Codegen target: UPLC or TPLC or PIR?

[klntsky]

This topic is to track our tree of thought w.r.t. choice of compiler target.

UPLC

UPLC was our initial target to consider that is the most low-level, and so gives us the most freedom.

TPLC

What advantages could TPLC give us?

There are none known in comparison with PIR.

What exactly are types in typed plutus core?

TBD

How well do they map to PS type system?

TBD

At which stage of the compilation process can we go from PS types to TPLC types?

TBD

PIR

Citing @`effectfully:

Plutus IR = Typed Plutus Core + multi-let + (mutually) recursive functions + (mutually) recursive data types + optional laziness (you don't care). The AST is defined here

Citing @`michaelpj:

What is the relationship between Haskell's laziness and PIR?

PIR is strict, but has "non-strict" bindings. Since PS is strict you probably just won't use those

How are type classes encoded in PIR?

PIR doesn't have type-classes. For Haskell we encode them with explicit dictionary passing as it is in GHC Core

How are mutually-recursive binding groups implemented?

[...] That's fairly complicated, but again, if there's an easy way to improve it we should do it in the PIR compiler and benefit everyone using it

How is linking of modules implemented downstream?

It isn't, PIR works on a whole program. You can glue programs together with simple function application, potentially. I suspect that you could encode your preferred module linking scheme in PIR, though

[klntsky]

(UPLC or TPLC) vs PIR:

If we go with PIR, we can skip encoding of recursive functions, like it is done here and here