1   Introduction

A common pattern when using diagrams is to make a program that generates a single diagram. Often there are details about the diagram that we would like to delay in fixing. In particular the output file, image format, and final size of the diagram are all standard options we would expect for making a diagram that are relatively independent of the image being made. With the Diagrams.Backend.SVG.CmdLine module (and related modules in other backends) we provide easy creation of a command-line interface that supports standard options as well as easy customization for additional parameters.

To give some concrete examples, the following are programs one can write using the tools already provided, and their interaction on the command-line. For an example of a custom extension to this framework, see the User Extensions section below.

First is the simplest case of generating a single diagram:

> -- Simple
>
> d :: Diagram SVG R2
> d = ...
>
> main = mainWith d

Here we just have a diagram and the standard options. We can invoke with just a width and the height is made to match the diagram.

$ ./Simple -o simple.svg -w 100

If we have multiple diagrams with names we can use mainWith to give an interface that allows the selection of a particular diagram by name.

> -- Multiple
>
> d1, d2, d3 :: Diagram SVG R2
> ...
>
> main = mainWith [("First", d1),("Second", d2),("Third", d3)]

The --list option just lists the available diagrams to render and the -s selection option takes a name and renders the associated diagram with the standard options.

$ ./Multiple --list
Available diagrams:
  First Second Third
$ ./Multiple -o d1.svg -w 100 -s First

Some backends support rendering animations (typically as individually indexed files of frames).

> -- Animation
>
> a :: Animation Cairo R2
> a = ...
>
> main = mainWith a

The default options are expected, but the output file has an index appended to the name for each frame. The --fpu option indicates the desired number of frames per unit of time. If the a animation above is one second long then the following will create files a01.png through a24.png.

$ ./Animation -o a.png -w 100 --fpu 24

In backends that support multiple pages we can list all the diagrams and have each render on its own page.

> -- Pages
>
> d1, d2, d3 :: Diagram Postscript R2
> ...
>
> main = mainWith [d1,d2,d3]

We only need the default options here and the interface is the same as a single diagram.

$ ./Pages -o pages.ps -w 400

To make things more interesting we could require additional arguments to build a diagram. We can take a function to build a diagram from some parameters and create an interface that fills those parameters with arguments from the command-line.

> -- Function
>
> f :: Colour Double -> Double -> Diagram SVG R2
> f c x = ...
>
> main = mainWith f

In addition to the standard flags, we can provide arguments blue and 42.0 which will be passed along to f.

$ ./Function -o blue.svg -w 400 blue 42.0

In addition to Colours and Doubles, default command-line parsers are provided for Int, String, and AlphaColour arguments. You can also easily define your own parsers for additional argument types; see the User Extensions section below.

Diagrams that are the result of executing an IO action can also be handled by mainWith. This can be useful for reading input files or doing other IO that the diagram depends on.

> -- IO-diagram
>
> d :: FilePath -> IO (Diagram SVG R2)
> d file = do
>     f <- readFile file
>     ...
>
> main = mainWith d

The program will expect a file name on the command-line which it reads to generate a diagram.

$ ./IO-diagram -o file.svg -w 400 ~/data.log

See the Clock Example section below.

2   Standard Options

The standard options for diagram creation are found in the Diagrams.Backend.CmdLine module of diagrams-lib and are represented with the following record:

> data DiagramOpts = DiagramOpts
>   { _width     :: Maybe Int
>   , _height    :: Maybe Int
>   , _output    :: FilePath
>   }

This models having an optional width and height and a required FilePath. We also need to have a parser for command-line arguments that results in a value of this record. We use the optparse-applicative package for command-line parsing and the parser for the standard options is the following:

> diagramOpts :: Parser DiagramOpts
> diagramOpts = DiagramOpts
>     <$> (optional . option)
>         ( long "width" <> short 'w'
>        <> value 400
>        <> metavar "WIDTH"
>        <> help "Desired WIDTH of the output image (default 400)")
>     <*> (optional . option)
>         ( long "height" <> short 'h'
>        <> value 400
>        <> metavar "HEIGHT"
>        <> help "Desired HEIGHT of the output image (default 400)")
>     <*> strOption
>         ( long "output" <> short 'o'
>        <> value ""
>        <> metavar "OUTPUT"
>        <> help "OUTPUT file")

This is written in applicative form, Constructor <$> ... <*> ... <*> ..., where the values we fill in are the parsers for the fields of the record. The parsers in optparse-applicative take as an argument a collection of parameters.

3   Abstracting Main

This section walks through and motivates the design of the abstraction mechanisms that make possible the examples shown above. If you want to create your own custom command-line-driven diagram generation executables, you will likely find it helpful to understand this section. The truly impatient, however, may wish to skip directly to User Extensions and return to this section as necessary.

What work does the backend need to do to render a diagram? It depends on the backend, of course, but there are several common tasks given the standard options. To start with we need to parse the command-line arguments. The optparse-applicative package provides all the tools we need for this. Next we will need to translate the standard arguments into something backend specific. Typically the extension on the output filename will drive the format of the output and some combination of the supplied width and height will dictate the final scale of the diagram. Let's look at a full example of a backend doing this work and try to see what parts we can abstract out for general use (we will use the Cairo backend for this example).

> defaultMain :: Diagram Cairo R2 -> IO ()
> defaultMain d = do
>   prog <- getProgName
>   let p = info (helper' <*> diagramOpts)
>               ( fullDesc
>              <> progDesc "Command-line diagram generation."
>              <> header prog)
>   opts <- execParser p
>   chooseRender opts d
>
> chooseRender :: DiagramOpts -> Diagram Cairo R2 -> IO ()
> chooseRender opts d =
>   case splitOn "." (output opts) of
>     [""] -> putStrLn "No output file given."
>     ps | last ps `elem` ["png", "ps", "pdf", "svg"] -> do
>            let outTy = case last ps of
>                  "png" -> PNG
>                  "ps"  -> PS
>                  "pdf" -> PDF
>                  "svg" -> SVG
>                  _     -> PDF
>            fst $ renderDia
>                    Cairo
>                    ( CairoOptions
>                      (output opts)
>                      (mkSizeSpec
>                        (fromIntegral <$> width opts)
>                        (fromIntegral <$> height opts)
>                      )
>                      outTy
>                      False
>                    )
>                    d
>        | otherwise -> putStrLn $ "Unknown file type: " ++ last ps

There are several things that make this structuring of the program inflexible. Let's consider building a main where we accept a function that can produce a diagram.

> functionMain :: (a -> Diagram Cairo R2) -> IO ()

Clearly we cannot use the given function as we have no way to produce an a. So we provide a type class called Parseable for associating a parser with the type that it parses:

> class Parseable a where
>    parser :: Parser a

Now we can make more progress.

> functionMain :: Parseable a => (a -> Diagram Cairo R2) -> IO ()
> functionMain f = do
>   prog <- getProgName
>   let p = info (helper' <*> ((,) <$> diagramOpts <*> parser))
>               ( fullDesc
>              <> progDesc "Command-line diagram generation."
>              <> header prog)
>   (opts,a) <- execParser p
>   chooseRender opts (f a)

The only parts so far that are backend specific are the type of the final diagram and chooseRender, though we may want other parts may be subject to customization. We will split this into four parts, the type of the options needed, the action of parsing the command-line, the backend specific rendering, and an entry point for the library consumer. We will give this the brilliant name Mainable.

> class Mainable d where
>    type MainOpts d :: *
>
>    mainArgs   :: Parseable (MainOpts d) => d -> IO (MainOpts d)
>    mainRender :: MainOpts d -> d -> IO ()
>    mainWith   :: Parseable (MainOpts d) => d -> IO ()

There is one associated type and three class methods. Let's consider the instance of Mainable for a simple diagram with type Diagram Cairo R2:

> instance Mainable (Diagram Cairo R2) where

The associated type indicates what options we will want to be parsed from the command-line. In this case we will just use the standard options:

>     type MainOpts (Diagram Cairo R2) = DiagramOpts

The mainArgs method is nearly what we had before. In this case there isn't anything backend specific, so instead of an instance implementation we will show the default implementation for mainArgs. Instead of a specific parser diagramOpts we have a constraint Parseable (MainOpts d) allowing us to use parser where we had diagramsOpts. The parser from the constraint is combined with some additional standard configuration for the program name and the right kind of help message. Running the mainArgs IO action results in either the program quitting with a parse error or help message, or the program continuing with the parsed value. Also note that we need the diagram to be passed to mainArgs only to fix the type so we can use our associated type function MainOpts.

>     mainArgs :: Parseable (MainOpts d) => d -> IO (MainOpts d)
>     mainArgs _ = do
>       prog <- getProgName
>       let p = info (helper' <*> parser)
>                   ( fullDesc
>                  <> progDesc "Command-line diagram generation."
>                  <> header prog)
>       execParser p

The next method to implement is the mainRender method. Here we can just use the chooseRender function we had before, handling all the backend specific interpretation of the standard arguments.

>     mainRender :: DiagramOpts -> Diagram Cairo R2 -> IO ()
>     mainRender = chooseRender

Finally we have mainWith which joins the previous parts to make an entry point for users of the backend to build their programs. In this example we take as an argument the Diagram Cairo R2 and result in a complete program. Again, we can get away with the default implementation.

>     mainWith :: Parseable (MainOpts d) => d -> IO ()
>     mainWith d = do
>         opts <- mainArgs d
>         mainRender opts d

Now let's try a much harder instance. We want to be able to handle functions whose final result has a Mainable instance, but require some Parseable arguments first. The tricky part of this instance is that we need to know up front what all our arguments are going to be in order to be able to parse them. It sounds tempting to peel off one argument at a time, parse, apply, and recurse with one less argument; but this does not work. To facilitate collecting the arguments, we make a new type class that has associated types for all the arguments of the type and the final result of the type. It will also contain a function to perform the application of all the arguments and give the final result.

> class ToResult d where
>     type Args d :: *
>     type ResultOf d :: *
>
>     toResult :: d -> Args d -> ResultOf d

We will need a base case for when we have reached the final result. It needs no arguments so we use the unit type for Args and the final result is just the diagram itself.

> instance ToResult (Diagram b v) where
>     type Args (Diagram b v) = ()
>     type ResultOf (Diagram b v) = Diagram b v
>
>     toResult d _ = d

Now we can write the inductive case of a function resulting in something with a ToResult instance.

> instance ToResult d => ToResult (a -> d) where
>     type Args (a -> d) = (a, Args d)
>     type ResultOf (a -> d) = ResultOf d
>
>     toResult f (a,args) = toResult (f a) args

Here Args is the product of the argument and any arguments that d demands. The final result is the final result of d and to produce a result we apply one argument and recurse to d's ToResult instance.

Now that we have ToResult to work with, we can write the type for the instance of Mainable that we want:

> instance (Parseable a, Parseable (Args d), ToResult d, Mainable (ResultOf d))
>         => Mainable (a -> d) where
  1. Think about this type for a bit.

Now we need a type for MainOpts (a -> d) and at least an implementation for mainRender. Remember the purpose of MainOpts is to give a type for all the arguments needed. We will need the MainOpts from the final result and some structure containing all the function arguments. Note that we rely on having a Parseable instance for products.

>     type MainOpts (a -> d) = (MainOpts (ResultOf (a -> d)), Args (a -> d))

Our mainRender will be handed a value of this type and a function of our instance type. We can use toResult to apply the second part of the pair to the function and hand the final result off to the final result's Mainable instance along with its required options.

>     mainRender (opts, a) f = mainRender opts (toResult f a)

Now we compile and cross our fingers!

We can also handle IO with a couple more instances. First we will need a ToResult instance that handles IO actions:

> instance ToResult d => ToResult (IO d) where
>    type Args (IO d) = Args d
>    type ResultOf (IO d) = IO (ResultOf d)
>
>    toResult d args = flip toResult args <$> d

This states that the needed arguments are not affected by this being an IO action and the final result is an IO action producing the final result of the action's result type. Our Mainable instance can now be written:

> instance Mainable d => Mainable (IO d) where
>     type MainOpts (IO d) = MainOpts d
>
>     mainRender opts dio = dio >>= mainRender opts

Here we merely perform the diagram creating action and bind its value to the Mainable instance that can handle it. For an example of using these instances see the Clock Example section below.

4   User Extensions

You can easily build on top of this framework to create executables taking your own custom command-line flags. This section walks through a simple example.

Suppose we want to make "flippable" diagrams: a single executable that can render either a diagram or its mirror image, depending on a command-line flag. Of course we also want to support all the usual command-line options like --width, --height, --output, and so on. The framework described above—together with the composability of optparse-applicative-based command-line parsers—makes this very easy to do.

First, some pragmas and imports:

> {-# LANGUAGE FlexibleInstances         #-}
> {-# LANGUAGE NoMonomorphismRestriction #-}
> {-# LANGUAGE TypeFamilies              #-}
>
> import           Diagrams.Backend.CmdLine
> import           Diagrams.Backend.SVG.CmdLine
> import           Diagrams.Prelude             hiding ((<>))
> import           Options.Applicative

(Unfortunately, Options.Applicative re-exports the (<>) from Data.Monoid, whereas Diagrams.Prelude re-exports the one from Data.Semigroup.)

We now create a newtype for "flippable" things:

> newtype Flippable a = Flippable a

We need a newtype since we need to make a Mainable instance which is different than the default instance for Diagram SVG R2.

We create a data structure to contain our new command-line options, along with a Parseable instance for it. In this case we just want a single Bool value, corresponding to a new command-line switch --flipped along with an appropriate help message

> data FlipOpts = FlipOpts Bool
>
> instance Parseable FlipOpts where
>   parser = FlipOpts <$> switch (long "flipped" <> help "Flip the diagram L-R")

For help on constructing command-line parsers, see the documentation for the optparse-applicative package; you can also look at the source code of Diagrams.Backend.CmdLine for some examples.

Finally, we create a Mainable instance for flippable diagrams. The MainOpts for flippable diagrams consists of a pair of our new FlipOpts along with the MainOpts for diagrams. To implement mainRender, we take in our options and a flippable diagram, and pass the diagram-specific options along to the mainRender method for diagrams, flipping the diagram appropriately.

> instance Mainable (Flippable (Diagram SVG R2)) where
>   type MainOpts (Flippable (Diagram SVG R2)) = (MainOpts (Diagram SVG R2), FlipOpts)
>
>   mainRender (opts, FlipOpts f) (Flippable d) = mainRender opts ((if f then reflectX else id) d)

Let's try it out!

> d :: Diagram SVG R2
> d = square 1 # fc red ||| square 1 # fc blue
>
> main = mainWith (Flippable d)

Note the --flipped option in the help message:

$ ./Flippable --help

Flippable

Usage: Flippable [-w|--width WIDTH] [-h|--height HEIGHT] [-o|--output OUTPUT] [-l|--loop] [-s|--src ARG] [-i|--interval INTERVAL] [--flipped]
  Command-line diagram generation.

Available options:
  -?,--help                Show this help text
  -w,--width WIDTH         Desired WIDTH of the output image (default 400)
  -h,--height HEIGHT       Desired HEIGHT of the output image (default 400)
  -o,--output OUTPUT       OUTPUT file
  -l,--loop                Run in a self-recompiling loop
  -s,--src ARG             Source file to watch
  -i,--interval INTERVAL   When running in a loop, check for changes every INTERVAL seconds.
  --flipped                Flip the diagram L-R

And running it yields:

$ ./Flippable -o Flippable.svg -w 400
$ ./Flippable -o Flippable.svg -w 400 --flipped

It works!

It is also worth noting that for this simple example, we actually did not need the Flippable wrapper or Mainable instance at all! Given only the FlipOpts type and its Parseable instance, we can simply write

> main = mainWith (\(FlipOpts f) -> (if f then reflectX else id) d)

which gives us exactly the same program! Indeed, if you squint at the function instance for Mainable and the instance we wrote for Flippable, you can see that our instance is a direct specialization of the more general one.

5   Clock Example

We may want to build diagrams based on the state of the world. For instance, if we want to build a diagram of a clock we will want to know what time it is. Consider the following program.

> import Diagrams.Prelude
> import Diagrams.Coordinates
> import Data.Time
>
> clock :: UTCTime -> Diagram B R2
> clock t = circle 0.35 # fc silver # lwG 0
>        <> bigHand # f 12 h <> littleHand # f 60 m
>        <> circle 1  # fc black # lwG 0
>        <> circle 11 # lwG 1.5 # lc slategray # fc lightsteelblue
>   where
>     s = realToFrac $ utctDayTime t :: Double
>     m = s / 60
>     h = m / 60
>
>     bigHand    = (0 ^& (-1.5)) ~~ (0 ^& 7.5) # lwG 0.5
>     littleHand = (0 ^& (-2))   ~~ (0 ^& 9.5) # lwG 0.2
>     f n v = rotate (- v / n @@ turn)
>
> main = mainWith (clock <$> getCurrentTime)

Running we get:

This uses the Mainable d => Mainable (IO d) instance to allow our effectful clock generator. However, we could have just as well avoided this instance, writing instead:

> main = do
>     t <- getCurrentTime
>     mainWith (clock t)

This instance is quite convenient, however, especially when we want our IO action to depend on some command-line option. The following exercises should be helpful in gaining practice working with IO and options by modifying the clock example into a useful clock making program.

  1. Modify the example so it can take a time as a command-line option, but if one is not given, it uses the current time.

  2. Modify clock to take a ClockStyle argument that includes options for various visual styles for the clock. For instance ClockStyle could include a color for the clock background, a flag for turning on hour marks, or a flag for including a second hand.