Code style guide


These are a few guidelines for Ganeti code and documentation.

In simple terms: try to stay consistent with the existing code. PEP 8 says:

A style guide is about consistency. Consistency with this style guide is important. Consistency within a project is more important. Consistency within one module or function is most important.


You might also want to take a look at the Google style guide, since we have some things in common with it.


In general, always indent using two (2) spaces and don’t use tabs.

The two spaces should always be relative to the previous level of indentation, even if this means that the final number of spaces is not a multiple of 2.

When going on a new line inside an open parenthesis, align with the content of the parenthesis on the previous line.

Valid example:

v = (somevalue,
       list_elem, # 7 spaces, but 2 from the previous indentation level

Formatting strings

Always use double quotes (""), never single quotes (''), except for existing code. Examples for formatting strings:

var = "value"

# Note: The space character is always on the second line
var = ("The quick brown fox jumps over the lazy dog. The quick brown fox"
       " jumps over the lazy dog. The quick brown fox jumps over the lazy"
       " dog.")

fn("The quick brown fox jumps over the lazy dog. The quick brown fox jumps"
   " over the lazy dog.")

   ("The quick brown fox jumps over the lazy dog. The quick brown fox"
    " jumps over the lazy dog. The quick brown fox jumps over the lazy"
    " dog."))

Don’t format strings like this:

# Don't use single quotes
var = 'value'

# Don't use backslash for line continuation
var = "The quick brown fox jumps over the lazy dog. The quick brown fox"\
      " jumps over the lazy dog."

# Space character goes to the beginning of a new line
var = ("The quick brown fox jumps over the lazy dog. The quick brown fox "
       "jumps over the lazy dog. The quick brown fox jumps over the lazy "

Formatting sequences

Built-in sequence types are list ([]), tuple (()) and dict ({}). When splitting to multiple lines, each item should be on its own line and a comma must be added on the last line. Don’t write multiline dictionaries in function calls, except when it’s the only parameter. Always indent items by two spaces.

# Short lists
var = ["foo", "bar"]
var = ("foo", "bar")

# Longer sequences and dictionary
var = [
var = {
  "key": func(),
  "otherkey": None,

# Multiline tuples as dictionary values
var = {
    ("long value taking the whole line, requiring you to go to a new one",

# Function calls
var = frozenset([1, 2, 3])
var = F({
  "xyz": constants.XYZ,
  "abc": constants.ABC,

# Wrong
F(123, "Hello World",
  { "xyz": constants.XYZ })

We consider tuples as data structures, not containers. So in general please use lists when dealing with a sequence of homogeneous items, and tuples when dealing with heterogeneous items.

Passing arguments

Positional arguments must be passed as positional arguments, keyword arguments must be passed as keyword arguments. Everything else will be difficult to maintain.

# Function signature
def F(data, key, salt=None, key_selector=None):

# Yes
F("The quick brown fox", "123456")
F("The quick brown fox", "123456", salt="abc")
F("The quick brown fox", "123456", key_selector="xyz")
F("The quick brown fox", "123456", salt="foo", key_selector="xyz")

# No: Passing keyword arguments as positional argument
F("The quick brown fox", "123456", "xyz", "bar")

# No: Passing positional arguments as keyword argument
F(salt="xyz", data="The quick brown fox", key="123456", key_selector="xyz")



PEP 257 is the canonical document, unless epydoc overrules it (e.g. in how to document the type of an argument).

For docstrings, the recommended format is epytext, to be processed via epydoc. There is an apidoc target that builds the documentation and puts it into the doc/api subdir. Note that we currently use epydoc version 3.0.

Note that one-line docstrings are only accepted in the unittests.

Rules for writing the docstrings (mostly standard Python rules):

  • the docstring should start with a sentence, with punctuation at the end, summarizing the the aim of what is being described. This sentence cannot be longer than one line
  • the second line should be blank
  • afterwards the rest of the docstring
  • special epytext tags should come at the end
  • multi-line docstrings must finish with an empty line
  • do not try to make a table using lots of whitespace
  • use L{} and C{} where appropriate

Here’s an example:

def fn(foo, bar):
  """Compute the sum of foo and bar.

  This functions builds the sum of foo and bar. It's a simple function.

  @type foo: int
  @param foo: First parameter.
  @type bar: float
  @param bar: The second parameter. This line is longer
    to show wrapping.
  @rtype: float
  @return: the sum of the two numbers

  return foo + bar

Some rules of thumb which should be applied with good judgement on a case-to- case basis:

  • If the meaning of parameters is already obvious given its name and the methods description, don’t document it again. Just add a @type tag.
  • Refer to the base methods documentation when overwriting methods. Only document more if it applies to the current subclass only, or if you want to clarify on the meaning of parameters for the special subclass.

Rules for classes and modules

As PEP 257 says, the docstrings of classes should document their attributes and the docstrings of modules should shortly document the exported functions/variables/etc.

See for example the pydoc output for the os or ConfigParser standard modules.


The most important consideration is, as usual, to stay consistent with the existing code.

As there’s no “canonical” style guide for Haskell, this code style has been inspired from a few online resources, including the style guide for the Snap framework, this style guide and this other style guide.


Use ordinary, non-literate Haskell .hs files.

Use proper copyright headers, and proper Haddock style documentation headers:

{-| Short module summary.

Longer module description.



Copyright (C) ...

This program is free software ...


If there are module-level pragmas add them right at the top, before the short summary.


Imports should be grouped into the following groups and inside each group they should be sorted alphabetically:

  1. import of non-Ganeti libaries
  2. import of Ganeti libraries

It is allowed to use qualified imports with short names for:

  • standard library (e.g. import qualified Data.Map as M)
  • local imports (e.g. import qualified Ganeti.Constants as C)

Whenever possible, prefer explicit imports, either in form of qualified imports, or by naming the imported functions (e.g., import Control.Arrow ((&&&)), import Data.Foldable(fold, toList))


Use only spaces, never tabs. Indentation level is 2 characters. For Emacs, this means setting the variable haskell-indent-offset to 2.

Line length should be at most 78 chars, and 72 chars inside comments.

Use indentation-based structure, and not braces/semicolons.


Special indendation of if/then/else construct

For the do notation, the if-then-else construct has a non-intuitive behaviour. As such, the indentation of if-then-else (both in do blocks and in normal blocks) should be as follows:

if condition
  then expr1
  else expr2

i.e. indent the then/else lines with another level. This can be accomplished in Emacs by setting the variable haskell-indent-thenelse to 2 (from the default of zero).

If you have more than one line of code please newline/indent after the “=”. Do not do:

f x = let y = x + 1
      in  y

Instead do:

f x =
  let y = x + 1
  in  y

or if it is just one line:

f x = x + 1

Multiline strings

Multiline strings are created by closing a line with a backslash and starting the following line with a backslash, keeping the indentation level constant. Whitespaces go on the new line, right after the backslash.

longString :: String
longString = "This is a very very very long string that\
             \ needs to be split in two lines"

Data declarations


Note that this is different from the Python style!

When declaring either data types, or using list literals, etc., the columns should be aligned, and for lists use a comma at the start of the line, not at the end. Examples:

data OpCode = OpStartupInstance ...
            | OpShutdownInstance ...
            | ...

data Node = Node { name :: String
                 , ip   :: String
                 , ...

myList = [ value1
         , value2
         , value3

The choice of whether to wrap the first element or not is up to you; the following is also allowed:

myList =
  [ value1
  , value2

For records, always add spaces around the braces and the equality sign.

foo = Foo { fBar = "bar", fBaz = 4711 }

foo' = Foo { fBar = "bar 2"
           , fBaz = 4712

node' = node { ip = "" }

White space

Like in Python, surround binary operators with one space on either side. Do no insert a space after a lamda:

-- bad
map (\ n -> ...) lst
-- good
foldl (\x y -> ...) ...

Use a blank line between top-level definitions, but no blank lines between either the comment and the type signature or between the type signature and the actual function definition.


Ideally it would be two blank lines between top-level definitions, but the code only has one now.

As always, no trailing spaces. Ever.

Spaces after comma

Instead of:



("a", "b")


Functions should be named in mixedCase style, and types in CamelCase. Function arguments and local variables should be mixedCase.

When using acronyms, ones longer than 2 characters should be typed capitalised, not fully upper-cased (e.g. Http, not HTTP).

For variable names, use descriptive names; it is only allowed to use very short names (e.g. a, b, i, j, etc.) when:

  • the function is trivial, e.g.:

    sum x y = x + y
  • we talk about some very specific cases, e.g. iterators or accumulators in folds:

    map (\v -> v + 1) lst
  • using x:xs for list elements and lists, etc.

In general, short/one-letter names are allowed when we deal with polymorphic values; for example the standard map definition from Prelude:

map :: (a -> b) -> [a] -> [b]
map _ []     = []
map f (x:xs) = f x : map f xs

In this example, neither the a nor b types are known to the map function, so we cannot give them more explicit names. Since the body of the function is trivial, the variables used are longer.

However, if we deal with explicit types or values, their names should be descriptive.

Finally, the naming should look familiar to people who just read the Prelude/standard libraries.

Naming for updated values

Since one cannot update a value in Haskell, this presents a particular problem on the naming of new versions of the same value. For example, the following code in Python:

def failover(pri, sec, inst):

becomes in Haskell something like the following:

failover pri sec inst =
  let pri'  = removePrimary pri inst
      pri'' = addSecondary pri' inst
      sec'  = removeSecondary sec inst
      sec'' = addPrimary sec' inst
  in (pri'', sec'')

When updating values, one should add single quotes to the name for up to three new names (e.g. inst, inst', inst'', inst''') and otherwise use numeric suffixes (inst1, inst2, inst3, ..., inst8), but that many updates is already bad style and thus should be avoided.

Type signatures

Always declare types for functions (and any other top-level bindings).

If in doubt, feel free to declare the type of the variables/bindings in a complex expression; this usually means the expression is too complex, however.

Similarly, provide Haddock-style comments for top-level definitions.

Use sum types instead of exceptions

Exceptions make it hard to write functional code, as alternative control flows need to be considered and compiler support is limited. Therefore, Ganeti functions should never allow exceptions to escape. Function that can fail should report failure by returning an appropriate sum type (Either or one of its glorified variants like Maybe or Result); the preferred sum type for reporting errors is Result.

As other Ganeti functions also follow these guide lines, they can safely be composed. However, be careful when using functions from other libraries; if they can raise exceptions, catch them, preferably as close to their origin as reasonably possible.

Parentheses, point free style

Prefer the so-called point-free style style when declaring functions, if applicable:

-- bad
let a x = f (g (h x))
-- good
let a = f . g . h

Also use function composition in a similar manner in expressions to avoid extra parentheses:

-- bad
f (g (h x))
-- better
f $ g $ h x
-- best
f . g . h $ x

Language features


It is recommended to keep the use of extensions to a minimum, so that the code can be understood even if one is familiar with just Haskel98/Haskell2010. That said, some extensions are very common and useful, so they are recommended:

  • Bang patterns: useful when you want to enforce strict evaluation (and better than repeated use of seq)
  • CPP: a few modules need this in order to account for configure-time options; don’t overuse it, since it breaks multi-line strings
  • Template Haskell: we use this for automatically deriving JSON instances and other similar boiler-plate

Such extensions should be declared using the Language pragma:

{-# Language BangPatterns #-}

{-| This is a small module... -}


Always use proper sentences; start with a capital letter and use punctuation in top level comments:

-- | A function that does something.
f :: ...

For inline comments, start with a capital letter but no ending punctuation. Furthermore, align the comments together with a 2-space width from the end of the item being commented:

data Maybe a = Nothing  -- ^ Represents empty container
             | Just a   -- ^ Represents a single value

The comments should be clear enough so that one doesn’t need to look at the code to understand what the item does/is.

Use -- | to write doc strings rather than bare comment with --.


We generate the API documentation via Haddock, and as such the comments should be correct (syntax-wise) for it. Use markup, but sparingly.

We use hlint as a lint checker; the code is currently lint-clean, so you must not add any warnings/errors.

Use these two commands during development:

make hs-apidoc
make hlint

QuickCheck best practices

If you have big type that takes time to generate and several properties to test on that, by default 500 of those big instances are generated for each property. In many cases, it would be sufficient to only generate those 500 instances once and test all properties on those. To do this, create a property that uses conjoin to combine several properties into one. Use counterexample to add expressive error messages. For example:

prop_myMegaProp :: myBigType -> Property
prop_myMegaProp b =
    [ counterexample
        ("Something failed horribly here: " ++ show b) (subProperty1 b)
    , counterexample
        ("Something else failed horribly here: " ++ show b)
        (subProperty2 b)
    , -- more properties here ...

subProperty1 :: myBigType -> Bool
subProperty1 b = ...

subProperty2 :: myBigType -> Property
subProperty2 b = ...

Maybe Generation

Use genMaybe genSomething to create Maybe instances of something including some Nothing instances.

Use Just <$> genSomething to generate only Just instances of something.

String Generation

To generate strings, consider using genName instead of arbitrary. arbitrary has the tendency to generate strings that are too long.