Michael McCandless created LUCENE-10211:
-------------------------------------------
Summary: Move IntBlockPool's slice allocator and
SliceReader/Writer out to MemoryIndex
Key: LUCENE-10211
URL: https://issues.apache.org/jira/browse/LUCENE-10211
Project: Lucene - Core
Issue Type: Improvement
Reporter: Michael McCandless
Spinoff from exciting issue LUCENE-9673.
Lucene's {{ByteBlockPool}} and {{IntBlockPool}} are sort of like a little
malloc implementation, embedding many logical growing {{int[]}} into a series
of block buffers. This is needed to compactly/efficiently store postings data,
since there can be many unique terms, each with their own growing {{byte[]}}
holding postings.
{{IntBlockPool}} also has these slices, however, nowhere in {{core}} do we use
those. Rather, the allocation needs for {{int[]}} is simpler: just allocated a
fixed length 1, 2 or 3 {{int[]}} per unique term. We can greatly simplify
{{IntBlockPool}}.
However, {{MemoryIndex}} does use these slices from {{IntBlockPool}}. I think
we should move the complex slice logic in {{IntBlockPool}} out to
{{MemoryIndex}}, simplifying core?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
