Github user takuti commented on a diff in the pull request:
https://github.com/apache/incubator-hivemall/pull/145#discussion_r182611062
--- Diff: nlp/src/main/java/hivemall/nlp/tokenizer/KuromojiUDF.java ---
@@ -69,13 +69,10 @@
private static final int READ_TIMEOUT_MS = 60000; // 60 sec
private static final long MAX_INPUT_STREAM_SIZE = 32L * 1024L * 1024L;
// ~32MB
- private Mode _mode;
--- End diff --
As far as we can observe from the result of test case
`KuromojiUDFTest#testSerialization`, there is no serialization issue here.
What this modification does is to defensively avoid early instantiation of
Lucene's unclear classes (`Mode`, `CharArraySet` and `UserDictionary`).
---
