jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2651214933
> Computing a schema for a wildcard differs from expanding the actual
expressions that match the wildcard
I think their logic is basically the same, otherwise there would
findepi commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2650821795
> but some users do not want the casts introduced by `TypeCoercion`
This is very broad statement.
I don't like the DataFusion coercion logic per se because it has bugs (
jayzhan211 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2649641630
> Since `exprlist_to_fields` is called in the builder, it seems that
wildcard expansion still hasn't been delayed.
Computing schema for wildcard is different from expan
jayzhan211 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2649605687
> but some users do not want the casts introduced by `TypeCoercion`
We can make it optional
--
This is an automated message from the Apache Git Service.
To respond to
jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2648181791
> Should we move `TypeCoercion` into builder 🤔 ?
I also thought about that, but some users do not want the casts introduced
by `TypeCoercion`. See
https://github.co
findepi commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2648109414
I appreciate the conciseness of representing all columns of a table with a
wildcard.
This doesn't change the fact that `Expr::Wildcard` is not an expression.
If there are a
jayzhan211 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2646822350
> That could potentially unlock removing Expr::Wildcard. It's not really an
expression (just like Expr::Alias isn't
https://github.com/apache/datafusion/issues/1468 and Sort
jayzhan211 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2646820171
> > Hmm, my immediate thought here is that if we cannot guarantee correctly
coerced inputs during the first pass, then it's probably not worth attempting
to coerce in the bui
goldmedal commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2646780333
> Since `exprlist_to_fields` is called in the builder, it seems that
wildcard expansion still hasn't been delayed.
>
I see. I think we can revert the change
https://gith
jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2645780946
> Hmm, my immediate thought here is that if we cannot guarantee correctly
coerced inputs during the first pass, then it's probably not worth attempting
to coerce in the builder
jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2645778218
> The original discussion about wildcard expansion is [#11639
(comment)](https://github.com/apache/datafusion/issues/11639#issuecomment-2250895868).
It's used to delay the wild
goldmedal commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2644487765
> Perhaps [@goldmedal](https://github.com/goldmedal) have some thoughts on
it.
The original discussion about wildcard expansion is
https://github.com/apache/datafusion/
rkrishn7 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2644143759
+1 on moving `ExpandWildcardRule`!
> We might need to perform union coercion twice: once in the builder, and
once in TypeCoercion(after coercion of exprs).
Hmm, my
findepi commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2643654770
> I also think it's necessary to move the `ExpandWildcardRule` to the
builder.
That would be great!
That could potentially unlock removing `Expr::Wildcard`. It's no
jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2643116010
I also think it's necessary to move the `ExpandWildcardRule` to the builder.
This helps to solve the current problem. We won't need to perform wildcard
expansion when [computin
jonahgao commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2643083761
We need to rely on the `TypeCoercion` analyzer rule to obtain the correct
schema of union inputs. See #11961 for some details.
We might need to perform union coercion twi
findepi commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2639110011
You have good point about the wildcards
> is there benefit to producing a plan that has a correct schema but may
have non-coerced input plans?
i don't think there
rkrishn7 commented on issue #14380:
URL: https://github.com/apache/datafusion/issues/14380#issuecomment-2638760053
Hello @findepi!
I'd be happy to work on this. I believe it's as simple as shifting union
schema coercion from the analyzer to logical plan building.
However, since
18 matches
Mail list logo
