Copilot commented on code in PR #13124:
URL: https://github.com/apache/gluten/pull/13124#discussion_r4132113241
##########
docs/developers/VeloxUDF.md:
##########
@@ -192,18 +192,21 @@ VeloxColumnarToRow
+- Scan hive spark_catalog.default.tbl [col1#11], HiveTableRelation
[`spark_catalog`.`default`.`tbl`,
org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, Data Cols: [col1#11],
Partition Cols: []]
```
-## Natively Only UDF Registration
+## Natively Only UDF/UDAF Registration
Review Comment:
This section now covers by-name UDAFs, but the public configuration text
remains UDF-only in `VeloxConfig.scala:728-731`, this guide's configuration
table at line 219, and the generated `docs/velox-configuration.md:65`. Please
update the source config description and regenerate the configuration docs so
users can discover that this flag also applies to UDAFs.
##########
docs/developers/VeloxUDF.md:
##########
@@ -192,18 +192,21 @@ VeloxColumnarToRow
+- Scan hive spark_catalog.default.tbl [col1#11], HiveTableRelation
[`spark_catalog`.`default`.`tbl`,
org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, Data Cols: [col1#11],
Partition Cols: []]
```
-## Natively Only UDF Registration
+## Natively Only UDF/UDAF Registration
-This is an alternative to the registration described above, for a UDF that is
implemented only in Velox and has no Java counterpart.
+This is an alternative to the registration described above, for a UDF or UDAF
that is implemented only in Velox and has no Java counterpart.
This is off by default. Set
`spark.gluten.sql.columnar.backend.velox.nativeUDF.bypassRegistration=true` to
turn it on.
-Once enabled, a UDF whose registered name contains no dot is added to the
session's function registry under that name, provided the name is not already a
Spark built-in and no other loaded UDF differs from it only in case. It needs
no matching Hive UDF class, no jar on the classpath, and no `CREATE TEMPORARY
FUNCTION` — register it under a name with no dot, such as `my_udf`, and call it
directly:
+Once enabled, a UDF or UDAF whose registered name contains no dot is added to
the session's function registry under that name, provided the name is not
already a Spark built-in, no other loaded function differs from it only in
case, and the name is not loaded as both a UDF and a UDAF. It needs no matching
Hive class, no jar on the classpath, and no `CREATE TEMPORARY FUNCTION` —
register it under a name with no dot, such as `my_udf`, and call it directly:
```
spark-sql (default)> select my_udf(col1) from tbl;
+spark-sql (default)> select k, my_udaf(v) from tbl group by k;
```
+A name loaded as both a UDF and a UDAF is skipped. Velox keeps its scalar and
aggregate registries separately, so a library may declare one of each, but a
Spark function name resolves to a single builder and a call site gives nothing
to choose with. Both remain reachable through a Hive class name.
Review Comment:
“Both remain reachable through a Hive class name” is only true when the
library separately registers dotted aliases. The collision handled here
necessarily involves an undotted name, so if that is the only registration,
skipping injection makes both implementations unreachable. Please qualify the
statement to avoid promising a fallback that may not exist.
--
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]
