Incorrect deserialization for BigDecimal numbers

[cristian-mocanu-mob]

If the number in JSON cannot be represented as Double and DeserializationFeature.USE_BIG_DECIMAL_FOR_FLOATS is enabled, big numbers are returned as Infinity.

The test below should pass:

        final ObjectMapper objectMapper = new ObjectMapper();
        objectMapper.enable(DeserializationFeature.USE_BIG_DECIMAL_FOR_FLOATS);

        final JsonNode jsonNode = objectMapper.readTree("7976931348623157e309");
        Assert.assertTrue(jsonNode instanceof NumericNode);

        // the following fails with NumberFormatException, because jsonNode is a DoubleNode with a value of POSITIVE_INFINITY
        Assert.assertTrue(jsonNode.decimalValue().compareTo(new BigDecimal("7976931348623157e309")) == 0);

The problem is reproducible in the latest released version, 2.9.1.

[cowtowncoder]

Thank you for reporting this, does sound like a bug.

[cowtowncoder]

Hmmh. This may be tricky to handle... problem being that detection of NaN case is triggered, and since value itself does round up to positive infinity (when attempted with double), value is not coerced as BigDecimal as it has no NaN options.

Will add failing test, see if I can figure out a way to handle this as expected.

[cowtowncoder]

@GedMarc maybe this comment is for different issue... ?

[GedMarc]

Oh Geez! This is for the double deserialization as float into big decimal!
This week has taken it out of me sigh, apologies!

[cowtowncoder]

For context, looks like fix for #1028 possibly caused this. So code that should coerce value results in parser trying to determine if it got a NaN and... in a way it is, wrt range of Double. But it would not be for BigDecimal.

This is tricky to solve, JsonParser (actual JSON implementation) will need to be careful not to effectively coerce "too big"/"too small" numbers into positive/negative infinity, but only report true for isNaN() when actual marker is used. Or I need to figure out some other logic...
Seems likely, however, that this requires help from streaming level and can not be solved purely by databind.

[cowtowncoder]

@pjfanning This issue is something where I was hoping lazy parsing might work. I do not remember all the details on exact failure but seems like this could maybe be revisited to see it was now solvable in 2.16.