[GitHub] [calcite] Aaaaaaron commented on a change in pull request #2253: [CALCITE-4385] Extracts and eliminates/merges the condition that has common expressions (Jiatao Tao)

[GitBox]

Aaaaaaron commented on a change in pull request #2253:
URL: https://github.com/apache/calcite/pull/2253#discussion_r535309649



##########
File path: core/src/main/java/org/apache/calcite/rex/RexSimplify.java
##########
@@ -1698,6 +1699,113 @@ RexNode simplifyAnd2ForUnknownAsFalse(List<RexNode> 
terms,
     }
   }
 
+  /**
+   * <p>Simplifies AND/OR condition that has common expressions,
+   * extracts and eliminates/merges them as much as possible.</p>
+   * <ul>
+   * <li>(a OR b) AND (a OR b OR c OR d) =&gt; (a OR b)
+   * <li>(a OR b OR c OR d) AND (a OR b OR e OR f) =&gt; (a OR b) OR ((c OR d) 
AND (e OR f))
+
+   * <li>(a AND b) OR (a AND b AND c) =&gt; (a AND b)
+   * <li>(a AND b AND c AND d) OR (a AND b AND e AND f) =&gt;
+   * (a AND b) AND ((c AND d) OR (e AND f))
+   * </ul>
+   * <p> The difference between {@link #simplifyAnd} is that {@link 
#simplifyAnd} mainly
+   * simplifies expressions whose answer can be determined without evaluating 
both sides,
+   * like: FALSE AND (xxx) =&gt; FALSE</p>
+   */
+  public RexNode eliminateCommonExprInCondition(RexNode node) {
+    if (!(node instanceof RexCall)) {
+      return node;
+    }
+    // Simplify children recursively
+    // so that when we simplify parent node
+    // the child node should be simplified
+    List<RexNode> operands = ((RexCall) node).getOperands();
+    List<RexNode> simplifiedChildren = operands.stream()
+        .map(this::eliminateCommonExprInCondition)
+        .collect(Collectors.toList());
+
+    switch (node.getKind()) {
+    case AND: {
+      // Split the children to get the disjunctive predicates.
+      List<List<RexNode>> disjunctions = simplifiedChildren.stream()
+          .map(RelOptUtil::disjunctions)
+          .collect(Collectors.toList());
+
+      // Find the common predict among the list by retainAll.
+      List<RexNode> common = new ArrayList<>(disjunctions.get(0));
+      disjunctions.forEach(common::retainAll);
+
+      if (common.isEmpty()) {
+        // no common, return the original but with child simplified
+        return and(simplifiedChildren);
+      }
+
+      // Get the predicates that without the common part.
+      disjunctions.forEach(p -> p.removeAll(common));
+
+      if (disjunctions.stream().anyMatch(List::isEmpty)) {
+        // (a || b) && (a || b || c || d) => (a || b)
+        return or(common);
+      }
+      // (a || b || c || d) && (a || b || e || f) =>
+      // (a || b) || ((c || d) && (e || f))
+      List<RexNode> splice = disjunctions.stream()
+          .map(this::or)
+          .collect(Collectors.toList());
+      return or(or(common), and(splice));

Review comment:
       `or(or(...), ...)`'s advance is more readable, I found 
`RexUtil.composeDisjunction` will o the flatten anyway(flattenOr), do you think 
we can keep this here?




----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org