v2/pkg/engine/datasource/grpc_datasource/execution_plan_requires_test.go (1)

1807-1819: Add one direct-interface @requires case.

These new cases only exercise fragment-local selections, so a regression that drops shared interface fields like name or __typename from the composite request would still pass. Please add one case like primaryItem { name ... on PalletItem { palletCount } ... on ContainerItem { containerSize } } so the planner path for common interface fields is covered too.

Based on learnings: composite @requires handling carries common interface fields through fragment-based selections.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@v2/pkg/engine/datasource/grpc_datasource/execution_plan_requires_test.go`
around lines 1807 - 1819, Add a test case that exercises a direct-interface
`@requires` path which preserves common interface fields in fragment-local
selections: in execution_plan_requires_test.go add a new case alongside
"requires with interface type in selection set" that uses the same query
structure but sets the federationConfigs SelectionSet for the
Storage.itemInfo/primaryItem to "primaryItem { name ... on PalletItem {
palletCount } ... on ContainerItem { containerSize } }" (or the equivalent
selection used in the comment) so the planner is forced to carry through the
shared interface field "name" and __typename across the composite request;
update the test vector (the case name and mapping/testMapping()) accordingly so
this new case runs with the existing test harness.

v2/pkg/engine/datasource/grpc_datasource/grpc_datasource_federation_test.go (1)

1724-1810: Please add one nested abstract-path datasource case.

The planner suite now covers nested shapes like securitySummary, itemHandlerInfo, itemSpecsInfo, and deepItemInfo, but this end-to-end suite only runs the flat itemInfo / operationReport paths. A single nested case here would protect the compiler + JSON execution path that is most likely to regress.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@v2/pkg/engine/datasource/grpc_datasource/grpc_datasource_federation_test.go`
around lines 1724 - 1810, Add one nested abstract-path test case to the existing
testCases slice: create a case named like "Query Storage with deepItemInfo
nested abstract path" that queries _entities for Storage returning id, name and
a nested abstract field (use deepItemInfo or securitySummary/itemHandlerInfo as
the nested path) and supply vars with representations that include the nested
__typename structures (e.g., Storage -> deepItemInfo -> primaryItem with
PalletItem/ContainerItem). In federationConfigs add the base Storage selection
(id) and a FieldName entry for the nested abstract field with a SelectionSet
that drills into the nested object and includes inline fragments for concrete
types (e.g., primaryItem { ... on PalletItem { name palletCount } ... on
ContainerItem { name containerSize } }). Implement validate to assert _entities
length, Storage id/name and that the nested field string is formatted as
expected, and set validateError to require.Empty; reuse existing pattern from
the "itemInfo" and "operationReport" cases so test structure and names
(testCases, federationConfigs, validate, validateError) match the surrounding
tests.

v2/pkg/engine/datasource/grpc_datasource/compiler.go (1)

1131-1147: Consider logging when oneof resolution falls back to empty message.

When FragmentFields[typeName] is not found (line 1139) or findOneOfFieldDescriptor returns nil (line 1145), the function returns an empty message without any indication. This could make debugging schema misconfigurations difficult in production.

💡 Optional: Add debug logging for silent fallbacks

If debugging visibility is desired, consider adding optional logging when these fallbacks occur:

 	fragmentFields, ok := rpcMessage.FragmentFields[typeName]
 	if !ok {
+		// Consider: log.Debug("no fragment fields found for typename", "typename", typeName)
 		return message, nil
 	}

 	// Find the matching oneof field descriptor for the expected concrete type.
 	oneofFieldDesc := p.findOneOfFieldDescriptor(inputMessage.Desc.Oneofs(), protoref.Name(rpcMessage.OneOfType.FieldName()), protoref.Name(typeName))
 	if oneofFieldDesc == nil {
+		// Consider: log.Debug("no oneof field descriptor found", "typename", typeName)
 		return message, nil
 	}

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@v2/pkg/engine/datasource/grpc_datasource/compiler.go` around lines 1131 -
1147, In buildOneOfMessage on RPCCompiler, add debug/error logging when the
function currently silently returns an empty message: log when
rpcMessage.FragmentFields[typeName] is missing (include typeName and rpcMessage
identity) and when findOneOfFieldDescriptor(...) returns nil (include typeName
and rpcMessage.OneOfType.FieldName()); ensure logs give enough context to trace
schema/resolution issues (use existing logger/ctx in this package) before
returning the empty message.