xiaoh1024 opened a new pull request, #58940:
URL: https://github.com/apache/spark/pull/58940

   ### What changes were proposed in this pull request?
   
   Extend `BooleanSimplification` to remove redundant integral range 
comparisons:
   
   ```sql
   a > 5 AND a > 0             --> a > 5
   a > 1 OR a > 2              --> a > 1
   a > 1 OR (a > 2 AND a < 4)  --> a > 1
   ```
   
   Support `<`, `<=`, `>`, and `>=` between direct Byte/Short/Int/Long 
attributes and non-null literals of the same type, including reversed operands 
and one level of nested absorption. Bounds are compared without arithmetic to 
avoid overflow. Casts and arbitrary expressions are excluded.
   
   ### Why are the changes needed?
   
   The existing rule retains these redundant comparisons. Removing them reduces 
per-row evaluation. This implements the integral range subset of 
[SPARK-31760](https://issues.apache.org/jira/browse/SPARK-31760).
   
   ### Does this PR introduce _any_ user-facing change?
   
   Yes: eligible queries have simpler optimized plans and may execute faster. 
Query results, including SQL NULL semantics, are preserved. No public API or 
configuration changes.
   
   ### How was this patch tested?
   
   Added Catalyst and SQL tests covering operand order, inclusive/exclusive 
bounds, nested absorption, NULLs, integer extremes, unsupported expressions, 
and generated evaluation. The new contained-range regression test fails with 
the original rule.
   
   On Spark 5.0.0-SNAPSHOT with Scala 2.13.18 and JDK 17.0.18:
   
   - 37 Catalyst tests passed across the new suite and the existing 
BooleanSimplification and BinaryComparisonSimplification suites.
   - The SQL suite passed all 16 filter/projection and AQE on/off combinations.
   - Catalyst source/test and SQL test scalastyle checks passed.
   
   Local synthetic-data benchmarks used 30 million cached rows and `local[2]`, 
covering filter/projection queries with 0% and 10% nulls. AND placed the wider 
condition first, OR placed the narrower condition first, and nested absorption 
placed the complex branch first. Four JVMs ran in 
baseline/candidate/candidate/baseline order. Median query time decreased by 
3–5% for AND, 30–34% for OR, and 35–43% for nested absorption. A separate rule 
microbenchmark measured up to 3.2 microseconds additional time per 
32-expression projection (about 8%).
   
   ### Was this patch authored or co-authored using generative AI tooling?
   
   Yes.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to