Respect fixity declarations inside where/let/class in applyFixities.

[barrucadu]

Both here,

foo = pure 1 <^> pure 2 <^> pure (+) where
  (<^>) = flip (<*>)
  infixr 4 <^>

and here,

class Foo f where
  (<^>) :: Applicative f => f a -> f (a -> b) -> f b
  infixr 4 <^>

foo = pure 1 <^> pure 2 <^> pure (+)

In the body of foo, (<^>) is right-associative (and indeed GHC parses it as such). This patch modifies applyFixities to grab fixities in Binds and ClsDecls and take them into account.

---

I'm not sure how to write tests for this, as none of the test groups use applyFixities. Should I add another group?

[pjonsson]

How well do you understand the fixity resolution code in HSE? What is the risk that this change will break something else?

I think I merged a pull request that will break building of this code. Could you please update it against latest HEAD?

Style issue: your let expressions doesn't indent the in an extra column like the surrounding code.

Define extraFix so you can write the code for Let like this: liftM2 Let (extraFix bs bs) $ extraFix bs e

Don't hijack the name fix in the instance for AppFixity Matchto do something else than the other definitions of fix in the file.

[pjonsson]

@barrucadu: I apologize for not getting back on this code sooner. If you're busy, can you please respond to the first question and we'll take care of the rest to get this merged for 1.17.

[barrucadu]

I'm not intimately familiar with the fixity resolution, but have tested this by parsing some complete modules, applying fixities, and checking everything was expected. I would be surprised if anything broke.

[pjonsson]

Merged, thanks!