LUCENE-15108: use writer.getReader instead of DirectoryReader.open

Author: VivekKumarNeu

lucene/CHANGES.txt

@@ -96,6 +96,8 @@ Bug Fixes
 * GITHUB#14847: Allow Faiss vector format to index >2GB of vectors per-field per-segment by using MemorySegment APIs
   (instead of ByteBuffer) to copy bytes to native memory. (Kaival Parikh)
 
+* GITHUB#15108: Fixed test failure caused by DirectoryReader.open() reading an index with uncommitted changes. (Vivek Kumar)
+
 * GITHUB#15213: Don't normalize centroids with zero samples, which caused divide-by-zero. (Mike Sokolov)
 
 * GITHUB#15125: Handle inconsistent schema on flush with index sorts (Nhat Nguyen)

lucene/test-framework/src/java/org/apache/lucene/tests/index/LegacyBaseDocValuesFormatTestCase.java

@@ -1544,14 +1544,14 @@ private void doTestSortedNumericsVsStoredFields(LongSupplier counts, LongSupplie
       int id = random().nextInt(numDocs);
       writer.deleteDocuments(new Term("id", Integer.toString(id)));
     }
-    try (DirectoryReader reader = maybeWrapWithMergingReader(DirectoryReader.open(dir))) {
+    try (DirectoryReader reader = maybeWrapWithMergingReader(writer.getReader())) {
       TestUtil.checkReader(reader);
       compareStoredFieldWithSortedNumericsDV(reader, "stored", "dv");
     }
     // merge some segments and ensure that at least one of them has more than
     // 256 values
     writer.forceMerge(numDocs / 256);
-    try (DirectoryReader reader = maybeWrapWithMergingReader(DirectoryReader.open(dir))) {
+    try (DirectoryReader reader = maybeWrapWithMergingReader(writer.getReader())) {
       TestUtil.checkReader(reader);
       compareStoredFieldWithSortedNumericsDV(reader, "stored", "dv");
     }