kosiew opened a new issue, #19673: URL: https://github.com/apache/datafusion/issues/19673
In [`row_filter.rs`](https://github.com/apache/datafusion/blob/main/datafusion/datasource-parquet/src/row_filter.rs), the `required_columns` field in both `PushdownChecker` and `PushdownColumns` structs uses `BTreeSet<usize>` to track column indices required for filter evaluation. **Current implementation:** ```rust struct PushdownChecker<'schema> { required_columns: BTreeSet<usize>, // ... other fields } struct PushdownColumns { required_columns: BTreeSet<usize>, nested: NestedColumnSupport, } ``` ### Motivation For typical filter predicates, the number of columns referenced is usually small (1-5 columns). Using `BTreeSet<usize>` adds overhead: - Memory overhead from tree structure - Insertion cost is O(log n) vs O(1) for append to `Vec` - The only operation that benefits from `BTreeSet` is deduplication, but for small sets, a simple `Vec` with linear scan would be faster - The data is immediately converted to `Vec` after collection anyway (line 244) ### Proposed Solution Replace `BTreeSet<usize>` with `Vec<usize>` and handle deduplication explicitly if needed source: https://github.com/apache/datafusion/pull/19545#discussion_r2665094677 -- 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]
