2023-04-06 03:17:57 +01:00
|
|
|
from tokens import Token, TokenType, TokenIterator
|
2023-04-08 18:21:37 +01:00
|
|
|
from position import Position, Span, Node
|
|
|
|
from parsed import Assign, AssignType, Binary, BinaryType, Block, Break, Call, Continue, Match, MatchArm, PatternError, Expr, For, Id, If, Index, Int, Char, Loop, Pattern, Return, String, ExprError, StructMember, Tuple, TupleMember, Unary, UnaryType, Unit, While
|
2023-04-06 03:17:57 +01:00
|
|
|
from typing import List, Optional
|
2023-04-08 18:21:37 +01:00
|
|
|
from utils import Result, Ok, Err
|
2023-04-06 03:17:57 +01:00
|
|
|
|
|
|
|
|
|
|
|
class Parser:
|
|
|
|
def __init__(self, text: str, tokens: TokenIterator) -> None:
|
|
|
|
self.text = text
|
|
|
|
self.tokens = tokens
|
|
|
|
self.current_token = tokens.next()
|
|
|
|
|
|
|
|
def parse(self) -> List[Node[Expr]]:
|
|
|
|
statements: List[Node[Expr]] = []
|
|
|
|
while not self.done():
|
|
|
|
statements.append(self.parse_statement())
|
|
|
|
return statements
|
|
|
|
|
|
|
|
def parse_statement(self) -> Node[Expr]:
|
|
|
|
if self.current_is(TokenType.KwIf):
|
|
|
|
return self.parse_if()
|
|
|
|
else:
|
|
|
|
return self.parse_assign()
|
|
|
|
|
|
|
|
def parse_assign(self) -> Node[Expr]:
|
|
|
|
subject = self.parse_expr()
|
|
|
|
if self.current_is(TokenType.Equal):
|
|
|
|
self.step()
|
|
|
|
value = self.parse_expr()
|
|
|
|
return Node(Assign(AssignType.Assign, subject, value), subject.span.to(value.span))
|
|
|
|
else:
|
|
|
|
return subject
|
|
|
|
|
|
|
|
def parse_expr(self) -> Node[Expr]:
|
|
|
|
return self.parse_or()
|
|
|
|
|
|
|
|
def parse_or(self) -> Node[Expr]:
|
|
|
|
left = self.parse_and()
|
|
|
|
while self.current_is(TokenType.KwOr):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_and()
|
|
|
|
left = Node(Binary(BinaryType.Or, left, right), left.span.to(right.span))
|
|
|
|
return left
|
|
|
|
|
|
|
|
|
|
|
|
def parse_and(self) -> Node[Expr]:
|
|
|
|
left = self.parse_equal()
|
|
|
|
while self.current_is(TokenType.KwOr):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_equal()
|
|
|
|
left = Node(Binary(BinaryType.And, left, right), left.span.to(right.span))
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_equal(self) -> Node[Expr]:
|
|
|
|
left = self.parse_compare()
|
|
|
|
while not self.done():
|
|
|
|
if self.current_is(TokenType.EqualEqual):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_compare()
|
|
|
|
left = Node(Binary(BinaryType.Equal, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.ExclamationEqual):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_compare()
|
|
|
|
left = Node(Binary(BinaryType.Inequal, left, right), left.span.to(right.span))
|
|
|
|
else:
|
|
|
|
break
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_compare(self) -> Node[Expr]:
|
|
|
|
left = self.parse_add_subtract()
|
|
|
|
while not self.done():
|
|
|
|
if self.current_is(TokenType.LT):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_add_subtract()
|
|
|
|
left = Node(Binary(BinaryType.LT, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.GT):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_add_subtract()
|
|
|
|
left = Node(Binary(BinaryType.LT, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.LTEqual):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_add_subtract()
|
|
|
|
left = Node(Binary(BinaryType.LTEqual, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.GTEqual):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_add_subtract()
|
|
|
|
left = Node(Binary(BinaryType.GTEqual, left, right), left.span.to(right.span))
|
|
|
|
else:
|
|
|
|
break
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_add_subtract(self) -> Node[Expr]:
|
|
|
|
left = self.parse_multiply_divide_modulo()
|
|
|
|
while not self.done():
|
|
|
|
if self.current_is(TokenType.Plus):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_multiply_divide_modulo()
|
|
|
|
left = Node(Binary(BinaryType.Add, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.Minus):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_multiply_divide_modulo()
|
|
|
|
left = Node(Binary(BinaryType.Subtract, left, right), left.span.to(right.span))
|
|
|
|
else:
|
|
|
|
break
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_multiply_divide_modulo(self) -> Node[Expr]:
|
|
|
|
left = self.parse_negate()
|
|
|
|
while not self.done():
|
|
|
|
if self.current_is(TokenType.Asterisk):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_negate()
|
|
|
|
left = Node(Binary(BinaryType.Multiply, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.Slash):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_negate()
|
|
|
|
left = Node(Binary(BinaryType.Divide, left, right), left.span.to(right.span))
|
|
|
|
elif self.current_is(TokenType.Percent):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_negate()
|
|
|
|
left = Node(Binary(BinaryType.Modulo, left, right), left.span.to(right.span))
|
|
|
|
else:
|
|
|
|
break
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_negate(self) -> Node[Expr]:
|
|
|
|
if self.current_is(TokenType.Minus):
|
|
|
|
token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = self.parse_exponent()
|
|
|
|
return Node(Unary(UnaryType.Negate, subject), token_span.to(subject.span))
|
|
|
|
else:
|
|
|
|
return self.parse_exponent()
|
|
|
|
|
|
|
|
def parse_exponent(self) -> Node[Expr]:
|
|
|
|
left = self.parse_unary()
|
|
|
|
if self.current_is(TokenType.AsteriskAsterisk):
|
|
|
|
self.step()
|
|
|
|
right = self.parse_exponent()
|
|
|
|
return Node(Binary(BinaryType.Exponent, left, right), left.span.to(right.span))
|
|
|
|
else:
|
|
|
|
return left
|
|
|
|
|
|
|
|
def parse_unary(self) -> Node[Expr]:
|
|
|
|
if self.current_is(TokenType.KwNot):
|
|
|
|
token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = self.parse_unary()
|
|
|
|
return Node(Unary(UnaryType.Not, subject), token_span.to(subject.span))
|
|
|
|
elif self.current_is(TokenType.Asterisk):
|
|
|
|
token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = self.parse_unary()
|
|
|
|
return Node(Unary(UnaryType.Dereference, subject), token_span.to(subject.span))
|
|
|
|
elif self.current_is(TokenType.Ampersand):
|
|
|
|
token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
if self.current_is(TokenType.KwMut):
|
|
|
|
self.step()
|
|
|
|
subject = self.parse_unary()
|
|
|
|
return Node(Unary(UnaryType.ReferenceMut, subject), token_span.to(subject.span))
|
|
|
|
else:
|
|
|
|
subject = self.parse_unary()
|
|
|
|
return Node(Unary(UnaryType.ReferenceMut, subject), token_span.to(subject.span))
|
|
|
|
else:
|
|
|
|
return self.parse_member_index_call()
|
|
|
|
|
|
|
|
def parse_member_index_call(self) -> Node[Expr]:
|
|
|
|
subject: Node[Expr] = self.parse_operand()
|
|
|
|
while not self.done():
|
|
|
|
if self.current_is(TokenType.Dot):
|
|
|
|
self.step()
|
|
|
|
if self.current_is(TokenType.Id):
|
|
|
|
id_token = self.current()
|
|
|
|
self.step()
|
|
|
|
text = id_token.text_slice(self.text)
|
|
|
|
subject = Node(StructMember(subject, text), subject.span.to(id_token.span))
|
|
|
|
elif self.current_is(TokenType.Int):
|
|
|
|
int_token = self.current()
|
|
|
|
self.step()
|
|
|
|
value = int(int_token.text_slice(self.text))
|
|
|
|
subject = Node(TupleMember(subject, value), subject.span.to(int_token.span))
|
|
|
|
else:
|
|
|
|
return Node(ExprError("expected Int or Id"), subject.span)
|
|
|
|
elif self.current_is(TokenType.LBracket):
|
|
|
|
self.step()
|
|
|
|
value = self.parse_expr()
|
|
|
|
if not self.current_is(TokenType.RBracket):
|
|
|
|
return Node(ExprError("expected ']'"), subject.span.to(value.span))
|
|
|
|
rbracket_token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = Node(Index(subject, value), subject.span.to(rbracket_token_span))
|
|
|
|
elif self.current_is(TokenType.LParen):
|
|
|
|
self.step()
|
|
|
|
arguments: List[Node[Expr]] = []
|
|
|
|
if not self.done() and self.current() != TokenType.RParen:
|
|
|
|
arguments.append(self.parse_expr())
|
|
|
|
while not self.done() and self.current() == TokenType.Comma:
|
|
|
|
self.step()
|
|
|
|
if self.done() or self.current() == TokenType.RParen:
|
|
|
|
break
|
|
|
|
arguments.append(self.parse_expr())
|
|
|
|
if not self.current_is(TokenType.RParen):
|
|
|
|
if len(arguments) > 0:
|
|
|
|
end = arguments[-1].span
|
|
|
|
else:
|
|
|
|
end = subject.span
|
|
|
|
return Node(ExprError("expected ')'"), subject.span.to(end))
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = Node(Call(subject, arguments), subject.span.to(end))
|
|
|
|
else:
|
|
|
|
break
|
|
|
|
return subject
|
|
|
|
|
|
|
|
def parse_operand(self) -> Node[Expr]:
|
|
|
|
if self.current_is(TokenType.Id):
|
|
|
|
token = self.current()
|
|
|
|
value = token.text_slice(self.text)
|
|
|
|
self.step()
|
|
|
|
return Node(Id(value), token.span)
|
|
|
|
elif self.current_is(TokenType.Int):
|
|
|
|
token = self.current()
|
|
|
|
value = int(token.text_slice(self.text))
|
|
|
|
self.step()
|
|
|
|
return Node(Int(value), token.span)
|
|
|
|
elif self.current_is(TokenType.Char):
|
|
|
|
token = self.current()
|
|
|
|
value = token.text_slice(self.text)
|
|
|
|
self.step()
|
|
|
|
return Node(Char(value), token.span)
|
|
|
|
elif self.current_is(TokenType.String):
|
|
|
|
token = self.current()
|
|
|
|
value = token.text_slice(self.text)
|
|
|
|
self.step()
|
|
|
|
return Node(String(value), token.span)
|
2023-04-08 18:21:37 +01:00
|
|
|
elif self.current_is(TokenType.LParen):
|
|
|
|
return self.parse_unit_group_tuple()
|
2023-04-06 03:17:57 +01:00
|
|
|
elif self.current_is(TokenType.LBrace):
|
|
|
|
return self.parse_block()
|
|
|
|
elif self.current_is(TokenType.KwIf):
|
|
|
|
return self.parse_if()
|
|
|
|
elif self.current_is(TokenType.KwLoop):
|
|
|
|
return self.parse_loop()
|
|
|
|
elif self.current_is(TokenType.KwWhile):
|
|
|
|
return self.parse_while()
|
|
|
|
elif self.current_is(TokenType.KwFor):
|
|
|
|
return self.parse_for()
|
|
|
|
else:
|
|
|
|
token = self.current()
|
|
|
|
self.step()
|
|
|
|
return Node(ExprError("expected value"), token.span)
|
|
|
|
|
2023-04-08 18:21:37 +01:00
|
|
|
def parse_unit_group_tuple(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
if self.current_is(TokenType.RParen):
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
return Node(Unit(), begin.to(end))
|
|
|
|
else:
|
|
|
|
first_expr = self.parse_expr()
|
|
|
|
if self.current_is(TokenType.RParen):
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
return Node(first_expr.value, begin.to(end))
|
|
|
|
elif self.current_is(TokenType.Comma):
|
|
|
|
values = [first_expr]
|
|
|
|
end = self.current().span
|
|
|
|
while self.current_is(TokenType.Comma):
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
if self.done() or self.current().token_type == TokenType.RParen:
|
|
|
|
break
|
|
|
|
value = self.parse_expr()
|
|
|
|
end = value.span
|
|
|
|
values.append(value)
|
|
|
|
if not self.current_is(TokenType.RParen):
|
|
|
|
return Node(ExprError("expected ')'"), begin.to(end))
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
return Node(Tuple(values), begin.to(end))
|
|
|
|
else:
|
|
|
|
return Node(ExprError("expected ')' or ','"), begin.to(first_expr.span))
|
|
|
|
|
2023-04-06 03:17:57 +01:00
|
|
|
def parse_block(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
statements: List[Node[Expr]] = []
|
|
|
|
value: Optional[Node[Expr]] = None
|
|
|
|
while not self.done() and self.current().token_type != TokenType.RBrace:
|
|
|
|
if value is not None:
|
|
|
|
statements.append(value)
|
|
|
|
value = self.parse_statement()
|
|
|
|
if not self.current_is(TokenType.RBrace):
|
|
|
|
if value is not None:
|
|
|
|
end = value.span
|
|
|
|
else:
|
|
|
|
end = begin
|
|
|
|
return Node(ExprError("expected '}'"), begin.to(end))
|
|
|
|
end = self.current().span
|
|
|
|
self.step()
|
|
|
|
return Node(Block(statements, value), begin.to(end))
|
|
|
|
|
|
|
|
def parse_if(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
condition = self.parse_expr()
|
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin.to(condition.span))
|
|
|
|
truthy = self.parse_block()
|
|
|
|
if self.current_is(TokenType.KwElse):
|
|
|
|
else_token_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin.to(else_token_span))
|
|
|
|
falsy = self.parse_block()
|
|
|
|
return Node(If(condition, truthy, falsy), begin.to(falsy.span))
|
|
|
|
else:
|
|
|
|
return Node(If(condition, truthy, None), begin.to(truthy.span))
|
|
|
|
|
|
|
|
def parse_match(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
2023-04-08 18:21:37 +01:00
|
|
|
value = self.parse_expr()
|
2023-04-06 03:17:57 +01:00
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin)
|
2023-04-08 18:21:37 +01:00
|
|
|
lbrace_span = self.current().span
|
2023-04-06 03:17:57 +01:00
|
|
|
self.step()
|
|
|
|
arms: List[Node[MatchArm]] = []
|
|
|
|
if not self.done() and self.current() != TokenType.RBrace:
|
2023-04-08 18:21:37 +01:00
|
|
|
arm = self.parse_match_arm()
|
|
|
|
if not arm.ok():
|
|
|
|
return Node(ExprError(arm.error().value), begin.to(arm.error().span))
|
|
|
|
arms.append(arm.value())
|
|
|
|
while not self.done() and self.current() != TokenType.RBrace:
|
|
|
|
arm = self.parse_match_arm()
|
|
|
|
if not arm.ok():
|
|
|
|
return Node(ExprError(arm.error().value), begin.to(arm.error().span))
|
|
|
|
arms.append(arm.value())
|
|
|
|
if not self.current_is(TokenType.RBrace):
|
|
|
|
if len(arms) > 0:
|
|
|
|
end = arms[-1].span
|
|
|
|
else:
|
|
|
|
end = lbrace_span
|
|
|
|
return Node(ExprError("expected '}'"), begin.to(end))
|
|
|
|
rbrace_span = self.current().span
|
|
|
|
self.step()
|
|
|
|
return Node(Match(value, arms), begin.to(rbrace_span))
|
2023-04-06 03:17:57 +01:00
|
|
|
|
|
|
|
|
2023-04-08 18:21:37 +01:00
|
|
|
def parse_match_arm(self) -> Result[Node[MatchArm], Node[str]]:
|
|
|
|
pattern = self.parse_pattern()
|
|
|
|
if not self.current_is(TokenType.EqualLT):
|
|
|
|
return Err(Node("expected '=>'", pattern.span.to(pattern.span)))
|
|
|
|
self.step()
|
|
|
|
if self.current_is(TokenType.LParen):
|
|
|
|
expr = self.parse_block()
|
|
|
|
if self.current_is(TokenType.Comma):
|
|
|
|
self.step()
|
|
|
|
else:
|
|
|
|
expr = self.parse_match_arm_expr()
|
|
|
|
if not self.current_is(TokenType.Comma):
|
|
|
|
return Err(Node("expected ','", pattern.span.to(expr.span)))
|
|
|
|
self.step()
|
|
|
|
return Ok(Node(MatchArm(pattern, expr), pattern.span.to(expr.span)))
|
|
|
|
|
|
|
|
def parse_match_arm_expr(self) -> Node[Expr]:
|
|
|
|
if self.current_is(TokenType.KwReturn):
|
|
|
|
return self.parse_return()
|
|
|
|
elif self.current_is(TokenType.KwBreak):
|
|
|
|
return self.parse_break()
|
|
|
|
elif self.current_is(TokenType.KwContinue):
|
|
|
|
return self.parse_continue()
|
|
|
|
else:
|
|
|
|
return self.parse_expr()
|
2023-04-06 03:17:57 +01:00
|
|
|
|
|
|
|
def parse_loop(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin)
|
|
|
|
body = self.parse_block()
|
|
|
|
return Node(Loop(body), begin.to(body.span))
|
|
|
|
|
|
|
|
def parse_while(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
condition = self.parse_expr()
|
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin.to(condition.span))
|
|
|
|
self.step()
|
|
|
|
body = self.parse_block()
|
|
|
|
return Node(While(condition, body), begin.to(body.span))
|
|
|
|
|
|
|
|
def parse_for(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
subject = self.parse_pattern()
|
|
|
|
if not self.current_is(TokenType.KwIn):
|
|
|
|
return Node(ExprError("expected 'in'"), begin.to(subject.span))
|
|
|
|
self.step()
|
|
|
|
value = self.parse_expr()
|
|
|
|
if not self.current_is(TokenType.LBrace):
|
|
|
|
return Node(ExprError("expected '{'"), begin.to(value.span))
|
|
|
|
self.step()
|
|
|
|
body = self.parse_block()
|
|
|
|
return Node(For(subject, value, body), begin.to(body.span))
|
|
|
|
|
2023-04-08 18:21:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
def parse_return(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
value: Optional[Node[Expr]] = None
|
|
|
|
if not self.done() and self.current() not in [TokenType.Comma, TokenType.Semicolon]:
|
|
|
|
value = self.parse_expr()
|
|
|
|
return Node(Return(value), begin.to(value.span if value is not None else begin))
|
|
|
|
|
|
|
|
def parse_break(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
value: Optional[Node[Expr]] = None
|
|
|
|
if not self.done() and self.current() not in [TokenType.Comma, TokenType.Semicolon]:
|
|
|
|
value = self.parse_expr()
|
|
|
|
return Node(Break(value), begin.to(value.span if value is not None else begin))
|
|
|
|
|
|
|
|
def parse_continue(self) -> Node[Expr]:
|
|
|
|
begin = self.current().span
|
|
|
|
self.step()
|
|
|
|
value: Optional[Node[Expr]] = None
|
|
|
|
if not self.done() and self.current() not in [TokenType.Comma, TokenType.Semicolon]:
|
|
|
|
value = self.parse_expr()
|
|
|
|
return Node(Continue(value), begin.to(value.span if value is not None else begin))
|
2023-04-06 03:17:57 +01:00
|
|
|
def parse_pattern(self) -> Node[Pattern]:
|
|
|
|
return Node(PatternError("not implemented"), self.current().span)
|
|
|
|
|
|
|
|
def step(self) -> None:
|
|
|
|
self.current_token = self.tokens.next()
|
|
|
|
|
|
|
|
def current_is(self, token_type: TokenType) -> bool:
|
|
|
|
return not self.done() and self.current().token_type == token_type
|
|
|
|
|
|
|
|
def done(self) -> bool:
|
|
|
|
return self.current_token.token_type == TokenType.Eof
|
|
|
|
|
|
|
|
def current(self) -> Token:
|
|
|
|
return self.current_token
|