Is it possible to resume after a panic?

[Michael-F-Bryan]

I know the wasmer runtime will detect when a host function (i.e. something passed in using imports!()) panics and translate that to a RuntimeError (presumably around here). Is it possible for the top-level caller to use std::panic::resume_unwind() to continue panicking?

The data in RuntimeError::Error is Box<Any>, whereas std::panic::resume_unwind() expects Box<dyn Any + Send>.

I guess one possible solution is to match on the result of call() and then explicitly panic!() if a runtime error was detected. That would let us continue crashing, but would also mean we lose the original backtrace.

This may be related to #986 or #827.

[nlewycky]

I think as a library we don't want to take down the program that's using us by default, though we should make it easy to turn that RuntimeError back into a panic! with the stack information if, as the host program, that's what you want to do.

Maybe it's as simple as making RuntimeError::Error a Box<Any + Send>?

[Michael-F-Bryan]

I think it would be. Or maybe add an extra HostFunctionPanicked(Box<dyn Any + Send>) variant to RuntimeError? That way you won't accidentally try to resume a Box<dyn Any + Send> which isn't actually a panic, and can differentiate between whether a function panicked or just returned an error.