Cannot Count(Expr:Wildcard) with DataFrame API

[Jefffrey]

Describe the bug
A clear and concise description of what the bug is.

Cannot do count(Expr::Wildcard) aggregate with DataFrame API

To Reproduce
Steps to reproduce the behavior:

    ctx.table("alltypes_plain")
        .await?
        .aggregate(vec![], vec![count(Expr::Wildcard)])?
        .explain(false, false)?
        .show()
        .await?;

Throws error:

Error: Internal("Wildcard expressions are not valid in a logical query plan")

Expected behavior
A clear and concise description of what you expected to happen.

Should be like when doing SQL:

    ctx.sql("select count(*) from alltypes_plain")
        .await?
        .explain(false, false)?
        .show()
        .await?;

Gives plan:

+---------------+---------------------------------------------------+
| plan_type     | plan                                              |
+---------------+---------------------------------------------------+
| logical_plan  | Aggregate: groupBy=[[]], aggr=[[COUNT(UInt8(1))]] |
|               |   TableScan: alltypes_plain projection=[id]       |
| physical_plan | ProjectionExec: expr=[7300 as COUNT(UInt8(1))]    |
|               |   EmptyExec: produce_one_row=true                 |
|               |                                                   |
+---------------+---------------------------------------------------+

Where count(*) has been transformed to count(1) essentially.

Additional context
Add any other context about the problem here.

Expansion substitute:

https://github.com/apache/arrow-datafusion/blob/e9852074bacd8c891d84eba38b3417aa16a2d18c/datafusion/expr/src/utils.rs#L43-L45

See SQL substitute function:

https://github.com/apache/arrow-datafusion/blob/e9852074bacd8c891d84eba38b3417aa16a2d18c/datafusion/sql/src/expr/function.rs#L193-L218

[Jefffrey]

Ideally see if can push this logic to the LogicalPlanBuilder, so as to not duplicate the logic b/w DataFrame API and SQL planning?

[jiangzhx]

Ideally see if can push this logic to the LogicalPlanBuilder, so as to not duplicate the logic b/w DataFrame API and SQL planning?

how about add a new OptimizerRule? to rewrite count(*) to count(1)

[Jefffrey]

how about add a new OptimizerRule? to rewrite count(*) to count(1)

Hmm, does the Logical Optimizer require that the plan be fully resolved (no wildcards), I wonder? I think it might require it so, but I could be mistaken. Could be worth exploring 👍