Sorry I didn’t fully read your message, Gonzalo. I’m very busy this week, so I
am focusing on strategy rather than implementation.
The right solution is probably to link the map of type coercions to the
connection’s conformance. That will allow Calcite to receive SQL and pretend to
be (say)
+1 for a release.
Our users rely on a regular cadence of releases. If someone submits a patch and
doesn’t see it in a release until 6 months later (or it sits un-reviewed) they
will be less inclined to submit a patch.
Yeah, I know Open Source Doesn’t Require Providing Builds [1] but it makes
It looks to me like whatever semantics Pinot desires, it would still be useful
to have pluggable semantics for casts. So perhaps a new issue should be opened
for that; it looks like the existing framework isn't sufficient.
In the meantime we can probably also accept the solution in CALCITE-6210
Calcite supports infinities, but has some trouble computing using infinities as
literals, because literals are represented as decimal values internally. The
following PR https://github.com/apache/calcite/pull/3663 fixes this problem,
but has not been reviewed. The net outcome of the PR is that
Caican Cai created CALCITE-6279:
---
Summary: Use Null to replace the results of some Math functions
that return NaN.
Key: CALCITE-6279
URL: https://issues.apache.org/jira/browse/CALCITE-6279
Project:
Hi, Tanner Clary
I also found that some sql functions return java's double.NaN when it is
empty, instead of sql's null. Maybe I will raise a jira to discuss this.
Best wishes,
Cancai Cai
Cancai Cai 于2024年2月24日周六 00:32写道：
> Tanner Clary,
>
> Thank you very much for your help. I didn't expect
Tanner Clary,
Thank you very much for your help. I didn't expect it to work like this. I
learned interesting knowledge.
It seems that the infinity conversion is indeed not converted to double,
but converted to BigDecimal type. I have left a message on jira and I will
try to solve it.
Thank you
Cancai,
I have authored a commit that shows you how to allow functions with
non-null operands to return null [1]. Feel free to use this in your PR.
As for infinity, Calcite supports infinity, although I believe you need to
wrap it in a cast to double (at least in my experience). If some dialect
