Implement previews

[nimec01]

This pull request implements previews that allow us to show preview images of incomplete programs.

Example:

import CodeWorld

{-| Draw something -}
main :: IO ()
main = drawingOf (translated 5 undefined (circle 3) & undefined)

will produce this image:

The feature can be enabled per-request by adding a enablePreview field to the /compile request that has value true. It can also be enabled by default via a config setting. The field value in the request will then override the default.

The config path can be set via the CONFIG_PATH environment variable. The content of the config.yaml would then look like this:

preview:
  enabledByDefault: false
  defaultHoleValues:
    Double: "0"
    Color: "black"
    Picture: "blank"

(My further goal is to replace the extensions.yaml config file by a field in this new consolidated config.)

A few questions left:

• Should this also work in the non-Haskell mode? (It currently does not)
• Should the user be able to configure this itself in the editor?

[jvoigtlaender]

• Should this also work in the non-Haskell mode? (It currently does not)

No, that’s not needed.

• Should the user be able to configure this itself in the editor?

I wasn’t envisioning this in the editor-and-compile-view, at least not originally. My idea was to use this feature in the post-to-API-to-get-a-display-without-code.

About the configurable hole-filling: does this also allow for „undefined :: Double -> Picture“? (which could be replaced by „const blank“)

[nimec01]

About the configurable hole-filling: does this also allow for „undefined :: Double -> Picture“? (which could be replaced by „const blank“)

The regular expression should not cover that currently. I'll try to update it.

[nimec01]

Function holes are now supported as well. Config would then look like this:

preview:
  defaultHoleValues:
    "Double -> Picture": "const blank"
    "Double": "0"

[jvoigtlaender]

Wie robust ist das gegen Vorkommen von „undefined“ im Programmtext, die gar nicht für Prelude.undefined stehen, sondern etwa „let undefinedNumber = 42 in whatever“ oder einen String, der „undefined“ enthält?

Sollte vielleicht zumindest gecheckt werden, dass der GHC genau so viele typed holes meldet, wie man vorher „undefined“ durch „_“ ersetzt hat?

[jvoigtlaender]

Das heißt, ein weiterer Compile wird selbst dann durchgeführt, wenn überhaupt keine Ersetzungen vorgenommen wurden (weil keine undefineds vorkamen)?

Das sollte sicher optimiert werden.

[jvoigtlaender]

Anschließend: fmidue/haskell-template-task#40