futilehdl/src/parser.rs

pub mod module;
pub mod proc;

use nom::{
    branch::alt,
    bytes::complete::tag,
    character::complete::{alpha1, alphanumeric1, char, multispace0, u64 as decimal},
    combinator::{map, opt, recognize},
    error::{ParseError, VerboseError},
    multi::{many0, separated_list0},
    sequence::{delimited, pair, preceded, separated_pair, terminated, tuple},
};

use nom_locate::LocatedSpan;

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

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

use crate::literals::hexadecimal;
pub use crate::parser::module::{module, Module, ModuleItem, PortDirection};

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
fn typename(input: Span) -> IResult<Span, TypeName> {
    map(
        tuple((
            identifier,
            opt(delimited(char('<'), ws0(expression), char('>')))
        )),
        |(ident, _)| {
            TypeName {
                name: ident,
                generics: ()
            }
        }
    )(input)
}

fn widthspec(input: Span) -> IResult<Span, u64> {
    delimited(char('['), ws0(decimal), char(']'))(input)
}

fn intliteral(input: Span) -> IResult<Span, (u64, u64)> {
    tuple((terminated(decimal, char('\'')), alt((decimal, hexadecimal))))(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<(u64, u64)>,
}

#[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>>),
}

fn declaration(i: Span) -> IResult<Span, NetDecl> {
    map(
        tuple((
            separated_pair(identifier, ws0(char(':')), typename),
            opt(preceded(ws0(char('=')), intliteral)),
        )),
        |((ident, typ), value)| NetDecl {
            name: ident,
            typ,
            value,
        },
    )(i)
}

fn operation(input: Span) -> IResult<Span, Operation> {
    // temporarily given up on before I learn the shunting yard algorithm
    alt((
        map(
            separated_pair(ws0(expression_nonrecurse), char('&'), ws0(expression)),
            |(a, b)| Operation::And { a, b },
        ),
        map(
            separated_pair(ws0(expression_nonrecurse), char('|'), ws0(expression)),
            |(a, b)| Operation::Or { a, b },
        ),
        map(
            separated_pair(ws0(expression_nonrecurse), char('^'), ws0(expression)),
            |(a, b)| Operation::Xor { a, b },
        ),
        map(preceded(char('~'), expression), Operation::Not),
    ))(input)
}

fn call_item(input: Span) -> IResult<Span, Call> {
    map(
        tuple((
            ws0(identifier),
            delimited(
                char('('),
                ws0(separated_list0(char(','), expression)),
                char(')'),
            ),
        )),
        |(name, args)| Call { name, 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: Span) -> IResult<Span, Expression> {
    alt((
        map(ws0(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: Span) -> IResult<Span, Expression> {
    alt((
        map(ws0(decimal), Expression::Literal),
        map(ws0(call_item), |call| Expression::Call(Box::new(call))),
        map(ws0(identifier), |ident| {
            Expression::Ident(*ident.fragment())
        }),
        delimited(char('('), expression, char(')')),
    ))(input)
}

fn assign_statement(input: Span) -> IResult<Span, Assign> {
    map(
        separated_pair(ws0(identifier), char('='), ws0(expression)),
        |(lhs, expr)| Assign {
            lhs: (*lhs.fragment()),
            expr,
        },
    )(input)
}

pub fn parse(input: Span) -> IResult<Span, Module> {
    ws0(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();
    }
}