fix not(null) with constant null

[WinkerDu]

Which issue does this PR close?

Closes #1191.

Rationale for this change

df not expression can not take NULL like

> select not(null);
Internal("NOT 'Literal { value: Utf8(NULL) }' can't be evaluated because the expression's type is Utf8, not boolean")

the correct output should be NULL

What changes are included in this PR?

NotExpr accepts NULL input presented as ScalarValue::Utf8(None)

Are there any user-facing changes?

NO.

[WinkerDu]

cc @alamb @Dandandan @yjshen
plz have a review, thank you.

[Ted-Jiang]

Is there any Null transform rule (doc) in datafusion?

[WinkerDu]

@Ted-Jiang I'm afraid there is not.
Maybe most of expressions (except for string expressions like concat, concat_ws, etc) should return NULL when accept NULL, correct me if wrong @alamb

[alamb]

Thank you for starting to take this on @WinkerDu ❤️

While the general purpose handling of null literals will likely be a more substantial change, I think this is a good step forward. If we can tighten up the checks a little I think we could merge this and improve things incrementally in follow on PRs.

[alamb]

I wonder if it would be helpful to add a test in datafusion/core/tests/sql somewhere to show a query like SELECT null, not(null) working?

[WinkerDu]

I wonder if it would be helpful to add a test in datafusion/core/tests/sql somewhere to show a query like SELECT null, not(null) working?

@alamb OK, will add more tests:

1. SELECT not(true), not(false) = false, true
2. SELECT null, not(null) = NULL, NULL
3. SELECT NOT('hi'), type check error would raise.

[WinkerDu]

@alamb PTAL, thank you.

[alamb]

Looks good to me -- thank you @WinkerDu

[alamb]

👍

[WinkerDu]

cc @yjshen @xudong963 @liukun4515 @Dandandan PTAL, thank you!

[xudong963]

❯ select not null;
+----------------+
| NOT Utf8(NULL) |
+----------------+
|                |
+----------------+
1 row in set. Query took 0.003 seconds.
❯ select not(null);
+----------------+
| NOT Utf8(NULL) |
+----------------+
|                |
+----------------+
1 row in set. Query took 0.003 seconds.

I tested in my local, very good! @WinkerDu