lyne7-sc opened a new pull request, #23148:
URL: https://github.com/apache/datafusion/pull/23148
## Which issue does this PR close?
<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases. You can
link an issue to this PR using the GitHub syntax. For example `Closes #123`
indicates that this PR will close issue #123.
-->
- Related to #13232.
## Rationale for this change
<!--
Why are you proposing this change? If this is already explained clearly in
the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand your
changes and offer better suggestions for fixes.
-->
Outer join elimination relies on proving that a filter rejects the
NULL-padding rows introduced by an outer join. DataFusion already handles many
built-in NULL-propagating expressions, but scalar UDFs did not expose whether
they preserve the same property.
```sql
SELECT t1.a
FROM t1 LEFT JOIN t2 ON t1.a = t2.x
WHERE abs(t2.y) > 5;
```
Rows produced by the unmatched side of the left join have `t2.y = NULL`.
Since `abs(NULL)` is also `NULL`, the predicate `abs(t2.y) > 5` cannot evaluate
to true for those rows. That means the left join can be safely rewritten as an
inner join. Without function-level null propagation metadata, the optimizer has
to treat scalar functions conservatively and misses this rewrite.
This PR adds `ScalarUDFImpl::is_strict()` to let scalar UDF implementations
declare that they always return NULL when any argument is NULL. The default is
`false` so existing UDFs remain conservative. Optimizer rules can then use this
metadata when reasoning about expression nullability and null-rejecting
predicates.
This design follows a pattern used by other query engines. PostgreSQL
exposes `STRICT / RETURNS NULL ON NULL INPUT` on functions, and documents that
such functions are not executed when any argument is NULL; a NULL result is
assumed automatically. DuckDB similarly has function null-handling metadata,
with default NULL-in/NULL-out behavior and special handling for functions that
do not follow that rule.
References:
- PostgreSQL `CREATE FUNCTION: STRICT / RETURNS NULL ON NULL INPUT`
https://www.postgresql.org/docs/current/sql-createfunction.html
- DuckDB `FunctionNullHandling` definition
https://github.com/duckdb/duckdb/blob/main/src/include/duckdb/function/function.hpp
- DuckDB scalar function null-handling API
https://github.com/duckdb/duckdb/blob/main/src/include/duckdb/function/scalar_function.hpp
## What changes are included in this PR?
- Adds `ScalarUDFImpl::is_strict()`, defaulting to false.
- Adds `ScalarUDF::is_strict()` as the public forwarding API.
- Marks `abs` as strict.
- Uses strict scalar functions in predicate/nullability reasoning and outer
join elimination.
- Propagates `is_strict` through `datafusion-ffi::FFI_ScalarUDF`.
- Adds unit and slt coverage for strict and non-strict scalar UDF behavior.
<!--
There is no need to duplicate the description in the issue here but it is
sometimes worth providing a summary of the individual changes in this PR.
-->
## Are these changes tested?
Yes.
<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code
If tests are not included in your PR, please explain why (for example, are
they covered by existing tests)?
-->
## Are there any user-facing changes?
Yes. Scalar UDF authors can now override ScalarUDFImpl::is_strict() to tell
the optimizer that their function always returns NULL when any argument is NULL.
This PR also changes the FFI_ScalarUDF layout to propagate strictness across
the FFI boundary, so it should carry the api change label.
<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->
<!--
If there are any breaking changes to public APIs, please add the `api
change` label.
-->
## Future work
- Mark more built-in scalar functions as strict where the behavior is
clearly NULL-in/NULL-out.
- Reuse `is_strict()` in other optimizer rules to infer `arg IS NOT NULL`
from predicates like `strict_func(arg) > 0`.
- Use inferred non-null predicates to simplify redundant `IS NOT NULL`
checks and push filters closer to scans.
- Use strictness in statistics/selectivity estimation by deriving tighter
nullability information for function outputs.
--
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]
