Panic while formatting date in beta and nightly compilers

[Nerixyz]

The following code panics while handling a panic in beta and nightly versions of Rust:

main.rs:

use std::str::FromStr;

use icu_calendar::Date;
use icu_datetime::{options::length, DateFormatter};
use icu_locid::Locale;

fn main() {
    let locale = Locale::from_str("en-u-ca-japanese").unwrap();
    let formatter = DateFormatter::try_new_with_length(&locale.into(), length::Date::Full).unwrap();
    let date = Date::try_new_iso_date(2020, 5, 30).unwrap().to_any();
    dbg!(formatter.format_to_string(&date).unwrap());
}

cargo.toml

[package]
name = "bar"
version = "0.1.0"
edition = "2021"

[dependencies]
icu_datetime = { version = "1.5.1", features = ["std"] }
icu_calendar = { version = "1.5.1", features = ["std"] }
icu_locid = { version = "1.5.0", features = ["std"] }

output:

panicked at library\core\src\fmt\mod.rs:2431:38:
thread panicked while processing panic. aborting.
error: process didn't exit successfully: `target\debug\bar.exe` (exit code: 0xc0000409, STATUS_STACK_BUFFER_OVERRUN)

(library\core\src\fmt\mod.rs:2431:38)

It's not really easy to debug, but it seems like the original panic is because of this debug_assert failing and the next panic seems to happen while trying to format icu_calendar::Era - so the original error was probably DateTimeWriteError::MissingEraSymbol. In release mode, this causes invalid UTF-8 to be written, because the era is wrong (from debugging I noticed it seems to be garbage - at least until the next nul).

---

test_japanese panics on nightly as well (when changing the toolchain in rust-toolchain.toml to nightly).

---

Update: It seems like rustc decided to swap the order of the tuple elements inside dates_to_eras, so it thinks the string comes first, but that one doesn't contain valid UTF-8.

[Manishearth]

I'm surprised this wasn't caught by our CI

The test I've been using to check this is

use icu::calendar::{AnyCalendar, DateTime, Time};
use icu::datetime::{options::length, DateTimeFormatter};
use icu::locale::locale;
use writeable::Writeable;

fn main() {
    let locale = locale!("en-u-ca-japanese").into(); // English with the Japanese calendar

    let calendar = AnyCalendar::new_for_locale(&locale);

    let iso = DateTime::try_new_iso_datetime(2020, 10, 5, 13, 33, 15).unwrap();
    let datetime = iso.to_calendar(calendar);
    let options = length::Bag::from_date_time_style(length::Date::Full, length::Time::Short);

    let dtf = DateTimeFormatter::try_new(&locale, options.into())
        .expect("Failed to create DateTimeFormatter instance.");
    let manual_value = dtf
        .format(&datetime)
        .expect("Calendars should match");
    let s = manual_value.write_to_string();
    println!("{:?}", s);

    assert_eq!(s, "Monday, October 5, 2 Reiwa, 1:33 PM")
}

[Manishearth]

test_japanese panics on nightly as well (when changing the toolchain in rust-toolchain.toml to nightly).

@robertbastian would you be able to verify why the nightly CI didn't catch this?

[Manishearth]

[cargo-make] INFO - Running Task: set-ci-toolchain
forced-toolchain environment not required 

in both scheduled CI jobs https://github.com/unicode-org/icu4x/actions/runs/9501131062/job/26185924275 , https://github.com/unicode-org/icu4x/actions/runs/9501092498/job/26185787376

I know we tweaked how the nightly toolchain got set a bunch of times, I suspect we broke it?