items_after_test_module interacts badly with rstest and submodules

[asomers]

Summary

I like to organize my tests into multiple modules just to provide namespacing. For example, I might use one module for each function under test, and each of those modules will contain one function for each test case, like this:

mod func1 {
    use super::*;
    #[test]
    fn test_case1() {...}
    #[test]
    fn test_case2() {...}
}
mod func2 {
    use super::*;
    #[test]
    fn test_case1() {...}
    #[test]
    fn test_case2() {...}
}

The problem is that if anything above the final module includes a #[cfg(test)], then the items_after_test_module lint will fire. That's even true in integration tests, which aren't built with and without #[cfg(test)]. The problem is exacerbated by macro crates like rstest, which sometimes emit a #[cfg(test)] in the generated code. And it's frustrating to debug, because the user can't even see the offending #[cfg(test)].
I don't know what a perfect solution would be. But skipping this lint in integration tests would be a good start. Or, skipping this lint within an outer module which is itself #[cfg(test)].

Lint Name

items_after_test_module

Reproducer

I tried this code:

use rstest::rstest;

#[rstest]
#[case(42)]
fn t(#[case] _x: u32) {}

mod func1 {
    struct SomethingUsedByTheTest{}

    #[test]
    fn t() {}
}

I saw this happen:

warning: items were found after the testing module
  --> tests/t.rs:3:1
   |
3  |   #[rstest]
   |   ^--------
   |   |
   |  _in this procedural macro expansion
   | |
4  | | #[case(42)]
5  | | fn t(#[case] _x: u32) {}
6  | |
...  |
10 | |     #[test]
11 | |     fn t() {}
   | |_____________^
   |
   = help: move the items to before the testing module was defined
   = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#items_after_test_module
   = note: `#[warn(clippy::items_after_test_module)]` on by default
   = note: this warning originates in the attribute macro `rstest` (in Nightly builds, run with -Z macro-backtrace for more info)

I expected to see this happen:

No errors (except for SomethingUsedByTheTest is never constructed, of course).

Version

rustc 1.72.0-nightly (5bd28f5ea 2023-06-28)
binary: rustc
commit-hash: 5bd28f5eac1ba3569bfa8d49ec3f5acbdfdff7a0
commit-date: 2023-06-28
host: x86_64-unknown-freebsd
release: 1.72.0-nightly
LLVM version: 16.0.5

Additional Labels

No response