BigDecimal invalid support

[dybuk87]

Hi,
I think BigDecimal support is misleading in current version, currently json :
new JSONObject(" { "value" : 34300000000000000000000000.14 }").getBigDecimal("value").toPlainString()
will return 34300000000000000000000000 instead of correct value 34300000000000000000000000.14

I found in code

// if we want full Big Number support this block can be replaced with:
// return stringToNumber(string);
if (isDecimalNotation(string)) {

After uncommenting this line support is valid and I believe this should be default instead of current one because there is a lot of situation where you need full precision like money amount, or at least there should be some configuration for json parser.

[johnjaylward]

I'll take a look at this

[johnjaylward]

@stleary , while this is "working as designed", I do agree with @dybuk87 that we should consider changing the default handling of numbers. What's happening in this example is that the stringToValue parser is always using a Double for the representation. That is causing the value to be truncated here.

Let me know your thoughts.

[stleary]

@johnjaylward Haven't looked at the code yet, but this sounds reasonable given that BigNumber is a recent enhancement.

[urfuchs]

There is a difference between getBigDecimal and optBigDecimal.

JSONObject json = new JSONObject("{ \"key\" : 72.35 }");
BigDecimal bd1 = json.getBigDecimal(„key“);
BigDecimal bd2 = json.optBigDecimal(„key“, null);
System.out.println(bd1);
System.out.println(bd2);

prints:

72.35
72.349999999999994315658113919198513031005859375

Why is there a difference?

getBigDecimal converts the double to a String and than to a BigDecimal:
new BigDecimal(Double.valueOf(72.35).toString())
optBigDecimal converts the double directly to BigDecimal: new BigDecimal(Double.valueOf(72.35))

[johnjaylward]

@urfuchs can you open a new issue for this? it's a different problem than the original request

[urfuchs]

There is a difference between getBigDecimal and optBigDecimal.

JSONObject json = new JSONObject("{ \"key\" : 72.35 }");
BigDecimal bd1 = json.getBigDecimal(„key“);
BigDecimal bd2 = json.optBigDecimal(„key“, null);
System.out.println(bd1);
System.out.println(bd2);

prints:

72.35
72.349999999999994315658113919198513031005859375

Why is there a difference?

getBigDecimal converts the double to a String and than to a BigDecimal:
new BigDecimal(Double.valueOf(72.35).toString())
optBigDecimal converts the double directly to BigDecimal: new BigDecimal(Double.valueOf(72.35))

moved to #438

[stleary]

Working as designed.

[borgogelli]

Same code, but the output for me is:

72.349999999999994315658113919198513031005859375
72.349999999999994315658113919198513031005859375

I use the 20200518 version

[johnjaylward]

@borgogelli , I'm not sure I understand what action you are expecting from your comment. Are you confirming the fix worked, or do you expect the behaviour to be different?

[borgogelli]

Hi @johnjaylward , both the things. First the output of my code is different from the code posted by @urfuchs . And then I have expected the result to be 72.35

[johnjaylward]

The recommendation when using the *BigDecimal functions is to specify your scale. We purposefully expose the full double with rounding errors due to use cases like this:

double myDouble = someComplexCalculation();
JSONObject jo = new JSONObject();
jo.put("myDouble", myDouble);
// ... some other code
someFunction(jo.getBigDecimal("myDouble"));

The old method of using the string value of the underlying double did preserve the value when parsed from a string, but it lost some complexities if the double was originally stored from a calculation.

PR #453 basically solves the issue for both use cases by preserving the textual representation when parsing (by using a BigDecimal for storage), but allowing complex calculations to be stored as a double or float.

Until #453 is accepted, the expected use would be to do something like (use setScale):

double myDouble = someComplexCalculation();
JSONObject jo = new JSONObject();
jo.put("myDouble", myDouble);
// ... some other code
someFunction(jo.getBigDecimal("myDouble").setScale(5, RoundingMode.ROUND_HALF_EVEN));