Applicative instances of IsBlock, IsInline etc.

[AshleyYakeley]

No description provided.

[jgm]

Just trying to think about the implications of UndecideableInstances and the overlap.

I've never really been clear about the potential dangers of the former, but I was under the impression that the common wisdom was to avoid it. Maybe you can tell me more?

As for the overlapping, I take it that this would still allow you to define your own newtype wrapper that is an instance of Applicative but that diverges from this default in the way it renders certain elements?

[AshleyYakeley]

• Ordinarily GHC requires that the superclass of an instance is "smaller" than the instance head. This is to prevent non-termination during compilation (although GHC will actually stop anyway after 200 reductions). UndecideableInstances switches this restriction off. Note that you are already using UndecideableInstances in some files prior to this PR.
• For overlapping, given that the more general instances (f a) are marked OVERLAPPABLE, you can have more specific instances e.g. (SourceMap a) which take priority. When GHC is trying to solve e.g. the constraint (IsInline (SourceMap a)), while both the instances match, GHC will pick the one that isn't marked OVERLAPPABLE.

[jgm]

Note that you are already using UndecideableInstances in some files prior to this PR.

I don't see that in any of the code in this repository. It could be that I've used it in some other project.

Anyway, I'm persuaded that this won't cause problems.

[AshleyYakeley]

I spelled it wrong, it's UndecidableInstances.

[jgm]

Ah yes, you're right, we already use that!

[AshleyYakeley]

Thanks