[CALCITE-3411] Incorrect code generated for BigDecimal ConstantExpression

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.21.0
• Fix Version/s: 1.22.0
• Labels: pull-request-available
• Assignee: Feng Zhu
• Reporter: Feng Zhu
• Created: 14/Oct/19 14:23
• Updated: 06/Mar/20 04:36
• Resolved: 15/Oct/19 04:26

Description

In current implementation

Expressions.constant("104", BigDecimal.class) -> new java.math.BigDecimal(104L)
Expressions.constant("1000", BigDecimal.class) -> new java.math.BigDecimal(1L, -3)

However, in java.math.BigDecimal, there is no constructor:

java.math.BigDecimal(long exact, int scale) 

Instead we should call method

 java.math.BigDecimal.valueOf(long exact, int scale)

Why this issue did not exposed before?

BlockBuilder conducts optimizations that can bury this issue.
For example, when we disable the optimization. MaterializationTest#testAggregateMaterializationAggregateFuncs11 will produce the code below. However, it can't pass the compilation.

final long v = org.apache.calcite.runtime.SqlFunctions.floor(853763696000L, ((Long) new java.math.BigDecimal(1L, -3)).longValue());

Issue Links

• links to: GitHub Pull Request #1506