[SOLR-15943] Simplify HDFS cache to account for LUCENE-10366 / LUCENE-10376

• Type: Task
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 9.0
• Component/s: Hadoop Integration, hdfs
• Assignee: Kevin Risden
• Reporter: Kevin Risden
• Created: 20/Jan/22 20:38
• Updated: 12/May/22 00:35
• Resolved: 18/Feb/22 15:55

Description

LUCENE-10366 / LUCENE-10376 found that the loop unrolling isn't necessary anymore. SOLR-15405 implemented the little endian loop unrolling, but this can be simplified to avoid issues in the future.

Example of what is needed: https://github.com/apache/lucene/pull/592 and https://github.com/apache/lucene/pull/602

  @Override
  public final int readVInt() throws IOException {
    // this can make JVM less confused (see LUCENE-10366)
    return super.readVInt();
  }

  @Override
  public final long readVLong() throws IOException {
    // this can make JVM less confused (see LUCENE-10366)
    return super.readVLong();
  }

Issue Links

• relates to: SOLR-15405 Make HDFS handle little endian LUCENE-9047
• relates to: LUCENE-10366 Reduce the number of valid checks for ByteBufferIndexInput#readVInt
• relates to: LUCENE-10376 Roll up the loop in vint/vlong in DataInput
• links to: GitHub Pull Request #650