[
https://issues.apache.org/jira/browse/CALCITE-5615?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17711181#comment-17711181
]
Mihai Budiu commented on CALCITE-5615:
--------------------------------------
I believe that the suite was built to test correctness of sqlite. But I have
run it on mysql and postgres and alll tests passed. It tests such a narrow side
of SQL that almost any failure probably signals a bug. It is designed to stay
within the standard portable side of SQL.
The second question is more complicated. Calcite is a framework for building
SQL compilers. We have built a SQL compiler which has passed all tests, but it
only uses a narrow subset of the existing query plan optimizations. The
"default" Calcite compiler fails several tests (and I have run only a few
thousands of the tests so far). Some failures are crashes, so they are clearly
bugs. Other are incorrect results; some (or all) of these are most likely bugs
as well. They probably indicate incorrect query optimization passes.
Take a look at the above URL; Sql logic test is open-source and the
documentation is good too, it's not particularly complicated.
> Run SQLLogicTests using Calcite
> -------------------------------
>
> Key: CALCITE-5615
> URL: https://issues.apache.org/jira/browse/CALCITE-5615
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Affects Versions: 1.34.0
> Reporter: Mihai Budiu
> Priority: Minor
>
> Sqllogictest is a program designed to verify that an SQL database engine
> computes correct results by comparing the results to identical queries from
> other SQL database engines.
> https://www.sqlite.org/sqllogictest/doc/trunk/about.wiki
> The nice thing about SLT is that it contains more than 7 million tests. The
> tests only cover the core of SQL, ideally the portable part across all
> engines. They only test integers, doubles, and strings. So they could
> probably be part of the Calcite slow tests.
> The tests should be structured so that any query execution engine can be used.
> I plan to contribute such an implementation if people think it is useful, but
> I haven't yet worked out all the details.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
