Personal tools

Means of expression

From HaskellWiki

(Difference between revisions)
Jump to: navigation, search
(extend list of means of expression)
(special values)
Line 63: Line 63:
 
<haskell>
 
<haskell>
 
foo :: [a] -> Maybe a
 
foo :: [a] -> Maybe a
  +
</haskell>
  +
  +
=== Proper types instead of special values ===
  +
  +
In languages like C you often see declarations like
  +
<code>
  +
/*
  +
This function returns the number of objects
  +
or -1 if there was an I/O error.
  +
*/
  +
int count (description d);
  +
</code>
  +
It is very easy to miss an error indicated by the <code>count</code> function
  +
and the language let you do arithmetic even with the error indicating value <code>-1</code> as in
  +
<code>
  +
c = count(d0)+count(d1);
  +
</code>
  +
But C only allows one return value per function.
  +
Further results have to be written to a pointer argument (cumbersome)
  +
or to a global variable (easy to miss, too, and not thread-safe).
  +
Haskells type system supports such situations much better
  +
and you should use this support.
  +
Try hard to avoid values that have a special meaning
  +
by using appropriate types for indicating special conditions as in
  +
<haskell>
  +
count :: Description -> IO (Maybe Int)
 
</haskell>
 
</haskell>
   

Revision as of 11:15, 28 January 2011

Programming languages have different types of means of expression:

  • Primary means of expression: variables, types, parentheses, modules, etc. in general all things that are relevant to the compiler
  • Secondary means of expression: language irrelevant layout and spacing, comments, conventions, etc. that is language elements that are irrelevant for the compiler, and thus are made exclusively for the human reader

It is generally good style to remember the rule:

Prefer primary means of expression to secondary ones!

The reason is, that elements that are relevant to the compiler are checked by the compiler and can be processed by documentation, analysis and refactoring tools. Thus it is not good style to comment intensively if the language provides primary ways to express our ideas.

Contents

1 Examples

1.1 Expressive names instead of comments

I recently saw code like

-- | first name
fn :: Person -> String
 
-- | surname
sn :: Person -> String

It was hard to read the program because there were several other two-character function names to remember, and the comment was only attached to the function definition, not the calls. The comments would be better replaced by expressive function names like

firstName :: Person -> String
surname :: Person -> String

1.2 Functions instead of comments

solveSLE2 :: Fractional a => ((a,a), (a,a)) -> (a,a) -> (a,a)
solveSLE2 ((a00,a10),(a01,a11)) (b0,b1) =
   let det = a00*a11 - a10*a01
   in  (-- determinant with first column replaced by b
        (b0*a11 - b1*a01) / det,
        -- determinant with second column replaced by b
        (a00*b1 - a10*b0) / det)

It is likely that logical units like the 2x2 determinant are reused later (e.g. for the vector product) or that they should be tested separately. Thus it is better to factor them out into separate functions.

solveSLE2 :: Fractional a => ((a,a), (a,a)) -> (a,a) -> (a,a)
solveSLE2 a@(a0,a1) b =
   let det = det2 a
   in  (det2 (b, a1) / det,
        det2 (a0, b) / det)
 
det2 :: Num a => ((a,a), (a,a)) -> a
det2 ((a00,a10),(a01,a11)) = a00*a11 - a10*a01

1.3 Types instead of comments

-- | returned list contains at most one element
foo :: [a] -> [a]

This can be expressed without comments more safely and concisely

foo :: [a] -> Maybe a

1.4 Proper types instead of special values

In languages like C you often see declarations like

  /*
  This function returns the number of objects
  or -1 if there was an I/O error.
  */
  int count (description d);

It is very easy to miss an error indicated by the count function and the language let you do arithmetic even with the error indicating value -1 as in c = count(d0)+count(d1); But C only allows one return value per function. Further results have to be written to a pointer argument (cumbersome) or to a global variable (easy to miss, too, and not thread-safe). Haskells type system supports such situations much better and you should use this support. Try hard to avoid values that have a special meaning by using appropriate types for indicating special conditions as in

count :: Description -> IO (Maybe Int)

1.5 Qualification instead of identifier prefix/suffix conventions

You may define

module File where
 
openFile :: FilePath -> IO Handle

and intend to import that unqualified

import File
 
main :: IO ()
main = do
   h <- openFile "myfile"
   ...

in other modules, or you can define

module File where
 
open :: FilePath -> IO Handle

that would fit for qualified import as in

import qualified File
 
main :: IO ()
main = do
   h <- File.open "myfile"
   ...

The second way would use primary means of the language (modules and qualified imports) whereas the first way uses secondary means of the language (naming convention). The naming convention cannot be checked,

that is
fileOpen
and
openFill

would be accepted by any Haskell compiler and it forces the user to follow this convention, leading to respect different conventions from different packages that are imported in one module, e.g. type name in the identifier as prefix vs. suffix. Using qualified imports, the importer can choose an abbreviation that he likes, say:

import qualified File as F
 
main :: IO ()
main = do
   h <- F.open "myfile"
   ...

and thus assert a consistent naming style in his module.

Find out about more style questions on importing and qualified names.

2 See also

Johannes Waldmann: Haskell mit Stil