weiting-chen commented on code in PR #13245:
URL: https://github.com/apache/gluten/pull/13245#discussion_r4271688203
##########
gluten-ut/spark42/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala:
##########
@@ -152,11 +162,15 @@ object VeloxSQLQueryTestSettings extends
SQLQueryTestSettings {
"postgreSQL/boolean.sql",
"postgreSQL/case.sql",
"postgreSQL/comments.sql",
+ "postgreSQL/create_view.sql",
"postgreSQL/date.sql",
"postgreSQL/float4.sql",
+ "postgreSQL/float8.sql",
"postgreSQL/groupingsets.sql",
"postgreSQL/insert.sql",
"postgreSQL/int2.sql",
+ "postgreSQL/int4.sql",
+ "postgreSQL/int8.sql",
Review Comment:
**Refresh the Java 21 shadow goldens for the newly enabled overflow tests**
**Target Location:**
`gluten-ut/spark42/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala:173,194`;
corresponding `results/postgreSQL/int8.sql.out.java21` and
`results/try_aggregates.sql.out.java21` under
`gluten-ut/spark42/src/test/resources/backends-velox/sql-tests`.
**Problem:** These newly enabled cases still select Spark 4.1-era expected
output on JDK 21/25. `GlutenSQLQueryTestSuite` prefers an existing
`.out.java21` file whenever `Utils.isJavaVersionAtLeast21`; this PR refreshes
the ordinary `.out` files but leaves both higher-priority companions unchanged.
They expect `"long overflow"` where Spark 4.2 produces `"overflow"`, so these
restored cases fail golden comparison on those supported JDK profiles. This is
a test-compatibility defect, not a production wrong-result claim. Confidence:
high; the mismatch is source-proven, not a locally executed JDK 21 failure.
**Evidence:**
```scala
// Newly added allowlist entries:
"postgreSQL/int8.sql",
// ...
"try_aggregates.sql",
```
The [result-file
selector](https://github.com/apache/gluten/blob/56d9f1c54b64a683f11ddf067be4f36c2f3ac1da/gluten-ut/spark42/src/test/scala/org/apache/spark/sql/GlutenSQLQueryTestSuite.scala#L170-L173)
selects the stale companions. Their `"message" : "long overflow"` values
remain at lines 398/921/961/1001 (`int8`) and 160/368 (`try_aggregates`). Spark
4.2's
[arithmeticOverflowError](https://github.com/apache/spark/blob/32f7299601108917fb01920a54e084595b7b3bf8/sql/api/src/main/scala/org/apache/spark/sql/errors/ExecutionErrors.scala#L118-L143)
explicitly canonicalizes `\w+ overflow` to `overflow`; the exact upstream Java
21 goldens contain that normalized value. The passing SQL CI XML records JDK
17, which selects the ordinary files and does not exercise this branch.
**Suggested Fix:** Refresh both `.out.java21` resources from Spark
`32f7299601108917fb01920a54e084595b7b3bf8`, just as for their ordinary
counterparts. The complete substantive delta is these six parameter
replacements:
```diff
- "message" : "long overflow"
+ "message" : "overflow"
```
Preserve the other Java-specific output rather than deleting the overrides.
Run the two restored SQL cases on JDK 21 with golden generation unset; include
alternate result-file suffixes in the import/provenance 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]
