The specification leaves the implementor guessing as to the semantics of numeric expression, and the type assigned to numeric literals.

§4.6.17.1 lists the arithmetic operators and says:

Arithmetic operations use numeric promotion.

§4.6.1 says:

Exact numeric literals support the use of Java integer literal syntax as well as SQL exact numeric literal syntax.
Approximate literals support the use Java floating point literal syntax as well as SQL approximate numeric literal syntax.

But this leaves a lot unspecified. For example, what type does the following query return:

select 1.23 from Entity

Double? Float? BigDecimal?

Worse, what value does this query return:

select 1/2 from Entity

That is, does "use numeric promotion" mean that this is an integer division or is it promoted to float? Or to double? (Note that different SQL databases disagree on this question.)

I propose that we nail this stuff down by simply saying that:

• the semantics of numeric literals and arithmetic expressions are the same as in Java (so that 1/2 is integer division),
• with the additional rule that (a) integral types may be widened to BigInteger or BigDecimal, (b) BigInteger may be widened to BigDecimal, and (c) floating point types may be widened to BigDecimal.

Actually, I'm not completely clear on whether FP types should be widened to BigDecimal, but I'm thinking they should be, simply because that's something every SQL database allows.