vadimpiven commented on code in PR #18848:
URL: https://github.com/apache/datafusion/pull/18848#discussion_r2547313934
##########
datafusion/optimizer/src/push_down_filter.rs:
##########
@@ -418,6 +418,204 @@ fn extract_or_clause(expr: &Expr, schema_columns:
&HashSet<Column>) -> Option<Ex
predicate
}
+/// Tracks coalesce predicates that can be pushed to each side of a FULL JOIN.
+struct PushDownCoalesceFilterHelper {
+ join_keys: Vec<(Column, Column)>,
+ left_filters: Vec<Expr>,
+ right_filters: Vec<Expr>,
+ remaining_filters: Vec<Expr>,
+}
+
+impl PushDownCoalesceFilterHelper {
+ fn new(join_keys: &[(Expr, Expr)]) -> Self {
+ let join_keys = join_keys
+ .iter()
+ .filter_map(|(lhs, rhs)| {
+ Some((lhs.try_as_col()?.clone(), rhs.try_as_col()?.clone()))
+ })
+ .collect();
+ Self {
+ join_keys,
+ left_filters: Vec::new(),
+ right_filters: Vec::new(),
+ remaining_filters: Vec::new(),
+ }
+ }
+
+ fn push_columns<F: FnMut(Expr) -> Expr>(
+ &mut self,
+ columns: (Column, Column),
+ mut build_filter: F,
+ ) {
+ self.left_filters
+ .push(build_filter(Expr::Column(columns.0)));
+ self.right_filters
+ .push(build_filter(Expr::Column(columns.1)));
+ }
+
+ fn extract_join_columns(&self, expr: &Expr) -> Option<(Column, Column)> {
+ if let Expr::ScalarFunction(ScalarFunction { func, args }) = expr {
+ if func.name() != "coalesce" {
Review Comment:
I think the fact that coalesce is basically identity function over kept side
of the join allows such optimization...
I will add end-to-end test tomorrow. I have a problem on my data where I
need this optimization, so I will try to reproduce it in end-to-end test.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
