Johan B.W. de Vries 38e43944c7 Replaces type3 with type5

type5 is much more first principles based, so we get a lot
of weird quirks removed:

- FromLiteral no longer needs to understand AST
- Type unifications works more like Haskell
- Function types are just ordinary types, saving a lot of
  manual busywork

and more.

2025-08-10 14:51:17 +02:00

1.5 KiB

Raw Blame History

TODO

Decide between lineair types / uniqueness vs garbage collector
- https://borretti.me/article/how-australs-linear-type-checker-works
Rename constant to literal
Implement a trace() builtin for debugging
Check if we can use DataView in the Javascript examples, e.g. with setUint32
Implement a FizzBuzz example
Also, check the codes for FIXME and TODO
Allocation is done using pointers for members, is this desired?
See if we want to replace Fractional with Real, and add Rational, Irrationl, Algebraic, Transendental
- Implement q32? q64? Two i32/i64 divided?
Have a set of rules or guidelines for the constraint comments, they're messy.
calculate_alloc_size can be reworked; is_member isn't useful with TYPE_INFO_MAP
Parser is putting stuff in ModuleDataBlock
- Surely the compiler should build data blocks
- Also put the struct.pack constants in TYPE_INFO_MAP
Implemented Bounded: https://hackage.haskell.org/package/base-4.21.0.0/docs/Prelude.html#t:Bounded
Try to implement the min and max functions using select
Read https://bytecodealliance.org/articles/multi-value-all-the-wasm
GRose :: (* -> *) -> * -> *
skolem => variable that cannot be unified

Limitations (for now):

no type level nats

only support first order kinds Do not support yet:

data Record f = Record {
  field: f Int
}
Record :: (* -> *) -> *

(Nested arrows)

only support rank 1 types

mapRecord :: forall f g. (forall a. f a -> f b) -> Record f -> Record g

(Nested forall)

1.5 KiB Raw Blame History

TODO

1.5 KiB

Raw Blame History