Baymine opened a new issue, #66183: URL: https://github.com/apache/doris/issues/66183
### Search before asking - [X] I had searched in the [issues](https://github.com/apache/doris/issues?q=is%3Aissue) and found no similar issues. ### Description Today ORC/Parquet dictionary filtering only applies when a filter conjunct is a bare column reference compared to constants. A predicate whose value side is a deterministic single-slot string expression — e.g. `split_by_string(col, sep)[n] = 'x'`, `element_at(...)`, `substring_index(...)`, `regexp_extract(...)` — is evaluated once per row even though it depends only on the column value. Such predicates can instead be evaluated once per distinct dictionary value. The FE (Nereids) is the authoritative admission check: it stamps a conjunct root as dict-filterable when the predicate is an equality/IN over a single slot, is deterministic, and preserves NULL semantics (null-in implies null-out — the dictionary carries no NULL entry, so a NULL source must map to a NULL result to match per-row semantics). The verdict is carried to BE via `TExprNode.can_dict_filter`; when unset (older FE) BE falls back to a conservative allowlist. A session variable `enable_dict_filter_for_expr` (default true, forwarded as a query option) gates the expression case. ### Use case Scans that filter on a heavy string expression of a low-cardinality dictionary-encoded ORC/Parquet column evaluate the expression per row today; evaluating per distinct dictionary value is a large reduction for low-NDV columns. ### Related issues _No response_ ### Are you willing to submit PR? - [X] Yes I am willing to submit a PR! ### Code of Conduct - [X] I agree to follow this project's [Code of Conduct](https://www.apache.org/foundation/policies/conduct) -- 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]
