yujun777 commented on code in PR #62742:
URL: https://github.com/apache/doris/pull/62742#discussion_r3356155213
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/PushDownFilterThroughSetOperation.java:
##########
@@ -61,11 +63,47 @@ public Rule build() {
.when(s -> s.arity() > 0
|| (s instanceof LogicalUnion && !((LogicalUnion)
s).getConstantExprsList().isEmpty())))
.thenApply(ctx -> {
- LogicalFilter<LogicalSetOperation> filter = ctx.root;
- LogicalSetOperation setOperation = filter.child();
+ LogicalFilter<LogicalSetOperation> origFilter = ctx.root;
+ LogicalSetOperation setOperation = origFilter.child();
+
+ // Pushing a conjunct that contains a volatile expression
(rand/uuid/random_bytes/...)
+ // into each branch changes semantics for every set-op except
UNION ALL.
+ // - UNION ALL: each branch row = exactly one output row
(1:1), so evaluating
+ // rand() once per branch row still matches the
per-output-row semantic.
+ // - UNION DISTINCT / INTERSECT / EXCEPT: the set-op semantics
depend on the
+ // full branch row sets before dedup/intersect/except.
Sampling rows in each
+ // branch independently changes which rows participate (e.g.
INTERSECT becomes
+ // "half of A intersect half of B" instead of "half of (A
intersect B)").
+ boolean canPushVolatileExpr = setOperation instanceof
LogicalUnion
+ && setOperation.getQualifier() == Qualifier.ALL;
+ Set<Expression> pushableConjuncts;
+ Set<Expression> keptAboveConjuncts;
+ if (canPushVolatileExpr) {
+ pushableConjuncts = origFilter.getConjuncts();
+ keptAboveConjuncts = ImmutableSet.of();
+ } else {
+ pushableConjuncts = new LinkedHashSet<>();
+ Set<Expression> kept = new LinkedHashSet<>();
+ for (Expression c : origFilter.getConjuncts()) {
+ if (c.containsVolatileExpression()) {
+ kept.add(c);
+ } else {
+ pushableConjuncts.add(c);
+ }
+ }
+ keptAboveConjuncts = kept;
+ if (pushableConjuncts.isEmpty()) {
+ return null;
+ }
+ }
+ LogicalFilter<LogicalSetOperation> filter = pushableConjuncts
== origFilter.getConjuncts()
Review Comment:
This is a robustness cleanup only. The previous reference-equality check
worked because the UNION ALL branch reused the original set instance, but that
dependency was implicit and brittle. Using an explicit boolean makes the intent
clear and avoids depending on getConjuncts() preserving reference identity.
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/AddProjectForVolatileExpression.java:
##########
@@ -269,6 +271,85 @@ public <T extends Expression> Optional<Pair<List<T>,
LogicalProject<Plan>>> rewr
return Optional.of(Pair.of(newTargetsBuilder.build(), new
LogicalProject<>(projects, plan.child(0))));
}
+ private Optional<JoinRewriteResult>
rewriteJoinExpressions(LogicalJoin<Plan, Plan> join,
+ Collection<Expression> targets) {
+ Map<Expression, Integer> volatileExpressionCounter =
Maps.newLinkedHashMap();
+ Map<Expression, Set<Slot>> volatileExpressionSlots =
Maps.newLinkedHashMap();
+ for (Expression target : targets) {
+ target.foreach(e -> {
+ Expression expr = (Expression) e;
+ if (expr.isVolatile()) {
+ volatileExpressionCounter.merge(expr, 1, Integer::sum);
+ Set<Slot> inputSlots = expr.getInputSlots();
+ volatileExpressionSlots
+ .computeIfAbsent(expr, ignored ->
Sets.newLinkedHashSet())
+ .addAll(inputSlots.isEmpty() ?
target.getInputSlots() : inputSlots);
Review Comment:
This rename only removes the local naming ambiguity in the nested loop that
collects repeated volatile expressions. No behavior changes.
--
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]
