Vas Shabu created FLINK-40844:
---------------------------------
Summary: Planner inserting a ghost column when all rows are
partition keys
Key: FLINK-40844
URL: https://issues.apache.org/jira/browse/FLINK-40844
Project: Flink
Issue Type: Bug
Reporter: Vas Shabu
While adding EXPR$0 handling for PTFs, we found SystemTypeInference always
falls back to a phantom EXPR$0 column whenever the function's own output is
empty — even when PARTITION BY or rowtime columns already populate the row:
- deriveFunctionOutputFields returns EXPR$0 unconditionally for an empty
function output, without checking whether pass-through/rowtime columns already
contributed fields.
- This produces a wrong resolved schema (extra column) and a codegen arity
mismatch whenever an empty-output PTF is combined with PARTITION BY.
- Fix: only fall back to EXPR$0 when the whole assembled output row
(pass-through + function output + rowtime) is empty.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
