Biao Geng created FLINK-40986:
---------------------------------
Summary: Equal named Rows can have different hash values in PyFlink
Key: FLINK-40986
URL: https://issues.apache.org/jira/browse/FLINK-40986
Project: Flink
Issue Type: Bug
Components: API / Python
Reporter: Biao Geng
h3. Problem
Named PyFlink {{Row}} instances compare equal when their field names and
corresponding values match, regardless of field order. However,
{{Row._{_}hash{_}_}} hashes values in their original storage order.
Consequently, equal rows can have different hashes, violating Python's
equality/hash contract.
This causes incorrect deduplication in Python sets and failed lookups or
duplicate entries in Python dictionaries when equal named rows have different
field orders.
h3. Reproduce
{code:python}
from pyflink.common import Row
left = Row(a=1, b=2)
right = Row(b=2, a=1)
print(left == right)
print(hash(left) == hash(right))
print(len({left, right}))
print({left: "value"}.get(right, "NOT FOUND"))
{code}
Actual output:
{code:java}
True
False
2
NOT FOUND
{code}
Expected output:
{code:java}
True
True
1
value
{code}
h3. Root cause
In {{{}flink-python/pyflink/common/types.py{}}}, {{Row._{_}eq{_}}} \{_}compares
named-row values in sorted field-name order, while
{{Row.}}{_}{{{}{_}hash{_}_{}}} hashes {{tuple(self)}} in the original value
order.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
