[
https://issues.apache.org/jira/browse/LUCENE-9444?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17192918#comment-17192918
]
Michael McCandless commented on LUCENE-9444:
--------------------------------------------
Overall the patch looks good! Some small feedback:
* Shouldn't {{FacetLabelReader}} be public so callers can interact with it?
This is the danger of putting unit tests in the same package as the classes
they are testing...
* I think {{FacetLabelReader}} is not thread safe (since {{decodedOrds}} is
reused)? So, each thread must pull its own instance? Can you add javadocs
explaining the thread safety?
* Instead of returning {{null}} if caller calls {{next()}} when {{hasNext()}}
had returned {{false}}, can you throw an exception?
* I think the caller must provide {{docId}} in non-decreasing order every
time? Can you 1) add unit test confirming you get an exception if you break
that? And 2) advertise this requirement in the javadocs (for both
{{lookupLabels}} methods)?
* Could you add {{@lucene.experimental}} to the class level javadocs? This
notifies the caller that these APIs are allowed to suddenly change, even in
feature release.
* Shouldn't the comparison be {{== INVALID_ORDINAL}} not {{<=}}?
* I think {{hasNext}} might incorrectly return {{true}} when there are in fact
no more labels, for the variant of {{lookupLabels}} taking a
{{facetDimension}}, when all remaining ordinals are from a different dimension?
Maybe 1) add a unit test showing this issue (failing on the current patch),
then 2) fix it, and 3) confirm the test then passes? {{Iterator}} is annoying
because of this {{hasNext}}/{{next}} dynamic!
* Maybe, instead of {{Iterator}}, we should simply return
{{List<FacetLabel>}}? Or, maybe we just do {{nextFacetLabel}} (and it can
return {{null}} when done)?
> Need an API to easily fetch facet labels for a field in a document
> ------------------------------------------------------------------
>
> Key: LUCENE-9444
> URL: https://issues.apache.org/jira/browse/LUCENE-9444
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/facet
> Affects Versions: 8.6
> Reporter: Ankur
> Priority: Major
> Labels: facet
> Attachments: LUCENE-9444.patch
>
>
> A facet field may be included in the list of fields whose values are to be
> returned for each hit.
> In order to get the facet labels for each hit we need to
> # Create an instance of _DocValuesOrdinalsReader_ and invoke
> _getReader(LeafReaderContext context)_ method to obtain an instance of
> _OrdinalsSegmentReader()_
> # _OrdinalsSegmentReader.get(int docID, IntsRef ordinals)_ method is then
> used to fetch and decode the binary payload in the document's BinaryDocValues
> field. This provides the ordinals that refer to facet labels in the
> taxonomy.**
> # Lastly TaxonomyReader.getPath(ord) is used to fetch the labels to be
> returned.
>
> Ideally there should be a simple API - *String[] getLabels(docId)* that hides
> all the above details and gives us the string labels. This can be part of
> *TaxonomyFacets* but that's just one idea.
> I am opening this issue to get community feedback and suggestions.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
