<from JIRA-user mailing list post>

I'd like to make two improvements to the way Jira handles searching
for issues:

• currently, when a search is performed, the entire set of found
  issues is loaded into memory. in some cases this might be needed,
  but when a user is doing a search, only the issues for the first
  "frame" of hits needs to be loaded - the next can be loaded when
  (if) the user clicks to the next page of results.

• Lucene 1.4 includes sorting code now, so that the sorting can
  be done efficiently inside Lucene, instead of afterwards in the
  application.

Changing both of these will dramatically speed up the response time
of doing issue searches. For us, it would mean at least a ten-fold
improvement.

Looking into doing this, I've got some questions about the Lucene
code in Jira:

Why is there a comment index and an issue index? It would seem more
efficient to have only one index, where the information needed to
find issues by comments is in the issue index. Unless I'm not seeing it, there don't appear to be any searches specifically interested in the comment records themselves - only the issues that the comments
are in.

Is this correct?

Most of the values used to index in Lucene are stored as
Field.Keyword or Field.Text, which besides just indexing the field,
stores the value in the index. Then, when a hit is retrieved from
the index, those values are retrieved (meaning all the text for
the Field is returned). It does not appear, however, that any of
those retrieved values are ever used - all searches seem to go
through DefaultSearchProvider, which only uses a single field,
DocumentConstants.ISSUE_ID, and throws the rest away. The ISSUE_ID
is used to re-retrieve the information out of the database (or
out of the issue cache, if it is there).

Is this also correct?

It seems like what needs to happen is for there to be some kind
of "IssueSearchResult" object which contains the values for found
issues, and that object used in the generation of the JSP instead
of a GenericEntity. That way, the IssueSearchResult object could
get its values straight from Lucene, bypassing the database.

Does this seem a reasonable approach?

Thanks -

Tim