Add `MonadFail` instance to `Gen` data type

[chshersh]

Without this instance it's impossible to perform partial pattern-matching inside property-based tests. Maybe this instance can be somehow smart, like, discard all cases that doesn't match the case?

[moodmosaic]

If you switch to GHC 8.6 you'll also get a compiler error because of the MonadFail proposal, as discussed in #233. In fact, @HuwCampbell proposed

instance Monad m => MonadFail (GenT m) where
  fail _ =
    mzero

which actually discards ━ a change we don't consider making at the moment, as it requires a major version bump. So, for this reason (still in #233) we lean towards

instance Monad m => MonadFail (GenT m) where
  fail =
    error

which at least won't break your code when switching to GHC 8.6. ━ It'd be nice to have your thoughts around this.

[chshersh]

@moodmosaic I'm actually okay with having error for now, works for me as well 👍

[ocharles]

I think discarding is the correct thing to do, personally. One could imagine doing something like

Just foo <- forAll ...

And having tests run, rather than the test crash.

[moodmosaic]

Yes, discarding looks good, but also requires a major version bump, because it's a breaking change. In reality though, it's more or less what @sol tries to explain here:

MonadFail is still fairly new and we as a community don't have the collective experience yet on how things will pan out in practice.

That you get compile-time errors instead of runtime errors in some situations now is an intentional change. It's the whole point of introducing MonadFail.

I'm really curious to see what happens with nick8325/quickcheck#228.

[fosskers]

Any more thoughts here? Or should we just consider it incorrect to pattern match in this way in tests?

[moodmosaic]

Follow #257 for updates.

[fosskers]

Thanks!