Pass to GHC visible modules for instance filtering

[harpocrates]

The GHC-side getNameToInstancesIndex filters out incorrectly some
instances because it is not aware of what modules are visible. On the
Haddock side, we need to pass in the modules we are processing.

On the GHC side, we need to check against those modules when checking
if an instance is visible.

Here is the GHC patch: https://phabricator.haskell.org/D4290

[alexbiehl]

Very good catch Alec! Looking forward to this!

[harpocrates]

Me too! This sort of bug is the worst from a user perspective - you don't know that there are instances you aren't seeing, and once you do, you cease to trust Haddock.

Please do let me know if there is anything I should/could do to help this along - I understand the scheduling is a bit tight. I will endeavor to be responsive. 😃

[harpocrates]

Alright. I circled back and figured out why #469 (the Read instance of Array) was missing. The problem turned out to be only tangentially related to getNameToInstancesIndex.

We simply weren't updating the GlobalRdrEnv when processing modules, which meant that loadUnqualIfaces in getNameToInstancesIndex would load nothing (in fact, that was my clue: I noticed that it was loading stuff when I would use :info in GHCi, but it wasn't loading anything when called from Haddock).

The fix is just to update the environment every time we process a module. There are no extra GHC-side changes.

AFAICT, there are no more missing instances anywhere.

[alexbiehl]

I will merge this as is, could you open another ghc diff updating the haddock submodule?

[harpocrates]

I'll open a new diff as soon as this is merged. (Anything wrong with just updating my existing diff: https://phabricator.haskell.org/D4290?)