Inconsistent reporting of [E0733] between cargo check and cargo run

[bragov4ik]

Problem

I encountered a weird issue where cargo check reports success, whereas cargo run (or build, test, etc.) results in compilation error (of course with exactly the same feature flags and cfg(test)).

I expected them to be consistent, (check should report all build failures as far as I undestand)

Steps

1. Use the following minimal reproducible example

pub trait Foo {
    type Dependency: Foo;

    async fn foo() {
        Self::Dependency::foo().await;
    }
}

impl Foo for () {
    type Dependency = ();
}

#[tokio::main]
async fn main() {
    <()>::foo().await;
}

2. Run cargo check - only warning is shown:

warning: `test-play` (bin "test-play") generated 1 warning
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.08s

3. cargo build results in error:

error[E0733]: recursion in an async fn requires boxing
 --> src/main.rs:4:5
  |
4 |     async fn foo() {
  |     ^^^^^^^^^^^^^^
5 |         Self::Dependency::foo().await;
  |         ----------------------------- recursive call here
  |
  = note: a recursive `async fn` call must introduce indirection such as `Box::pin` to avoid an infinitely sized future

For more information about this error, try `rustc --explain E0733`.
error: could not compile `test-play` (bin "test-play") due to 1 previous error

Possible Solution(s)

Usage of associated type is an important part of repro. So the reason is somehow related to this

Notes

No response

Version

cargo 1.78.0 (54d8815d0 2024-03-26)
release: 1.78.0
commit-hash: 54d8815d04fa3816edc207bbc4dd36bf18014dbc
commit-date: 2024-03-26
host: aarch64-apple-darwin
libgit2: 1.7.2 (sys:0.18.2 vendored)
libcurl: 8.1.2 (sys:0.4.72+curl-8.6.0 system ssl:(SecureTransport) LibreSSL/3.3.6)
ssl: OpenSSL 1.1.1w  11 Sep 2023
os: Mac OS 14.1.0 [64-bit]

[epage]

I expected them to be consistent, (check should report all build failures as far as I undestand)

This is not actually true. There are post-monomorphization errors that do not get reported by check.

[bragov4ik]

I expected them to be consistent, (check should report all build failures as far as I undestand)

This is not actually true. There are post-monomorphization errors that do not get reported by check.

So this is a feature rather than bug?
Did not expect this, thanks for explaining!

Should I close the issue then?..

[epage]

You aren't the only one; I had the same perception until rust-lang/rfcs#3477 was posted.

The challenge is knowing whether this is a "bug" from that or something else. I think its likely safe to assume it is so I'll close this.