mbutrovich opened a new issue, #27276: URL: https://github.com/apache/datafusion/issues/27276
**Describe the bug** `split_part(s, delimiter, 0)` on a row where `s` is NULL returns NULL when the position comes from a column, and errors with "field position must not be zero" when the position is a literal. The [fast path](https://github.com/apache/datafusion/blob/14b3edd9a350b983fae6d2f194df1a3ae343ba15/datafusion/functions/src/string/split_part.rs#L97-L105) for an array string with a scalar delimiter and position [rejects position 0](https://github.com/apache/datafusion/blob/14b3edd9a350b983fae6d2f194df1a3ae343ba15/datafusion/functions/src/string/split_part.rs#L269-L271) before it looks at any row. The array path [skips rows where any argument is NULL](https://github.com/apache/datafusion/blob/14b3edd9a350b983fae6d2f194df1a3ae343ba15/datafusion/functions/src/string/split_part.rs#L530-L541), so it errors only on a row whose three arguments are all non-NULL. The two paths disagree whenever no row has a non-NULL string, such as a batch whose strings are all NULL. **To Reproduce** ```sql -- Literal delimiter and position: the fast path SELECT split_part(s, '.', 0) AS r FROM (VALUES (NULL::text)) AS t(s); -- Execution error: field position must not be zero -- The same row with the position from a column: the array path SELECT split_part(s, '.', p) AS r FROM (VALUES (NULL::text, 0)) AS t(s, p); -- NULL -- Only literals SELECT split_part(NULL::text, '.', 0) AS r; -- NULL ``` **Expected behavior** All three return NULL. PostgreSQL's `split_part` is strict: its [`pg_proc` entry](https://github.com/postgres/postgres/blob/884aca9b96989b9bc1b65d42e731e90f098ee2ee/src/include/catalog/pg_proc.dat#L3887-L3889) leaves [`proisstrict`](https://github.com/postgres/postgres/blob/884aca9b96989b9bc1b65d42e731e90f098ee2ee/src/include/catalog/pg_proc.h#L70) at its default of true, so a NULL argument returns NULL without calling the function. [`expr.slt`](https://github.com/apache/datafusion/blob/14b3edd9a350b983fae6d2f194df1a3ae343ba15/datafusion/sqllogictest/test_files/expr.slt#L714-L716) expects the first query's error ("Position 0 with column input errors even for empty/null inputs"), while the review of #21238 asked for the fast path to match the array path ([discussion_r3030106101](https://github.com/apache/datafusion/pull/21238#discussion_r3030106101)). Which behavior is intended? If it is NULL, the fast path could return all NULLs when the string array has no valid row before checking the position, and the `expr.slt` case would expect NULL. **Additional context** Property-based tests found this on `main` at 14b3edd9a3: calling `split_part` with an array string and scalar delimiter and position gave an error, and the same call with every argument as an array gave NULL. The reproduction runs on `main` at 14b3edd9a3. The fast path came with #21238. -- 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]
