sunchao opened a new issue, #5402:
URL: https://github.com/apache/datafusion-comet/issues/5402
## Describe the bug
#5393 fixes decimal-coerced negative-zero literals, but a few SQL assertions
still do not fully exercise signed-zero behavior:
- In `arrays_zip.sql`, replacing the bare decimal zero with `float('-0.0')`
changes the second result field from `DOUBLE` to `FLOAT`. The first field
remains double, but an explicit double signed-zero assertion is still needed.
- In the float and double `array_intersect.sql` fixtures, `[+0.0, -0.0]`
intersected with `[+0.0]` returns `[+0.0]` whether membership distinguishes the
signs. Opposite-sign singleton inputs in both directions, plus a same-sign
negative-zero control, exercise the distinction.
- In `percentile.sql`, the special-value query asks for probabilities `0`,
`0.5`, `0.8`, and `1`. Its results are `[-Infinity, 0.5, Infinity, NaN]`, so
losing the negative-zero input would not fail the test.
## Expected coverage
Preserve the existing double `arrays_zip` test shape and add explicit
float/double signed-zero column cases. Add the singleton intersections using
the SQL harness's Spark oracle, since Spark 4.2 changes floating-point
set-operation normalization. For percentile, use separate groups with only one
zero sign and compare exact median results; mixed-sign zero ties can introduce
ordering-dependent assertions.
These are test-coverage follow-ups to #5271 and #5393. They do not establish
a new production regression.
--
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]
