jayzhan211 opened a new issue, #26497:
URL: https://github.com/apache/datafusion/issues/26497
### Describe the bug
`nth_value` and `approx_percentile_cont` return the same result with
`DISTINCT` as without it. This is the same problem #26471 reports for `corr`,
`covar_*` and `regr_*`, which #26484 fixes. Both functions are tagged
`DistinctHandling::Unsupported`, but their accumulators never read
`is_distinct`.
### To Reproduce
```sql
CREATE TABLE t(x DOUBLE) AS VALUES (1.0), (1.0), (1.0), (2.0), (3.0);
SELECT nth_value(x, 2 ORDER BY x), nth_value(DISTINCT x, 2 ORDER BY x) FROM
t;
-- 1 1
SELECT approx_percentile_cont(x, 0.5), approx_percentile_cont(DISTINCT x,
0.5) FROM t;
-- 1 1
SELECT approx_percentile_cont(0.5) WITHIN GROUP (ORDER BY x),
approx_percentile_cont(DISTINCT 0.5) WITHIN GROUP (ORDER BY x) FROM t;
-- 1 1
```
### Expected behavior
The distinct values are `{1, 2, 3}`, so all three `DISTINCT` forms should
return 2. If DataFusion does not support `DISTINCT` for these functions, the
query should fail with an error instead of returning the non-distinct answer.
### Additional context
#26484 adds a `not_impl_err!` check to each accumulator. Another option is a
single check in `AggregateExprBuilder::build`: return `not_impl_err!` when
`is_distinct` is set and the function reports `DistinctHandling::Unsupported`.
Both aggregates and window functions go through that builder, and it runs after
the logical optimizer, so `SingleDistinctToGroupBy` can still rewrite cases
like `corr(DISTINCT x, x)`. The default for `distinct_handling` is `Sensitive`,
so user-defined functions that don't opt in are not affected. Future functions
tagged `Unsupported` would then be covered automatically, and the error would
come at planning time instead of execution 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]
