newt/TODO.md

TODO

NOW - two things

1. I broke Tree.newt by removing some over-aggressive reduction. I need to make it happen during unification. Not sure if I have all of the bits to vappspine there.

2. Need to collect constraints and re-run, I think that may get me to typeclasses

• Remember operators from imports
• Default cases for non-primitives (currently gets expanded to all constructors)
• Case for primitives
• aoc2023 translation
  • day1
  • day2
  • some "real world" examples
• Maybe Eq and stuff would work for typeclass without dealing with unification issues yet
• unsolved meta errors repeat (need to freeze or only report at end)
• Sanitize JS idents, e.g. _+_
• Generate some programs that do stuff
• import
• consider making meta application implicit in term, so its more readable when printed
  • Currently we have explicit App surrounding Meta when inserting metas. Some people leave that implicit for efficiency. I think it would also make printing more readable.
  • When printing Value, I now print the spine size instead of spine.
• eval for case (see order.newt)
• dynamic pattern unification (add test case first)
• switch from commit/mustWork to checking progress
• type constructors are no longer generated? And seem to have 0 arity.
• raw let is not yet implemented (although define used by case tree building)
• there is some zero argument application in generated code
• get equality.newt to work
  • broken again because I added J, probably need to constrain scrutinee to value
• inline metas. Maybe zonk after TC/elab
• implicit patterns
• operators
• pair syntax (via comma operator)
• matching on operators
  • top level
  • case statements
• Lean / Agda ⟨ ⟩
• Lean-like .map, etc? (resolve name in namespace of target type, etc)
• SKIP list syntax
  • Agda doesn't have it, clashes with pair syntax
• autos / typeclass resolution
  • very primitive version in place, not higher order, search at end
  • We need special handling in unification to make this possible for typeclasses on U -> U, we could still do Eq or Show with the current unification
  • options
    • keep as implicit and do auto if the type constructor is flagged auto
    • keep as implicit and mark auto, behavior overlaps a lot with implicit
    • have separate type of implict with {{}}
    • TypeClass.newt is the exercise for this
• do blocks (needs typeclass, overloaded functions, or constrain to IO)
• some solution for + problem (classes? ambiguity?)
• show compiler failure in the editor (exit code != 0)
• write output to file
  • uses -o option
• detect extra clauses in case statements
• add test framework
• decide what to do for erasure
• type at point in vscode
• repl
• LSP
• don't match forced constructors at runtime
  • I think we got this by not switching for single cases
• magic nat (codegen as number with appropriate pattern matching)
• magic tuple? (codegen as array)
• magic newtype? (drop in codegen)
• records / copatterns
• vscode: syntax highlighting for String
• add pop or variant of pfunc that maps to an operator, giving the js operator and precedence on RHS

Parsing

• consider allowing σ etc in identifiers
  • Probably need to merge oper / ident first and sort out mixfix in parsing.

Background

• Read Ulf Norell thesis