LuciferYang opened a new issue, #7699:
URL: https://github.com/apache/kyuubi/issues/7699
### Description
`DynMethods.UnboundMethod.invokeChecked` branches only on whether the
matched method is varargs:
```java
if (argLength < 0) {
return (R) method.invoke(target, args);
} else {
return (R) method.invoke(target, Arrays.copyOfRange(args, 0,
argLength));
}
```
For any fixed-arity method it copies the argument array, including when
`args.length` already equals `argLength`. There the copy carries the same
element references into `Method.invoke`, and `Arrays.copyOfRange(T[], int,
int)` preserves the source array's runtime component type, so it is not
normalizing the carrier's type either. It is discarded as soon as the call
returns.
The copy exists to change arity: to drop trailing arguments when the caller
passes more than the matched method takes, and to null-pad when it passes
fewer. When the counts are equal neither applies, and the varargs branch
already passes the caller's array straight through.
So every exact-arity call allocates one array for no effect. Nobody has
measured how hot the affected call sites are.
Two behaviours of the copying branch are relied on and have no test:
`HiveConnectorUtils.newStorageFormat` and `copyStorageFormat` invoke with 7
arguments against the 6-argument `CatalogStorageFormat.apply` that Spark below
4.2 provides, and depend on the trailing `serdeName` being dropped. A comment
above that call site spells the dependency out. `DynMethodsTest` covers exact
arity and varargs only, so nothing pins it.
A caller passing fewer arguments than the matched method takes reaches the
target with trailing nulls rather than being rejected. Also untested, and it
constrains any change to this branch.
### Affects Version(s)
master (1.13.0-SNAPSHOT). `DynMethods` is copied from parquet-common by way
of iceberg-common, and both upstreams still copy unconditionally, so this is
not a regression from a sync.
### Are you willing to submit PR?
Yes.
--
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]
