Functions have slightly different behaviors on different databases. Solving the problem starts with logging a jira case, and having a discussion about the specification.
Rather than say (for example) ’SOUNDEX returns the wrong result’, it’s helpful to say ’SOUNDEX result for empty string is inconsistent with MySQL’. Then we can discover that SOUNDEX has different behaviors on MySQL and Postgres, decide whether we want to support both behaviors, and if so, decide how we will achieve that behavior. As I have said many times: log a jira first, and only then get to work on a PR. Julian > On Feb 6, 2024, at 8:38 AM, Mihai Budiu <mbu...@gmail.com> wrote: > > These are two different functions with the same name. > You need two different implementations. > There are other similar functions in Calcite, e.g., SOUNDEX. > https://calcite.apache.org/docs/reference.html#dialect-specific-operators > > Mihai > ________________________________ > From: Cancai Cai <can...@apache.org> > Sent: Tuesday, February 6, 2024 8:26 AM > To: dev@calcite.apache.org <dev@calcite.apache.org> > Subject: About Negative testing for adapting to different databases > > Hello Calcite Community, > I'm currently completing CALCITE-6224 > <https://issues.apache.org/jira/browse/CALCITE-6224>. > During the completion process, we found that the return result of log2(0) > in calcite is different from that in mysql and spark. The return result of > log2(0) in calcite is -Infinity. But mysql and spark return NULL. We are > considering whether we need to adapt the return results of log2(0) in mysql > and spark. But a new problem comes out, if I adapt log2(0). Then do I also > need to consider the negative test of the two functions log10 and log in > other databases? > I believe this happens to me not only in the log function, I observed the > same situation in CALCITE-6241 > <https://issues.apache.org/jira/browse/CALCITE-6241>. > Now my doubt is whether we need to adapt the negative test of each function > in each database (because there are examples of negative test results > caused by different databases and different versions) . > I think we can add a negative test in calcite to tell users and developers > how calcite handles the negatvie test of this kind of function, but I am > not sure whether it needs to be adapted. > > Best wishes, > Cancai Cai
