I know that this has been discussed before in #225, but after the policy change in RFC-3477 (which will probably be approved shortly), cargo check effectively doesn't guarantee the amount of warnings that it produces.

So maybe it now makes sense to reconsider using cargo build instead?

I recognize that unfortunately this will make trybuild slower, but then again, the project is called trybuild; I would argue it's the desired behaviour to completely try to build the files, including the steps noted in the above RFC (linking, const errors, and so on).