Should transformers-compat provide Control.Monad.Trans.Instances?

[RyanGlScott]

In #9, I requested the Control.Monad.Trans.Instances module to backport instances that were introduced in more recent versions of transformers, by way of orphan instances. Some time later, I'm starting to think this isn't a good idea, for a couple of reasons:

• The PVP mandates that adding any orphan instances requires a major version bump, so we'd be forced to churn through major versions much more often than before.
• It's an utter pain in the behind to maintain. Seriously, just look at this morass of CPP. I get a headache every time I have to change it, so just imagine how other people would feel about this :)
• It's difficult to get right. Our Travis build offers some confidence that it's currently correct, but transformers-compat offers such a dazzling array of flag combinations that it's hard to know if we've covered every possibility. Plus, given the enormous number of reverse dependencies on transformers-compat and how painful it is to make new releases of this package, I'd hate to have to discover these problems only after it's been shipped into the wild.

Luckily, Control.Monad.Trans.Instances only exists in transformers-compat HEAD, so there's still time to remove it. My inclination is to do just that, and if we really need these instances somewhere, to put them in a separate transformers-orphans package.

[RyanGlScott]

On second thought, I change my mind. Upon further reflection:

• I was wrong: Control.Monad.Trans.Instances is already released, so removing it would itself be a breaking change.
• transformers-compat is already slow enough to make new releases that the PVP argument perhaps isn't so convincing.
• transformers-compat is already a huge mess of CPP—Control.Monad.Trans.Instances isn't much different in that regard.

Closing.