note: I guess this is a matter of taste, I personally liked the "try moto then error_type is X except Exception then error is Y", but both are equivalent so it does not matter 😄

I switched it to avoid a fail-fast situation which could happen if an exception is raised in the first assignment, leading to the second member attribute not being defined at all.

note: here it is seems that we do not really care what type the exception really is (I mean betweenServiceException andRESTError).

It might be easier to declare a tuple of type once? So that it would be easier to extend if we were to add yet another error type? That way, we also don't need theEllipsis type here. But this is just personal preference, really no need to act on this note.

Something like:

try:importmoto.core.exceptionsself._service_error_types= (ServiceException,moto.core.exceptions.ServiceException,moto.core.exceptions.RESTError,            )except (ModuleNotFoundError,AttributeError):# Moto may not be available in stripped-down versions of LocalStack, like LocalStack S3 image.self._service_error_types= (ServiceException, )...ifisinstance(exception,self._service_error_types):return

What do you think?

That's a good idea. Implemented in6236f11

nit: if the internal implementation of theInvalidKeyPairNameError is switched to be ofServiceException, then we won't test theRESTError behavior anymore.

Should we manually define a subclass ofRESTError in the unit test to be sure we're handling that case properly?

That's a very good point, if Moto were to migrate EC2 exceptions toServiceException, this would silently invalidate the test.

I redefined the exceptions within the tests ine326aaf, what do you think?