Description Martin Oberhuber 2011-11-28 12:33:47 EST

Build ID: Eclipse 3.7.1 / Lucene 2.9.1

The build process for generating a pre-built help index is not reproducable.

This is a problem for software delivery, since deployed binary plugins with identical version (and identical source content) may have different binary content - thus jeopardizing trust in the software delivery mechanism.

For example, comparing the user docs of Eclipse 3.7.0 and Eclipse 3.7.1:
  plugins/org.eclipse.platform.doc.user_3.7.0.v20110601-0800.jar
shows following 4 differences under the index/ folder:

1.) Following files have a timestamp embedded, so re-building changes them:
    - indexed_contributions
    - indexed_dependencies
    - indexed_docs

2.) The binary Lucene segments file differs in bytes 9-12:
    - segments_2
    Bytes 9-12 are the Lucene "NameCounter" field as per the Lucene file format;
    The last 4 bytes also differ, since they are a checksum over the file:
    http://lucene.apache.org/java/2_9_1/fileformats.html#Segments%20File 

I'm not sure to what extent this is an Eclipse problem or a Lucene problem... if it's a Lucene problem it should be brought up with upstream. Suggestion would be to avoid dumping timestamps, and using a "NameCounter" algorithm that's not sensitive to the timestamp or folder being built in.