schala/src/schala_lang/parsing.rs

201 lines
5.4 KiB
Rust

extern crate itertools;
use language::{TokenError, ParseError};
use std::rc::Rc;
use std::iter::{Enumerate, Peekable};
use self::itertools::Itertools;
use std::str::Chars;
#[allow(dead_code)]
#[derive(Debug)]
pub enum TokenType {
Newline, Semicolon,
LParen, RParen,
LSquareBracket, RSquareBracket,
LAngleBracket, RAngleBracket,
LCurlyBrace, RCurlyBrace,
Comma, Period, Colon, Underscore,
Operator(Rc<String>),
DigitGroup(Rc<String>), HexNumberSigil, BinNumberSigil,
StrLiteral(Rc<String>),
Identifier(Rc<String>),
Keyword(Kw),
Error(String),
}
#[derive(Debug)]
pub enum Kw {
If,
Else,
Func,
For,
Loop,
}
#[derive(Debug)]
pub struct Token {
token_type: TokenType,
offset: usize,
}
fn is_digit(c: &char) -> bool {
c.is_digit(10)
}
type CharIter<'a> = Peekable<Enumerate<Chars<'a>>>;
pub fn tokenize(input: &str) -> Result<Vec<Token>, TokenError> {
use self::TokenType::*;
let mut tokens: Vec<Token> = Vec::new();
let mut input: CharIter = input.chars().enumerate().peekable();
while let Some((idx, c)) = input.next() {
let cur_tok_type = match c {
'#' => {
if let Some(&(_, '{')) = input.peek() {
} else {
while let Some((_, c)) = input.next() {
if c == '\n' {
break;
}
}
}
continue;
},
c if char::is_whitespace(c) && c != '\n' => continue,
'\n' => Newline, ';' => Semicolon,
':' => Colon, ',' => Comma, '.' => Period,
'(' => LParen, ')' => RParen,
'{' => LCurlyBrace, '}' => RCurlyBrace,
'<' => LAngleBracket, '>' => RAngleBracket,
'[' => LSquareBracket, ']' => RSquareBracket,
'"' => handle_quote(&mut input),
c if is_digit(&c) => handle_digit(c, &mut input),
c @ '_' | c if c.is_alphabetic() => handle_alphabetic(c, &mut input),
c => handle_operator(c, &mut input),
};
tokens.push(Token { token_type: cur_tok_type, offset: idx });
}
Ok(tokens)
}
fn handle_digit(c: char, input: &mut CharIter) -> TokenType {
use self::TokenType::*;
if c == '0' && input.peek().map_or(false, |&(_, c)| { c == 'x' }) {
input.next();
HexNumberSigil
} else if c == '0' && input.peek().map_or(false, |&(_, c)| { c == 'b' }) {
input.next();
BinNumberSigil
} else {
let mut buf = c.to_string();
buf.extend(input.peeking_take_while(|&(_, ref c)| is_digit(c)).map(|(_, c)| { c }));
DigitGroup(Rc::new(buf))
}
}
fn handle_quote(input: &mut CharIter) -> TokenType {
let mut buf = String::new();
while let Some(c) = input.next().map(|(_, c)| { c }) {
if c == '"' {
break;
} else if c == '\\' {
let next = input.peek().map(|&(_, c)| { c });
if next == Some('n') {
input.next();
buf.push('\n')
} else if next == Some('"') {
input.next();
buf.push('"');
}
//TODO handle more escapes
} else {
buf.push(c);
}
}
TokenType::StrLiteral(Rc::new(buf))
}
fn handle_alphabetic(c: char, input: &mut CharIter) -> TokenType {
unimplemented!()
}
fn handle_operator(c: char, input: &mut CharIter) -> TokenType {
unimplemented!()
}
/*
Schala EBNF grammar
type alias <name> = <other type>
type <name> = struct { <field> : <type>,* }
type <name> = Variant1 | Variant2(type, type) | Variant3 struct { }
'' = literal, all other symbols are nonterminals
program := (statement delimiter ?)*
delimiter := 'Newline' | ';'
statement := declaration | expression
declaration := module | function | type_decl
type_decl := 'type' type_format
type_format := 'alias' '=' type | type_constructor
type_constructor := capital_ident '=' type_rhs
type_rhs := struct_decl | type_variant ('|' type_variant)*
struct_decl := 'struct' '{' (ident ':' type)* '}'
type_variant := capital_ident | tuple_type | capital_ident struct_decl
tuple_type := // something like Variant(a,b)
type := // something like Type[A[b]]
ascription := expression (':' type)+
function := 'fn' prototype '{' (statement)* '}'
prototype := identifier '(' identlist ')'
identlist := identifier (',' identifier)* | ε
declaration := FN prototype LCurlyBrace (statement)* RCurlyBrace
prototype := identifier LParen identlist RParen
identlist := Ident (Comma Ident)* | ε
exprlist := Expression (Comma Expression)* | ε
itemlist := Ident COLON Expression (Comma Ident COLON Expression)* | ε
expression := postop_expression (op postop_expression)*
postop_expression := primary_expression postop
primary_expression := number_expr | String | identifier_expr | paren_expr | conditional_expr | while_expr | lambda_expr | list_expr | struct_expr
number_expr := (PLUS | MINUS ) number_expr | Number
identifier_expr := call_expression | Variable
list_expr := LSquareBracket exprlist RSquareBracket
struct_expr := LCurlyBrace itemlist RCurlyBrace
call_expression := Identifier LParen exprlist RParen
while_expr := WHILE primary_expression LCurlyBrace (expression delimiter)* RCurlyBrace
paren_expr := LParen expression RParen
conditional_expr := IF expression LCurlyBrace (expression delimiter)* RCurlyBrace (LCurlyBrace (expresion delimiter)* RCurlyBrace)?
lambda_expr := FN LParen identlist RParen LCurlyBrace (expression delimiter)* RCurlyBrace
lambda_call := | LParen exprlist RParen
postop := ε | LParen exprlist RParen | LBracket expression RBracket
op := '+', '-', etc.
*/
#[allow(dead_code)]
#[derive(Debug)]
pub struct AST { }
#[allow(dead_code)]
pub fn parse(_input: Vec<Token>) -> Result<AST, ParseError> {
Ok(AST { })
}