Use m ~ IO to improve type inference

[ocharles]

Currently, one has to use the hedgehog identity function to force the type of the test to be PropertyT IO (). A better way to do this is to instead use an equality constraint in the instance head. This means that when we typecheck it we encounter PropertyT m () as before, but we're now able to find an instance that matches. Refining the context will the force m to be IO and everything works out fine.

[ocharles]

I think with this you could actually deprecate hedgehog and change the docs to not suggest pure () :: PropertyT m (), but I wanted to present just this change first.

[sol]

This looks great 👍

@parsonsmatt can we get this into a release?

[sol]

@parsonsmatt ping.

[parsonsmatt]

Looks good! Mind updating changelog and doing a major version bump?

[parsonsmatt]

So the main issue here is that end users won't be able to define their own instances of Example (PropertyT TheirNeatMonad ())

Which I'm pretty cool with, I think? But worth noting that it's a breaking change.

Wondering if putting an inference constraint on the (), too:

Suggested change

	instance m ~ IO => Example (PropertyT m ()) where
	instance (a ~ (), m ~ IO) => Example (PropertyT m a) where

This would forbid users from writing instances like:

instance (m ~ IO) => Example (PropertyT m (SqlPersistT m (TestT m ()))) where
  type Arg _ = SqlBackend

  -- run the property, 
  -- run the action,
  -- run the asserts

Which - idk - maybe it's good and proper to make that less magical?

[sol]

Wondering if putting an inference constraint on the (), too

I'm trying to mentally piece together what the tangible benefits would be. I guess you could put an undefined as the last statement of a do-block and things may still type check. You may also get better type errors at times.

From what I understand, adding this constrained does not change the picture fundamentally. But it could be nice to have.

For me, the essential thing is to constrain the base monad to IO. This is what I have been using for the last couple of months. Getting this on Hackage will remove a lot of friction points for me.

[sol]

Looks good! Mind updating changelog and doing a major version bump?

Done as a separate PR (#27) to not unnecessarily cause conflicts.

[ocharles]

Thanks both!