🚸 Broaden name deduplication in HasType to search all contexts if no type is passed

lamindb/models/schema.py

@@ -738,6 +738,8 @@ def __init__(
             coerce=coerce_dtype,
             n_features=n_features,
         )
+        # pop before update_attributes/super so it never reaches Django fields or getattr
+        type_val = validated_kwargs.pop("_type_val")
         if not features and not slots and not is_type and not itype:
             raise InvalidArgument(
                 "Please pass features or slots or itype or set is_type=True"
@@ -775,7 +777,7 @@ def __init__(
             validated_kwargs["uid"] = base62_16()
 
         validated_kwargs.update(space_branch_kwargs)
-        super().__init__(**validated_kwargs)
+        super().__init__(**validated_kwargs, _type_val=type_val)
 
     def query_schemas(self) -> QuerySet:
         """Query schemas of sub types.
@@ -859,6 +861,7 @@ def _validate_kwargs_calculate_hash(
             "name": name,
             "description": description,
             "type": None if type is UNSET else type,
+            "_type_val": type,
             "is_type": is_type,
             "_dtype_str": dtype,
             "otype": otype,

lamindb/models/sqlrecord.py

@@ -610,7 +610,7 @@ def get_branch_id_for_create(
 
 
 def suggest_records_with_similar_names(
-    record: SQLRecord, name_field: str, kwargs
+    record: SQLRecord, name_field: str, kwargs, type_val=UNSET
 ) -> SQLRecord | None:
     """Returns a record if found exact match, otherwise None.
 
@@ -624,10 +624,17 @@ def suggest_records_with_similar_names(
     # the below needs to be .first() because there might be multiple records with the same
     # name field in case the record is versioned (e.g. for Transform key)
     if isinstance(record, HasType):
-        if kwargs.get("type", None) is None:
-            subset = record.__class__.filter(type__isnull=True)
-        else:
-            subset = record.__class__.filter(type=kwargs["type"])
+        if type_val is None:
+            # explicit type=None → always create new at root, skip dedup
+            if not kwargs.get("is_type", False):
+                logger.warning(
+                    f"Creating a root-level {record.__class__.__name__.lower()} without"
+                    " a type. In most cases, objects should be created under a type."
+                )
+            return None
+        # UNSET: search all type contexts (catches typed records with same name, fixes silent dup bug)
+        # <object>: search within that type context only
+        subset = record.__class__ if type_val is UNSET else record.__class__.filter(type=type_val)
     else:
         subset = record.__class__
     exact_match = subset.filter(**{name_field: kwargs[name_field]}).first()
@@ -1162,10 +1169,15 @@ class Meta:
 
     def __init__(self, *args, **kwargs):
         skip_validation = kwargs.pop("_skip_validation", False)
-        # strip sentinel before validate_fields and Django's Model.__init__ see it
+        # capture user intent before stripping: UNSET=not passed, None=explicit None, obj=typed
+        # Schema pre-computes this (converts UNSET→None for hashing) and passes it via _type_val
         # `is` never calls __eq__, so FeaturePredicate objects are safe
-        if isinstance(self, HasType) and kwargs.get("type", UNSET) is UNSET:
-            kwargs.pop("type", None)
+        if isinstance(self, HasType):
+            type_val = kwargs.pop("_type_val", kwargs.get("type", UNSET))
+            if type_val is UNSET:
+                kwargs.pop("type", None)
+        else:
+            type_val = UNSET
         if not args:
 
             def resolve_fk_or_id(field_name: str) -> bool:
@@ -1240,7 +1252,7 @@ def resolve_fk_or_id(field_name: str) -> bool:
                 ):
                     name_field = getattr(self, "_name_field", "name")
                     exact_match = suggest_records_with_similar_names(
-                        self, name_field, kwargs
+                        self, name_field, kwargs, type_val
                     )
                     if exact_match is not None:
                         if "version_tag" in kwargs:

tests/pydata/test_record_basics.py

@@ -690,14 +690,17 @@ def test_feature_manager_raise_not_validated_values():
 def test_name_lookup():
     my_type = ln.Record(name="MyType", is_type=True).save()
     label1 = ln.Record(name="label 1", type=my_type).save()
+    # same type → returns existing typed record
     label2 = ln.Record(name="label 1", type=my_type)
     assert label2 == label1
+    # no type passed, only typed record exists → fallback returns the typed one
     label2 = ln.Record(name="label 1")
-    assert label2 != label1
-    label2.save()
-    label3 = ln.Record(name="label 1")
-    assert label3 == label2
-    label2.delete(permanent=True)
+    assert label2 == label1
+    # root-level record exists → no-type search finds root first before typed
+    root_label = ln.Record(name="root label 1").save()
+    label3 = ln.Record(name="root label 1")
+    assert label3 == root_label
+    root_label.delete(permanent=True)
     label1.delete(permanent=True)
     my_type.delete(permanent=True)
 
@@ -1194,7 +1197,7 @@ def test_record_features_add_remove_values():
 
     # test passing ISO-format date string for date
 
-    test_record2 = ln.Record(name="test_record").save()
+    test_record2 = ln.Record(name="test_record", type=None).save()
     # we could also test different ways of formatting but don't yet do that
     # in to_dataframe() we enforce ISO format already
     feature_date = ln.Feature.get(name="feature_date")

tests/pydata/test_sqlrecord.py

@@ -12,6 +12,7 @@
 from lamindb.errors import FieldValidationError
 from lamindb.models import sqlrecord as sqlrecord_module
 from lamindb.models.sqlrecord import (
+    UNSET,
     _get_record_kwargs,
     _search,
     check_key,
@@ -287,10 +288,10 @@ def test_suggest_similar_names():
     assert ln.Record(name="Test experiment 1").uid == record1.uid
 
     assert suggest_records_with_similar_names(
-        record1, "name", {"name": "Test experiment 1"}
+        record1, "name", {"name": "Test experiment 1"}, type_val=UNSET
     )
     assert not suggest_records_with_similar_names(
-        record2, "name", {"name": "Test experiment 123"}
+        record2, "name", {"name": "Test experiment 123"}, type_val=UNSET
     )
 
     queryset = _search(