GH-2926: Fix issue reading buggy parquet files with incorrect compressed size

parquet-avro/src/test/java/org/apache/parquet/avro/TestParquetReaderBadCompressedSize.java

@@ -0,0 +1,29 @@
+package org.apache.parquet.avro;
+
+import com.google.common.io.Resources;
+import org.apache.avro.generic.GenericData;
+import org.apache.hadoop.conf.Configuration;
+import org.apache.hadoop.fs.Path;
+import org.apache.parquet.hadoop.ParquetReader;
+import org.junit.Test;
+
+/**
+ * Validates that old files which have compressed size excluding the dictionary page size are readable
+ */
+public class TestParquetReaderBadCompressedSize {
+  private static final String STATIC_FILE_BAD_COMPRESSED_SIZE = "test-bad-compressed-size.parquet";
+
+  @Test
+  public void testBadCompressedSize() throws Exception {
+    Path testFile =
+        new Path(Resources.getResource(STATIC_FILE_BAD_COMPRESSED_SIZE).getFile());
+    Configuration configuration = new Configuration();
+    try (ParquetReader<Object> parquet = AvroParquetReader.builder(testFile)
+        .disableCompatibility()
+        .withDataModel(GenericData.get())
+        .withConf(configuration)
+        .build()) {
+      parquet.read();
+    }
+  }
+}

parquet-hadoop/src/main/java/org/apache/parquet/hadoop/ParquetFileReader.java

@@ -2056,6 +2056,9 @@ public BytesInput readAsBytesInput(int size) throws IOException {
 
         ByteBuffer lastBuffer = ByteBuffer.allocate(missingBytes);
         f.readFully(lastBuffer);
+        lastBuffer.flip();
+        LOG.warn(
+            "compressedLength underflow occurred, possibly due to older files not including dict header len");
 
         List<ByteBuffer> buffers = new ArrayList<>(streamBuffers.size() + 1);
         buffers.addAll(streamBuffers);