vranes opened a new pull request, #56674:
URL: https://github.com/apache/spark/pull/56674
### What changes were proposed in this pull request?
This is a follow-up to the `BIN BY` relation operator (SPARK-57133) that
cleans up its error conditions in `error-conditions.json` and the related
analyzer code. Four groups of changes:
1. **SQLSTATE corrections** for conditions that carried placeholder states
that did not match their semantics:
- `BIN_BY_ALIGN_TO_TYPE_MISMATCH`: `42804` -> `42K09`
- `BIN_BY_RANGE_TYPE_MISMATCH`: `42804` -> `42K09`
- `BIN_BY_DUPLICATE_DISTRIBUTE_COLUMN`: `42701` -> `42713`
- `BIN_BY_INVALID_ALIGN_TO`: `42K09` -> `42K08`
- `BIN_BY_REQUIRES_TOP_LEVEL_COLUMN`: `42K09` -> `0A000`
2. **Rename** `BIN_BY_DISTRIBUTE_TYPE_MISMATCH` ->
`BIN_BY_INVALID_DISTRIBUTE_COLUMN_TYPE`. The check is an absolute required-type
check with no second operand, so `*_TYPE_MISMATCH` was the wrong shape; this
follows the `INVALID_<thing>_TYPE` convention (precedent:
`INVALID_BUCKET_COLUMN_DATA_TYPE`). The genuine two-operand mismatches keep
`*_TYPE_MISMATCH`.
3. **Split** the `BIN_BY_INVALID_BIN_WIDTH` catch-all, which fired at three
sites (eval-failed, non-positive, wrong-type) under one SQLSTATE. SubClasses
cannot carry per-subclass SQLSTATEs, so this splits into separate conditions
mirroring the `ALIGN TO` structure:
- `BIN_BY_INVALID_BIN_WIDTH`: narrowed to constant-folding failure;
message and SQLSTATE (`42K08`) now match `BIN_BY_INVALID_ALIGN_TO`.
- new `BIN_BY_NON_POSITIVE_BIN_WIDTH` (`42816`)
- new `BIN_BY_INVALID_BIN_WIDTH_TYPE` (`42K09`)
4. **Remove** `BIN_BY_COLUMN_NOT_FOUND`. A missing column now reports the
standard `UNRESOLVED_COLUMN.WITH_SUGGESTION` (with column suggestions),
matching every other operator. `BIN_BY_REQUIRES_TOP_LEVEL_COLUMN` still covers
the nested-field case.
### Why are the changes needed?
The conditions were added with the operator in SPARK-57133 and some carried
SQLSTATEs that do not describe the error (`42804` is "CASE result expressions
incompatible"; `42701` is "same assignment target twice"). The `BIN WIDTH`
catch-all reported three distinct failures under one state and message, and a
missing column raised a bespoke condition where the standard
`UNRESOLVED_COLUMN` already exists. This aligns the operator's error surface
with the rest of Spark SQL before it ships.
### Does this PR introduce _any_ user-facing change?
Yes, but only within unreleased `master`. `BIN BY` was introduced on
`master` (5.0.0), is gated off by default, and has not shipped in any released
version. This changes the SQLSTATEs and condition names of analysis-time errors
raised by `BIN BY`, and replaces `BIN_BY_COLUMN_NOT_FOUND` with
`UNRESOLVED_COLUMN.WITH_SUGGESTION`. No released version is affected.
### How was this patch tested?
Updated and ran `ResolveBinBySuite` (bin-width split, distribute-type
rename, missing-column now expects `UNRESOLVED_COLUMN.WITH_SUGGESTION`) and
`SparkThrowableSuite`, which validates `error-conditions.json` (ordering,
SQLSTATE invariants, no duplicates). Both suites pass.
### Was this patch authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Claude Opus 4.8)
--
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]
