Minor-Fixes Support 7x segments as archive in 8x / 9x

docs/changelog/125666.yaml

@@ -0,0 +1,5 @@
+pr: 125666
+summary: Minor-Fixes Support 7x segments as archive in 8x / 9x
+area: Search
+type: bug
+issues: []

x-pack/plugin/old-lucene-versions/src/main/java/org/elasticsearch/xpack/lucene/bwc/codecs/BWCCodec.java

@@ -29,7 +29,6 @@
 import org.apache.lucene.store.IOContext;
 import org.apache.lucene.util.Version;
 import org.elasticsearch.core.UpdateForV10;
-import org.elasticsearch.xpack.lucene.bwc.codecs.lucene70.BWCLucene70Codec;
 import org.elasticsearch.xpack.lucene.bwc.codecs.lucene80.BWCLucene80Codec;
 import org.elasticsearch.xpack.lucene.bwc.codecs.lucene84.BWCLucene84Codec;
 import org.elasticsearch.xpack.lucene.bwc.codecs.lucene86.BWCLucene86Codec;
@@ -229,7 +228,6 @@ private static Codec getBackwardCompatibleCodec(Codec codec) {
         if (codec == null) return null;
 
         return switch (codec.getClass().getSimpleName()) {
-            case "Lucene70Codec" -> new BWCLucene70Codec();
             case "Lucene80Codec" -> new BWCLucene80Codec();
             case "Lucene84Codec" -> new BWCLucene84Codec();
             case "Lucene86Codec" -> new BWCLucene86Codec();

x-pack/plugin/old-lucene-versions/src/main/java/org/elasticsearch/xpack/lucene/bwc/codecs/lucene60/MetadataOnlyBKDReader.java

@@ -86,8 +86,16 @@ public MetadataOnlyBKDReader(IndexInput metaIn, boolean isVersionPost86) throws
         pointCount = metaIn.readVLong();
         docCount = metaIn.readVInt();
 
-        // This code has been introduced to process IndexInput created with Lucene86Codec+. This is not necessary
-        // in the read-only version for older formats.
+        // The pre-8.6 code does not read the following fields that its standard Lucene counterpart does. After experimenting with the
+        // code, we got to the assumption that these are the last fields being read, which are not needed in the metadata-only reader, and
+        // we can safely ignore them when loading the file. Although by coincidence, nothing breaks if we read a couple of VLongs, as long
+        // as some bytes are available to read.
+        //
+        // The extra reads have been introduced to process IndexInput created with Lucene86Codec+, where a new BKD format has been
+        // introduced. It seems that we have stricter checks around the header and footer starting from the 86 formats hence we do need to
+        // consume all the data input there but not in previous formats.
+        //
+        // For correctness, we added version checking here. If and only if, the version is 8.6 or higher, we read the additional fields.
         if (isVersionPost86) {
             metaIn.readVInt();
             metaIn.readLong();