Rounding necessary with BigDecimal numbers

Question

I want to set scale of two BigDecimal numbers a and b . as in this example :

BigDecimal a = new BigDecimal("2.6E-1095");
        BigDecimal b = new BigDecimal("2.7E-1105");
        int i = 112, j=1;

        BigDecimal aa = a.setScale(i+j);
        BigDecimal bb = b.setScale(i+j);

and when i run i have this exception:

java.lang.ArithmeticException: Rounding necessary
    at java.math.BigDecimal.divideAndRound(BigDecimal.java:1439)
    at java.math.BigDecimal.setScale(BigDecimal.java:2394)
    at java.math.BigDecimal.setScale(BigDecimal.java:2437)

Why rounding is necessary ? if i don't want to make around, what is solution please ?

Thanks

Indeed it is a surprise. Without looking at docs I expected setting the scale to truncate at the scale... — Brian Knoblauch, Apr 26 '16 at 19:24
I expectedthe same thing. This exception even comes when specifying the RoundingMode.ROUNDING_UNNECESSARY. I didn't see a truncate option, but wish there was. — wheeleruniverse, Jul 11 '18 at 21:05

score 45 · Accepted Answer · edited Nov 13 '17 at 21:00

45

You have two BigDecimal numbers both of which require over a 1000 decimal places. Trying to set the scale to only have 113 decimal places means you will lose precision and therefore you need to round.

You can use the setScale methods that take a RoundingMode to prevent the exception but not the rounding.

edited Nov 13 '17 at 21:00

Jus12

17,824
28
99
157

answered May 24 '12 at 09:07

brain

5,496
1
26
29

score 31 · Answer 2 · edited Feb 21 '23 at 20:24

31

Try to use rounding mode of setScale method.

Some thing like:

BigDecimal aa = a.setScale(i+j, RoundingMode.HALF_DOWN);

edited Feb 21 '23 at 20:24

Michael Piefel

18,660
9
81
112

answered Jun 15 '16 at 18:20

Akostha

679
7
8

4

I think the enum should be RoundingMode.HALF_DOWN – abbas Jun 27 '19 at 10:02

Haozhun · Answer 3 · 2016-09-09T20:02:07.857

9

Rounding is necessary.

In javadoc for BigDecimal, it says BigDecimal is represented as (unscaledValue × 10^-scale), where unscaledValue is an arbitatrily long integer and scale is a 32-bit integer.

2.6*10^-1095 requires a scale of at least 1096 to represent accurately. It can not be represented accurately with (any integer)*10^-113. Therefore, you need to providing a roundingMode.

edited Sep 09 '16 at 20:02

answered May 24 '12 at 09:11

Haozhun

6,331
3
29
50

Rounding necessary with BigDecimal numbers

3 Answers3

Linked