tokio::io::Take overflows on 32bit target

[svenrademakers]

Version

cargo tree | grep tokio
│   │   │   ├── tokio v1.32.0
│   │   │   │   └── tokio-macros v2.1.0 (proc-macro)
│   │   │   ├── tokio-util v0.7.9
│   │   │   │   ├── tokio v1.32.0 (*)
│   │   │   └── tokio v1.32.0 (*)
│   │   │   ├── tokio v1.32.0 (*)
│   │   │   ├── tokio-openssl v0.6.3
│   │   │   │   └── tokio v1.32.0 (*)
│   │   │   ├── tokio-util v0.7.9 (*)
│   │   │   ├── tokio v1.32.0 (*)
│   │   │   ├── tokio-util v0.7.9 (*)
│   │   ├── tokio v1.32.0 (*)
│   │   ├── tokio-util v0.7.9 (*)
│   │   │   ├── tokio v1.32.0 (*)
│   └── tokio v1.32.0 (*)
├── tokio v1.32.0 (*)
├── tokio-serial v5.4.4
│   ├── tokio v1.32.0 (*)
│   └── tokio-util v0.7.9 (*)
├── tokio-stream v0.1.14
│   └── tokio v1.32.0 (*)
└── tokio-util v0.7.9 (*)

Platform

• host = x86_64 GNU/Linux
• target = QEMU?! armel, armv7-unknown-linux-gnueabi
  (tested on real armv7 target as well)

Description

calling tokio::io::copy on my armv7 target returned exactly with u32::MAX bytes left to copy. I distilled the problem into a generic tokio test, but the behavior is still not as expected.

given the following test-case:

    use tokio::io::{sink, AsyncReadExt};

    #[tokio::test]
    async fn big_copy_test() -> std::io::Result<()> {
        // make sure it overflows usize on 32bit targets
        let mut reader = tokio::io::repeat(0b101).take(u32::MAX as u64 + 1u64);
        let copied = tokio::io::copy(&mut reader, &mut sink()).await?;
        assert_eq!(copied, u32::MAX as u64 + 1u64);
        Ok(())
    }

Completes successfully, as expected, on x64 executed via:
cargo test big_copy_test

FAILS on armv7-linux-gnueabi using cross-rs:
cross test big_copy --target armv7-unknown-linux-gnueabi --release

Outcome:

running 1 test
test utils::io::test::big_copy_test ... thread 'utils::io::test::big_copy_test' panicked at 'assertion failed: `(left == right)`
  left: `0`,
 right: `4294967296`', src/utils/io.rs:88:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
FAILED

Expected:
left == right, test passed.

[Darksonn]

Thanks for reporting this. I've filed a PR with a fix.

[Darksonn]

I've renamed the issue since the problem is actually in take, as far as I can tell.

[svenrademakers]

No thank you for looking into this! Keep up the good work!