[CALCITE-7409] MERGE JOIN condition cannot contain IS NOT DISTINCT FROM

[xiedeyantu]

See CALCITE-7409

[mihaibudiu]

there is another comment about this right below; are both comments necessary?

[xiedeyantu]

I'd like to keep these two: one explaining the reason, and the other describing a reasonable approach. However, supporting "IS NOT DISTINCT FROM" would require modifying Linq4j, which doesn't seem straightforward. So, I'll add an extra TODO for now. WDYT?

[mihaibudiu]

You can check for a JIRA issue about it, and if there is one add the link.
I think the two comments could be combined into one.

[xiedeyantu]

Is it acceptable to change it to the following format?

    // TODO: support IS NOT DISTINCT FROM condition as join keys of MergeJoin.
    //  MergeJoin cannot handle IS NOT DISTINCT FROM because it stops at NULL values
    //  while IS NOT DISTINCT FROM treats NULL = NULL as true.

I couldn't find anything similar on Jira, or maybe I'm just not very good at using Jira.

[xiedeyantu]

This SQL query actually comes from CALCITE-6452. Jira should currently be working correctly, but due to a cost model issue, it's selecting MergeJoin. From the DAG, it's clear that MergeJoin has about 7 fewer rows than HashJoin, but consumes 6 times more CPU. The current cost model ignores CPU usage, hence selecting MergeJoin. Although the cost model is flawed, this error should still be fixed.

rel#207 (EnumerableHashJoin):
  rows=14.209999999999999
  cost={222.6832026146136 rows, 204.2 cpu, 0.0 io}

rel#213 (EnumerableMergeJoin):
  rows=14.209999999999999
  cost={215.13639999999998 rows, 1272.3104732091813 cpu, 0.0 io}

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud