Evgeny Stanilovsky created IGNITE-18401:
-------------------------------------------

             Summary: Sql. Different behavior of float cost approach leads to 
failed test. 
                 Key: IGNITE-18401
                 URL: https://issues.apache.org/jira/browse/IGNITE-18401
             Project: Ignite
          Issue Type: Bug
          Components: sql
    Affects Versions: 3.0.0-beta1
            Reporter: Evgeny Stanilovsky


 

Test like:
{code:java}
query IIIII
SELECT 127::DECIMAL(3,0)::FLOAT, 
-17014118346046923173168730371588410572::DECIMAL(38,0)::FLOAT, 
-7::DECIMAL(9,1)::FLOAT, 27::DECIMAL(18,1)::FLOAT, 33::DECIMAL(38,1)::FLOAT
----
127  -1.7014119E37  -7 27 33 {code}
failed in ai-3 and passed in ai-2 cause in ai-2 in case of one fragment it 
always plans into local node _without_ any serialization\deserialization usage 
against ai-3 approach: through serialization, thus we obtain in ai-2 approach 
only 
ExpressionFactoryImpl#compile -{-}> ConstantExpression#write (case FLOAT) 
branch, while  in ai{-}3: float was trimmed a bit in 
org.apache.calcite.rex.RexBuilder#clean and after goes into 
ExpressionFactoryImpl and so on .. 

Need further investigation.

 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to