[
https://issues.apache.org/jira/browse/LUCENE-5708?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14013477#comment-14013477
]
Robert Muir commented on LUCENE-5708:
-------------------------------------
All these "PostingsFormat" tests wire their codec to themselves. This allows us
to implement generic base classes with tests that all codecs should pass, as
well as codec-specific tests (e.g. in lucene41) that test particular corner
cases of importance.
this worked well in knocking out bugs for postings, so the whole scheme was
duplicated to docvalues, storedfields, vectors, everything.
Now here comes this commit, and these tests (which are important to ensure the
index format is working) are no longer testing what they are supposed to. For
example the tests in Lucene41 package explicitly test special cases of that
codec that would otherwise be extraordinarily rare in the existing random
tests. If they are executing against random codecs or even random
configurations then they just became useless.
So thats why I'm concerned: I see this commit causing these failures, and I
know we just experienced a significant loss of test coverage to the index
format. We are relying upon tests to *fail* to detect this, but unfortunately
'loss of test coverage' doesn't always trigger a jenkins build.
So maybe instead of playing whack-a-mole with jenkins tests failures, we should
pay more attention reviewing *all* changes to unit tests where clone() was
previously used. The patch is buggy here, and I just want to ensure its taken
seriously so that we do not lose coverage
> Remove IndexWriterConfig.clone
> ------------------------------
>
> Key: LUCENE-5708
> URL: https://issues.apache.org/jira/browse/LUCENE-5708
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/index
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: 4.9, 5.0
>
> Attachments: LUCENE-5708.patch, LUCENE-5708.patch, LUCENE-5708.patch
>
>
> We originally added this clone to allow a single IWC to be re-used against
> more than one IndexWriter, but I think this is a mis-feature: it adds
> complexity to hairy classes (merge policy/scheduler, DW thread pool, etc.), I
> think it's buggy today.
> I think we should just disallow sharing: you must make a new IWC for a new
> IndexWriter.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
