bharos opened a new issue, #15882: URL: https://github.com/apache/iceberg/issues/15882
### Feature Request / Improvement StrictMetricsEvaluator.notStartsWith currently always returns ROWS_MIGHT_NOT_MATCH, ignoring file-level column bounds entirely. There is an existing [TODO in the code](https://github.com/apache/iceberg/blob/9a939d68358de9dac2c6ba9b236b675ebe477490/api/src/main/java/org/apache/iceberg/expressions/StrictMetricsEvaluator.java#L470): ``` // TODO: Handle cases that definitely cannot match, such as notStartsWith("x") when the bounds // are ["a", "b"]. ``` When the column's lower and upper bounds prove that no value in the file can start with the given prefix, the evaluator should return ROWS_MUST_MATCH. This allows the engine to skip unnecessary row-level filtering for NOT STARTS WITH predicates, improving query performance on string-heavy workloads. Specifically, the evaluator can determine that all rows match NOT STARTS WITH <prefix> when: - The column contains only null values - The lower bound (truncated to the shorter of prefix/bound length) is strictly greater than the prefix — all values are above the prefix range - The upper bound (truncated to the shorter of prefix/bound length) is strictly less than the prefix — all values are below the prefix range This mirrors how notEq and notIn already use bounds in the same evaluator. ### Query engine None ### Willingness to contribute - [x] I can contribute this improvement/feature independently - [ ] I would be willing to contribute this improvement/feature with guidance from the Iceberg community - [ ] I cannot contribute this improvement/feature at this time -- 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]
