My intention was to make the code more understandable for readers. Because if
we do explicitly casting and leave less guess room for readers, readers will
know the same type of values are compared, and they do not need to question the
correctness because of implicitly casting somewhere. At least, readers do not
need to search how `assertEquals` is implemented for different types. From
this perspective, `assertEquals(Byte.valueof((byte) 1),
nestedRow.getByte("aByte"))` is ugly, but necessary.
After I read the code, seems like the family of `assertEquals` in JUnit is
simpler. `assertEquals(Byte.valueof((byte) 1), nestedRow.getByte("aByte"))`
will be converted `assertEquals(object, object)`, and `object.equals(object)`
will be called. In `Byte` implementation, as Anton pointed out, `Byte.equals()`
accepts `Object` and does a casting anyway. So from implementation perspective,
there is no difference among approaches in this discussion.
Since there isn't a perfect way to make the code clear, so I am ok with either
the original way in this PR, or other discussed way to implement it.
[ Full content available at: https://github.com/apache/beam/pull/6383 ]
This message was relayed via gitbox.apache.org for [email protected]
