I would guess this is due to the updated libffi, and probably it's a case that is not sufficiently tested.

Adding tests is the first step. It'll probably have to enumerate many parameter types (in case there's something weird here like misaligning the double after the int and not clearing it up, as opposed to a constant 4-byte offset).

Hopefully then the issue is on our side and not part of libffi, but it could be anywhere.