Silent overflow on debug beta/nightly

[talchas]

The standard wishlist

fn main() {
    let mut it = 255u8..;
    println!("{:?}", it.next());
}

no longer panics, but AFAICT from the code this is not due to any deliberate change. The Iterator implementation is start.add_one()/swap, and add_one() is self + 1. This sounds like some overflow checks are being lost somehow.

[TimNN]

related: #35589

[durka]

This was indeed caused by MIR:

seas779:cargo-script alex$ RUSTFLAGS=-Zorbit=off mr ru nightly-2016-08-23 cargo script --force --use-shared-binary-cache no --debug -e '(255u8..).next()'
   Compiling expr v0.1.0 (file:///Users/alex/.multirust/toolchains/nightly-2016-08-23/cargo/.cargo/script-cache/expr-24db503e5185c33f)
    Finished debug [unoptimized + debuginfo] target(s) in 0.41 secs
thread 'main' panicked at 'attempt to add with overflow', ../src/libcore/iter/range.rs:103
note: Run with `RUST_BACKTRACE=1` for a backtrace.
seas779:cargo-script alex$ RUSTFLAGS=-Zorbit=on mr ru nightly-2016-08-23 cargo script --force --use-shared-binary-cache no --debug -e '(255u8..).next()'
   Compiling expr v0.1.0 (file:///Users/alex/.multirust/toolchains/nightly-2016-08-23/cargo/.cargo/script-cache/expr-24db503e5185c33f)
    Finished debug [unoptimized + debuginfo] target(s) in 0.39 secs
Some(255)

cc @eddyb

[eddyb]

Missing #[rustc_inherit_overflow_checks] on the Step methods that do arithmetic (add_one and sub_one), introduced in #34530.

[ashrwin]

Can I work on this ?

[eddyb]

@ashrko619 Sure! The only change needed here is adding #[rustc_inherit_overflow_checks] before those 2 methods in all 3 implementations of Step and writing a test (not sure if it can be #[test] - cc @rust-lang/libs).

[ashrwin]

@eddyb Thanks. About the test - why shouldn't it be #[test] ?

[eddyb]

I don't know where we can put a #[test] for this, that's all. Also, it's dependent on having it in a separate crate, so #[test] might invalidate that requirement (i.e. it would have a local copy of Step).

However, it looks like libcoretest has the tests and it's a separate crate, so you should be able to add a test in that file.

[matthew-piziak]

@ashrko619 Any update on this issue? If not I might like to work on it.

[ashrwin]

@matthew-piziak Feel free to work on it. I have been caught up at work this week :)

[cramertj]

@matthew-piziak Is issue fixed? I noticed your PR said "part of"-- is there more that needs to be done?

[matthew-piziak]

@cramertj I think we're done here. The "part of" idiom is just to identify the parent of the PR.

@talchas @eddyb What still needs to be done for this issue, if anything?

[brson]

Closing per @matthew-piziak !