butcher

Chops a command or program invocation into digestable pieces.

Similar to the optparse-applicative package, but less features, more flexibility and more evil.

The main differences are:

• Provides a pure interface by default

• Exposes an evil monadic interface, which allows for much nicer binding of command part results to some variable name.

  In optparse-applicative you easily lose track of what field you are modifying after the 5th <*> (admittedly, i think -XRecordWildCards improves on that issue already.)

  Evil, because you are not allowed to use the monad's full power in this case, i.e. there is a constraint that is not statically enforced. See below.

• The monadic interface allows much clearer definitions of commandparses with (nested) subcommands. No pesky sum-types are necessary.

Examples

The minimal example is

main = mainFromCmdParser $ addCmdImpl $ putStrLn "Hello, World!"

But lets look at a more feature-complete example:

main = mainFromCmdParserWithHelpDesc $ \helpDesc -> do

  addCmdSynopsis "a simple butcher example program"
  addCmdHelpStr "a very long help document"

  addCmd "version" $ do
    porcelain <- addSimpleBoolFlag "" ["porcelain"]
      (flagHelpStr "print nothing but the numeric version")
    addCmdHelpStr "prints the version of this program"
    addCmdImpl $ putStrLn $ if porcelain
      then "0.0.0.999"
      else "example, version 0.0.0.999"

  addCmd "help" $ addCmdImpl $ print $ ppHelpShallow helpDesc

  short <- addSimpleBoolFlag "" ["short"]
    (flagHelpStr "make the greeting short")
  name <- addStringParam "NAME"
    (paramHelpStr "your name, so you can be greeted properly")

  addCmdImpl $ do
    if short
      then putStrLn $ "hi, " ++ name ++ "!"
      else putStrLn $ "hello, " ++ name ++ ", welcome from butcher!"

Further:

• Full description of the above example, including sample behaviour
• Example of a pure usage of a CmdParser
• Example of using a CmdParser on interactive input
• The brittany formatting tool is a program that uses butcher for implementing its commandline interface. See its main module source or the config flag parser.

The evil monadic interface

As long as you only use Applicative or (Kleisli) Arrow, you can use the interface freely. When you use Monad, there is one rule: Whenever you read any command-parts like in

f <- addFlag ...
p <- addParam ...

you are only allowed to use bindings bound thusly in any command's implemenation, i.e. inside the parameter to addCmdImpl. You are not allowed to force/inspect/patternmatch on them before that. good usage is:

addCmdImpl $ do
  print x
  print y

while bad would be

f <- addFlag
when f $ do
  p <- addParam
  -- evil: the existence of the param `p`
  -- depends on parse result for the flag `f`.

That means that checking if a combination of flags is allowed must be done after parsing. (But different commands and their subcommands (can) have separate sets of flags.)

(abstract) Package intentions

Consider a commandline invocation like "ghc -O -i src -Main.hs -o Main". This package provides a way for the programmer to simultaneously define the semantics of your program based on its arguments and retrieve documentation for the user. More specifically, i had three goals in mind:

1. Straight-forward description of (sub)command and flag-specific behaviour
2. Extract understandable usage/help commandline documents/texts from that descriptions, think of ghc --help or stack init --help.
3. Extract necessary information to compute commandline completion results from any partial input. (This is not implemented to any serious degree.)

Semantics

(Sorry, this description is severely lacking, I know.)

Basic elements of a command are flags, parameters and subcommands. These can be composed in certain ways, i.e. flags can have a (or possibly multiple?) parameters; parameters can be grouped into sequences, and commands can have subcommands.

Commands are essentially String -> Either ParseError out where out can be chosen by the user. It could for example be IO ().

To allow more flexible composition, the parts of a command have the "classic" parser's type: String -> Maybe (p, String) where p depends on the part. Parse a prefix of the input and return something and the remaining input, or fail with Nothing.

A command-parser contains a sequence of parts and then a number of subcommands and/or some implementation.