[
https://issues.apache.org/jira/browse/LUCENE-7456?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15526416#comment-15526416
]
Julien MASSENET commented on LUCENE-7456:
-----------------------------------------
I agree that being this sneaky is not ideal, but the only alternative I see
would be change the Postings API.
In this patch, I tried to keep modifications constrained to the
{{org.apache.lucene.codecs.perfield}} package, but I can give a shot at trying
to a come up with a cleaner implementation that updates the other APIs if you
want.
I've uploaded an updated version of the patch which does not remove the
sneakiness but makes the {{PerFieldMergeState}} more robust:
* The {{FilterFieldInfos}} class now correctly computes and exposes the
{{hasXXX}} properties.
* Calls to {{FilterFieldInfos.fieldInfo()}} and
{{FilterFieldsProducer.terms()}} now throw {{IllegalArgumentException}} if
called with fields outside if the current merge context
* I've tweaked the unit tests to work with the latest 6.2.1 since the
{{Legacy*}} field types are not accessible in this module anymore.
As for why we're overriding merge() calls in our codec:
* We're using a customized codec to emulate nested documents.
* It works with the same idea as BlockJoin, but is less generic (it's tailored
to our use case).
* The main difference is that the maxDoc() for segments remain equal to the
number of "parent" documents, with only the nested fields having larger posting
lists.
* For it to work, when merging, we need to rebase correctly the "docIds" for
the nested documents (using the same idea as the docMap in the general use
case).
> PerField(DocValues|Postings)Format do not call the per-field merge methods
> --------------------------------------------------------------------------
>
> Key: LUCENE-7456
> URL: https://issues.apache.org/jira/browse/LUCENE-7456
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/codecs
> Affects Versions: 6.2.1
> Reporter: Julien MASSENET
> Attachments: LUCENE-7456-v2.patch, LUCENE-7456.patch
>
>
> While porting some old codec code from Lucene 4.3.1, I couldn't get the
> per-field formats to call upon the per-field merge methods; the default merge
> method was always being called.
> I think this is a side-effect of LUCENE-5894.
> Attached is a patch with a test that reproduces the error and an associated
> fix that pass the unit tests.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
