Fix Windowing/scanAndSort query issues on top of Joins.

[kgyrtkirk]

• allow a hashjoin result to be converted to RowsAndColumns
  • added StorageAdapterRowsAndColumns
• fix incorrect isConcrete() return values during early phase of planning

Fixes #XXXX.

Description

Fixed the bug ...

Renamed the class ...

Added a forbidden-apis entry ...

Release note

---

Key changed/added classes in this PR

• MyFoo
• OurBar
• TheirBaz

---

This PR has:

• been self-reviewed.
  • using the concurrency checklist (Remove this item if the PR doesn't have any relation to concurrency.)
• added documentation for new or modified features or behaviors.
• a release note entry in the PR description.
• added Javadocs for most classes and all non-trivial methods. Linked related entities via Javadoc links.
• added or updated version, license, or notice information in licenses.yaml
• added comments explaining the "why" and the intent of the code wherever would not be obvious for an unfamiliar reader.
• added unit tests or modified existing tests to cover new code paths, ensuring the threshold for code coverage is met.
• added integration tests.
• been tested in a test Druid cluster.

[imply-cheddar]

This looks fine to me. Given that this code path should hopefully only run on top of the results of another query, I suspect that it will be fine. I'll admit that I find StorageAdapterRowsAndColumns to be a little gross (it's like... doing something but trying not to do anything), but it should just work for the intended purpose here.

[abhishekagarwal87]

what is the reason for this change?

[kgyrtkirk]

• I think even in case we are defaulting null-s to something somewhere; we may still have the null value.
  running the following query with druid-27.0.0 with druid.generic.useDefaultValueForNull: true

select null,'',0,cast(null as integer)
null |  empty | 0 | null 

which suggests that the column value can be null

• in this case the incoming column also reported that it has nulls - I think it shouldn't be allowed to narrow the scope

[clintropolis]

the cast thing returning null is probably a bug (though its been a bug for so long that imo it isn't worth fixing in a deprecated mode), the native cast expression would spit out 0 from an expression selector if it actually ran...

I think to not risk any inadvertent changes we shouldn't make this change for now if possible. Default value mode is deprecated and will be removed in a future release, so we should probably just leave things as they are whenever possible.

[kgyrtkirk]

I think that:

• the system do allow null -s - via by a simple cast - so null exists even in defaultValue mode - so I don't see it bad...
• I believe this codepath is only triggered for windowoperator related stuff - but I'll check it (I'll update this comment when it finishes)
• the incoming frame declares that it has null values (which is being covered)
• we also have the ClusteredGroupPartitionerTest#testDefaultClusteredGroupPartitionerWithNulls which essentially inputs null-s as inputs...should I disable that testcase for defaultValues ? it will most likely just hide the issue I was facing - but I can live with that...should I just do that?

I believe the original logic incorrectly narrows to doesn't have nulls (NullHandling.sqlCompatible() && frameCol.hasNulls) - which could be false irrsepectively that the backing one does have (frameCol.hasNulls) ...if that's not allowed an exception should be thrown right away and not take further risks...

I see the following options:

• allow nulls - as the incoming frame already contained them
• raise an exception and mark the testcase ClusteredGroupPartitionerTest#testDefaultClusteredGroupPartitionerWithNulls not apply when useDefaultValue is enabled

[kgyrtkirk]

I've taken the second approach:

• restored these classes to the previous state
• added an assumeTrue(sqlCompatible) to the testcase

I still think that it might not be a good idea to retain/make tweaks like this...null does exists in the system even when defaultValue is enabled...so it may naturally happen as a value in a column as well.