nwb-linkml/nwb_linkml/tests/test_adapters/test_adapter_namespaces.py

163 lines
5.7 KiB
Python
Raw Normal View History

from pathlib import Path
import pytest
from nwb_linkml.adapters import NamespacesAdapter, SchemaAdapter
from nwb_schema_language import Attribute, Dataset, FlatDtype, Group, Namespace, Namespaces, Schema
2024-07-02 04:23:31 +00:00
@pytest.mark.parametrize(
2024-07-02 04:23:31 +00:00
["class_name", "schema_file", "namespace_name"],
[
2024-07-02 04:23:31 +00:00
("DynamicTable", "table.yaml", "hdmf-common"),
("Container", "base.yaml", "hdmf-common"),
("TimeSeries", "nwb.base.yaml", "core"),
("ImageSeries", "nwb.image.yaml", "core"),
],
)
def test_find_type_source(nwb_core_fixture, class_name, schema_file, namespace_name):
defining_sch = nwb_core_fixture.find_type_source(class_name)
assert defining_sch.path.name == schema_file
assert namespace_name == defining_sch.namespace
def test_populate_imports(nwb_core_fixture):
nwb_core_fixture._populate_imports()
schema: SchemaAdapter
assert len(nwb_core_fixture.schemas) > 0
for schema in nwb_core_fixture.schemas:
2024-07-02 04:23:31 +00:00
need_imports = [
nwb_core_fixture.find_type_source(cls.neurodata_type_def).namespace
for cls in schema.created_classes
if cls.neurodata_type_inc is not None
]
need_imports = [i for i in need_imports if i != schema.namespace]
for i in need_imports:
assert i in schema.imports
def test_build(nwb_core_fixture):
pass
2024-07-02 04:23:31 +00:00
def test_skip_imports(nwb_core_fixture):
"""
We can build just the namespace in question without also building the other namespaces that it imports
"""
res = nwb_core_fixture.build(skip_imports=True)
# we shouldn't have any of the hdmf-common schema in with us
2024-07-02 04:23:31 +00:00
namespaces = [sch.annotations["namespace"].value for sch in res.schemas]
assert all([ns == "core" for ns in namespaces])
def test_roll_down_inheritance():
"""
Classes should receive and override the properties of their parents
when they have neurodata_type_inc
Args:
nwb_core_fixture:
Returns:
"""
parent_cls = Group(
neurodata_type_def="Parent",
doc="parent",
attributes=[
Attribute(name="a", dims=["a", "b"], shape=[1, 2], doc="a", value="a"),
Attribute(name="b", dims=["c", "d"], shape=[3, 4], doc="b", value="b"),
],
datasets=[
Dataset(
name="data",
dims=["a", "b"],
shape=[1, 2],
doc="data",
attributes=[
Attribute(name="c", dtype=FlatDtype.int32, doc="c"),
Attribute(name="d", dtype=FlatDtype.int32, doc="d"),
],
)
],
)
parent_sch = Schema(source="parent.yaml")
parent_ns = Namespaces(
namespaces=[
Namespace(
author="hey",
contact="sup",
name="parent",
doc="a parent",
version="1",
schema=[parent_sch],
)
]
)
child_cls = Group(
neurodata_type_def="Child",
neurodata_type_inc="Parent",
doc="child",
attributes=[Attribute(name="a", doc="a", value="z")],
datasets=[
Dataset(
name="data",
doc="data again",
attributes=[Attribute(name="c", doc="c", value="z"), Attribute(name="e", doc="e")],
),
],
groups=[Group(name="untyped_child", neurodata_type_inc="Parent", doc="untyped child")],
)
child_sch = Schema(source="child.yaml")
child_ns = Namespaces(
namespaces=[
Namespace(
author="hey",
contact="sup",
name="child",
doc="a child",
version="1",
schema=[child_sch, Schema(namespace="parent")],
)
]
)
parent_schema_adapter = SchemaAdapter(path=Path("parent.yaml"), groups=[parent_cls])
parent_ns_adapter = NamespacesAdapter(namespaces=parent_ns, schemas=[parent_schema_adapter])
child_schema_adapter = SchemaAdapter(path=Path("child.yaml"), groups=[child_cls])
child_ns_adapter = NamespacesAdapter(
namespaces=child_ns, schemas=[child_schema_adapter], imported=[parent_ns_adapter]
)
child_ns_adapter.complete_namespaces()
child = child_ns_adapter.get("Child")
# overrides simple attrs
assert child.doc == "child"
# gets unassigned parent attrs
assert "b" in [attr.name for attr in child.attributes]
# overrides values while preserving remaining values when set
attr_a = [attr for attr in child.attributes if attr.name == "a"][0]
assert attr_a.value == "z"
assert attr_a.dims == parent_cls.attributes[0].dims
assert [attr.value for attr in child.attributes if attr.name == "a"][0] == "z"
# preserve unset values in child datasets
assert child.datasets[0].dtype == parent_cls.datasets[0].dtype
assert child.datasets[0].dims == parent_cls.datasets[0].dims
# gets undeclared attrs in child datasets
assert "d" in [attr.name for attr in child.datasets[0].attributes]
# overrides set values in child datasets while preserving unset
c_attr = [attr for attr in child.datasets[0].attributes if attr.name == "c"][0]
assert c_attr.value == "z"
assert c_attr.dtype == FlatDtype.int32
# preserves new attrs
assert "e" in [attr.name for attr in child.datasets[0].attributes]
# neurodata_type_def is not included in untyped children
assert child.groups[0].neurodata_type_def is None
# we don't set any of the attrs from the parent class here because we don't override them,
# so we don't need to merge them, and we don't want to clutter our linkml models unnecessarily
assert child.groups[0].attributes is None