Xia Li-yao

I was probably very very tired when I submitted this :)

That an input contains no `NaN` seems like a reasonable precondition to assume implicitly, similar to inputs not being `undefined`.

Isn't the original comment referring to `QCGen` as the seed?

#158 seems relevant.

These instances seem weird to me, including the already existing `Functor`s. We shouldn't add them "just because we can". Do you have a use case for these? My objection is...

How about providing counterexamples as `Dynamic` values? Just throwing the idea out there for the moment.

`Control.Monad.Fail` is a relatively recent addition to `base`, you need to depend on `fail` in other to support versions as old as `base-4.3`

How about putting this instance in a separate module? That provides the option to opt in while keeping total pattern matches available by default. Libraries might have a hard time...

Orphans are problematic because they lead to incoherence. In this case there seems to be only one `MonadFail` instance might one ever want to write for `Gen`.

The tag coq8.10 just identifies a commit. Since you already have found a commit that builds, you can leave it be for now and come back to it later if...