[HIVE-19523] Decimal truncation for trailing zeros in Hive 1.2.1

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 1.2.1, 2.3.1
• Component/s: Hive
• Assignee: Unassigned
• Reporter: vinay kant garg
• Created: 14/May/18 07:57
• Updated: 25/May/18 04:32

Description

This issue is related to the unnecessary truncation of zero's while serializing BigDecimal object to HiveDecimal object.

BigDecimal object has info about scale and precision still serializing ended up modifying scale  and precision metadata as well truncation of zero's.

Eg: if our BigDecimal val is : 47.6300, scale = 4 and precision = 6. If I am serializing this to HiveDecimal by using API hive expose :

static HiveDecimal create(BigDecimal b)

static HiveDecimal create(BigDecimal b, boolean allowRounding)

our output will be : val = 47.63, scale = 2 and precision = 4.

or if IP: val = 47.00, scale = 2 and precision = 4 then

       OP: val = 47, scale = 0 and precision = 2

In above example if we see there is no DATA CORRUPTION because 47.6300 is equivalent to 47.63 or 47.00 is equivalent to 47 but later on if this data is used to identify whether it is integer or decimal then we may run into weird issues because 47.00 is decimal but 47 is an integer.

I am able to reproduce this issue even with standalone program:

import java.math.BigDecimal;
import org.apache.hadoop.hive.common.type.HiveDecimal;

public class testHivedecimal {

public static void main(String[] argv){
BigDecimal bd = new BigDecimal ("47.6300");
bd.setScale(4);
System.out.println("bigdecimal object created and value is : " + bd.toString());
System.out.println("precision : "+ bd.precision());
System.out.println("scale : "+ bd.scale());
HiveDecimal hv = HiveDecimal.create(bd);
String str = hv.toString();
System.out.println("value after serialization of bigdecimal to hivedecimal : " + str);
System.out.println("precision after hivedecimal : "+ hv.precision());
System.out.println("scale after hivedecimal : "+ hv.scale());
}
}

you can use above program to reproduce issue:

To Compile:  

1) use jdk8_u72

2) command : javac -cp hive-common-1.2.1000.2.6.0.3-8.jar:. testHivedecimal.java 

To Run:

1) use jdk8_u72

2) command : java -cp hive-common-1.2.1000.2.6.0.3-8.jar:. testHivedecimal

 

Issue Links

• relates to: HIVE-12772 Beeline/JDBC output of decimal values is not 0-padded, does not match with CLI output
• relates to: HIVE-12063 Pad Decimal numbers with trailing zeros to the scale of the column