[Haskell-cafe] about Haskell code written to be "too smart"

Jonathan Cast jonathanccast at fastmail.fm
Tue Mar 24 14:48:42 EDT 2009


On Tue, 2009-03-24 at 19:42 +0100, Manlio Perillo wrote:
> Tim Newsham ha scritto:
> >> These friends are very interested in Haskell, but it seems that the 
> >> main reason why they don't start to seriously learning it, is that 
> >> when they start reading some code, they feel the "Perl syndrome".
> >>
> >> That is, code written to be "too smart", and that end up being totally 
> >> illegible by Haskell novice.
> >>
> >> I too have this feeling, from time to time.
> >>
> >> Since someone is starting to write the Haskell coding style, I really 
> >> suggest him to take this "problem" into strong consideration.
> > 
> > When you think about it, what you are saying is that Haskell programmers 
> > shouldn't take advantage of the extra tools that Haskell provides. 
> 
> No, I'm not saying this.
> 
> But, as an example, when you read a function like:
> 
> buildPartitions xs ns = zipWith take ns . init $ scanl (flip drop) xs ns
> 
> that can be rewritten (argument reversed) as:
> 
> takeList :: [Int] -> [a] -> [[a]]
> takeList [] _         =  []
> takeList _ []         =  []
> takeList (n : ns) xs  =  head : takeList ns tail
>      where (head, tail) = splitAt n xs

Huh?  This is ugly and un-readable.  Seriously.

> I think that there is a problem.

Damn straight.  It should be:

> buildPartitions xs ns =
>     zipWith take ns $ init $ scanl (flip drop) xs ns

Or, if you're really worried about blocks/line, you can increase the
line count a bit (I do this regularly):

> buildPartitions xs ns =
>     zipWith take ns $       -- Select just the indicated prefix of
each element
>     init $                  -- Skip the last (empty) element
>     scanl (flip drop) xs $  -- Cumulatively remove prefixes of
indicated length
>     ns

> The buildPartition contains too many "blocks".
> And I have read code with even more "blocks" in one line.
> 
> It may not be a problem for a "seasoned" Haskell programmer, but when 
> you write some code, you should never forget that your code will be read 
> by programmers that can not be at your same level.

Not if I can help it.

More seriously, beginner code belongs in the first two-three chapters of
Haskell programming textbooks, not anywhere else.  It's like putting Fun
with Dick & Jane-speak in an adult novel.[1]

> I think that many Haskell programmers forget this detail, and IMHO this 
> is wrong.
> 
> > Haskell provides the ability to abstract code beyond what many other 
> > programming systems allow.  This abstraction gives you the ability to 
> > express things much more tersely.  This makes the code a lot harder to 
> > read for people who are not familiar with the abstractions being used.  
> 
> The problem is that I have still problems at reading and understanding 
> code that is too much terse...
> Because I have to assemble in my mind each block, and if there are too 
> many blocks I have problems.

jcc

[1] Well, not that bad.  Beginner-level code is useful for teaching the
basics of the language; Fun with Dick & Jane is child abuse.




More information about the Haskell-Cafe mailing list