Github user paul-rogers commented on a diff in the pull request:
https://github.com/apache/drill/pull/797#discussion_r108048071
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/planner/physical/SubsetTransformer.java
---
@@ -45,15 +48,20 @@ public RelTraitSet newTraitSet(RelTrait... traits) {
}
- boolean go(T n, RelNode candidateSet) throws E {
+ public boolean go(T n, RelNode candidateSet) throws E {
if ( !(candidateSet instanceof RelSubset) ) {
return false;
}
boolean transform = false;
+ Set<RelNode> transformedRels = Sets.newHashSet();
for (RelNode rel : ((RelSubset)candidateSet).getRelList()) {
if (isPhysical(rel)) {
RelNode newRel = RelOptRule.convert(candidateSet,
rel.getTraitSet().plus(Prel.DRILL_PHYSICAL));
+ if(transformedRels.contains(newRel)) {
--- End diff --
Does this work? A hash map depends on `hashCode()` and `equals()`. A random
check of one `DrillRelNode`, `DrillFilterRelBase`, shows that the above two
methods are not implemented. As a result, set membership may be based on object
identity. But, surely each call to `convert()` will create a new object?
Or, does this rule apply to only certain kinds of nodes which do have
`hashCode()` and `equals()` implemented? If so, perhaps add some comments to
identify the subtree of nodes with which this code works.
Do we have a test case for this code? Or, did you step through it to see if
it was doing what we want? If so, and it worked, then likely this code does
work with a specific subset of nodes that have the needed methods.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
