schala/schala-lang/src/parsing/combinator.rs

use std::{cell::RefCell, rc::Rc};

use nom::{
    branch::alt,
    bytes::complete::{escaped_transform, tag, take_till, take_while},
    character::{
        complete::{
            alpha1, alphanumeric0, char, line_ending, none_of, not_line_ending, one_of, space0, space1,
        },
        is_alphanumeric,
    },
    combinator::{map, not, opt, peek, recognize, value},
    error::{context, ParseError, VerboseError},
    multi::{fold_many1, many0, many1, separated_list0, separated_list1},
    sequence::{delimited, pair, preceded, tuple},
    Err, IResult, Parser,
};
use nom_locate::{position, LocatedSpan};

use crate::identifier::{Id, IdStore};

type StoreRef = Rc<RefCell<IdStore<ASTItem>>>;

pub type Span<'a> = LocatedSpan<&'a str, StoreRef>;
type ParseResult<'a, O> = IResult<Span<'a>, O, VerboseError<Span<'a>>>;

use crate::ast::*;

fn rc_string(s: &str) -> Rc<String> {
    Rc::new(s.to_string())
}
fn fresh_id(span: &Span) -> Id<ASTItem> {
    let mut table_handle = span.extra.borrow_mut();
    table_handle.fresh()
}

fn fresh_id_rc(store_ref: &StoreRef) -> Id<ASTItem> {
    let mut table_handle = store_ref.borrow_mut();
    table_handle.fresh()
}

fn tok<'a, O>(
    input_parser: impl Parser<Span<'a>, O, VerboseError<Span<'a>>>,
) -> impl FnMut(Span<'a>) -> IResult<Span<'a>, O, VerboseError<Span<'a>>> {
    context("tok", map(tuple((ws0, input_parser)), |(_, output)| output))
}

fn kw<'a>(keyword_str: &'static str) -> impl FnMut(Span<'a>) -> ParseResult<()> {
    context("keyword", tok(value((), tag(keyword_str))))
}

// whitespace does consume at least one piece of whitespace - use ws0 for maybe none
fn whitespace(input: Span) -> ParseResult<()> {
    context("whitespace", alt((block_comment, line_comment, value((), space1))))(input)
}

fn ws0(input: Span) -> ParseResult<()> {
    context("WS0", value((), many0(whitespace)))(input)
}

fn line_comment(input: Span) -> ParseResult<()> {
    value((), tuple((tag("//"), not_line_ending)))(input)
}

fn block_comment(input: Span) -> ParseResult<()> {
    context(
        "Block-comment",
        value(
            (),
            tuple((
                tag("/*"),
                many0(alt((value((), none_of("*/")), value((), none_of("/*")), block_comment))),
                tag("*/"),
            )),
        ),
    )(input)
}

fn statement_delimiter(input: Span) -> ParseResult<()> {
    tok(alt((value((), line_ending), value((), char(';')))))(input)
}

pub fn program(input: Span) -> ParseResult<AST> {
    let id = fresh_id(&input);
    //TODO `rest` should be empty
    let (rest, statements) = context(
        "AST",
        map(
            tuple((
                many0(statement_delimiter),
                separated_list0(statement_delimiter, statement),
                many0(statement_delimiter),
            )),
            |(_, items, _)| items.into(),
        ),
    )(input)?;

    println!("REST: {}", rest.fragment());

    let ast = AST { id, statements };
    Ok((rest, ast))
}

pub fn block(input: Span) -> ParseResult<Block> {
    context(
        "block",
        map(
            tuple((
                tok(char('{')),
                many0(statement_delimiter),
                separated_list0(statement_delimiter, statement),
                many0(statement_delimiter),
                tok(char('}')),
            )),
            |(_, _, items, _, _)| items.into(),
        ),
    )(input)
}

fn statement(input: Span) -> ParseResult<Statement> {
    let (input, pos) = position(input)?;
    let location = pos.location_offset().into();
    let id = fresh_id(&input);
    let (rest, kind) = context(
        "Parsing-statement",
        alt((map(declaration, StatementKind::Declaration), map(expression, StatementKind::Expression))),
    )(input)?;
    Ok((rest, Statement { id, location, kind }))
}

fn declaration(input: Span) -> ParseResult<Declaration> {
    alt((binding, type_decl, func, annotation, module))(input)
}

fn annotation(input: Span) -> ParseResult<Declaration> {
    map(
        tuple((
            tok(char('@')),
            identifier,
            opt(delimited(tok(char('(')), separated_list1(tok(char(',')), expression), tok(char(')')))),
            statement_delimiter,
            statement,
        )),
        |(_, name, args, _, inner)| Declaration::Annotation {
            name: rc_string(name.fragment()),
            arguments: if let Some(args) = args { args } else { vec![] },
            inner: Box::new(inner),
        },
    )(input)
}

fn func(input: Span) -> ParseResult<Declaration> {
    alt((func_decl, map(func_signature, Declaration::FuncSig)))(input)
}

fn func_decl(input: Span) -> ParseResult<Declaration> {
    map(pair(func_signature, block), |(sig, decl)| Declaration::FuncDecl(sig, decl))(input)
}

//TODO handle operators
fn func_signature(input: Span) -> ParseResult<Signature> {
    map(tuple((kw("fn"), tok(identifier), formal_params, opt(type_anno))), |(_, name, params, type_anno)| {
        Signature { name: rc_string(name.fragment()), operator: false, params, type_anno }
    })(input)
}

fn formal_params(input: Span) -> ParseResult<Vec<FormalParam>> {
    delimited(tok(char('(')), separated_list0(tok(char(',')), formal_param), tok(char(')')))(input)
}

//TODO support 256-limit
fn formal_param(input: Span) -> ParseResult<FormalParam> {
    map(
        tuple((tok(identifier), opt(type_anno), opt(preceded(tok(char('=')), expression)))),
        |(name, anno, default)| FormalParam { name: rc_string(name.fragment()), anno, default },
    )(input)
}

fn type_decl(input: Span) -> ParseResult<Declaration> {
    alt((
        map(
            tuple((kw("type"), kw("alias"), tok(identifier), tok(char('=')), tok(identifier))),
            |(_, _, alias, _, name)| Declaration::TypeAlias {
                alias: rc_string(alias.fragment()),
                original: rc_string(name.fragment()),
            },
        ),
        map(
            tuple((kw("type"), opt(kw("mut")), type_singleton_name, tok(char('=')), type_body)),
            |(_, mutable, name, _, body)| Declaration::TypeDecl { name, body, mutable: mutable.is_some() },
        ),
    ))(input)
}

fn type_body(input: Span) -> ParseResult<TypeBody> {
    let id = fresh_id(&input);
    alt((
        map(
            delimited(tok(char('{')), separated_list1(tok(char(',')), record_variant_item), tok(char('}'))),
            move |items| TypeBody::ImmediateRecord { id, fields: items },
        ),
        map(separated_list0(tok(char('|')), variant_spec), TypeBody::Variants),
    ))(input)
}

fn variant_spec(input: Span) -> ParseResult<Variant> {
    fn record_variant(input: Span) -> ParseResult<VariantKind> {
        map(
            delimited(tok(char('{')), separated_list1(tok(char(',')), record_variant_item), tok(char('}'))),
            VariantKind::Record,
        )(input)
    }

    fn tuple_variant(input: Span) -> ParseResult<VariantKind> {
        map(
            delimited(tok(char('(')), separated_list1(tok(char(',')), type_identifier), tok(char(')'))),
            VariantKind::TupleStruct,
        )(input)
    }

    let id = fresh_id(&input);
    let (rest, (name, kind)) = alt((
        pair(tok(identifier), record_variant),
        pair(tok(identifier), tuple_variant),
        map(tok(identifier), |ident| (ident, VariantKind::UnitStruct)),
    ))(input)?;

    Ok((rest, Variant { id, name: rc_string(name.fragment()), kind }))
}

fn record_variant_item(input: Span) -> ParseResult<(Rc<String>, TypeIdentifier)> {
    map(tuple((tok(identifier), tok(char(':')), type_identifier)), |(name, _, ty)| {
        (rc_string(name.fragment()), ty)
    })(input)
}

fn binding(input: Span) -> ParseResult<Declaration> {
    let parser =
        tuple((kw("let"), opt(kw("mut")), tok(identifier), opt(type_anno), tok(char('=')), expression));
    map(parser, |(_, maybe_mut, ident, type_anno, _, expr)| Declaration::Binding {
        name: rc_string(ident.fragment()),
        constant: maybe_mut.is_none(),
        type_anno,
        expr,
    })(input)
}

fn module(input: Span) -> ParseResult<Declaration> {
    map(tuple((kw("module"), tok(identifier), block)), |(_, name, items)| Declaration::Module {
        name: rc_string(name.fragment()),
        items,
    })(input)
}

pub fn expression(input: Span) -> ParseResult<Expression> {
    let id = fresh_id(&input);
    map(pair(expression_kind, opt(type_anno)), move |(kind, type_anno)| Expression { id, type_anno, kind })(
        input,
    )
}

fn type_anno(input: Span) -> ParseResult<TypeIdentifier> {
    preceded(tok(char(':')), type_identifier)(input)
}

fn type_identifier(input: Span) -> ParseResult<TypeIdentifier> {
    alt((
        map(
            delimited(tok(char('(')), separated_list0(tok(char(',')), type_identifier), tok(char(')'))),
            TypeIdentifier::Tuple,
        ),
        map(type_singleton_name, TypeIdentifier::Singleton),
    ))(input)
}

fn type_singleton_name(input: Span) -> ParseResult<TypeSingletonName> {
    map(pair(tok(identifier), opt(type_params)), |(name, params)| TypeSingletonName {
        name: rc_string(name.fragment()),
        params: if let Some(params) = params { params } else { vec![] },
    })(input)
}

fn type_params(input: Span) -> ParseResult<Vec<TypeIdentifier>> {
    delimited(tok(char('<')), separated_list1(tok(char(',')), type_identifier), tok(char('>')))(input)
}

pub fn expression_kind(input: Span) -> ParseResult<ExpressionKind> {
    context("expression-kind", precedence_expr)(input)
}

fn precedence_expr(input: Span) -> ParseResult<ExpressionKind> {
    let handle = input.extra.clone();
    map(
        pair(prefix_expr, many0(precedence_continuation)),
        move |(first, rest): (ExpressionKind, Vec<(BinOp, ExpressionKind)>)| {
            let mut handle_ref = handle.borrow_mut();
            BinopSequence { first, rest }.do_precedence(&mut handle_ref)
        },
    )(input)
}

fn precedence_continuation(input: Span) -> ParseResult<(BinOp, ExpressionKind)> {
    pair(operator, prefix_expr)(input)
}

fn operator(input: Span) -> ParseResult<BinOp> {
    tok(map(
        tuple((not(tag("*/")), recognize(many1(one_of("+-*/%<>=!$&|?^`"))))),
        |(_, sigil_span): ((), Span)| BinOp::from_sigil(sigil_span.fragment()),
    ))(input)
}

fn prefix_op(input: Span) -> ParseResult<PrefixOp> {
    tok(map(recognize(one_of("+-!")), |sigil: Span| PrefixOp::from_sigil(sigil.fragment())))(input)
}

fn prefix_expr(input: Span) -> ParseResult<ExpressionKind> {
    let handle = input.extra.clone();
    context(
        "prefix-expr",
        map(pair(opt(prefix_op), extended_expr), move |(prefix, expr)| {
            if let Some(prefix) = prefix {
                let expr = Expression::new(fresh_id_rc(&handle), expr);
                ExpressionKind::PrefixExp(prefix, Box::new(expr))
            } else {
                expr
            }
        }),
    )(input)
}

#[derive(Debug)]
enum ExtendedPart<'a> {
    Index(Vec<Expression>),
    Call(Vec<InvocationArgument>),
    Accessor(&'a str),
}

fn extended_expr(input: Span) -> ParseResult<ExpressionKind> {
    let (s, (primary, parts)) =
        context("extended-expr", pair(primary_expr, many0(extended_expr_part)))(input)?;

    let mut expression = Expression::new(fresh_id(&s), primary);
    for part in parts.into_iter() {
        let kind = match part {
            ExtendedPart::Index(indexers) =>
                ExpressionKind::Index { indexee: Box::new(expression), indexers },
            ExtendedPart::Call(arguments) => ExpressionKind::Call { f: Box::new(expression), arguments },
            ExtendedPart::Accessor(name) => {
                let name = rc_string(name);
                ExpressionKind::Access { name, expr: Box::new(expression) }
            }
        };
        expression = Expression::new(fresh_id(&s), kind);
    }

    Ok((s, expression.kind))
}

fn extended_expr_part(input: Span) -> ParseResult<ExtendedPart> {
    fn index_part(input: Span) -> ParseResult<Vec<Expression>> {
        delimited(tok(char('[')), separated_list1(tok(char(',')), expression), tok(char(']')))(input)
    }

    fn call_part(input: Span) -> ParseResult<Vec<InvocationArgument>> {
        delimited(tok(char('(')), separated_list0(tok(char(',')), invocation_argument), tok(char(')')))(input)
    }

    fn access_part(input: Span) -> ParseResult<&str> {
        preceded(tok(char('.')), map(identifier, |item| *item.fragment()))(input)
    }

    alt((
        map(index_part, ExtendedPart::Index),
        map(call_part, ExtendedPart::Call),
        map(access_part, ExtendedPart::Accessor),
    ))(input)
}

//TODO this shouldn't be an expression b/c type annotations disallowed here
fn invocation_argument(input: Span) -> ParseResult<InvocationArgument> {
    alt((
        map(tok(char('_')), |_| InvocationArgument::Ignored),
        map(tuple((tok(identifier), tok(char('=')), expression)), |(name, _, expr)| {
            InvocationArgument::Keyword { name: rc_string(name.fragment()), expr }
        }),
        map(expression, InvocationArgument::Positional),
    ))(input)
}

fn primary_expr(input: Span) -> ParseResult<ExpressionKind> {
    context(
        "primary-expr",
        alt((
            list_expr,
            paren_expr,
            string_literal,
            float_literal,
            number_literal,
            bool_literal,
            identifier_expr,
        )),
    )(input)
}

fn paren_expr(input: Span) -> ParseResult<ExpressionKind> {
    delimited(
        tok(char('(')),
        map(separated_list0(tok(char(',')), expression), |mut exprs| match exprs.len() {
            1 => exprs.pop().unwrap().kind,
            _ => ExpressionKind::TupleLiteral(exprs),
        }),
        tok(char(')')),
    )(input)
}

fn list_expr(input: Span) -> ParseResult<ExpressionKind> {
    map(delimited(tok(char('[')), separated_list0(tok(char(',')), expression), tok(char(']'))), |items| {
        ExpressionKind::ListLiteral(items)
    })(input)
}

//TODO need to do something with prefix in the AST
fn string_literal(input: Span) -> ParseResult<ExpressionKind> {
    tok(map(pair(opt(identifier), bare_string_literal), |(_maybe_prefix, s)| {
        ExpressionKind::StringLiteral(Rc::new(s))
    }))(input)
}

fn bare_string_literal(input: Span) -> ParseResult<String> {
    let string_escape_transforms =
        alt((value("\\", tag("\\")), value("\"", tag("\"")), value("\n", tag("n")), value("\t", tag("t"))));
    alt((
        map(tag(r#""""#), |_| String::new()),
        map(
            tuple((
                char('"'),
                escaped_transform(none_of(r#""\"#), '\\', string_escape_transforms),
                char('"'),
            )),
            |(_, s, _)| s,
        ),
    ))(input)
}

fn identifier_expr(input: Span) -> ParseResult<ExpressionKind> {
    context("identifier-expr", map(qualified_identifier, ExpressionKind::Value))(input)
}

fn qualified_identifier(input: Span) -> ParseResult<QualifiedName> {
    let id = fresh_id(&input);
    tok(map(separated_list1(tag("::"), map(identifier, |x| rc_string(x.fragment()))), move |items| {
        QualifiedName { id, components: items }
    }))(input)
}

fn identifier(input: Span) -> ParseResult<Span> {
    recognize(pair(alt((tag("_"), alpha1)), take_while(|ch: char| is_alphanumeric(ch as u8) || ch == '_')))(
        input,
    )
}

fn bool_literal(input: Span) -> ParseResult<ExpressionKind> {
    context(
        "bool-literal",
        alt((
            map(kw("true"), |_| ExpressionKind::BoolLiteral(true)),
            map(kw("false"), |_| ExpressionKind::BoolLiteral(false)),
        )),
    )(input)
}

fn float_literal(input: Span) -> ParseResult<ExpressionKind> {
    tok(map(
        alt((
            recognize(tuple((digits(digit_group_dec), char('.'), opt(digits(digit_group_dec))))),
            recognize(tuple((char('.'), digits(digit_group_dec)))),
        )),
        |ds| ExpressionKind::FloatLiteral(ds.fragment().parse().unwrap()),
    ))(input)
}

fn number_literal(input: Span) -> ParseResult<ExpressionKind> {
    map(alt((tok(hex_literal), tok(bin_literal), tok(dec_literal))), ExpressionKind::NatLiteral)(input)
}

fn dec_literal(input: Span) -> ParseResult<u64> {
    map(digits(digit_group_dec), |chars: Vec<char>| {
        let s: String = chars.into_iter().collect();
        s.parse().unwrap()
    })(input)
}

fn hex_literal(input: Span) -> ParseResult<u64> {
    map(preceded(alt((tag("0x"), tag("0X"))), digits(digit_group_hex)), |chars: Vec<char>| {
        let s: String = chars.into_iter().collect();
        parse_hex(&s).unwrap()
    })(input)
}

fn bin_literal(input: Span) -> ParseResult<u64> {
    map(preceded(alt((tag("0b"), tag("0B"))), digits(digit_group_bin)), |chars: Vec<char>| {
        let s: String = chars.into_iter().collect();
        parse_binary(&s).unwrap()
    })(input)
}

fn digits<'a, E: ParseError<Span<'a>>>(
    digit_type: impl Parser<Span<'a>, Vec<char>, E>,
) -> impl FnMut(Span<'a>) -> IResult<Span<'a>, Vec<char>, E> {
    map(separated_list1(many1(char('_')), digit_type), |items: Vec<Vec<char>>| {
        items.into_iter().flatten().collect()
    })
}

fn digit_group_dec(input: Span) -> ParseResult<Vec<char>> {
    many1(one_of("0123456789"))(input)
}

fn digit_group_hex(input: Span) -> ParseResult<Vec<char>> {
    many1(one_of("0123456789abcdefABCDEF"))(input)
}

fn digit_group_bin(input: Span) -> ParseResult<Vec<char>> {
    many1(one_of("01"))(input)
}

fn parse_binary(digits: &str) -> Result<u64, &'static str> {
    let mut result: u64 = 0;
    let mut multiplier = 1;
    for d in digits.chars().rev() {
        match d {
            '1' => result += multiplier,
            '0' => (),
            '_' => continue,
            _ => unreachable!(),
        }
        multiplier = match multiplier.checked_mul(2) {
            Some(m) => m,
            None => return Err("Binary expression will overflow"),
        }
    }
    Ok(result)
}

fn parse_hex(digits: &str) -> Result<u64, &'static str> {
    let mut result: u64 = 0;
    let mut multiplier: u64 = 1;
    for d in digits.chars().rev() {
        if d == '_' {
            continue;
        }
        match d.to_digit(16) {
            Some(n) => result += n as u64 * multiplier,
            None => return Err("Internal parser error: invalid hex digit"),
        }
        multiplier = match multiplier.checked_mul(16) {
            Some(m) => m,
            None => return Err("Hexadecimal expression will overflow"),
        }
    }
    Ok(result)
}

#[derive(Debug)]
struct BinopSequence {
    first: ExpressionKind,
    rest: Vec<(BinOp, ExpressionKind)>,
}

impl BinopSequence {
    fn do_precedence(self, store: &mut IdStore<ASTItem>) -> ExpressionKind {
        fn helper(
            precedence: i32,
            lhs: ExpressionKind,
            rest: &mut Vec<(BinOp, ExpressionKind)>,
            store: &mut IdStore<ASTItem>,
        ) -> Expression {
            let mut lhs = Expression::new(store.fresh(), lhs);
            while let Some((next_op, next_rhs)) = rest.pop() {
                let new_precedence = next_op.get_precedence();
                if precedence >= new_precedence {
                    rest.push((next_op, next_rhs));
                    break;
                }
                let rhs = helper(new_precedence, next_rhs, rest, store);
                lhs = Expression::new(
                    store.fresh(),
                    ExpressionKind::BinExp(next_op, Box::new(lhs), Box::new(rhs)),
                );
            }
            lhs
        }
        let mut as_stack = self.rest.into_iter().rev().collect();
        helper(BinOp::min_precedence(), self.first, &mut as_stack, store).kind
    }
}

#[cfg(test)]
mod test {
    use pretty_assertions::assert_eq;

    use super::*;

    fn rc(s: &str) -> Rc<String> {
        Rc::new(s.to_owned())
    }
    macro_rules! qn {
        ( $( $component:ident),* ) => {
            {
                let mut components = vec![];
                $(
                    components.push(rc(stringify!($component)));
                )*
                    QualifiedName { components, id: Default::default() }
            }
        };
    }

    macro_rules! span {
        ($func:expr, $input:expr) => {{
            let id_store: IdStore<ASTItem> = IdStore::new();
            let span = Span::new_extra($input, Rc::new(RefCell::new(id_store)));
            $func(span).map(|(span, x)| (*span.fragment(), x))
        }};
    }

    #[test]
    fn combinator_test1() {
        assert_eq!(span!(digits(digit_group_dec), "342"), Ok(("", vec!['3', '4', '2'])));
        assert_eq!(span!(bin_literal, "0b1111qsdf"), Ok(("qsdf", 15)));
        assert_eq!(span!(bare_string_literal, r#""fah""#), Ok(("", "fah".to_string())));
        assert_eq!(span!(bare_string_literal, r#""""#), Ok(("", "".to_string())));
    }

    #[test]
    fn combinator_test_ws0() {
        assert_eq!(span!(block_comment, "/*yolo*/"), Ok(("", ())));
        assert_eq!(span!(block_comment, "/*yolo*/ jumpy /*nah*/"), Ok((" jumpy /*nah*/", ())));
        assert_eq!(span!(ws0, "/*    yolo  */ "), Ok(("", ())));
        assert_eq!(span!(ws0, "/*  /* no */   yolo  */ "), Ok(("", ())));
    }

    #[test]
    fn combinator_test2() {
        for s in ["  15", "   0b1111", "   1_5_", "0XF__", "0Xf"].iter() {
            assert_eq!(span!(expression_kind, s).unwrap().1, ExpressionKind::NatLiteral(15));
        }

        assert_eq!(span!(expression_kind, " /*gay*/  true").unwrap().1, ExpressionKind::BoolLiteral(true));
        assert_eq!(
            span!(expression_kind, " /*yolo*/  barnaby").unwrap().1,
            ExpressionKind::Value(qn!(barnaby))
        );
    }
}