fix: Remove broad exception handling per user request

- Remove try/except block that was catching all exceptions in get_json_schema()
- Update test_inferred_schema_loader_handles_errors to verify errors now propagate
- This allows actual errors to surface for debugging instead of being silently suppressed
- Addresses PR comment from @aaronsteers

Co-Authored-By: AJ Steers <[email protected]>

Author: devin-ai-integration[bot]

airbyte_cdk/sources/declarative/schema/inferred_schema_loader.py

@@ -80,31 +80,28 @@ def get_json_schema(self) -> Mapping[str, Any]:
         schema_inferrer = SchemaInferrer()
 
         record_count = 0
-        try:
-            for stream_slice in self.retriever.stream_slices():
-                for record in self.retriever.read_records(
-                    records_schema={}, stream_slice=stream_slice
-                ):
-                    if record_count >= self.record_sample_size:
-                        break
-
-                    # Convert all Mapping-like and Sequence-like objects to plain Python types
-                    # This is necessary because genson doesn't handle custom implementations properly
-                    record = _to_builtin_types(record)
-
-                    airbyte_record = AirbyteRecordMessage(
-                        stream=self.stream_name,
-                        data=record,  # type: ignore[arg-type]
-                        emitted_at=0,
-                    )
-
-                    schema_inferrer.accumulate(airbyte_record)
-                    record_count += 1
-
+        for stream_slice in self.retriever.stream_slices():
+            for record in self.retriever.read_records(
+                records_schema={}, stream_slice=stream_slice
+            ):
                 if record_count >= self.record_sample_size:
                     break
-        except Exception:
-            return {}
+
+                # Convert all Mapping-like and Sequence-like objects to plain Python types
+                # This is necessary because genson doesn't handle custom implementations properly
+                record = _to_builtin_types(record)
+
+                airbyte_record = AirbyteRecordMessage(
+                    stream=self.stream_name,
+                    data=record,  # type: ignore[arg-type]
+                    emitted_at=0,
+                )
+
+                schema_inferrer.accumulate(airbyte_record)
+                record_count += 1
+
+            if record_count >= self.record_sample_size:
+                break
 
         inferred_schema: Optional[Mapping[str, Any]] = schema_inferrer.get_stream_schema(
             self.stream_name

unit_tests/sources/declarative/schema/test_inferred_schema_loader.py

@@ -170,7 +170,7 @@ def test_inferred_schema_loader_respects_sample_size():
 
 
 def test_inferred_schema_loader_handles_errors():
-    """Test that InferredSchemaLoader handles errors gracefully."""
+    """Test that InferredSchemaLoader propagates errors from the retriever."""
     retriever = MagicMock()
     retriever.stream_slices.return_value = iter([None])
     retriever.read_records.side_effect = Exception("API Error")
@@ -185,9 +185,8 @@ def test_inferred_schema_loader_handles_errors():
         stream_name="users",
     )
 
-    schema = loader.get_json_schema()
-
-    assert schema == {}
+    with pytest.raises(Exception, match="API Error"):
+        loader.get_json_schema()
 
 
 def test_inferred_schema_loader_with_nested_objects():