`#[should_panic]` macro support

[epage]

See https://doc.rust-lang.org/reference/attributes/testing.html#the-should_panic-attribute

[skogseth]

I've begun working on this now. So far I've ended up using recursive parsing in the macro, and then generating the expanded code (with panic capture using catch_unwind) in Case::run itself. Any thoughts on how this should be implemented?

[epage]

Drat, you beat me to touch the macros :). I was planning on experimenting with ways to generalize the macros to make it easier to add new features. Was hoping to have it done before they needed to be touched to avoid conflicts. I did just get some test clean up in.

In catching the panic, I'd recommend checking out how should_panic is implemented in case there is any special nuance, see https://github.com/rust-lang/rust/tree/master/library/test/src

I wonder if we could expose an assert_panic macro and the attribute then wraps the call with that macro.

Looking at the reference:

Only the first instance of should_panic on a function is honored.

This assumes there is one specific panic that can happen which is fine in most cases. I do know that some errors are platform specific, like file not found (https://github.com/rust-lang/cargo/blob/c3fc265123e39c768e7a9de48544cd184232f82d/crates/cargo-test-support/src/compare.rs#L267-L277) so I could potentially see supporting multiple.

The return type of the test function must be ().

I'm assuming this is to avoid ambiguity with whether errors are covered by this. It could be nice to also cover errors but then that is weird with the name.

[skogseth]

Hehe, well it's up to you which goes first :) Perfectly understandable if you want to try to generalize the macros before adding any new ones.

[epage]

Go for it. You'll likely get it done before I do

[skogseth]

I've added the following tests to the testsuite in libtest2 now:

#[libtest2::test]
#[should_panic]
fn (_context: &libtest2::TestContext) {
    panic!("any message would do")
}

#[libtest2::test]
#[should_panic = "intentional"]
fn intentionally_panics_with_message(_context: &libtest2::TestContext) {
    panic!("this is intentional")
}

#[libtest2::test]
#[should_panic = "in a controlled manner"]
fn panics_with_the_wrong_message(_context: &libtest2::TestContext) {
    panic!("with the wrong message")
}

#[libtest2::test]
#[should_panic]
#[should_panic = "very long and specific message"]
fn passes_because_subsequent_panics_are_ignored(_context: &libtest2::TestContext) {
    panic!("any message should be valid")
}

I am, however, not sure how it makes sense to add equivalent tests to libtest2-mimic. There isn't really any functionality there to support panics, as far as I am aware.

The current design I've implemented for the #[should_panic] macro exposes an assert_panic macro (as you suggested), but this isn't exposed to libtest2-mimic. And even if it was, the final test (passes_because_subsequent_panics_are_ignored) makes no sense without the attribute macros.

[epage]

We've been abusing that test file. We should probably make a separate one for this and not worry about syncing them.

[skogseth]

Can I add a new file called macros.rs (which does this kind of macro testing) to the testsuite directory then? Or do you want to do something different?

[epage]

I'd create a should_panic.rs