dev: 251.4027 ns, lb 200.4399 ns, ub 335.3004 ns, ci 0.950

     benchmarking nf (map isLetter)
     mean: 26.95068 us, lb 26.91681 us, ub 26.99481 us, ci 0.950
     std dev: 197.5631 ns, lb 158.9950 ns, ub 239.4986 ns, ci 0.950

I'm curious what the difference is between the functions, and whether 
isLetter is ever preferable...

-- 
Live well,
~wren