bruno-roustant commented on a change in pull request #633: LUCENE-8753
UniformSplit PostingsFormat
URL: https://github.com/apache/lucene-solr/pull/633#discussion_r292959376
##########
File path:
lucene/core/src/java/org/apache/lucene/codecs/lucene80/Lucene80Codec.java
##########
@@ -91,7 +91,11 @@ public Lucene80Codec() {
* flushed/merged segments.
*/
public Lucene80Codec(Mode mode) {
- super("Lucene80");
+ this("Lucene80", mode);
+ }
+
+ protected Lucene80Codec(String name, Mode mode) {
+ super(name);
Review comment:
> Use FilterCodec instead?
Yes! Good point. I'll do that.
> Use Lucene50PostingsWriter instead?
DeltaBaseTermStateSerializer writes each TermState with file pointers delta
encoded relatively to a base pointer (which is the block base file pointer).
And it does only that thing. This is different from
Lucene50PostingsWriter.encodeTerm which writes the TermState with file pointers
delta encoded relatively to the previous TermState written.
This logic to delta encode relatively to the base pointer allows us to read
TermStates in random access, no sequential reading required. This allows us to
make a binary search inside the block itself and read only one TermState.
To summarize: No, we cannot use Lucene50PostingsWriter.encodeTerm. That said
DeltaBaseTermStateSerializer could be located elsewhere, the only issue is that
it writes IntBlockTermState to stay compatible and still use
Lucene50PostingsReader.postings() which requires IntBlockTermState because it
casts internally the provided BlockTermState.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
