bx123 created FLINK-24895:
-----------------------------
Summary: SqlTimestampSerializer#deserialize forgets to fully read
from source when the record is null.
Key: FLINK-24895
URL: https://issues.apache.org/jira/browse/FLINK-24895
Project: Flink
Issue Type: Bug
Components: API / Type Serialization System
Reporter: bx123
As we support null Timestamp in this Serializer. In the 2 deserialize()
methods, when we read the first long and get Long.MIN_VALUE, we directly return
null and forget to read the next int which responds to nanos.
If there is really a null Timestamp that has been serialized before, we may get
wrong value when we deserialize the Object after this timestamp. So before
return null, source.skipBytesToRead(4) is needed to make the source in the
right position.
By the way, I find the copy(source, target) is implemented by deserialize the
intermediate value, while its more efficient to just copy 12 bytes from source
to target like target.write(source, 12).
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
