Remove Min/Max references from AggregateStatistics

[alamb]

Is your feature request related to a problem or challenge?

Part of #11151 (where we are removing special case uses of Min/Max)

Describe the solution you'd like

Reemove these cases:

datafusion/datafusion/core/src/physical_optimizer/aggregate_statistics.rs

Lines 185 to 186 in 569be9e

	if let Some(casted_expr) =
	agg_expr.as_any().downcast_ref::<expressions::Min>()

datafusion/datafusion/core/src/physical_optimizer/aggregate_statistics.rs

Lines 197 to 198 in 569be9e

	if let Some(casted_expr) =
	agg_expr.as_any().downcast_ref::<expressions::Min>()

datafusion/datafusion/core/src/physical_optimizer/aggregate_statistics.rs

Lines 235 to 236 in 569be9e

	if let Some(casted_expr) =
	agg_expr.as_any().downcast_ref::<expressions::Max>()

datafusion/datafusion/core/src/physical_optimizer/aggregate_statistics.rs

Lines 247 to 248 in 569be9e

	if let Some(casted_expr) =
	agg_expr.as_any().downcast_ref::<expressions::Max>()

Specifically, the idea is to remove the pattern

       expr.as_any().downcast_ref::<Max>()

and

       expr.as_any().downcast_ref::<Min>()

Describe alternatives you've considered

I suggest adding a general purpose function to AggregateExec and then implementing it for Min and Max (so we can do the same for Min/Max UDAF)

impl AggregateExpr {

 /// Return the value of the aggregate function, if known, given the number of input rows.
 /// 
 /// Return None if the value can not be determined solely from the input.
 /// 
 /// # Examples
 /// * The `COUNT` aggregate would return `Some(11)` given `num_rows = 11`
 /// * The `MIN` aggregate would return `Some(Null)` given `num_rows = 0
 /// * The `MIN` aggregate would return `None` given num_rows = 11
 fn output_from_rows(&self, num_rows: usize) -> Option<ScalarValue> { None }
...
}

### Additional context

_No response_

[Rachelint]

take

[alamb]

Thanks @Rachelint -- I also have some time later this week to help with this issue (if it might help to have one or two examples as we work through #11151)

[Rachelint]

Hi @alamb , I have some questions about the alternatives mentioned in the issue.
It seems the interface should be like:

trait AggregateExpr {
	fn output_from_stats(&self, stats: &Statistics) -> Option<ScalarValue> { None }
	...
}

As I understand, it is a optimization takes effect when specific aggregate functions(such as max, min, count...) can get the results directly from stats?

[alamb]

As I understand, it is a optimization takes effect when specific aggregate functions(such as max, min, count...) can get the results directly from stats?

Yes that is the case

I think output_from_stats would also be reasonable and potentially more general. One challenge is that the AggregateExpr might not have access to its argument exprs (and thus it might not know what the argument is) 🤔

[Rachelint]

As I understand, it is a optimization takes effect when specific aggregate functions(such as max, min, count...) can get the results directly from stats?

Yes that is the case

I think output_from_stats would also be reasonable and potentially more general. One challenge is that the AggregateExpr might not have access to its argument exprs (and thus it might not know what the argument is) 🤔

It seems a function expressions exists, and can help？

datafusion/datafusion/physical-expr-common/src/aggregate/mod.rs

Line 121 in 569be9e

fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>>;

and it is used to get min/max from stats now?

datafusion/datafusion/core/src/physical_optimizer/aggregate_statistics.rs

Lines 250 to 257 in 569be9e

	if casted_expr.expressions().len() == 1 {
	// TODO optimize with exprs other than Column
	if let Some(col_expr) = casted_expr.expressions()[0]
	.as_any()
	.downcast_ref::<expressions::Column>()
	{
	if let Precision::Exact(val) =
	&col_stats[col_expr.index()].max_value

[alamb]

Sounds like it might work -- I think the only way to really know for sure would be to try it out

[Rachelint]

Sounds like it might work -- I think the only way to really know for sure would be to try it out

Thanks, I try it now.

[edmondop]

Sounds like it might work -- I think the only way to really know for sure would be to try it out

Thanks, I try it now.

@Rachelint if you stop working on this could you please let me know?

[Rachelint]

Sounds like it might work -- I think the only way to really know for sure would be to try it out

Thanks, I try it now.

@Rachelint if you stop working on this could you please let me know?

Sorry for delay, still working on now, will try to push codes today...

[edmondop]

@alamb we can close this too as fixed via #12296 right?

[alamb]

Thanks @edmondop