[jira] [Created] (IMPALA-7753) Invalid logic when rewriting ORDER BY clause expressions

["Paul Rogers (JIRA)" <ji...@apache.org>]

Paul Rogers created IMPALA-7753:
-----------------------------------

             Summary: Invalid logic when rewriting ORDER BY clause expressions
                 Key: IMPALA-7753
                 URL: https://issues.apache.org/jira/browse/IMPALA-7753
             Project: IMPALA
          Issue Type: Bug
          Components: Frontend
    Affects Versions: Impala 3.0
            Reporter: Paul Rogers
            Assignee: Paul Rogers


The select statement represents the ORDER BY clause in two distinct ways. First, there is a list of the "original" ordering expressions, {{orderByElements_}}. Second, there is an analyzed list in {{sortInfo_}}. The explanation is:

{code:java}
      // create copies, we don't want to modify the original parse node, in case
      // we need to print it
{code}

Later, we apply rewrite rules to the {{ORDER BY}} expression, but we do so using the original version, not the copy:

{code:java}
      for (OrderByElement orderByElem: orderByElements_) {
        orderByElem.setExpr(rewriteCheckOrdinalResult(rewriter, orderByElem.getExpr()));
      }
{code}

The result is that we apply rewrite rules to expressions which have not been analyzed, triggering the assertion mentioned above. This assertion is telling us something: we skipped a step. Here, it turns out we are rewriting the wrong set of expressions. Modifying the code to rewrite those in {{sortInfo_}} solves the problem. The current behavior is a bug as the rewrites currently do nothing, and the expressions we thought we were rewriting are never touched.

The correct code would rewrite the expressions which are actually used when analyzing the query:

{code}    if (orderByElements_ != null) {
      List<Expr> sortExprs = sortInfo_.getSortExprs();
      for (int i = 0; i < sortExprs.size(); i++) {
        sortExprs.set(i, rewriteCheckOrdinalResult(rewriter, sortExprs.get(i)));
      }
{code}

We can, in addition, ask a more basic question: do we even need to do rewrites for {{ORDER BY}} expressions? The only valid expressions are column references, aren't they? Or, does Impala allow expressions in the {{ORDER BY}} clause?



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org