pepijnve commented on code in PR #17813:
URL: https://github.com/apache/datafusion/pull/17813#discussion_r2515827494
##########
datafusion/physical-expr/src/expressions/case.rs:
##########
@@ -1282,15 +1283,57 @@ impl PhysicalExpr for CaseExpr {
}
fn nullable(&self, input_schema: &Schema) -> Result<bool> {
- // this expression is nullable if any of the input expressions are
nullable
- let then_nullable = self
+ let nullable_then = self
.body
.when_then_expr
.iter()
- .map(|(_, t)| t.nullable(input_schema))
- .collect::<Result<Vec<_>>>()?;
- if then_nullable.contains(&true) {
- Ok(true)
+ .filter_map(|(w, t)| {
+ let is_nullable = match t.nullable(input_schema) {
+ // Pass on error determining nullability verbatim
+ Err(e) => return Some(Err(e)),
+ Ok(n) => n,
+ };
+
+ // Branches with a then expression that is not nullable do not
impact the
+ // nullability of the case expression.
+ if !is_nullable {
+ return None;
+ }
+
+ // For case-with-expression assume all 'then' expressions are
reachable
+ if self.body.expr.is_some() {
+ return Some(Ok(()));
+ }
+
+ // For branches with a nullable 'then' expression, try to
determine
Review Comment:
That's actually what I'm investigating. I was thinking about moving that
file over to `expr-common` and then rewriting using the `when` expression using
a guarantee of `then -> NullableInterval::Null`. That would remove the need for
the `certainly_null_expr: Option<&Expr>` argument.
The downside is that I then have to materialise the rewritten `Expr` where
that's now done implicitly.
If you agree with the move to `expr-common` I can give this a go and then we
can measure the planning impact.
--
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]
