Data.Foldable `Maybe` instance mapM_ inlining and/or specialisation?

Motivation

Wisely, or otherwise, I've been using mapM_ action as idiom that performs action input given a Just input value, and does nothing given a Nothing value. I rather expected that all the abstractions involved are erased through inlining and specialisation, and that this would be a more concise and equivalent in cost variant of maybe (return ()) action.

However, benchmarking this shows that the generated code is not the same in both cases. For trivial actions and input, measuring just the overhead, the mapM_ variant was ~1.5x slower. Is this expected? It seems that the Maybe instance of Foldable is not inlined in a manner that allows the code to be reduced to the logically equivalent maybe (return ()) action form.

Proposal

If there are no downsides or obstacles, perhaps inline the foldr instance of Maybe, and, as appropriate, mapM_ and/or traverse_. There are perhaps implementation constraints imposed by list fusion that I don't understand, so this is more of a question at this stage than a concrete proposal.

https://gitlab.haskell.org/ghc/ghc/-/blob/master/libraries/base/Data/Foldable.hs#L587-592 https://gitlab.haskell.org/ghc/ghc/-/blob/master/libraries/base/Data/Foldable.hs#L905-909 https://gitlab.haskell.org/ghc/ghc/-/blob/master/libraries/base/Data/Foldable.hs#L1181-1257

Cc: @hsyl20