- (negation) with NULL literals does not work: can't be evaluated because the expression's type is Utf8, not signed

[alamb]

Reproducer:

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

The correct output is null

I think the solution will likely be to teach Expr::Negative about type coercion (as is done in BinaryExpr and function calls) once coercion knows about the null type (and that it can be cast to any other type). See #1184 for more details

[WinkerDu]

@alamb plz assign to me, thank you.

[xudong963]

I tested select -null; in mysql and postgres: mysql returned null but

postgres=# select -null;
ERROR:  operator is not unique: - unknown
LINE 1: select -null;
               ^
HINT:  Could not choose a best candidate operator. You might need to add explicit type casts.

Do you need this syntax in iox? @alamb

[WinkerDu]

I tested select -null; in mysql and postgres: mysql returned null but

postgres=# select -null;
ERROR:  operator is not unique: - unknown
LINE 1: select -null;
               ^
HINT:  Could not choose a best candidate operator. You might need to add explicit type casts.

Do you need this syntax in iox? @alamb

@xudong963 Thanks for the test! I agree we can discuss more on it.

[alamb]

Do you need this syntax in iox? @alamb

We do not need this syntax in IOx (and if we do I can always add it at that time)

[WinkerDu]

@alamb @xudong963 Maybe we can introduce this syntax, and keep the contract that operators comes out result NULL when accepts NULL.

[liukun4515]

@alamb @xudong963 Maybe we can introduce this syntax, and keep the contract that operators comes out result NULL when accepts NULL.

It's better to make it consistent with PG in SQL dialect.

[WinkerDu]

I feel free to close the pr #2249 since more conservative opinions of introducing -NULL SQL dialect
Thank you @liukun4515 @xudong963

[liukun4515]

I think this issue can be closed.
@alamb @andygrove

[alamb]

🤔 I just tried the query again, and it now gives a different error:

❯ select -null;
Internal("(- 'Literal { value: NULL }') can't be evaluated because the expression's type is Null, not signed numeric")

I think we are getting closer...

[alamb]

I bet this is a fairly straightforward matter of adding a case in the evaluation kernels

[drrtuy]

It is indeed an additional condition in expressions/negative.rs. Here is the PR #3771.