Github user cloud-fan commented on a diff in the pull request:
https://github.com/apache/spark/pull/19488#discussion_r145017429
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
---
@@ -205,14 +205,17 @@ object PhysicalAggregation {
case logical.Aggregate(groupingExpressions, resultExpressions, child)
=>
// A single aggregate expression might appear multiple times in
resultExpressions.
// In order to avoid evaluating an individual aggregate function
multiple times, we'll
- // build a set of the distinct aggregate expressions and build a
function which can
+ // build a map of the distinct aggregate expressions and build a
function which can
// be used to re-write expressions so that they reference the single
copy of the
- // aggregate function which actually gets computed.
- val aggregateExpressions = resultExpressions.flatMap { expr =>
+ // aggregate function which actually gets computed. Note that
aggregate expressions
+ // should always be deterministic, so we can use its canonicalized
expression as its
--- End diff --
That's not true, see Postgres:
```
cloud=# select random(), random();
random | random
-------------------+------------------
0.360924747306854 | 0.90011026058346
(1 row)
cloud=# select random() - random();
?column?
---------------------
-0.0260558221489191
(1 row)
```
Seems we need a way to distinguish FIRST_VALUE and RANDOM. Both of them are
nondeterministic, but `FIRST_VALUE` has consistent values in the same
project/aggregate.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
