airborne12 commented on code in PR #66449:
URL: https://github.com/apache/doris/pull/66449#discussion_r3719079134
##########
be/src/exprs/vsearch.cpp:
##########
@@ -60,6 +58,58 @@ void add_search_binding_diagnostic(const IndexExecContext*
index_context,
}
}
+Status collect_slot_search_input(const VSearchExpr& expr, const VSlotRef&
slot_ref,
+ const TSearchFieldBinding* binding,
+ IndexExecContext* index_context,
SearchInputBundle* bundle) {
+ DCHECK(index_context != nullptr);
+ DCHECK(bundle != nullptr);
+
+ // VSlotRef::column_id() is the scan-schema position used by
IndexExecContext.
+ const int column_index = slot_ref.column_id();
+ const std::string field_name =
+ binding != nullptr ? binding->field_name : slot_ref.column_name();
+ const bool is_variant_subcolumn = binding != nullptr &&
binding->__isset.is_variant_subcolumn &&
+ binding->is_variant_subcolumn;
+
+ bundle->field_name_to_column_id[field_name] = column_index;
+
+ auto* iterator =
index_context->get_inverted_index_iterator_by_column_id(column_index);
+ if (iterator == nullptr) {
+ // For example, `data.items.message` has its own SlotRef in the scan
schema. The
+ // storage layer may inherit index metadata from `data`, but it still
constructs a
+ // child iterator whose stored field name contains the complete
Variant path.
+ if (is_variant_subcolumn) {
+ add_search_binding_diagnostic(
+ index_context,
+ fmt::format("[VariantSearchBinding] phase=collect_inputs "
+ "result=no_iterator logical_field={}
column_index={} "
+ "reason=slot_iterator_missing",
+ field_name, column_index));
+ }
+ return Status::OK();
Review Comment:
**[P1] Preserve UNKNOWN for a missing Variant child iterator**
Returning here leaves `bundle.iterators` empty when this is the only SEARCH
field. The fast path in `VSearchExpr::evaluate_inverted_index()` then stores
empty data and empty null bitmaps, which represents `FALSE`. An outer `NOT
search(...)` complements that result to the entire segment.
This conflicts with the existing missing-Variant invariant in
`FunctionSearch::build_leaf_query()` and
`TestBuildLeafQueryVariantMissingFieldReturnsUnknown`: an unbound Variant field
is `UNKNOWN` (`data = {}`, `null = all segment rows`).
Please keep the child-only physical binding, but allow missing Variant
bindings to reach `FieldReaderResolver`. Both empty-input shortcuts in
`vsearch.cpp` and `function_search.cpp` need to preserve `UNKNOWN`. The new
test should also assert data cardinality `0`, null cardinality `32`, and verify
that an outer `NOT` produces no true rows.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
