futilehdl/src/parser.rs

pub mod error;
mod literals;
pub mod module;
pub mod proc;
pub mod tokens;

use nom::{
    branch::alt,
    bytes::complete::tag,
    character::complete::{alpha1, alphanumeric1, multispace0},
    combinator::{map, opt, recognize},
    error::{ErrorKind, ParseError},
    multi::{many0, separated_list0},
    sequence::{delimited, pair, preceded, separated_pair, tuple},
};
use nom_greedyerror::GreedyError;
use nom_locate::LocatedSpan;

// custom span type for nom_locate
pub type Span<'a> = LocatedSpan<&'a str>;

pub type IErr<I> = GreedyError<I, ErrorKind>;
// custom IResult type for VerboseError
pub type IResult<I, O, E = IErr<I>> = nom::IResult<I, O, E>;

pub use crate::parser::module::{module, Module, ModuleItem, PortDirection};
use crate::parser::tokens::{token, TokenKind as tk, TokenSpan};

fn ws0<'a, F: 'a, O, E: ParseError<Span<'a>>>(
    inner: F,
) -> impl FnMut(Span<'a>) -> IResult<Span<'a>, O, E>
where
    F: FnMut(Span<'a>) -> IResult<Span<'a>, O, E>,
{
    delimited(multispace0, inner, multispace0)
}

fn identifier(input: Span) -> IResult<Span, Span> {
    recognize(pair(
        alt((alpha1, tag("_"))),
        many0(alt((alphanumeric1, tag("_")))),
    ))(input)
}

// TODO: allow recursive generics
// TODO: allow expressions again
fn typename(input: TokenSpan) -> IResult<TokenSpan, TypeName> {
    map(
        tuple((
            token(tk::Ident),
            opt(delimited(token(tk::LAngle), expression, token(tk::RAngle))),
        )),
        |(ident, _)| TypeName {
            name: ident.span(),
            generics: (),
        },
    )(input)
}

#[derive(Debug)]
pub struct TypeName<'a> {
    name: Span<'a>,
    generics: (),
}

#[derive(Debug)]
pub struct NetDecl<'a> {
    pub name: Span<'a>,
    pub typ: TypeName<'a>,
    pub value: Option<Span<'a>>,
}

#[derive(Debug)]
pub struct Assign<'a> {
    pub lhs: &'a str,
    pub expr: Expression<'a>,
}

#[derive(Debug, Clone)]
pub enum Operation<'a> {
    And {
        a: Expression<'a>,
        b: Expression<'a>,
    },
    Or {
        a: Expression<'a>,
        b: Expression<'a>,
    },
    Xor {
        a: Expression<'a>,
        b: Expression<'a>,
    },
    Not(Expression<'a>),
}

#[derive(Debug, Clone)]
pub struct Call<'a> {
    pub name: Span<'a>,
    pub args: Vec<Expression<'a>>,
}

#[derive(Debug, Clone)]
pub enum Expression<'a> {
    Ident(&'a str),
    Literal(u64),
    Call(Box<Call<'a>>),
    Operation(Box<Operation<'a>>),
}

// TODO: reallow assignments
fn declaration(i: TokenSpan) -> IResult<TokenSpan, NetDecl> {
    map(
        tuple((
            separated_pair(token(tk::Ident), token(tk::Colon), typename),
            opt(preceded(token(tk::Assign), token(tk::Number))),
        )),
        |((ident, typ), _value)| NetDecl {
            name: ident.span(),
            typ,
            value: None,
        },
    )(i)
}

fn operation(input: TokenSpan) -> IResult<TokenSpan, Operation> {
    // temporarily given up on before I learn the shunting yard algorithm
    alt((
        map(
            separated_pair(expression_nonrecurse, token(tk::BitAnd), expression),
            |(a, b)| Operation::And { a, b },
        ),
        map(
            separated_pair(expression_nonrecurse, token(tk::BitOr), expression),
            |(a, b)| Operation::Or { a, b },
        ),
        map(
            separated_pair(expression_nonrecurse, token(tk::BitXor), expression),
            |(a, b)| Operation::Xor { a, b },
        ),
        map(preceded(token(tk::BitNot), expression), Operation::Not),
    ))(input)
}

fn call_item(input: TokenSpan) -> IResult<TokenSpan, Call> {
    map(
        tuple((
            token(tk::Ident),
            delimited(
                token(tk::LParen),
                separated_list0(token(tk::Comma), expression),
                token(tk::RParen),
            ),
        )),
        |(name, args)| Call {
            name: name.span(),
            args,
        },
    )(input)
}

/// parser combinators can not parse left-recursive grammars. To work around this, we split
/// expressions into a recursive and non-recursive portion.
/// Parsers reachable from this point must call expression_nonrecurse instead
fn expression(input: TokenSpan) -> IResult<TokenSpan, Expression> {
    alt((
        map(operation, |op| Expression::Operation(Box::new(op))),
        expression_nonrecurse,
    ))(input)
}

/// the portion of the expression grammar that can be parsed without left recursion
fn expression_nonrecurse(input: TokenSpan) -> IResult<TokenSpan, Expression> {
    alt((
        map(token(tk::Number), |_| Expression::Literal(42)),
        map(call_item, |call| Expression::Call(Box::new(call))),
        map(token(tk::Ident), |ident| {
            Expression::Ident(*ident.span().fragment())
        }),
        delimited(token(tk::LParen), expression, token(tk::RParen)),
    ))(input)
}

fn assign_statement(input: TokenSpan) -> IResult<TokenSpan, Assign> {
    map(
        separated_pair(token(tk::Ident), token(tk::EqAssign), expression),
        |(lhs, expr)| Assign {
            lhs: (*lhs.span().fragment()),
            expr,
        },
    )(input)
}

pub fn parse(input: TokenSpan) -> IResult<TokenSpan, Module> {
    module(input)
}

#[cfg(test)]
mod test {
    use super::*;
    use nom::combinator::all_consuming;

    #[test]
    fn test_operation() {
        operation(" a | b ".into()).unwrap();
        operation(" a & b ".into()).unwrap();
    }

    #[test]
    fn test_expression() {
        expression(" a ".into()).unwrap();
        expression(" a | b ".into()).unwrap();
        expression(" a | b | c ".into()).unwrap();
    }

    #[test]
    fn test_assignment() {
        // TODO: make wrapper and use for all tests
        all_consuming(assign_statement)(" a = b ".into()).unwrap();
        all_consuming(assign_statement)(" a = b | c ".into()).unwrap();
    }

    #[test]
    fn test_call() {
        call_item("thing ( )".into()).unwrap();
        call_item("thing ( a , b , c )".into()).unwrap();
        call_item("thing(a,b,c)".into()).unwrap();
    }
}