ORC-1482: Adaptation to read ORC files created by CUDF

### What changes were proposed in this pull request?

This pr is aimed at adapting to read ORC files created by CUDF, which may have missing statistics in their DOUBLE/FLOAT columns.

### Why are the changes needed?

Official ORC readers can't read CUDF-created ORC files properly.

### How was this patch tested?

Added UT.

Closes apache#1598 from guiyanakuang/ORC-1482-to-1.8.

Authored-by: Yiqun Zhang <guiyanakuang@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>

java/core/src/java/org/apache/orc/impl/RecordReaderImpl.java

@@ -695,8 +695,8 @@ static TruthValue evaluatePredicateProto(OrcProto.ColumnStatistics statsProto,
                    " include ORC-517. Writer version: {}",
           predicate.getColumnName(), writerVersion);
       return TruthValue.YES_NO_NULL;
-    } else if (category == TypeDescription.Category.DOUBLE
-        || category == TypeDescription.Category.FLOAT) {
+    } else if ((category == TypeDescription.Category.DOUBLE ||
+        category == TypeDescription.Category.FLOAT) && cs instanceof DoubleColumnStatistics) {
       DoubleColumnStatistics dstas = (DoubleColumnStatistics) cs;
       if (Double.isNaN(dstas.getSum())) {
         LOG.debug("Not using predication pushdown on {} because stats contain NaN values",
@@ -1708,4 +1708,12 @@ public CompressionCodec getCompressionCodec() {
   public int getMaxDiskRangeChunkLimit() {
     return maxDiskRangeChunkLimit;
   }
+
+  /**
+   * Get sargApplier for testing.
+   * @return sargApplier in record reader.
+   */
+  SargApplier getSargApp() {
+    return sargApp;
+  }
 }

java/core/src/test/org/apache/orc/impl/TestRecordReaderImpl.java

@@ -41,6 +41,7 @@
 import org.apache.orc.CompressionCodec;
 import org.apache.orc.CompressionKind;
 import org.apache.orc.DataReader;
+import org.apache.orc.DoubleColumnStatistics;
 import org.apache.orc.OrcConf;
 import org.apache.orc.OrcFile;
 import org.apache.orc.OrcProto;
@@ -2474,6 +2475,47 @@ public void testWithoutStatistics() {
     assertEquals(TruthValue.YES_NO_NULL, truthValue);
   }
 
+  @Test
+  public void testDoubleColumnWithoutDoubleStatistics() throws Exception {
+    // orc-file-no-double-statistic.orc is an orc file created by cudf with a schema of
+    // struct<x:double>, one row and a value of null.
+    // Test file source https://issues.apache.org/jira/projects/ORC/issues/ORC-1482
+    Path filePath = new Path(ClassLoader.getSystemResource("orc-file-no-double-statistic.orc")
+        .getPath());
+
+    Configuration conf = new Configuration();
+    FileSystem fs = FileSystem.get(conf);
+
+    Reader reader = OrcFile.createReader(filePath,
+        OrcFile.readerOptions(conf).filesystem(fs));
+
+    TypeDescription schema = TypeDescription.fromString("struct<x:double>");
+
+    assertEquals(schema, reader.getSchema());
+    assertFalse(reader.getStatistics()[0] instanceof DoubleColumnStatistics);
+
+    SearchArgument sarg = SearchArgumentFactory.newBuilder()
+        .isNull("x", PredicateLeaf.Type.FLOAT)
+        .build();
+
+    Reader.Options options = reader.options()
+        .searchArgument(sarg, new String[] {"x"})
+        .useSelected(true)
+        .allowSARGToFilter(true);
+
+    VectorizedRowBatch batch = schema.createRowBatch();
+    long rowCount = 0;
+    try (RecordReader rr = reader.rows(options)) {
+      assertTrue(rr.nextBatch(batch));
+      rowCount += batch.size;
+      assertFalse(rr.nextBatch(batch));
+      if (rr instanceof RecordReaderImpl) {
+        assertEquals(0, ((RecordReaderImpl) rr).getSargApp().getExceptionCount()[0]);
+      }
+    }
+    assertEquals(1, rowCount);
+  }
+
   @Test
   public void testMissMinOrMaxInStatistics() {
     OrcProto.ColumnEncoding encoding = OrcProto.ColumnEncoding.newBuilder()