aoto-tech opened a new issue, #25266:
URL: https://github.com/apache/datafusion/issues/25266
### Describe the bug
I found that `percentile_cont` quantizes its Float64 interpolation weight in
steps of `1e-6`. Small nonzero weights therefore become zero, producing the
lower input value instead of an interpolated result.
### To Reproduce
I reproduced this on 55.1.0 and `main` at `9082d6b10`.
```sql
SELECT percentile_cont(x, CAST(2.5e-7 AS DOUBLE))
FROM (VALUES
(CAST(0.0 AS DOUBLE)),
(CAST(1.0 AS DOUBLE)),
(CAST(2.0 AS DOUBLE))
) AS t(x);
```
```text
actual: 0.0
expected: 5e-7
### Expected behavior
The rank is `(3 - 1) * 2.5e-7 = 5e-7`, so linear interpolation between `0.0`
and `1.0` should return approximately `5e-7`.
### Additional context
The Float64 path converts the fraction to an integer using
[`INTERPOLATION_PRECISION =
1_000_000`](https://github.com/apache/datafusion/blob/9082d6b10c29b72d56bede3d8e353d9d61fde542/datafusion/functions-aggregate/src/percentile_cont.rs#L935-L945).
That truncates this weight to zero. I couldn't find an existing issue for the
Float64 behavior.
--
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]
