Release build taking forever on 1.34.1

[fizyk20]

I'm trying to build my crate at this commit: https://github.com/fizyk20/atm-refraction/tree/a7f38e582d507345f0f6bddbf769244caee0fe70

cargo build --release ran on Rust 1.34.1 doesn't finish for at least a few minutes (I admit I haven't been too patient with letting it run for a long time, as I knew it used to compile quite quickly, but I can safely say that it wasn't able to finish for me in at least 5 minutes). A debug build (cargo build) finishes in about 15 seconds.

I checked on another Rust version to make sure that it's different, and indeed, compilation of the same commit on 1.31.0 in release mode takes 24 seconds. Nightly 1.36.0 2019-05-03 can't finish just like 1.34.1.

I haven't tried versions other than 1.31.0, 1.34.1 and 1.36.0 nightly so far, but I can do so if needed.

I'm on Gentoo Linux on x86_64. All tested Rust versions installed via rustup.

If any other information is required, please let me know.

[ohsayan]

@fizyk20 Can you reproduce this frequently?

[fizyk20]

It happens every time I try to build this specific crate using 1.34.1. I haven't encountered the problem with any other piece of code yet, though.

I'm not sure if I answered the question.

[jonas-schievink]

I've given the crate a cursory look and nothing stood out as being particularly difficult for the compiler (except a few nom parsers perhaps?). Build is stuck somewhere in LLVM, probably running module passes on a cgu. I'll try to bisect, but it's likely that it will just point us to an LLVM upgrade.

[jonas-schievink]

Good: 1.33.0
Bad: 1.34.0

Good: rustc 1.34.0-nightly (097c04cf4 2019-02-24)
Bad: rustc 1.34.0-nightly (00aae71f5 2019-02-25)

Commits in range: 097c04c...00aae71

Likely cause: #57609 (since it touches MIR lowering)
cc @matthewjasper

[jonas-schievink]

It's something in air/parser.rs – if parse_atmospheres body is replaced with unimplemented!() the issue disappears.

[ohsayan]

@jonas-schievink
✔️ Verified on Rust 1.31.0 -> Build okay stable
✔️ Verified on Rust 1.33.0 -> Build okay stable
✔️ Verified on Rust 1.34.0 -> Slow Build nightly
✔️ Verified on Rust 1.34.1 -> Slow Build nightly
(Likely commits causing problems: c15437c...5ffc919

[jonas-schievink]

This is a proper hang, my build is still running after >5 hours.

[ohsayan]

This is a proper hang, my build is still running after >5 hours.

Yeah, I ran the build for the crate on EC2 and that's running for > 6 hours

[matthewjasper]

#60174 appears to fix this. Probably the issue is broken debug info. I'll try a build with llvm assertions tomorrow to see if that's really the case.

Smallish repro (pretty hard to meaning fully reduce further)

lib.rs:

use nom::sp;
use nom::types::CompleteStr;
use nom::Convert;
use nom::InputIter;
use nom::{need_more, Compare};
use nom::{Err, IResult, Needed};
pub fn float(i: CompleteStr) -> IResult<CompleteStr, f64> {
    regex::Regex::new("");
    Ok((i, 0.0))
}
pub struct TemperatureAt {
    pub start_h: f64,
    pub start_t: f64,
}
pub fn temperature_at(i: CompleteStr) -> IResult<CompleteStr, TemperatureAt> {
    match {
        let mut sep_res = sp(i);
        match sep_res {
            Ok((i1, _)) => {
                let res: IResult<CompleteStr, TemperatureAt> = match {
                    let sep_res = sp(i1);
                    match sep_res {
                        Ok((i2, _)) => {
                            i2.compare("");
                            let res2: IResult<_, CompleteStr> = need_more(i2, Needed::Size(0));
                            res2
                        }
                        Err(e07) => Err(Err::convert(e07)),
                    }
                } {
                    Err(e06) => Err(e06),
                    Ok(_) => match match sep_res {
                        Ok(_) => {
                            let res2: IResult<_, char> = match i1.iter_elements().next() {
                                Some('A') => {
                                    let e05 = nom::ErrorKind::Char;
                                    Err(Err::Error(nom::Context::Code(i1, e05)))
                                }
                                _ => need_more(i1, Needed::Size(1)),
                            };
                            res2
                        }
                        Err(e08) => Err(e08),
                    } {
                        Err(e09) => Err(e09),
                        _ => match {
                            sep_res = sp(i);
                            match sep_res {
                                Err(e10) => Err(e10),
                                _ => {
                                    let res = float(i1);
                                    res
                                }
                            }
                        } {
                            Err(e11) => Err(e11),
                            Ok((i7, o)) => match {
                                sep_res = sp(i7);
                                match sep_res {
                                    Ok((i5, _)) => {
                                        let res =
                                            need_more::<_, CompleteStr, u32>(i5, Needed::Size(1));
                                        res
                                    }
                                    Err(e12) => Err(Err::convert(e12)),
                                }
                            } {
                                Err(e01) => Err(e01),
                                _ => match {
                                    sep_res = sp(i7);
                                    sep_res
                                } {
                                    Err(e02) => Err(e02),
                                    _ => Ok((
                                        i,
                                        TemperatureAt {
                                            start_h: 0.0,
                                            start_t: 0.0,
                                        },
                                    )),
                                },
                            },
                        },
                    },
                };
                drop(&res);
                res
            }
            Err(e03) => Err(e03),
        }
    } {
        Err(e04) => Err(e04),
        Ok((i6, o01)) => loop {},
    }
}

Cargo.toml

[package]
name = ""
version = "0.1.0"
authors = [""]
edition = "2018"

[dependencies]
nom = "4.1"
regex = "1.0"

[matthewjasper]

This is fixed by d38eeacc86179a37e2bf35c1bd336df306b6edde.
Taking just the scope changes from there doesn't fix the issue, so this might just be a case that the code that's generated is different enough to avoid the issue.

[nikomatsakis]

@matthewjasper should we put "Fixes" on that PR for this issue?

[nikomatsakis]

triage: P-high

Regression -- assigning to @matthewjasper, who seems to have a pending fix

[matthewjasper]

@matthewjasper should we put "Fixes" on that PR for this issue?

I'd rather not use "Fixes" with no test and no idea what the actual problem is here.

[matthewjasper]

Changed labels now that this compiles successfully on beta.

[Mark-Simulacrum]

I don't think this is worth keeping open in the hopes of finding a test case.