[jira] [Created] (IMPALA-7951) Rewrites change meaning when converted to SQL

[Paul Rogers (JIRA)]

Paul Rogers created IMPALA-7951:
-----------------------------------

             Summary: Rewrites change meaning when converted to SQL
                 Key: IMPALA-7951
                 URL: https://issues.apache.org/jira/browse/IMPALA-7951
             Project: IMPALA
          Issue Type: Bug
          Components: Frontend
    Affects Versions: Impala 3.1.0
            Reporter: Paul Rogers
            Assignee: Paul Rogers


Consider this query:

{code:sql}
SELECT int_col not between 1 and 10 and int_col <> -1
FROM functional.alltypestiny
{code}

Then, render the query using toSql() after rewrites

{code:sql}
SELECT int_col < 1 OR int_col > 10 AND int_col != -1
FROM functional.alltypestiny

Notice the meaning has changed. Correct grouping:

{code:sql}
(int_col < 1 OR int_col > 10) AND int_col != -1
{code}

This means to accept all values {{(-infinity, -2), 0 (1, 10)}}.

Actual grouping based on precedence rules:

{code:sql}
int_col < 1 OR (int_col > 10 AND int_col != -1)
{code}

This means to accept all values {{(-infinity, 0)}}.

The rewrites must insert parens if the rewrite would render a tree that is 
incorrect when interpreted using default precedence rules. (That is, if a 
parent node has lower precedence than a child node.)

The quick fix is to set the “print sql in parens” flag, but this is less than 
idea as it adds parens where they are not required. Better would be to compute 
the parens based on precedence at toSql time, but that is out of scope of this 
fix.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)