[
https://issues.apache.org/jira/browse/LUCENE-7531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15625244#comment-15625244
]
Dawid Weiss commented on LUCENE-7531:
-------------------------------------
Well, it's sad to see the stuff I came up with (and Mike implemented) go
away... :) But more seriously -- this does seem to impact large automata. Can
you recode the existing automata and see how much we lose by removing packing?
Looking at the patch target addresses are still vint-encoded; if I recall right
the compression ratio gained by packing was significant (compared to baseline
fst), but a small fraction of overall input size. So a fst gain of a few
megabytes for data size that is several hundred megabytes is indeed worth
cutting the additional complexity of fst construction.
+1 to remove it, but some stats on dictionary sizes before/after would be nice.
> Remove packing support from FST
> -------------------------------
>
> Key: LUCENE-7531
> URL: https://issues.apache.org/jira/browse/LUCENE-7531
> Project: Lucene - Core
> Issue Type: Task
> Reporter: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-7531.patch
>
>
> This seems to be only used for the kuromoji dictionaries, but we could easily
> rebuild those dictionaries with packing disabled.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
