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

//! # Reduced AST
//! The reduced AST is a minimal AST designed to be built from the full AST after all possible
//! static checks have been done. Consequently, the AST reduction phase does very little error
//! checking itself - any errors should ideally be caught either by an earlier phase, or are
//! runtime errors that the evaluator should handle. That said, becuase it does do table lookups
//! that can in principle fail [especially at the moment with most static analysis not yet complete],
//! there is an Expr variant `ReductionError` to handle these cases.
//!
//! A design decision to make - should the ReducedAST types contain all information about
//! type/layout necessary for the evaluator to work? If so, then the evaluator should not
//! have access to the symbol table at all and ReducedAST should carry that information. If not,
//! then ReducedAST shouldn't be duplicating information that can be queried at runtime from the
//! symbol table. But I think the former might make sense since ultimately the bytecode will be
//! built from the ReducedAST.
use std::rc::Rc;
use std::str::FromStr;

use crate::ast::*;
use crate::symbol_table::{Symbol, SymbolSpec, SymbolTable, FullyQualifiedSymbolName};
use crate::builtin::Builtin;
use crate::util::deref_optional_box;

#[derive(Debug)]
pub struct ReducedAST(pub Vec<Stmt>);

#[derive(Debug, Clone)]
pub enum Stmt {
  PreBinding {
    name: Rc<String>,
    func: Func,
  },
  Binding {
    name: Rc<String>,
    constant: bool,
    expr: Expr,
  },
  Expr(Expr),
  Noop,
}

#[derive(Debug, Clone)]
pub enum Expr {
  Unit,
  Lit(Lit),
  Sym(Rc<String>), //a Sym is anything that can be looked up by name at runtime - i.e. a function or variable address
  Tuple(Vec<Expr>),
  Func(Func),
  Constructor {
    type_name: Rc<String>,
    name: Rc<String>,
    tag: usize,
    arity: usize, // n.b. arity here is always the value from the symbol table - if it doesn't match what it's being called with, that's an eval error, eval will handle it
  },
  Call {
    f: Box<Expr>,
    args: Vec<Expr>,
  },
  Assign {
    val: Box<Expr>, //TODO this probably can't be a val
    expr: Box<Expr>,
  },
  Conditional {
    cond: Box<Expr>,
    then_clause: Vec<Stmt>,
    else_clause: Vec<Stmt>,
  },
  ConditionalTargetSigilValue,
  CaseMatch {
    cond: Box<Expr>,
    alternatives: Vec<Alternative>
  },
  UnimplementedSigilValue,
  ReductionError(String),
}

pub type BoundVars = Vec<Option<Rc<String>>>; //remember that order matters here

#[derive(Debug, Clone)]
pub struct Alternative {
  pub matchable: Subpattern,
  pub item: Vec<Stmt>,
}

#[derive(Debug, Clone)]
pub struct Subpattern {
  pub tag: Option<usize>,
  pub subpatterns: Vec<Option<Subpattern>>,
  pub bound_vars: BoundVars,
  pub guard: Option<Expr>,
}

#[derive(Debug, Clone)]
pub enum Lit {
  Nat(u64),
  Int(i64),
  Float(f64),
  Bool(bool),
  StringLit(Rc<String>),
}

#[derive(Debug, Clone)]
pub enum Func {
  BuiltIn(Builtin),
  UserDefined {
    name: Option<Rc<String>>,
    params: Vec<Rc<String>>,
    body: Vec<Stmt>,
  }
}

pub fn reduce(ast: &AST, symbol_table: &SymbolTable) -> ReducedAST {
  let mut reducer = Reducer { symbol_table };
  reducer.ast(ast)
}

struct Reducer<'a> {
  symbol_table: &'a SymbolTable
}

impl<'a> Reducer<'a> {
  fn ast(&mut self, ast: &AST) -> ReducedAST {
    let mut output = vec![];
    for statement in ast.statements.iter() {
      output.push(self.statement(statement));
    }
    ReducedAST(output)
  }

  fn statement(&mut self, stmt: &Statement) -> Stmt {
    match &stmt.kind {
      StatementKind::Expression(expr) => Stmt::Expr(self.expression(&expr)),
      StatementKind::Declaration(decl) => self.declaration(&decl),
      StatementKind::Import(_) => Stmt::Noop,
      StatementKind::Module(modspec) => {
        for statement in modspec.contents.iter() {
          self.statement(&statement);
        }
        Stmt::Noop
      }
    }
  }

  fn block(&mut self, block: &Block) -> Vec<Stmt> {
    block.iter().map(|stmt| self.statement(stmt)).collect()
  }

  fn invocation_argument(&mut self, invoc: &InvocationArgument) -> Expr {
    use crate::ast::InvocationArgument::*;
    match invoc {
      Positional(ex) => self.expression(ex),
      Keyword { .. } => Expr::UnimplementedSigilValue,
      Ignored => Expr::UnimplementedSigilValue,
    }
  }

  fn expression(&mut self, expr: &Expression) -> Expr {
    use crate::ast::ExpressionKind::*;
    let symbol_table = self.symbol_table;
    let ref input = expr.kind;
    match input {
      NatLiteral(n) => Expr::Lit(Lit::Nat(*n)),
      FloatLiteral(f) => Expr::Lit(Lit::Float(*f)),
      StringLiteral(s) => Expr::Lit(Lit::StringLit(s.clone())),
      BoolLiteral(b) => Expr::Lit(Lit::Bool(*b)),
      BinExp(binop, lhs, rhs) => self.binop(binop, lhs, rhs),
      PrefixExp(op, arg) => self.prefix(op, arg),
      Value(qualified_name) => self.value(qualified_name),
      Call { f, arguments } =>  self.reduce_call_expression(f, arguments),
      TupleLiteral(exprs) => Expr::Tuple(exprs.iter().map(|e| self.expression(e)).collect()),
      IfExpression { discriminator, body } => self.reduce_if_expression(deref_optional_box(discriminator), body),
      Lambda { params, body, .. } => self.reduce_lambda(params, body),
      NamedStruct { name, fields } => self.reduce_named_struct(name, fields),
      Index { .. } => Expr::UnimplementedSigilValue,
      WhileExpression { .. } => Expr::UnimplementedSigilValue,
      ForExpression { .. } => Expr::UnimplementedSigilValue,
      ListLiteral { .. } => Expr::UnimplementedSigilValue,
    }
  }

  fn value(&mut self, qualified_name: &QualifiedName) -> Expr {
    let symbol_table = self.symbol_table;
    let ref id = qualified_name.id;
    let ref sym_name = match symbol_table.get_fqsn_from_id(id) {
      Some(fqsn) => fqsn,
      None => return Expr::ReductionError(format!("FQSN lookup for Value {:?} failed", qualified_name)),
    };

    //TODO this probably needs to change
    let FullyQualifiedSymbolName(ref v) = sym_name;
    let name = v.last().unwrap().name.clone();

    let Symbol { local_name, spec, .. } = match symbol_table.lookup_by_fqsn(&sym_name) {
      Some(s) => s,
      //None => return Expr::ReductionError(format!("Symbol {:?} not found", sym_name)),
      None => return Expr::Sym(name.clone())
    };

    match spec {
      SymbolSpec::RecordConstructor { .. } => Expr::ReductionError(format!("AST reducer doesn't expect a RecordConstructor here")),
      SymbolSpec::DataConstructor { index, type_args, type_name } => Expr::Constructor {
        type_name: type_name.clone(),
        name: name.clone(),
        tag: index.clone(),
        arity: type_args.len(),
      },
      SymbolSpec::Func(_) => Expr::Sym(local_name.clone()),
      SymbolSpec::Binding => Expr::Sym(local_name.clone()), //TODO not sure if this is right, probably needs to eventually be fqsn
      SymbolSpec::Type { .. } => Expr::ReductionError("AST reducer doesnt expect a type here".to_string())
    }
  }

  fn reduce_lambda(&mut self, params: &Vec<FormalParam>, body: &Block) -> Expr {
    Expr::Func(Func::UserDefined {
      name: None,
      params: params.iter().map(|param| param.name.clone()).collect(),
      body: self.block(body),
    })
  }

  fn reduce_named_struct(&mut self, name: &QualifiedName, fields: &Vec<(Rc<String>, Expression)>) -> Expr {
    let symbol_table = self.symbol_table;
    let ref sym_name = match symbol_table.get_fqsn_from_id(&name.id) {
      Some(fqsn) => fqsn,
      None => return Expr::ReductionError(format!("FQSN lookup for name {:?} failed", name)),
    };

    let FullyQualifiedSymbolName(ref v) = sym_name;
    let ref name = v.last().unwrap().name;
    let (type_name, index, members_from_table) = match symbol_table.lookup_by_fqsn(&sym_name) {
      Some(Symbol { spec: SymbolSpec::RecordConstructor { members, type_name, index },  .. }) => (type_name.clone(), index, members),
      _ => return Expr::ReductionError("Not a record constructor".to_string()),
    };
    let arity = members_from_table.len();

    let mut args: Vec<(Rc<String>, Expr)> = fields.iter()
      .map(|(name, expr)| (name.clone(), self.expression(expr)))
      .collect();

    args.as_mut_slice()
      .sort_unstable_by(|(name1, _), (name2, _)| name1.cmp(name2)); //arbitrary - sorting by alphabetical order

    let args = args.into_iter().map(|(_, expr)| expr).collect();

    //TODO make sure this sorting actually works
    let f = box Expr::Constructor { type_name, name: name.clone(), tag: *index, arity, };
    Expr::Call { f, args }
  }

  fn reduce_call_expression(&mut self, func: &Expression, arguments: &Vec<InvocationArgument>) -> Expr {
    Expr::Call {
      f: Box::new(self.expression(func)),
      args: arguments.iter().map(|arg| self.invocation_argument(arg)).collect(),
    }
  }

  fn reduce_if_expression(&mut self, discriminator: Option<&Expression>, body: &IfExpressionBody) -> Expr {
    let symbol_table = self.symbol_table;
    let cond = Box::new(match discriminator {
      Some(expr) => self.expression(expr),
      None => return Expr::ReductionError(format!("blank cond if-expr not supported")),
    });

    match body {
      IfExpressionBody::SimpleConditional { then_case, else_case } => {
        let then_clause = self.block(&then_case);
        let else_clause = match else_case.as_ref() {
          None => vec![],
          Some(stmts) => self.block(&stmts),
        };
        Expr::Conditional { cond, then_clause, else_clause }
      },
      IfExpressionBody::SimplePatternMatch { pattern, then_case, else_case } => {
        let then_clause = self.block(&then_case);
        let else_clause = match else_case.as_ref() {
          None => vec![],
          Some(stmts) => self.block(&stmts),
        };

        let alternatives = vec![
          pattern.to_alternative(then_clause, symbol_table),
          Alternative {
            matchable: Subpattern {
              tag: None,
              subpatterns: vec![],
              bound_vars: vec![],
              guard: None,
            },
            item: else_clause
          },
        ];

        Expr::CaseMatch {
          cond,
          alternatives,
        }
      },
      IfExpressionBody::CondList(ref condition_arms) => {
        let mut alternatives = vec![];
        for arm in condition_arms {
          match arm.condition {
            Condition::Expression(ref _expr) => {
              return Expr::UnimplementedSigilValue
            },
            Condition::Pattern(ref p) =>  {
              let item = self.block(&arm.body);
              let alt = p.to_alternative(item, symbol_table);
              alternatives.push(alt);
            },
            Condition::TruncatedOp(_, _) => {
              return Expr::UnimplementedSigilValue
            },
            Condition::Else => {
              return Expr::UnimplementedSigilValue
            }
          }
        }
        Expr::CaseMatch { cond, alternatives }
      }
    }
  }

  fn binop(&mut self, binop: &BinOp, lhs: &Box<Expression>, rhs: &Box<Expression>) -> Expr {
    let operation = Builtin::from_str(binop.sigil()).ok();
    match operation {
      Some(Builtin::Assignment) => Expr::Assign {
        val: Box::new(self.expression(&*lhs)),
        expr: Box::new(self.expression(&*rhs)),
      },
      Some(op) => {
        let f = Box::new(Expr::Func(Func::BuiltIn(op)));
        Expr::Call { f, args: vec![self.expression(&*lhs), self.expression(&*rhs)] }
      },
      None => {
        //TODO handle a user-defined operation
        Expr::UnimplementedSigilValue
      }
    }
  }

  fn prefix(&mut self, prefix: &PrefixOp, arg: &Box<Expression>) -> Expr {
    match prefix.builtin {
      Some(op) => {
        let f = Box::new(Expr::Func(Func::BuiltIn(op)));
        Expr::Call { f, args: vec![self.expression(arg)] }
      },
      None => { //TODO need this for custom prefix ops
        Expr::UnimplementedSigilValue
      }
    }
  }

  fn declaration(&mut self, declaration: &Declaration) -> Stmt {
    use self::Declaration::*;
    match declaration {
      Binding {name, constant, expr, .. } => Stmt::Binding { name: name.clone(), constant: *constant, expr: self.expression(expr) },
      FuncDecl(Signature { name, params, .. }, statements) => Stmt::PreBinding {
        name: name.clone(),
        func: Func::UserDefined {
          name: Some(name.clone()),
          params: params.iter().map(|param| param.name.clone()).collect(),
          body: self.block(&statements),
        }
      },
      TypeDecl { .. } => Stmt::Noop,
      TypeAlias{ .. } => Stmt::Noop,
      Interface { .. } => Stmt::Noop,
      Impl { .. } => Stmt::Expr(Expr::UnimplementedSigilValue),
      _ => Stmt::Expr(Expr::UnimplementedSigilValue)
    }
  }
}





/*                     ig var   pat
 * x is SomeBigOldEnum(_, x, Some(t))
 */

fn handle_symbol(symbol: Option<&Symbol>, inner_patterns: &Vec<Pattern>, symbol_table: &SymbolTable) -> Subpattern {
  use self::Pattern::*;
  let tag = symbol.map(|symbol| match symbol.spec {
    SymbolSpec::DataConstructor { index, .. } => index.clone(),
    _ => panic!("Symbol is not a data constructor - this should've been caught in type-checking"),
  });
  let bound_vars = inner_patterns.iter().map(|p| match p {
    VarOrName(qualified_name) => {
      let fqsn = symbol_table.get_fqsn_from_id(&qualified_name.id);
      let symbol_exists = fqsn.and_then(|fqsn| symbol_table.lookup_by_fqsn(&fqsn)).is_some();
      if symbol_exists {
        None
      } else {
        let QualifiedName { components, .. } = qualified_name;
        if components.len() == 1 {
          Some(components[0].clone())
        } else {
          panic!("Bad variable name in pattern");
        }
      }
    },
    _ => None,
  }).collect();

  let subpatterns = inner_patterns.iter().map(|p| match p {
    Ignored => None,
    VarOrName(_) => None,
    Literal(other) => Some(other.to_subpattern(symbol_table)),
    tp @ TuplePattern(_) => Some(tp.to_subpattern(symbol_table)),
    ts @ TupleStruct(_, _) => Some(ts.to_subpattern(symbol_table)),
    Record(..) => unimplemented!(),
  }).collect();

  let guard = None;
  /*
     let guard_equality_exprs: Vec<Expr> = subpatterns.iter().map(|p| match p {
     Literal(lit) => match lit {
     _ => unimplemented!()
     },
     _ => unimplemented!()
     }).collect();
     */

  Subpattern {
    tag,
    subpatterns,
    guard,
    bound_vars,
  }
}

impl Pattern {
  fn to_alternative(&self, item: Vec<Stmt>, symbol_table: &SymbolTable) -> Alternative {
    let s = self.to_subpattern(symbol_table);
    Alternative {
      matchable: Subpattern {
        tag: s.tag,
        subpatterns: s.subpatterns,
        bound_vars: s.bound_vars,
        guard: s.guard,
      },
      item
    }
  }

  fn to_subpattern(&self, symbol_table: &SymbolTable) -> Subpattern {
    use self::Pattern::*;
    match self {
      TupleStruct(QualifiedName{ components, id }, inner_patterns) => {
        let fqsn = symbol_table.get_fqsn_from_id(&id);
        match fqsn.and_then(|fqsn| symbol_table.lookup_by_fqsn(&fqsn)) {
          Some(symbol) => handle_symbol(Some(symbol), inner_patterns, symbol_table),
          None => {
            panic!("Symbol {:?} not found", components);
          }
        }
      },
      TuplePattern(inner_patterns) => handle_symbol(None, inner_patterns, symbol_table),
      Record(_name, _pairs) => {
        unimplemented!()
      },
      Ignored => Subpattern { tag: None, subpatterns: vec![], guard: None, bound_vars: vec![] },
      Literal(lit) => lit.to_subpattern(symbol_table),
      VarOrName(QualifiedName { components, id  }) => {
        // if fqsn is Some, treat this as a symbol pattern. If it's None, treat it
        // as a variable.
        let fqsn = symbol_table.get_fqsn_from_id(&id);
        match fqsn.and_then(|fqsn| symbol_table.lookup_by_fqsn(&fqsn)) {
          Some(symbol) => handle_symbol(Some(symbol), &vec![], symbol_table),
          None => {
            let name = if components.len() == 1 {
              components[0].clone()
            } else {
              panic!("check this line of code yo");
            };
            Subpattern {
              tag: None,
              subpatterns: vec![],
              guard: None,
              bound_vars: vec![Some(name.clone())],
            }
          }
        }
      },
    }
  }
}

impl PatternLiteral {
  fn to_subpattern(&self, _symbol_table: &SymbolTable) -> Subpattern {
    use self::PatternLiteral::*;
    match self {
      NumPattern { neg, num } => {
        let comparison = Expr::Lit(match (neg, num) {
          (false, ExpressionKind::NatLiteral(n)) => Lit::Nat(*n),
          (false, ExpressionKind::FloatLiteral(f)) => Lit::Float(*f),
          (true, ExpressionKind::NatLiteral(n)) => Lit::Int(-1*(*n as i64)),
          (true, ExpressionKind::FloatLiteral(f)) => Lit::Float(-1.0*f),
          _ => panic!("This should never happen")
        });
        let guard = Some(Expr::Call {
          f: Box::new(Expr::Func(Func::BuiltIn(Builtin::Equality))),
          args: vec![comparison, Expr::ConditionalTargetSigilValue],
        });
        Subpattern {
          tag: None,
          subpatterns: vec![],
          guard,
          bound_vars: vec![],
        }
      },
      StringPattern(s) => {
        let guard = Some(Expr::Call {
          f: Box::new(Expr::Func(Func::BuiltIn(Builtin::Equality))),
          args: vec![Expr::Lit(Lit::StringLit(s.clone())), Expr::ConditionalTargetSigilValue]
        });

        Subpattern {
          tag: None,
          subpatterns: vec![],
          guard,
          bound_vars: vec![],
        }
      },
      BoolPattern(b) =>  {
        let guard = Some(if *b {
          Expr::ConditionalTargetSigilValue
        } else {
          Expr::Call {
            f: Box::new(Expr::Func(Func::BuiltIn(Builtin::BooleanNot))),
            args: vec![Expr::ConditionalTargetSigilValue]
          }
        });
        Subpattern {
          tag: None,
          subpatterns: vec![],
          guard,
          bound_vars: vec![],
        }
      },
    }
  }

}