Open kewang1024 opened 1 month ago
I would say our result is more correct in this case (just replace INFINITY()
with a huge number to see why). No one should depend on such result though, if it is easy to fix we can change it to make it consistent, otherwise we can just leave it as it is.
@Yuhta Should we make Presto-Java match Velox then in this case? Since if Presto and Velox do not match, the fuzzer will just fail.
@kagamiori That would be ideal if we could
Bug description
create TABLE test_ke (name varchar, score double, score2 double, ds varchar) WITH (partitioned_by = ARRAY['ds'] );
INSERT INTO test_ke VALUES ('a', 1.1, 2.0, '2023-05-01'), ('a', INFINITY(), 1.0, '2023-05-01'),
select covar_pop(score, score2) FROM test_ke group by name;
[Presto java result] NaN
[DuckDB result] NaN
[Presto c++ result] -Infinity
System information
NA
Relevant logs
No response