Transform the arrow shaft style

[cchalmers]

Hopefully fixes #274.

It seems to be fixed by applying the transform to the shaft style. I've applied the transform before applying the global style since this (should) already have the transform applied to it.

Here's an example comparing arrow shafts to paths for different dashings:

import Diagrams.Backend.Rasterific.CmdLine
import Diagrams.Prelude

arrows :: Diagram Rasterific
arrows = mconcat
  [ arrowV' (with & shaftStyle %~ dashingL [0.25, 0.25] 0) unitX
  , arrowV' (with & shaftStyle %~ dashingO [5, 5] 0) unitY
  , arrowV' (with & shaftStyle %~ dashingN [0.05, 0.05] 0) unit_X
  , arrowV' (with & shaftStyle %~ dashingG [0.2, 0.2] 0) unit_Y
  , square 2
  ]

straights :: Diagram Rasterific
straights = mconcat
  [ v unitX  # dashingL [0.25, 0.25] 0
  , v unitY  # dashingO [5, 5] 0
  , v unit_X # dashingN [0.05, 0.05] 0
  , v unit_Y # dashingG [0.2, 0.2] 0
  , square 2
  ] where v = stroke . (`at` origin) . straight

main :: IO ()
main = defaultMain . frame 0.2 $
  (arrows    ||| arrows # scale 2)
             ===
  (straights ||| straights # scale 2)

[byorgey]

Ah, yes, that makes sense, thanks!

The travis failure seems to be an out-of-memory error while running a test suite, not sure if that is due to a bug in the test suite or just plain insufficient memory, but in any case it's clearly not caused by this PR.

[bergey]

That's an interesting failure. It looks stochastic (one build out of five, not always the same GHC version), which makes sense since it's in a QuickCheck test. We should try limiting the list length in the Arbitrary instance, and see if it goes away.

I think travis allocates more RAM to the new containerized builders. We should switch all the travis configs over at some point; this is another reason.

[bergey]

@dashhalpern points out that we're generating Arbitrary Directions' , notPath`s, so it's less obvious how to pick Directions that don't lead to huge lists.

[byorgey]

I made a new issue for the test suite failures: #277 . Let's continue discussing it there.