schala/schala-lang/language/src/typechecking.rs

use std::rc::Rc;
//THINGS TODO
// look at the haskell compiler, see where in its flow the typechecking happens
// -nope, ghc deliberately does typechecking before desugaring to core
// cf. a history of haskell, peyton-jones

use crate::ast::*;
use crate::util::ScopeStack;
use crate::builtin::{PrefixOp, BinOp};

pub type TypeName = Rc<String>;

pub struct TypeContext<'a> {
  variable_map: ScopeStack<'a, Rc<String>, Type>,
  //evar_count: u32
}

/// `InferResult` is the monad in which type inference takes place.
type InferResult<T> = Result<T, TypeError>;

#[derive(Debug, Clone)]
pub struct TypeError { pub msg: String }

impl TypeError {
  fn new<A, T>(msg: T) -> InferResult<A> where T: Into<String> { //TODO make these kinds of error-producing functions CoW-ready
    Err(TypeError { msg: msg.into() })
  }
}

#[derive(Debug, Clone, PartialEq)]
pub enum Type {
  Const(TypeConst),
  Var(TypeVar),
  Arrow(Box<Type>, Box<Type>),
  Compound {
    ty_name: String,
    args:Vec<Type>
  }
}

#[derive(Debug, Clone, PartialEq)]
pub struct TypeVar(String);

#[derive(Debug, Clone, PartialEq)]
pub enum TypeConst {
  Unit,
  Nat,
  Int,
  Float,
  StringT,
  Bool,
  Ordering,
  UserDefined
}

macro_rules! ty {
  ($type_name:ident) => { Type::Const(TypeConst::$type_name) };
  ($t1:ident -> $t2:ident) => { Type::Arrow(Box::new(ty!($t1)), Box::new(ty!($t2))) };
  ($t1:ident -> $t2:ident -> $t3:ident) => { Type::Arrow(Box::new(ty!($t1)), Box::new(ty!($t2 -> $t3))) };
}

//TODO find a better way to capture the to/from string logic
impl Type {
  pub fn to_string(&self) -> String {
    use self::Type::*;
    use self::TypeConst::*;
    match self {
      Const(Unit) => format!("()"),
      Const(Nat) => format!("Nat"),
      Const(Int) => format!("Int"),
      Const(Float) => format!("Float"),
      Const(StringT) => format!("String"),
      Const(Bool) => format!("Bool"),
      Const(Ordering) => format!("Ordering"),
      _ => format!("UNKNOWN TYPE"),
    }
  }

  fn from_string(string: &str) -> Option<Type> {
    Some(match string {
      "()" | "Unit" => ty!(Unit),
      "Nat" => ty!(Nat),
      "Int" => ty!(Int),
      "Float" => ty!(Float),
      "String" => ty!(StringT),
      "Bool" => ty!(Bool),
      "Ordering" => ty!(Ordering),
      _ => return None
    })
  }
}

/*
/// `Type` is parameterized by whether the type variables can be just universal, or universal or
/// existential.
#[derive(Debug, Clone)]
enum Type<A> {
  Var(A),
  Const(TConst),
  Arrow(Box<Type<A>>, Box<Type<A>>),
}

#[derive(Debug, Clone)]
enum TVar {
  Univ(UVar),
  Exist(ExistentialVar)
}

#[derive(Debug, Clone)]
struct UVar(Rc<String>);

#[derive(Debug, Clone)]
struct ExistentialVar(u32);

impl Type<UVar> {
  fn to_tvar(&self) -> Type<TVar> {
    match self {
      Type::Var(UVar(name)) => Type::Var(TVar::Univ(UVar(name.clone()))),
      Type::Const(ref c) => Type::Const(c.clone()),
      Type::Arrow(a, b) => Type::Arrow(
        Box::new(a.to_tvar()),
        Box::new(b.to_tvar())
      )
    }
  }
}

impl Type<TVar> {
  fn skolemize(&self) -> Type<UVar> {
    match self {
      Type::Var(TVar::Univ(uvar)) => Type::Var(uvar.clone()),
      Type::Var(TVar::Exist(_)) => Type::Var(UVar(Rc::new(format!("sk")))),
      Type::Const(ref c) => Type::Const(c.clone()),
      Type::Arrow(a, b) => Type::Arrow(
        Box::new(a.skolemize()),
        Box::new(b.skolemize())
      )
    }
  }
}

impl TypeIdentifier {
  fn to_monotype(&self) -> Type<UVar> {
    match self {
      TypeIdentifier::Tuple(_) => Type::Const(TConst::Nat),
      TypeIdentifier::Singleton(TypeSingletonName { name, .. }) => {
        match &name[..] {
          "Nat" => Type::Const(TConst::Nat),
          "Int" => Type::Const(TConst::Int),
          "Float" => Type::Const(TConst::Float),
          "Bool" => Type::Const(TConst::Bool),
          "String" => Type::Const(TConst::StringT),
          _ => Type::Const(TConst::Nat),
        }
      }
    }
  }
}

#[derive(Debug, Clone)]
enum TConst {
  User(Rc<String>),
  Unit,
  Nat,
  Int,
  Float,
  StringT,
  Bool,
}

impl TConst {
  fn user(name: &str) -> TConst {
    TConst::User(Rc::new(name.to_string()))
  }
}
*/

impl<'a> TypeContext<'a> {
  pub fn new() -> TypeContext<'a> {
    TypeContext { 
      variable_map: ScopeStack::new(None),
      //evar_count: 0
    }
  }

  fn get_type_from_name(&self, name: &TypeIdentifier) -> InferResult<Type> {
    use self::TypeIdentifier::*;
    Ok(match name {
      Singleton(TypeSingletonName { name, params }) =>  {
        match Type::from_string(&name) {
          Some(ty) => ty,
          None => return TypeError::new("Unknown type name")
        }
      },
      Tuple(_) => return TypeError::new("tuples aren't ready yet"),
    })

  }

  pub fn typecheck(&mut self, ast: &AST) -> Result<Type, TypeError> {
    let mut returned_type = Type::Const(TypeConst::Unit);
    for statement in ast.0.iter() {
      returned_type = self.statement(statement.node())?;
    }
    Ok(returned_type)
  }

  fn statement(&mut self, statement: &Statement) -> InferResult<Type> {
    match statement {
      Statement::ExpressionStatement(e) => self.expr(e.node()),
      Statement::Declaration(decl) => self.decl(decl),
    }
  }

  fn decl(&mut self, decl: &Declaration) -> InferResult<Type> {
    use self::Declaration::*;
    match decl {
      Binding { name, expr, .. } => {
        let ty = self.expr(expr)?;
        self.variable_map.insert(name.clone(), ty);
      },
      _ => (),
    }
    Ok(ty!(Unit))
  }

  fn expr(&mut self, expr: &Expression) -> InferResult<Type> {
    match expr {
      Expression(expr_type, Some(anno)) => {
        let t1 = self.expr_type(expr_type)?;
        let t2 = self.get_type_from_name(anno)?;
        self.unify(t2, t1)
      },
      Expression(expr_type, None) => self.expr_type(expr_type)
    }
  }

  fn expr_type(&mut self, expr: &ExpressionKind) -> InferResult<Type> {
    use self::ExpressionKind::*;
    Ok(match expr {
      NatLiteral(_) => ty!(Nat),
      BoolLiteral(_) => ty!(Bool),
      FloatLiteral(_) => ty!(Float),
      StringLiteral(_) => ty!(StringT),
      PrefixExp(op, expr) => self.prefix(op, expr.node())?,
      BinExp(op, lhs, rhs) => self.binexp(op, lhs.node(), rhs.node())?,
      IfExpression { discriminator, body } => self.if_expr(discriminator, body)?,
      Value(val) => self.handle_value(val)?,
      Lambda { params, type_anno, body } => self.lambda(params, type_anno, body)?,
      _ => ty!(Unit),
    })
  }

  fn prefix(&mut self, op: &PrefixOp, expr: &Expression) -> InferResult<Type> {
    let f = match op.get_type() {
      Ok(ty) => ty,
      Err(e) => return TypeError::new(e)
    };

    let x = self.expr(expr)?;
    self.handle_apply(f, x)
  }

  fn binexp(&mut self, op: &BinOp, lhs: &Expression, rhs: &Expression) -> InferResult<Type> {
    let tf = match op.get_type() {
      Ok(ty) => ty,
      Err(e) => return TypeError::new(e),
    };

    let t_lhs = self.expr(lhs)?;
    let t_curried = self.handle_apply(tf, t_lhs)?;
    let t_rhs = self.expr(rhs)?;
    self.handle_apply(t_curried, t_rhs)
  }

  fn handle_apply(&mut self, tf: Type, tx: Type) -> InferResult<Type> {
    Ok(match tf {
      Type::Arrow(box ref t1, box ref t2) => {
        let _ = self.unify(t1.clone(), tx)?;
        t2.clone()
      },
      _ => return TypeError::new(format!("Not a function"))
    })
  }

  fn if_expr(&mut self, discriminator: &Discriminator, body: &IfExpressionBody) -> InferResult<Type> {
    use self::Discriminator::*; use self::IfExpressionBody::*;
    match (discriminator, body) {
      (Simple(expr), SimpleConditional(then_clause, else_clause)) => self.handle_simple_if(expr, then_clause, else_clause),
      _ => TypeError::new(format!("Complex conditionals not supported"))
    }
  }

  fn handle_simple_if(&mut self, expr: &Expression, then_clause: &Block, else_clause: &Option<Block>) -> InferResult<Type> {
    let t1 = self.expr(expr)?;
    let t2 = self.block(then_clause)?;
    let t3 = match else_clause {
      Some(block) => self.block(block)?,
      None => ty!(Unit)
    };

    let _ = self.unify(ty!(Bool), t1)?;
    self.unify(t2, t3)
  }

  fn lambda(&mut self, params: &Vec<FormalParam>, type_anno: &Option<TypeIdentifier>, body: &Block) -> InferResult<Type> {
    Ok(ty!(Unit))
  }

  fn block(&mut self, block: &Block) -> InferResult<Type> {
    let mut output = ty!(Unit);
    for s in block.iter() {
      let statement = s.node();
      output = self.statement(statement)?;
    }
    Ok(output)
  }

  fn handle_value(&mut self, val: &Rc<String>) -> InferResult<Type> {
    match self.variable_map.lookup(val) {
      Some(ty) => Ok(ty.clone()),
      None => TypeError::new(format!("Couldn't find variable: {}", val))
    }
  }

  fn unify(&mut self, t1: Type, t2: Type) -> InferResult<Type> {
    use self::Type::*; use self::TypeConst::*;
    Ok(match (t1, t2) {
      (Const(ref c1), Const(ref c2)) if c1 == c2 => Const(c1.clone()), //choice of c1 is arbitrary I *think*
      (a, b) => return TypeError::new(format!("{:?} and {:?} do not unify", a, b)),
    })
  }
}

#[cfg(test)]
mod typechecking_tests {
  use super::*;
  use crate::ast::AST;

  fn parse(input: &str) -> AST {
    let tokens = crate::tokenizing::tokenize(input);
    let mut parser = crate::parsing::Parser::new(tokens);
    parser.parse().unwrap()
  }

  macro_rules! assert_type_in_fresh_context {
    ($string:expr, $type:expr) => {
      let mut tc = TypeContext::new();
      let ref ast = parse($string);
      let ty = tc.typecheck(ast).unwrap();
      assert_eq!(ty, $type)
    }
  }

  #[test]
  fn basic_test() {
    assert_type_in_fresh_context!("1", ty!(Nat));
    assert_type_in_fresh_context!(r#""drugs""#, ty!(StringT));
    assert_type_in_fresh_context!("true", ty!(Bool));
    assert_type_in_fresh_context!("-1", ty!(Int));
  }

  #[test]
  fn operators() {
    assert_type_in_fresh_context!("1 + 2", ty!(Nat));
    assert_type_in_fresh_context!("-2", ty!(Int));
    assert_type_in_fresh_context!("!true", ty!(Bool));
  }
}