Joseph C. Sible

AFAIK, of all the possibilities in your post, only `flip const` has the problem I described, and all of the other combinations will be fine.

Looks like your 4th line has a typo: it should be `f = (\f x y -> f y x) (\x _ -> x)`. Anyway, the performance problem stems from...

After a beta-reduction, `f = (\f x y -> f y x) (\x _ -> x)` becomes `f = (\x y -> (\x _ -> x) y x)`. Now pass...

> So both definitions of f have a problem? That I understand, but wasn't my understanding of your statement. I realize I was a bit unclear. These are fine: ```...

Worth noting that `-Wincomplete-patterns` does tell you exactly which pattern was missing. Perhaps we could leverage that somehow.

To clarify, I definitely don't want a new function called `readMaybe` with a different type. I definitely agree if we add one, it should have a different name.

A Maven repo would also be a good way to solve this.

I can confirm that it builds fine on 9.2.1 with that patch.

@treeowl I don't have a use of my own for these. I wrote them because you said you wanted them in #814.

@alexfmpe Hah, see https://github.com/haskell/containers/pull/817 and https://github.com/haskell/containers/issues/814#issuecomment-1023539195