If the issue can be reproduced using a SQL query on a standard data set, the best test case is often a Quidem fragment.
A recent commit that did that: https://github.com/apache/calcite/commit/1245db6201. They also added a test to the applicable unit test. I confess it’s not very easy to run Quidem tests from the command line. Maybe someone can fix that. Julian > On Nov 9, 2023, at 6:30 AM, Egor Ryashin <[email protected]> wrote: > > I wonder which *Test.java should I try to use as a template? I need smth that > parses the SQL and uses Volcano optimizer as the stacktrace shows it fails > during optimization. > > BTW, it happened during Apache Druid query execution and I saw the Druid > custom join rule was engaged so I thought initially it was on the Druid side. > But Apache Druid fixed that with Apache Calcite update to 1.35, so now I’m > not sure was it Calcite or Druid or both. > >> On 9 Nov 2023, at 14:52, Ran Tao <[email protected]> wrote: >> >> Hi, Egor, I agree with Julian's opinion, maybe you can use calcite's test >> set for this case, such as emp, dept, hr and other schema/tables. >> then developers will easily locate the problem. >> >> Some bugs may have persisted in earlier versions to this day, some were >> later fixed, and some were covered up. >> But as long as a specific standard case/sql and a certain calcite version >> are given, >> we can easily draw from the commit/code history which commit, which version >> introduced the problem or it has never worked correctly. >> >> Best Regards, >> Ran Tao >> >> >> Egor Ryashin <[email protected]> 于2023年11月9日周四 18:10写道： >> >>> FYI, it’s not reproduced with Calcite 1.35 >>> >>>> On 7 Nov 2023, at 03:10, Julian Hyde <[email protected]> wrote: >>>> >>>> Did you find an existing Jira case? If not, could you log one? You >>> should create a minimal test case and figure out what are the essential >>> features that are necessary to reproduce this problem. Ideally convert it >>> to run on a built-in schema such as Scott (emp and dept tables). This will >>> help someone fix the case, help someone suggest a workaround, and help >>> others identify that they are hitting the same issue. >>>> >>>>> On Nov 6, 2023, at 3:51 AM, Egor Ryashin <[email protected]> >>> wrote: >>>>> >>>>> Hi all, >>>>> >>>>> I’ve just run into this exception: >>>>> Type mismatch: rowtype of new rel: RecordType(VARCHAR pub, VARCHAR >>> pub0, DOUBLE NOT NULL $f2) NOT NULL rowtype of set: RecordType(VARCHAR pub, >>> VARCHAR pub0, DOUBLE rate) NOT NULL >>>>> >>>>> >>>>> executing this SQL: >>>>> SELECT base."pub”, >>>>> Any_value(base.”rate”), >>>>> Any_value(comparison."rate") >>>>> FROM ( >>>>> SELECT "pub", >>>>> Min(total)/Max(price) AS "rate" >>>>> FROM "test_data" >>>>> WHERE "__time" >= timestamp '2022-03-01 >>> 00:00:00' >>>>> GROUP BY "pub" >>>>> ORDER BY TRUE, >>>>> 2 limit 500 ) base >>>>> left outer join >>>>> ( >>>>> SELECT "pub", >>>>> min(total)/max(price) AS "rate" >>>>> FROM "test_data" >>>>> WHERE "__time" < timestamp '2022-03-01 >>> 00:00:00' >>>>> GROUP BY "pub" ) comparison >>>>> ON base."pub" = comparison."pub" >>>>> GROUP BY 1 >>>>> >>>>> looking through JIRA I concluded that bug is still an issue, I wonder >>> if there’s a known workaround for that? >>>>> >>>>> >>>>> >>>> >>> >>> >
