agdarsec VS ghc-proposals

I've shared my experiences, too, though I don't have specific examples. I find https://hackage.haskell.org/package/parsec-3.1.14.0/docs/src/Text.Parsec.Prim.html mush easier to read, modify, and use than https://github.com/gallais/agdarsec/blob/master/src/Text/Parser.agda . That's a specific example.

There is the RecursiveLet proposal.

I believe simplified subsumption is required to implement quick look impredicativity and that is the only practical reason for this change.

This led me to the proposal and I found with simplified subsumption:

We have modifiers with the % symbol, they are currently only used in linear types: a %1 -> b, but they are not limited to that.

Was this not implemented? https://github.com/ghc-proposals/ghc-proposals/pull/240

I understand that the linked Explicit Splice Imports could help with that, but in theory would there be anything preventing us from tracking automatically exactly which functions make it into a splice for optimally fine-grained recompilation avoidance?

These differences via eta expansion/reduction are probably due to simplified subsumption introduced in GHC 9.0 (also see the linked GHC proposal.

As for unifying type and term: I think we're headed in that direction. According to https://github.com/ghc-proposals/ghc-proposals/pull/448, we will accept your "foo" example, and so it makes sense also to do so in types.

Inferred variables as a concept seems to be against "Explicit Variable Principle (EVP)" (mentioned in https://github.com/goldfirere/ghc-proposals/blob/type-variables/principles.rst, linked from https://github.com/ghc-proposals/ghc-proposals/pull/448). I agree with that priciple.