xylaaaaa commented on issue #69884: URL: https://github.com/apache/doris/issues/69884#issuecomment-6660427309
The [CAST file-pruning regression](https://github.com/apache/doris/issues/69884#issuecomment-6659316728) now has a verified local fix. Commit `a20f5343e7` is local to the development worktree and has not been pushed to PR #69904. On the retained two-file Spark table, all three entry points (Delta filesystem, Delta HMS and Hive gateway) previously returned an empty result for `CAST(decimal_decimal_same_scale AS BIGINT) = 123`. The FE converter discarded the cast and passed a comparison against the underlying decimal to Delta Kernel. This incorrectly excluded the file containing `123.45` before a BE scanner could evaluate the SQL. The selected Delta scan provider now declines residuals containing CAST or TRY_CAST for file pruning. Doris evaluates the original predicate on rows. Independent ordinary conjuncts still reach Kernel. The per-table decision matters for heterogeneous Hive catalogs. The existing catalog policy still applies, and the connector SPI signatures and version are unchanged. - Across the three entry points in synchronous and streaming modes, the cast-only query now returns `123.45` with two of two files selected. Adding `decimal_decimal_same_scale < 124` returns the same row. All six completed execution profiles selected just one of two files. - The new regression passed both generation and an ordinary rerun. All 24 result blocks matched the physical fixture, including filtered COUNT on the second file and TRY_CAST under OR. - The FE build and 106 targeted FE tests passed. Existing Delta basic, file-skipping and Parquet deletion-vector regressions passed. This resolves the reproduced correctness case in local verification. The separate small-table streaming performance observation remains open. -- 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]
