regression: ICE failing to normalize

[Mark-Simulacrum]

https://crater-reports.s3.amazonaws.com/beta-1.69-3/beta-2023-03-18/gh/jwiegley.motoko_parse/log.txt

[INFO] [stdout] error: internal compiler error: compiler/rustc_middle/src/ty/normalize_erasing_regions.rs:195:90: Failed to normalize fn() -> <impl combine::Parser<combine::easy::Stream<&str>, Output = std::string::String> as combine::Parser<combine::easy::Stream<&str>>>::PartialState {<<impl combine::Parser<combine::easy::Stream<&str>, Output = std::string::String> as combine::Parser<combine::easy::Stream<&str>>>::PartialState as std::default::Default>::default}, maybe try to call `try_normalize_erasing_regions` instead

[lukas-code]

searched nightlies: from nightly-2023-01-24 to nightly-2023-03-30
regressed nightly: nightly-2023-02-05
searched commit range: 658fad6...50d3ba5
regressed commit: 9dee4e4

bisected with cargo-bisect-rustc v0.6.5

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --regress=ice -- build 

This looks related to #107678, but wasn't fixed by #107688.

[lqd]

Let's notify the people involved in that bisection, it's #107267 cc author @cjgillot and reviewer @oli-obk.

A minimization would be helpful.

[lqd]

#107267 probably just allowed another existing bug to show up, like the other projection issues.

I've reduced the motoko_parse regression to this

use combine::parser::char::spaces;
use combine::Parser;
use combine::{easy, error, stream};

fn mk_type_parser<'a>() -> impl Parser<easy::Stream<&'a str>, Output = ()> {
    mk_inner_type_parser().skip(spaces())
}

struct TypeParser<'a>(&'a ());
fn mk_inner_type_parser<'a>() -> TypeParser<'a> {
    unimplemented!()
}

impl<'a> Parser<easy::Stream<&'a str>> for TypeParser<'a>
where
    easy::Stream<&'a str>: stream::Stream,
{
    type Output = ();
    type PartialState = ();
    fn add_committed_expected_error(
        &mut self,
        _errors: &mut error::Tracked<<easy::Stream<&'a str> as stream::StreamOnce>::Error>,
    ) {
        let mut parser = mk_type_parser();
        let _: &mut dyn Parser<_, Output = _, PartialState = _> = &mut parser;
    }
}

it still depends on combine, if anyone has the time to also remove that.

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium

[lukas-code]

This seems to have disappeared in nightly. Bisect points to #108821. Maybe fixed by #108750?

[lqd]

Wasn't that PR already present in your previous bisection up to nightly-2023-03-30 ?

[lukas-code]

Ah, looks like the MCVE is fixed in #108821, but not the original code.

[lukas-code]

new mcve

use std::marker::PhantomData;

pub trait StreamOnce {
    type Token;
}

impl StreamOnce for &str {
    type Token = ();
}

pub trait Parser<Input: StreamOnce> {
    type PartialState: Default;
    fn parse_mode(&self, _state: &Self::PartialState) {}
    fn parse_mode_impl() {}
}

pub fn parse_bool<'a>() -> impl Parser<&'a str> {
    pub struct TokensCmp<C, Input>
    where
        Input: StreamOnce,
    {
        _cmp: C,
        _marker: PhantomData<Input>,
    }

    impl<Input, C> Parser<Input> for TokensCmp<C, Input>
    where
        C: FnMut(Input::Token),
        Input: StreamOnce,
    {
        type PartialState = ();
    }

    TokensCmp {
        _cmp: |_| (),
        _marker: PhantomData,
    }
}

pub struct ParseBool;

impl<'a> Parser<&'a str> for ParseBool
where
    &'a str: StreamOnce,
{
    type PartialState = ();

    fn parse_mode_impl() {
        parse_bool().parse_mode(&Default::default())
    }
}

@rustbot label -E-needs-mcve +S-has-mcve