On Thu, Jul 20, 2023 at 8:40 AM David Gow davidgow@google.com wrote:
The only difference with user-space tests is that instead of using `#[cfg(test)]`, `#[kunit_tests(kunit_test_suit_name)]` is used.
I may be missing something, but this does not appear to map the `assert*!`s to the KUnit APIs, is that correct? (i.e. like we do for `rustdoc`-tests).
I made an assertion fail, and it seems to use the standard library macros, thus panicking and ending up in `BUG()` (rather than a failed test):
rust_kernel: panicked at 'assertion failed: `(left == right)` left: `true`, right: `false`', rust/kernel/kunit.rs:329:1 ------------[ cut here ]------------ kernel BUG at rust/helpers.c:34!
Then the test times out eventually and things break:
# rust_test_kunit_kunit_tests: try timed out ------------[ cut here ]------------ refcount_t: addition on 0; use-after-free.
- // Add `#[cfg(CONFIG_KUNIT)]` before the module declaration.
- let config_kunit = "#[cfg(CONFIG_KUNIT)]".to_owned().parse().unwrap();
- tokens.insert(
0,TokenTree::Group(Group::new(Delimiter::None, config_kunit)),- );
I wonder about compile-time here with this approach. As far as I understand, having the `cfg` explicitly outside the proc macro would avoid invoking it.
Do we know the potential compile-time impact, especially when we will have many tests? ventually it would be ideal to have an approach closer to the `rustdoc` one, where the compiler finds the tests for us and we generate the needed code in the build system, i.e. outside a proc macro.
Cheers, Miguel