Saxon » SaxonJS

I'm testing various changes in handling of Atomic.js, and hit a test failure in K-NumericIntegerDivide-1. The test does

(xs:decimal(6) idiv xs:integer(2)) instance of xs:integer

which is supposed to return true, but it's coming out as false.

In the course of this I found code which I haven't changed and which looks wrong.

Calculate.js line 207 has

"c~c": function (lhs, rhs) {
            // TODO explicit divByZero check? FOAR0002?
            try {
                const a = lhs.toBig(), b = rhs.toBig();
                const x = a.div(b).round(0, 0);
                return isSafeInteger(x) ? INT.fromNumber(x) : DEC.fromBig(x);
            }
            catch (e) {
                divByZeroErr();
            }
        },

Here a and b are Big objects, so x is also a Big, but isSafeInteger is designed to operate on a JS number.

I suspect isSafeInteger() applied to a Big is always returning false (because Math.abs() returns NaN), so we are always following the DEC.fromBig() path. With the current Atomic.js, DEC.fromBig() always returns an xs:decimal, causing this test to fail. Perhaps in the released code, there was some compensating code that caused DEC.fromBig() to return an xs:integer in this case?