Programming guidelines

Programming guidelines shall help to make the code of a project better readable and maintainable by the varying number of contributors.

It takes some programming experience to develop something like a personal "coding style" and guidelines only serve as rough shape for code. Guidelines should be followed by all members working on the project even if they prefer (or are already used to) different guidelines.

These guidelines have been originally set up for the hets-project hets-project and are now put on the HaskellWiki gradually integrating parts of the old hawiki entries ThinksToAvoid and HaskellStyle (hopefully not hurting someone's copyrights). The other related entry TipsAndTricks treats more specific points that are left out here,

Surely same style choices are a bit arbitrary (or "religious") and too restrictive with respect to language extensions. Nevertheless I hope to keep up these guidelines (at least as a basis) for our project in order to avoid maintaining diverging guidelines. Of course I want to supply - partly tool-dependent - reasons for certain decisions and also show alternatives by possibly bad examples. At the time of writing I use ghc-6.4.1, haddock-0.7 and (GNU-) emacs with the latest haskell mode.

The following quote and links are taken from the old general comments:

We all have our own ideas about good Haskell style. There's More Than One Way To Do It. But some ways are better than others.

Some (unfinished) comments from the GHC team about their internal coding standards can be found at http://www.cse.unsw.edu.au/~chak/haskell/ghc/comm/the-beast/coding-style.html

Also http://research.microsoft.com/~simonpj/papers/haskell-retrospective/ contains some brief comments on syntax and style,

What now follows are descriptions of program documentation, file format, naming conventions and good programming practice (adapted form Matt's C/C++ Programming Guidelines and the Linux kernel coding style).

Documentation

Comments are to be written in application terms (i.e. user's point of view). Don't use technical terms - that's what the code is for!

Comments should be written using correct spelling and grammar in complete sentences with punctation (in English only).

"Generally, you want your comments to tell WHAT your code does, not HOW. Also, try to avoid putting comments inside a function body: if the function is so complex that you need to separately comment parts of it, you should probably" (... decompose it)

Put a haddock comment on top of every exported function and data type! Make sure haddock accepts these comments.

File Format

All Haskell source files start with a haddock header of the form:

{- |
Module      :  <File name, i.e. generated by \$Header\$>
Description :  <Short text displayed on contents page>
Copyright   :  (c) <You> and <Your affiliation>
License     :  similar to LGPL, see LICENSE.txt

Maintainer  :  maeder@tzi.de
Stability   :  provisional
Portability :  portable

<module description starting at first column>
-}

A possible compiler pragma (like {-# OPTIONS -cpp #-}) may precede this header. The following hierarchical module name must of course match the file name.

Make sure that the description is changed to meet the module (if the header was copied from elsewhere). Insert your email address as maintainer.

Try to write portable (Haskell98) code. If you (indirectly) import a module that uses i.e. multi-parameter type classes and functional dependencies the code becomes "non-portable (MPTC with FD)".

The \$Header\$ entry is automatically expanded by cvs (and will wrap around). All other lines should not be longer than 80 (preferably 75) characters to avoid wrapped lines (for casual readers)!

Expand all your tabs to spaces to avoid the danger of wrongly expanding them (or a different display of tabs versus eight spaces). Possibly put something like the following in your ~/.emacs file.

 (custom-set-variables '(indent-tabs-mode nil))

The last character in your file should be a newline! Under solaris you'll get a warning if this is not the case and sometimes last lines without newlines are ignored (i.e. "#endif" without newline). Emacs usually asks for a final newline.

The whole module should not be too long (about 400 lines)

Naming Conventions

In Haskell types start with capital and functions with lowercase letters, so only avoid infix identifiers! Defining symbolic infix identifiers should be left to library writers only.

(The infix identifier "\\" at the end of a line causes cpp preprocessor problems.)

Names (especially global ones) should be descriptive and if you need long names write them as mixed case words (aka camelCase). (but "tmp" is to be preferred over "thisVariableIsATemporaryCounter")

Also in the standard libraries, function names with multiple words are written using the camelCase convention. Similarly, type, typeclass and constructor names are written using the StudlyCaps convention.

Some parts of our code use underlimes (without unnecessary uppercase letters) for long identifiers to better reflect names given with hyphens in the requirement documentation. Also such names should be transliterated to camlCase identifiers possibly adding a (consistent) suffix or prefix to avoid conflicts with keywords. However, instead of a recurring prefix or suffix you may consider to use qualified imports and names.

Good Programming Practice

"Functions should be short and sweet, and do just one thing. They should fit on one or two screenfuls of text (the ISO/ANSI screen size is 80x24, as we all know), and do one thing and do that well."

Most haskell functions should be at most a few lines, only case expression over large data types (that should be avoided, too) may need corresponding space.

The code should be succinct (though not obfuscated), readable and easy to maintain (after unforeseeable changes). Don't exploit exotic language features without good reason.

It's not fixed how deep you indent (4 or 8 chars). You can break the line after "do", "let", "where", and "case .. of". Make sure that renamings don't destroy your layout. (If you get to far to the right, the code is unreadable anyway and needs to be decomposed.)

Bad:

 case foo of Foo -> "Foo"
             Bar -> "Bar"

Good:

 case <longer expression> of
       Foo -> "Foo"
       Bar -> "Bar"

Avoid the notation with braces and semicolons since the layout rule forces you to properly align your alternatives.

Respect compiler warnings. Supply type signatures, avoid shadowing and unused variables. Particularly avoid non-exhaustive and overlapping patterns. Missing unreachable cases can be filled in using "error" with a fixed string "<ModuleName>.<function>" to indicate the error position (in case the impossible should happen). Don't invest time to "show" the offending value, only do this temporarily when debugging the code.

Don't leave unused or commented-out code in your files! Readers don't know what to think of it.

Case expressions

Prefer case expressions over pattern binding declarations with multiple equations.

Not always nice:

 longFunctionName (Foo: _ : _) = e1
 longFunctionName (Bar: _) = e2

Better:

 longFunctionName arg = case arg of
       Foo : _ : _ -> e1
       Bar : _ -> e2
       _ -> error "ProgrammingGuidelines.longFunctionName"

In http://research.microsoft.com/~simonpj/papers/haskell-retrospective/ the first example is said to be written in "declaration style". The equations look like written for a rewrite system (although their order matters of course).

But this declarative style is only nice for toy examples and annoying if functions are renamed or if the number of arguments changes.

The other extreme (according to SPJ) is "expression style":

 longFunctionName = \ arg -> ...

We don't propose this style either. We propose to use as much pattern matching (including as-patterns) on a single left-hand-side as appropriate.

However, the expression style with a lambda term may come in handy, when setting record fields of a function type.

We avoid lambda expressions if this is easily possibly using the Prelude functions const, flip, curry, uncurry or section notation or plain partial application. We do not indroduce an auxiliary function only to avoid the lambda, though.

Partial functions

For partial functions do document their preconditions (if not obvious) and make sure that partial functions are only called when preconditions are obviously fulfilled (i.e. by a case statement or a previous test). Particularly the call of "head" should be used with care or (even better) be made obsolete by a case statement.

Usually a case statement (and the import of isJust and fromJust from Data.Maybe) can be avoided by using the "maybe" function:

   maybe (error "<ModuleName>.<function>") id $ Map.lookup key map

Generally we require you to be more explicit about failure cases. Surely a missing (or an irrefutable) pattern would precisely report the position of a runtime error, but these are not so obvious when reading the code.

Let or where expressions

Do avoid mixing and nesting "let" and "where". (I prefer the expression-stylistic "let".) Use auxiliary top-level functions that you do not export. Export lists also support the detection of unused functions.

Code reuse

If you notice that you're doing the same task again, try to generalize it in order to avoid duplicate code. It is frustrating to change the same error in several places.

Application notation

Many parentheses can be eliminated using the infix application operator "$" with lowest priority. Try at least to avoid unnecessary parentheses in standard infix expression.

       f x : g x ++ h x

       a == 1 && b == 1 || a == 0 && b == 0

Rather than putting a large final argument in parentheses (with a distant closing one) consider using "$" instead.

"f (g x)" becomes "f $ g x" and consecutive applications "f (g (h x))" can be written as "f $ g $ h x" or "f . g $ h x".

A function definition like "f x = g $ h x" can be abbreviated to "f = g . h".

Note that the final argument may even be an infix- or case expression:

       map id $ c : l

       filter (const True) . map id $ case l of ...

However, be aware that $-terms cannot be composed further in infix expressions.

Probably wrong:

       f $ x ++ g $ x

But the scope of an expression is also limited by the layout rule, so it is usually save to use "$" on right hand sides.

Ok:

       do y <- f $ l
       ++
       do y <- g $ l

Of course "$" can not be used in types. GHC has also some primitive functions involving the kind "#" that cannot be applied using "$".

Last warning: always leave spaces around "$" (and other mixfix operators) since a clash with template haskell is possible.

(Also write "\ t" instead of "\t" in lambda expressions)

List Comprehensions

Use these only when "short and sweet". Prefer map, filter, and foldr!

Instead of:

   [toUpper c | c <- s]

write:

   map toUpper s

Consider:

   [toUpper c | s <- strings, c <- s]

Here it takes some time for the reader to find out which value depends on what other value and it is not so clear how many times the interim values s and c are used. In contrast to that the following can't be clearer:

   map toUpper (concat strings)

When using higher order functions you can switch easier to data structures different from list. Compare:

   map (1+) list

and:

   Set.map (1+) set

Types

Prefer proper data types over type synonyms or tuples even if you have to do more constructing and unpacking. This will make it easier to supply class instances later on. Don't put class constraints on a data type, constraints belong only to the functions that manipulate the data.

Using type synonyms consistently is difficult over a longer time, because this is not checked by the compiler. (The types shown by the compiler may be unpredictable: i.e. FilePath, String or [Char])

Take care if your data type has many variants (unless it is an enumeration type.) Don't repeat common parts in every variant since this will cause code duplication.

Bad (to handle arguments in sync):

 data Mode f p = Box f p | Diamond f p

Good (to handle arguments only once):

 data BoxOrDiamond = Box | Diamond

 data Mode f p = Mode BoxOrDiamond f p

Consider (bad):

 data Tupel a b = Tupel a b | Undefined

versus (better):

 data Tupel a b = Tupel a b

and using:

  Maybe (Tupel a b)

(or another monad) whenever an undefined result needs to be propagated

Records

For (large) records avoid the use of the constructor directly and remember that the order and number of fields may change.

Take care with (the rare case of) depend polymorphic fields:

 data Fields a = VariantWithTwo 
     { field1 :: a
     , field2 :: a }

The type of a value v can not be changed by only setting field1:

   v { field1 = f }

Better construct a new value:

   VariantWithTwo { field1 = f }  -- leaving field2 undefined

Or use a polymorphic element that is instantiated by updating:

   empty = VariantWithTwo { field1 = [], field2 = [] }

   empty { field1 = [f] }

Several variants with identical fields may avoid some code duplication when selecting and updating, though possibly not in a few depended polymorphic cases.

However, I doubt if the following is a really good alternative to the above data Mode with data BoxOrDiamond.

 data Mode f p =
      Box     { formula :: f,  positions :: p }
    | Diamond { formula :: f,  positions :: p }

IO

Try to strictly separate IO, Monad and pure (without do) function programming (possibly via separate modules).

Bad:

       x <- return y
       ...

Good:

       let x = y
       ...

Don't use Prelude.interact and make sure your program does not depend on the (not always obvious) order of evaluation. I.e. don't read and write to the same file:

This will fail:

 do s <- readFile f
    writeFile f $ 'a' : s

because of lazy IO! (Writing is starting before reading is finished.)

Trace

Tracing is for debugging purposes only and should not be used as feedback for the user. Clean code is not cluttered by trace calls.

Imports

Standard library modules like Char. List, Maybe, Monad, etc should be imported by their hierarchical module name, i.e. the base package (so that haddock finds them):

 import Data.List
 import Control.Monad
 import System.Environment

The libraries for Set and Map are to be imported qualified:

 import qualified Data.Set as Set
 import qualified Data.Map as Map

Glasgow extensions and Classes

Stay away form extensions as long as possible. Also use classes with care because soon the desire for overlapping instances (like for lists and strings) may arise. Then you may want MPTC (multi-parameter type classes), functional dependencies (FD), undecidable and possibly incoherent instances and then you are "in the wild" (according to SPJ).

Final remarks

Despite guidelines, writing "correct code" (without formal proof support yet) still remains the major challenge. As motivation to follow these guidelines consider the points that are from the "C++ Coding Standard", where I replaced "C++" with "Haskell".

Good Points:

• programmers can go into any code and figure out what's going on

• new people can get up to speed quickly

• people new to Haskell are spared the need to develop a personal style and defend it to the death

• people new to Haskell are spared making the same mistakes over and over again

• people make fewer mistakes in consistent environments

• programmers have a common enemy :-)

Bad Points:

• the standard is usually stupid because it was made by someone who doesn't understand Haskell

• the standard is usually stupid because it's not what I do

• standards reduce creativity

• standards are unnecessary as long as people are consistent

• standards enforce too much structure

• people ignore standards anyway