sunchao commented on PR #5472:
URL:
https://github.com/apache/datafusion-comet/pull/5472#issuecomment-5735182352
@andygrove Addressed the remaining feedback and rebased onto current `main`
at
`076efbac`:
- Added an end-to-end `CometExpressionSuite` regression with another column
as an `IN`
candidate. It keeps an `In` node with a non-literal candidate and executes
both the
projection and filter paths, covering the dynamic equality path for signed
zero.
- Kept all-literal, non-NaN top-level `FLOAT` / `DOUBLE` lists static. When
such a list
contains one zero sign, serialization adds the missing sign so `IN (0, 1,
2)` keeps
static-filter/Parquet pruning while matching Spark. NaN-containing or
dynamic lists still
normalize both sides.
- Updated the serialization/fallback coverage for the new static-zero path
and updated
`docs/source/user-guide/latest/compatibility/floating-point.md` to
describe the scalar
comparison and membership coverage.
- Checked the nested-type references: #5507 is ordering/rank only, and #5191
covered
`arrays_overlap` / `array_position`. Neither tracks nested array/struct
membership, so
I filed #6019 for that separate pre-existing gap and kept this PR scoped
to top-level
floating-point operands.
Local validation on the rebased head:
- `make core` with JDK 17
- `./mvnw test -Dtest=none
-Dsuites="org.apache.comet.CometExpressionSuite,org.apache.comet.rules.CometExecRuleSuite"`
on the default Spark 4.1 profile: 186 passed
- the same focused suites with `-Pspark-3.4`: 184 passed, 2 version-gated
cancellations
- CI-equivalent Spark 4.0 semantic Scalafix check
- `git diff --check`
--
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]
