For convenience. Same as Enabled True.

For convenience. Same as Enabled False.

For convenience. Same as Long ["foo"] %+ InvLong ["no-foo"]

For convenience. Same as Short [x]

Join attribute mappings. E.g. Key1 %> Attr1 %+ Attr2 %% Key2 %> Attr3 %+ Attr4. Also possible is [ Key1 %> Attr1, Key2 %> Attr2 ] %% Key3 %> Attr3, or many other variations.

Attach a (list of) attributes to a key. The key is usually either an ADT constructor (for use with ADTFlag-style flags) or a record selector (for use with RecordFlags).

data RFlags = Flags { wibblify :: Int, simplify :: Bool }
data AFlag = Simplify | Wibblify Int
rattr = wibblify %> Help "Add a wibblification pass." (%% ...)
aattr = Wibblify %> Help "Add a wibblification pass." (%% ...)

"%+" can be used to chain multiple attributes:

attrs = wibblify %> Help "some help" %+ Default (3 :: Int) %+ ArgHelp "intensity"

But lists work just as fine:

attrs = wibblify %> [ Help "some help", Default (3 :: Int), ArgHelp "intensity" ]

Attach an attribute to multiple keys: written from right to left, i.e. Attribute <% Key1 +% Key2. Useful for setting up option groups (although using "group" may be more convenient in this case) and option enablement.

Join multiple attributes into a list. Available for convenience (using [Attribute] directly works just as well if preferred, although this is not the case with keys, see "+%").

Join multiple keys into a list, e.g. Key1 +% Key2. Useful with "<%" to list multiple (possibly heterogenously-typed) keys.

Set an attribute on all keys.

Create a group. This extracts all the keys that are (explicitly) mentioned in the body of the group and assigns the corresponding Group attribute to them. Normally used like this:

group "Group name" [ option %> Help "some help"
                   , another %> Help "some other help" ]

Do not let the type confuse you too much. :)

The ADT wrapper type allows use of classic ADTs (algebraic data types) for flag representation. The flags are then passed to the command as a list of values of this type. However, you need to make the type an instance of the Attributes first (if you do not wish to attach any attributes, you may keep the instance body empty). E.g.:

data Flag = Simplify | Wibblify Int
instance Attributes where
    attributes _ = Wibblify %> Help "Add a wibblification pass." %+ ArgHelp "intensity" %%
                   Simplify %> Help "Enable a two-pass simplifier."

The Command instances should then use (ADT Flag) for their second type parameter (the flag type).

This wrapper type allows use of record types (single or multi-constructor) for handling flags. Each field of the record is made into a single flag of the corresponding type. The record needs to be made an instance of the Attributes class. That way, attributes can be attached to the field selectors, although when used with RecordCommand, its "rec_options" method can be used as well and the Attributes instance left empty.

data Flags = Flags { wibblify :: Int, simplify :: Bool }
instance Attributes Flags where
    attributes _ =
       wibblify %> Help "Add a wibblification pass." %+ ArgHelp "intensity" %%
       simplify %> Help "Enable a two-pass simplifier."

A single value of the Flags type will then be passed to the Command instances (those that use Record Flags as their second type parameter), containing the value of the rightmost occurence for each of the flags.

TODO: List-based option types should be accumulated instead of overriden.

Use noAttributes specify an empty attribute set. Available since 0.3.2.

Chain commands into a list suitable for "dispatch" and "helpCommands". E.g.:

dispatch (Command1 %: Command2 %: Command3) opts

A class that describes a single (sub)command. The cmd type parameter is just for dispatch (and the default command name is derived from this type's name, but this can be overriden). It could be an empty data decl as far as this library is concerned, although you may choose to store information in it.

To parse the commandline for a given command, see "execute". The basic usage can look something like this:

data Flag = Summary | Unified Bool | LookForAdds Bool
instance ADTFlag Flag

[...]

data Whatsnew = Whatsnew deriving Typeable

instance Command Whatsnew (ADT Flag) where
 options _ =  enable <% Summary +% Unified +% LookForAdds
 summary _ = "Create a patch from unrecorded changes."

 run _ f opts = do putStrLn $ "Record."
                   putStrLn $ "Options: " ++ show f
                   putStrLn $ "Non-options: " ++ show opts

Given a list of commands (see "%:") and a list of commandline arguments, dispatch on the command name, parse the commandline options (see "execute") and transfer control to the command. This function also implements the help pseudocommand.

Like dispatch but with the ability to control what happens when there is an error on user input

Parse options for and execute a single command (see Command). May be useful for programs that do not need command-based "dispatch", but still make use of the Command class to describe themselves. Handles --help internally. You can use this as the entrypoint if your application is non-modal (i.e. it has no subcommands).

How to process options for a command. See "optionStyle" for details.

This could be used to implement a disambiguation function

Note that there isn't presently a notion of hidden commands, but we're taking them into account now for future API stability

A bridge that allows multi-constructor record types to be used as a description of a command set. In such a type, each constructor corresponds to a single command and its fields to its options. To describe a program with two commands, foo and bar, each taking a --wibble boolean option and bar also taking a --text=string option, you can write:

data Commands = Foo { wibble :: Bool }
              | Bar { wibble :: Bool, text :: String }

instance RecordCommand Commands where (...)

You should at least implement run', rec_options and mode_summary are optional.

Construct a command list (for "dispatch"/"helpCommands") from a multi-constructor record data type. See also RecordCommand. Alternatively, you can use "dispatchR" directly.

A command parsing & dispatch entry point for record-based commands. Ex. (see RecordCommand):

main = getArgs >>= dispatchR [] >>= \x -> case x of
  Foo {} -> putStrLn $ "You asked for foo. Wibble = " ++ show (wibble x)
  Bar {} -> putStrLn $ "You asked for bar. ..."

Like "execute", but you get the flags as a return value. This is useful to implement non-modal applications with record-based flags, eg.:

data Main = Main { greeting :: String, again :: Bool }
    deriving (Typeable, Data, Eq)
instance Attributes Main where -- (...)
instance RecordCommand Main
main = getArgs >>= executeR Main {} >>= \opts -> do
   putStrLn (greeting opts) -- (...)

Obtain a value that is an instance of Command, i.e. suitable for use with "defaultCommand" and other Command-based APIs.

Create a global setter/getter pair for a flag. The setter can be then passed to the Global attribute and the getter used globally to query value of that flag. Example:

data Flag = Wibblify Int | Verbose Bool
(setVerbose, isVerbose) = globalFlag False

instance Attributes Flag where
    attributes _ = Verbose %> Global setVerbose

putVerbose str = isVerbose >>= flip when (putStrLn str)

The default parser for option arguments. Handles strings, string lists (always produces single-element list), integers, booleans (yes|true|1 vs no|false|0), PathF and integer lists (--foo=1,2,3).

Helper for dying with an error message (nicely, at least compared to "fail" in IO).

The Data class comprehends a fundamental primitive gfoldl for folding over constructor applications, say terms. This primitive can be instantiated in several ways to map over the immediate subterms of a term; see the gmap combinators later in this class. Indeed, a generic programmer does not necessarily need to use the ingenious gfoldl primitive but rather the intuitive gmap combinators. The gfoldl primitive is completed by means to query top-level constructors, to turn constructor representations into proper terms, and to list all possible datatype constructors. This completion allows us to serve generic programming scenarios like read, show, equality, term generation.

The combinators gmapT, gmapQ, gmapM, etc are all provided with default definitions in terms of gfoldl, leaving open the opportunity to provide datatype-specific definitions. (The inclusion of the gmap combinators as members of class Data allows the programmer or the compiler to derive specialised, and maybe more efficient code per datatype. Note: gfoldl is more higher-order than the gmap combinators. This is subject to ongoing benchmarking experiments. It might turn out that the gmap combinators will be moved out of the class Data.)

Conceptually, the definition of the gmap combinators in terms of the primitive gfoldl requires the identification of the gfoldl function arguments. Technically, we also need to identify the type constructor c for the construction of the result type from the folded term type.

In the definition of gmapQx combinators, we use phantom type constructors for the c in the type of gfoldl because the result type of a query does not involve the (polymorphic) type of the term argument. In the definition of gmapQl we simply use the plain constant type constructor because gfoldl is left-associative anyway and so it is readily suited to fold a left-associative binary operation over the immediate subterms. In the definition of gmapQr, extra effort is needed. We use a higher-order accumulation trick to mediate between left-associative constructor application vs. right-associative binary operation (e.g., (:)). When the query is meant to compute a value of type r, then the result type withing generic folding is r -> r. So the result of folding is a function to which we finally pass the right unit.

With the -XDeriveDataTypeable option, GHC can generate instances of the Data class automatically. For example, given the declaration

data T a b = C1 a b | C2 deriving (Typeable, Data)

GHC will generate an instance that is equivalent to

instance (Data a, Data b) => Data (T a b) where
    gfoldl k z (C1 a b) = z C1 `k` a `k` b
    gfoldl k z C2       = z C2

    gunfold k z c = case constrIndex c of
                        1 -> k (k (z C1))
                        2 -> z C2

    toConstr (C1 _ _) = con_C1
    toConstr C2       = con_C2

    dataTypeOf _ = ty_T

con_C1 = mkConstr ty_T "C1" [] Prefix
con_C2 = mkConstr ty_T "C2" [] Prefix
ty_T   = mkDataType "Module.T" [con_C1, con_C2]

This is suitable for datatypes that are exported transparently.

The class Typeable allows a concrete representation of a type to be calculated.

Computation getArgs returns a list of the program's command line arguments (not including the program name).