Merge pull request #65 from B612-Asteroid-Institute/ak/validate-fix

Subtables do not validate during concatenation

quivr/columns.py

@@ -242,7 +242,12 @@ def __get__(self, obj: Optional[tables.Table], objtype: type) -> Union[Self, T]:
         schema = self.schema.with_metadata(metadata)
 
         subtable = pa.Table.from_arrays(array.flatten(), schema=schema)
-        return self.table_type.from_pyarrow(subtable, permit_nulls=self.nullable)
+        # We don't validate the subtable. If the parent table were validated,
+        # then the subtable would have been validated as part of that process.
+        # If the parent table is intentionally not validated, then we don't
+        # want to validate the subtable either as it will throw an error
+        # when accessing the subtable as a column (e.g. during concatenation).
+        return self.table_type.from_pyarrow(subtable, permit_nulls=self.nullable, validate=False)
 
     def _nulls(self, n: int) -> pa.Array:
         """Return an array of nulls of the appropriate size."""

test/test_concat.py

@@ -68,16 +68,31 @@ def test_concatenate_empty_tables():
 
 
 def test_concatenate_no_validate():
+
+    class OtherTable(qv.Table):
+        y = qv.Int64Column(validator=qv.le(0))
+
     class ValidationTable(qv.Table):
         x = qv.Int64Column(validator=qv.ge(0))
+        subtable = OtherTable.as_column(nullable=True)
 
-    t1 = ValidationTable.from_kwargs(x=[-1], validate=False)
-    t2 = ValidationTable.from_kwargs(x=[1], validate=False)
+    valid = ValidationTable.from_kwargs(
+        x=[1], subtable=OtherTable.from_kwargs(y=[-1], validate=False), validate=False
+    )
+    invalid_x = ValidationTable.from_kwargs(x=[-1], subtable=[None], validate=False)
+    invalid_subtable = ValidationTable.from_kwargs(
+        x=[1], subtable=OtherTable.from_kwargs(y=[1], validate=False), validate=False
+    )
 
     with pytest.raises(qv.ValidationError, match="Column x failed validation"):
-        qv.concatenate([t1, t2])
+        qv.concatenate([invalid_x, valid])
+
+    have = qv.concatenate([invalid_x, valid], validate=False)
+    assert len(have) == 2
 
-    have = qv.concatenate([t1, t2], validate=False)
+    # Subtables are not validated during concatenation, as the main table
+    # was initialized with validate=False
+    have = qv.concatenate([valid, invalid_subtable])
     assert len(have) == 2