Rethink should_panic and fail_uncoverable options as global conditions

[adpaco-aws]

This PR is the next step to rework/introduce the should_panic/fail_uncoverable options as global conditions.

Until now, we haven't had a concrete proposal to do so other than the implementation in #2532. This PR presents one for each option in their respective RFCs. I'd like to agree on this design before starting the code review for #2532.

Related to #1905 #2272 #2299 #2516

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 and MIT licenses.

[zhassan-aws]

Looks good. How will this interact with the cover_or_fail macro proposed in #2792?

[adpaco-aws]

Looks good. How will this interact with the cover_or_fail macro proposed in #2792?

Thank you!

Good question. First of all, the interaction would depend on how the macro is implemented. I'm not aware of any specific proposal to do that, but I'm happy to discuss it considering the two options I've thought about. So please take what follows with a grain of salt because I may not be considering all options/cases.

The first option is to implement cover_or_fail through a special property class (e.g., cover_fail) which can hold a Failure status. In this case, I don't think there would be any conflicts because should_panic only cares about assertion properties and fail_uncoverable does the same for cover properties.

Another option is to have cover_or_fail enable a different global condition which checks that all cover_or_fail properties are indeed covered. At present, failing any global condition implies a verification failure. Again, I don't think there would be conflicts (because of the properties that each condition cares about) but this really depends on the implementation.

Also, I wanted to point out that, assuming fail_uncoverable worked in the way it's being described, I believe one could use it to achieve the desired outcome in #2792 (i.e., the second test would fail). That's because fail_uncoverable would essentially transform all reachable cover macros into cover_or_fail ones. But it's not enough for the general case since other cover statements may be present.

[zhassan-aws]

Sounds good.