This kind of thing is why I'd favour two different types of numeric types:

- "Number", which would be the generic DWIM arithmetic type. In your example both x and the literals would be Number.

- Hyper-specified arithmetic types. Not just number of bits and signedness, but also overflow and trap behaviour. Less convenient to use - you can't easily add a "32bit signed wrap on overflow" to "16 bit unsigned saturate on overflow", but have the advantage that there is no unspecified behaviour. This also allows the user to make use of machine-provided saturate and overflow behaviour where present.