Repository navigation
Columnar faceting: expose the facets of a result set in both packages #290
Description
Part of #289 (columnar faceting in the Python bindings).
Scope
- Expose the facets of a result set in both packages: an accessor on the result set (for example
result_set.facets) returning the engine'sFacetResultsas Python data:meta(exact,patched,lag_ms) and one entry per facet alias withtype,source, optionalreason,buckets(valueorkey,from,to,count, plus themissinganddefaultflags),other_countfor terms andmissing. - Facets that are
unavailable(with areason) are surfaced, not dropped. - Non-finite numbers in facet values (
NaN,+/-Infinity) decode to the Python floats, as rows do. - Native package: check how the
ctypesbridge exposes a result set and add what is needed to read the facets (for example one entry point returning the facets as JSON). The JVM package readsgetFacets()through JPype. - Input parameters in
RANGESandINCLUDEwork fromquery(...)andcommand(...).
Tests
TERMS, RANGES, a calendar HISTOGRAM, a disjunctive FILTER and an unavailable facet, in the JVM and native lanes, comparing against the same query's GROUP BY counts.
Activity
tae898 commentedon Nov 6, 2026
I will start here after the CI change. For the native package,
ArcadeNative.batchOfcloses the cursor when the last batch is read, so I will add the facets to that last reply as JSON fromrs.getFacets(), and make the lazyPrefetchedcursors passgetFacets()on (without that every facet would be silently lost). NaN and the infinities will travel in the same tag rows use, throughFacetResults.toJSON(ValueEncoder), so they decode to Python floats. Forarcadedb-jvmI will readgetFacets()through a bridge call.The accessor is
result_set.facets, withmeta(exact,patched,lag_ms) and one entry per alias,unavailableones included. I follow the engine's rule: rows first, and asking for facets before the rows are read ends the rows. Tell me if you want a different behaviour.Two things I will check first: whether
getFacets()still works after the engine closes the result set, and whether the native image needs the new parser classes (FacetsClause,FacetItem,FacetRangeItem) in its reflection metadata. Both go into the tests.Facets: promise only what both packages guarantee; pick the type-leve…