Updated TODO file

This commit is contained in:
Greg Shuflin 2021-10-20 01:18:09 -07:00
parent 75935db9e6
commit 5eb743a8b5

131
TODO.md
View File

@ -1,4 +1,12 @@
# Plan of attack # Immediate TODOs / General Code Cleanup
## Symbols
* the symbol table should probably *only* be for global definitions (maybe rename it to reflect this?)
* dealing with variable lookup w/in functions/closures should probably happen in AST -> ReducedAST
* b/c that's where we go from a string name to a canonical ID (for e.g. 2nd param in 3rd enclosing scope)
* Old notes on a plan of attack:
1. modify visitor so it can handle scopes 1. modify visitor so it can handle scopes
-this is needed both to handle import scope correctly -this is needed both to handle import scope correctly
@ -6,66 +14,61 @@
2. Once FQSNs are aware of function parameters, most of the Rc<String> things in eval.rs can go away 2. Once FQSNs are aware of function parameters, most of the Rc<String> things in eval.rs can go away
# TODO items ## Parser
* I think I can restructure the parser to get rid of most instances of expect!, at least at the beginning of a rule
-use 'let' sigil in patterns for variables : ## Typechecking
* make a type to represent types rather than relying on string comparisons
* look at https://rickyhan.com/jekyll/update/2018/05/26/hindley-milner-tutorial-rust.html
## General code cleanup
* standardize on an error type that isn't String
* implement a visitor pattern for the use of scope_resolver
* maybe implement this twice: 1) the value-returning, no-default one in the haoyi blogpost,
* look at
* https://gitlab.haskell.org/ghc/ghc/wikis/pattern-synonyms
* the non-value-returning, default one like in rustc (cf. https://github.com/rust-unofficial/patterns/blob/master/patterns/visitor.md)
# Longer-term Ideas
## Language Syntax
* the `type` declaration should have some kind of GADT-like syntax
* use `let` sigil to indicate a variable in a pattern explicitly:
``` ```
q is MyStruct(let a, Chrono::Trigga) then { q is MyStruct(let a, Chrono::Trigga) then {
// a is in scope here
} }
``` ```
-idea: what if there was something like React jsx syntas built in? i.e. a way to automatically transform some kind of markup * if you have a pattern-match where one variant has a variable and the other
into a function call, cf. `<h1 prop="arg">` -> h1(prop=arg) lacks it instead of treating this as a type error, promote the bound variable
to an option type
## General code cleanup * what if there was something like React jsx syntas built in? i.e. a way to
- I think I can restructure the parser to get rid of most instances of expect!, at least at the beginning of a rule automatically transform some kind of markup into a function call, cf. `<h1
DONE -experiment with storing metadata via ItemIds on AST nodes (cf. https://rust-lang.github.io/rustc-guide/hir.html, https://github.com/rust-lang/rust/blob/master/src/librustc/hir/mod.rs ) prop="arg">` -> h1(prop=arg)
-implement and test open/use statements
-implement field access
- standardize on an error type that isn't String
-implement a visitor pattern for the use of scope_resolver
- maybe implement this twice: 1) the value-returning, no-default one in the haoyi blogpost,
-look at https://gitlab.haskell.org/ghc/ghc/wikis/pattern-synonyms
2) the non-value-returning, default one like in rustc (cf. https://github.com/rust-unofficial/patterns/blob/master/patterns/visitor.md)
-parser error - should report subset of AST parsed *so far* * implement and test open/use statements
- what if you used python 'def' syntax to define a function? what error message makes sense here?
## Reduction * Include extensible scala-style `html"string ${var}"` string interpolations
- make a good type for actual language builtins to avoid string comparisons
## Typechecking * A neat idea for pattern matching optimization would be if you could match on
one of several things in a list
- make a type to represent types rather than relying on string comparisons
- look at https://rickyhan.com/jekyll/update/2018/05/26/hindley-milner-tutorial-rust.html
- cf. the notation mentioned in the cardelli paper, the debug information for the `typechecking` pass should
print the generated type variable for every subexpression in an expression
- think about idris-related ideas of multiple implementations of a type for an interface (+ vs * impl for monoids, for preorder/inorder/postorder for Foldable)
-should have an Idris-like `cast To From` function
## Schala-lang syntax
-idea: the `type` declaration should have some kind of GADT-like syntax
- Idea: if you have a pattern-match where one variant has a variable and the other lacks it
instead of treating this as a type error, promote the bound variable to an option type
- Include extensible scala-style html"string ${var}" string interpolations
- A neat idea for pattern matching optimization would be if you could match on one of several things in a list
ex: ex:
```if x { ```
if x {
is (comp, LHSPat, RHSPat) if comp in ["==, "<"] -> ... is (comp, LHSPat, RHSPat) if comp in ["==, "<"] -> ...
}``` }
```
- Schala should have both currying *and* default arguments! * Schala should have both currying *and* default arguments!
```fn a(b: Int, c:Int, d:Int = 1) -> Int ```
fn a(b: Int, c:Int, d:Int = 1) -> Int
a(1,2) : Int a(1,2) : Int
a(1,2,d=2): Int a(1,2,d=2): Int
a(_,1,3) : Int -> Int a(_,1,3) : Int -> Int
@ -73,35 +76,47 @@ ex:
a(_,_,_) : Int -> Int -> Int -> Int a(_,_,_) : Int -> Int -> Int -> Int
``` ```
- scoped types - be able to define a quick enum type scoped to a function or other type for * scoped types - be able to define a quick enum type scoped to a function or other type for
something, that only is meant to be used as a quick bespoke interface between something, that only is meant to be used as a quick bespoke interface between
two other things two other things
ex. ex.
```type enum { ```
type enum {
type enum MySubVariant { type enum MySubVariant {
SubVariant1, SubVariant2, etc. SubVariant1, SubVariant2, etc.
} }
Variant1(MySubVariant), Variant1(MySubVariant),
Variant2(...), Variant2(...),
}``` }
```
* inclusive/exclusive range syntax like .. vs ..=
## Typechecking
* cf. the notation mentioned in the cardelli paper, the debug information for the `typechecking` pass should
* print the generated type variable for every subexpression in an expression
* think about idris-related ideas of multiple implementations of a type for an interface (+ vs * impl for monoids, for preorder/inorder/postorder for Foldable)
* should have an Idris-like `cast To From` function
* something like the swift `Never` type ( https://nshipster.com/never/ ) in the stdlib
- inclusive/exclusive range syntax like .. vs ..=
## Compilation ## Compilation
-look into Inkwell for rust LLVM bindings * look into Inkwell for rust LLVM bindings
* https://cranelift.readthedocs.io/en/latest/?badge=latest<Paste>
-https://cranelift.readthedocs.io/en/latest/?badge=latest<Paste> * look at https://gluon-lang.org/doc/nightly/book/embedding-api.html
## Other links of note # Syntax Playground
- https://nshipster.com/never/
-consult http://gluon-lang.org/book/embedding-api.html
## Trying if-syntax again ## Trying if-syntax again
```
//simple if expr //simple if expr
if x == 10 then "a" else "z" if x == 10 then "a" else "z"
@ -132,7 +147,7 @@ if x {
is Person(_, age) if age > 13 then "barmitzvah'd" is Person(_, age) if age > 13 then "barmitzvah'd"
else "foo" else "foo"
} }
```
## (OLD) Playing around with conditional syntax ideas ## (OLD) Playing around with conditional syntax ideas