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

lamindb/__init__.py

@@ -182,6 +182,7 @@
     DB,
 )
 from .models.save import save
+from .models.sqlrecord import UNSET
 from . import core
 from . import integrations
 from . import curators

lamindb/models/feature.py

@@ -1221,7 +1221,7 @@ def __init__(
         | Registry
         | list[Registry]
         | FieldAttr,
-        type: Feature | None = None,
+        type: Feature | None = UNSET,
         is_type: bool = False,
         unit: str | None = None,
         description: str | None = None,

lamindb/models/project.py

@@ -28,7 +28,7 @@
 from .record import Record
 from .run import Run, TracksRun, TracksUpdates, User
 from .schema import Schema
-from .sqlrecord import BaseSQLRecord, HasType, IsLink, SQLRecord, ValidateFields
+from .sqlrecord import UNSET, BaseSQLRecord, HasType, IsLink, SQLRecord, ValidateFields
 from .transform import Transform
 from .ulabel import ULabel
 
@@ -217,7 +217,7 @@ class Meta(SQLRecord.Meta, TracksRun.Meta, TracksUpdates.Meta):
     def __init__(
         self,
         name: str,
-        type: Reference | None = None,
+        type: Reference | None = UNSET,
         is_type: bool = False,
         abbr: str | None = None,
         url: str | None = None,
@@ -449,7 +449,7 @@ class Meta(SQLRecord.Meta, TracksRun.Meta, TracksUpdates.Meta):
     def __init__(
         self,
         name: str,
-        type: Project | None = None,
+        type: Project | None = UNSET,
         is_type: bool = False,
         abbr: str | None = None,
         url: str | None = None,

lamindb/models/record.py

@@ -873,7 +873,7 @@ class Meta(SQLRecord.Meta, TracksRun.Meta, TracksUpdates.Meta):
     def __init__(
         self,
         name: str | None = None,
-        type: Record | None = None,
+        type: Record | None = UNSET,
         is_type: bool = False,
         features: dict[str | Feature, Any] | None = None,
         description: str | None = None,

lamindb/models/schema.py

@@ -631,7 +631,7 @@ def __init__(
         name: str | None = None,
         description: str | None = None,
         itype: str | Registry | FieldAttr | None = None,
-        type: Schema | None = None,
+        type: Schema | None = UNSET,
         is_type: bool = False,
         index: Feature | None = None,
         flexible: bool | None = None,
@@ -738,6 +738,7 @@ def __init__(
             coerce=coerce_dtype,
             n_features=n_features,
         )
+        # pop before update_attributes/super so it never reaches Django fields or getattr
         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"
@@ -858,7 +859,7 @@ def _validate_kwargs_calculate_hash(
         validated_kwargs = {
             "name": name,
             "description": description,
-            "type": None if type is UNSET else type,
+            "type": type,
             "is_type": is_type,
             "_dtype_str": dtype,
             "otype": otype,

lamindb/models/sqlrecord.py

@@ -451,7 +451,7 @@ def init_self_from_db(
 
 def update_attributes(record: SQLRecord, attributes: dict[str, str]):
     for key, value in attributes.items():
-        if getattr(record, key) != value and value is not None:
+        if value is not None and value is not UNSET and getattr(record, key) != value:
             if key not in {"uid", "_dtype_str", "otype", "hash"}:
                 logger.warning(f"updated {key} from {getattr(record, key)} to {value}")
                 setattr(record, key, value)
@@ -624,10 +624,18 @@ 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:
+        # "type" is always present in kwargs at this point (Solution A contract)
+        type = kwargs["type"]
+        if type is UNSET:
+            # user passed nothing → search all type contexts
+            # catches typed records with same name, fixes silent dup bug
+            subset = record.__class__.filter()
+        elif type is None:
+            # explicit type=None → root-level dedup (type IS NULL)
             subset = record.__class__.filter(type__isnull=True)
         else:
-            subset = record.__class__.filter(type=kwargs["type"])
+            # specific type object → scoped dedup within that type
+            subset = record.__class__.filter(type=type)
     else:
         subset = record.__class__
     exact_match = subset.filter(**{name_field: kwargs[name_field]}).first()
@@ -1162,10 +1170,6 @@ 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
-        # `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 not args:
 
             def resolve_fk_or_id(field_name: str) -> bool:
@@ -1213,7 +1217,19 @@ def resolve_fk_or_id(field_name: str) -> bool:
                     # `kwargs["branch"]` is now always a non-None Branch (explicit or
                     # the current one), so the record is created on that branch.
                     kwargs["created_on"] = kwargs["branch"]
+            # HasType models like Record, ULabel, Feature, Schema each explicitly
+            # pop and re-inject "type" (defaulting to UNSET) in their own __init__,
+            # so "type" is already in kwargs for them.
+            # Project and Reference pass kwargs straight through without touching "type",
+            # so we inject UNSET here to uphold the contract that kwargs["type"] is
+            # always readable — and so the strip below can use kwargs["type"] directly.
+            if isinstance(self, HasType) and "type" not in kwargs:
+                kwargs["type"] = UNSET
+
             if skip_validation:
+                # strip UNSET just before Django sees kwargs — FK descriptors reject non-model values
+                if isinstance(self, HasType) and kwargs["type"] is UNSET:
+                    kwargs.pop("type")
                 super().__init__(**kwargs)
             else:
                 from ..core._settings import settings
@@ -1270,6 +1286,9 @@ def resolve_fk_or_id(field_name: str) -> bool:
                             # track original values after replacing with the existing record
                             self._populate_tracked_fields()
                             return None
+                # strip UNSET just before Django sees kwargs — FK descriptors reject non-model values
+                if isinstance(self, HasType) and kwargs["type"] is UNSET:
+                    kwargs.pop("type")
                 super().__init__(**kwargs)
                 if isinstance(self, ValidateFields):
                     # this will trigger validation against django validators

lamindb/models/ulabel.py

@@ -224,7 +224,7 @@ class Meta(SQLRecord.Meta, TracksRun.Meta, TracksUpdates.Meta):
     def __init__(
         self,
         name: str,
-        type: ULabel | None = None,
+        type: ULabel | None = UNSET,
         is_type: bool = False,
         description: str | None = None,
         reference: str | None = None,

tests/pydata/test_record_basics.py

@@ -690,14 +690,24 @@ 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
+    # explicit type=None → root-level dedup: label1 is typed so not found → new record
+    label_new = ln.Record(name="label 1", type=None)
+    assert label_new != label1
+    assert label_new._state.adding  # not yet saved, truly a new record
+    # explicit type=None, root-level record exists → root-level dedup finds it → returns it
+    root_label = ln.Record(name="root label 1").save()
+    label3 = ln.Record(name="root label 1", type=None)
+    assert label3 == root_label
+    # no type passed (UNSET) → search all → finds the existing root-level record
+    label4 = ln.Record(name="root label 1")
+    assert label4 == root_label
+    root_label.delete(permanent=True)
     label1.delete(permanent=True)
     my_type.delete(permanent=True)
 
@@ -1194,7 +1204,7 @@ 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_2").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": UNSET}
     )
     assert not suggest_records_with_similar_names(
-        record2, "name", {"name": "Test experiment 123"}
+        record2, "name", {"name": "Test experiment 123", "type": UNSET}
     )
 
     queryset = _search(