mihaibudiu commented on code in PR #5384:
URL: https://github.com/apache/calcite/pull/5384#discussion_r4446229057


##########
core/src/main/java/org/apache/calcite/rex/RexUtil.java:
##########
@@ -1823,20 +1832,44 @@ private static boolean isFlat(
    * calls to an associative operator such as AND and OR.
    */
   public static boolean isFlat(RexNode expr) {
-    if (!(expr instanceof RexCall)) {
-      return true;
+    // Use an explicit stack to avoid StackOverflowError on deeply nested
+    // expressions (e.g. CALCITE-5845)
+    final Deque<RexNode> stack = new ArrayDeque<>();
+    stack.push(expr);
+    while (!stack.isEmpty()) {
+      final RexNode current = stack.pop();
+      if (!(current instanceof RexCall)) {
+        continue;
+      }
+      final RexCall call = (RexCall) current;
+      if (!isFlat(call.getOperands(), call.getOperator())) {

Review Comment:
   I haven't checked if this function may be recursive.



##########
core/src/test/java/org/apache/calcite/rex/RexUtilTest.java:
##########
@@ -43,4 +54,116 @@ class RexUtilTest {
         RexUtil.SubQueryFinder.find(expression, SqlKind.SCALAR_QUERY));
     assertNull(RexUtil.SubQueryFinder.find(expression, SqlKind.EXISTS));
   }
+
+  /** Test case for
+   * <a 
href="https://issues.apache.org/jira/browse/CALCITE-5845";>[CALCITE-5845]
+   * StackOverflowError when SQL has a very large predicate with many ANDs</a>.
+   *
+   * <p>Verifies that {@link RexUtil#flattenAnd} does not recurse when given a

Review Comment:
   Strictly speaking it's not checking the lack of recursion, but the ability 
to compute with a small stack.



##########
core/src/test/java/org/apache/calcite/rex/RexUtilTest.java:
##########
@@ -43,4 +54,116 @@ class RexUtilTest {
         RexUtil.SubQueryFinder.find(expression, SqlKind.SCALAR_QUERY));
     assertNull(RexUtil.SubQueryFinder.find(expression, SqlKind.EXISTS));
   }
+
+  /** Test case for
+   * <a 
href="https://issues.apache.org/jira/browse/CALCITE-5845";>[CALCITE-5845]
+   * StackOverflowError when SQL has a very large predicate with many ANDs</a>.
+   *
+   * <p>Verifies that {@link RexUtil#flattenAnd} does not recurse when given a
+   * deeply nested binary AND tree. */
+  @Test void testFlattenAndDeeplyNested() {
+    final RelBuilder builder =
+        RelBuilder.create(Frameworks.newConfigBuilder().build());
+    final RexBuilder rexBuilder = builder.getCluster().getRexBuilder();
+    final RelDataType booleanType =
+        builder.getTypeFactory().createSqlType(SqlTypeName.BOOLEAN);
+    final int depth = 20_000;
+    // Build AND(AND(AND(..., $1), $2), $3, ...) as a left-deep binary tree,
+    // iteratively. Hash codes are pre-warmed bottom-up so that the
+    // de-duplication set inside flattenAnd does not itself recurse deeply.
+    RexNode node = rexBuilder.makeInputRef(booleanType, 0);
+    node.hashCode();
+    for (int i = 1; i < depth; i++) {
+      node =
+          rexBuilder.makeCall(SqlStdOperatorTable.AND, node,
+              rexBuilder.makeInputRef(booleanType, i));
+      node.hashCode();
+    }
+    final RexNode deepTree = node;
+    final List<RexNode> flat =
+        runInSmallStack(() -> RexUtil.flattenAnd(ImmutableList.of(deepTree)));
+    assertThat(flat, hasSize(depth));
+  }
+
+  /** Test case for
+   * <a 
href="https://issues.apache.org/jira/browse/CALCITE-5845";>[CALCITE-5845]
+   * StackOverflowError when SQL has a very large predicate with many ANDs</a>.
+   *
+   * <p>Verifies that {@link RexUtil#flatten} does not recurse when given a
+   * deeply nested binary AND tree. */
+  @Test void testFlattenDeeplyNested() {
+    final RelBuilder builder =
+        RelBuilder.create(Frameworks.newConfigBuilder().build());
+    final RexBuilder rexBuilder = builder.getCluster().getRexBuilder();
+    final RelDataType booleanType =
+        builder.getTypeFactory().createSqlType(SqlTypeName.BOOLEAN);
+    final int depth = 20_000;
+    RexNode node = rexBuilder.makeInputRef(booleanType, 0);
+    for (int i = 1; i < depth; i++) {
+      node =
+          rexBuilder.makeCall(SqlStdOperatorTable.AND, node,
+              rexBuilder.makeInputRef(booleanType, i));
+    }
+    final RexNode deepTree = node;
+    final List<RexNode> flat =
+        runInSmallStack(
+            () -> RexUtil.flatten(ImmutableList.of(deepTree),
+            SqlStdOperatorTable.AND));
+    assertThat(flat, hasSize(depth));
+  }
+
+  /** Test case for
+   * <a 
href="https://issues.apache.org/jira/browse/CALCITE-5845";>[CALCITE-5845]
+   * StackOverflowError when SQL has a very large predicate with many ANDs</a>.
+   *
+   * <p>Verifies that {@link RexUtil#isFlat(RexNode)} does not recurse when
+   * given a deeply nested tree of a non-associative operator (NOT), where
+   * short-circuiting at the top level is not possible. */
+  @Test void testIsFlatDeeplyNested() {
+    final RelBuilder builder =
+        RelBuilder.create(Frameworks.newConfigBuilder().build());
+    final RexBuilder rexBuilder = builder.getCluster().getRexBuilder();
+    final RelDataType booleanType =
+        builder.getTypeFactory().createSqlType(SqlTypeName.BOOLEAN);
+    final int depth = 20_000;
+    RexNode node = rexBuilder.makeInputRef(booleanType, 0);
+    for (int i = 0; i < depth; i++) {
+      node = rexBuilder.makeCall(SqlStdOperatorTable.NOT, node);
+    }
+    final RexNode deepTree = node;
+    assertThat(runInSmallStack(() -> RexUtil.isFlat(deepTree)), is(true));
+  }
+
+  /** Runs {@code callable} on a thread with a small stack, so that a recursive
+   * implementation of the callee overflows quickly, while an iterative
+   * implementation completes. */
+  private static <T> T runInSmallStack(Callable<T> callable) {

Review Comment:
   how does this differ from the other function with the same name?



##########
core/src/main/java/org/apache/calcite/plan/RelOptUtil.java:
##########
@@ -2593,47 +2605,59 @@ public static void decomposeConjunction(
       @Nullable RexNode rexPredicate,
       List<RexNode> rexList,
       List<RexNode> notList) {
-    if (rexPredicate == null || rexPredicate.isAlwaysTrue()) {
+    // Use an explicit stack to avoid StackOverflowError on deeply nested ANDs

Review Comment:
   only ANDs or deep expressions in general?



-- 
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]

Reply via email to