[BREAKING] FIX Reject empty dataset filter criteria

pyrit/datasets/seed_datasets/seed_metadata.py

@@ -269,7 +269,9 @@ def _validate(self) -> None:
         # over criteria in _match_filter_to_metadata vacuously False, so the filter
         # would silently match no dataset at all.
         if not self.criteria:
-            raise ValueError("'criteria' must contain at least one metadata criterion.")
+            raise ValueError(
+                "'criteria' must contain at least one criterion. Omit 'criteria' for an unconstrained filter."
+            )
 
         # An empty set is not a filter. Without strict_match nothing can overlap with it,
         # and with strict_match nothing can be outside it, so the same filter would match

tests/unit/datasets/test_seed_dataset_metadata.py

@@ -375,3 +375,18 @@ def test_tags_values(self):
     def test_harm_categories_values(self):
         f = SeedDatasetFilter(harm_categories={"violence", "cybercrime"})
         assert "violence" in f.criteria[0].harm_categories
+
+
+@pytest.mark.parametrize("strict_match", [False, True])
+def test_empty_criteria_raises(strict_match: bool) -> None:
+    with pytest.raises(ValueError, match="at least one criterion"):
+        SeedDatasetFilter(criteria=[], strict_match=strict_match)
+
+
+def test_empty_criteria_with_kwargs_preserves_conflicting_arguments_error() -> None:
+    with pytest.raises(ValueError, match="Cannot pass both"):
+        SeedDatasetFilter(criteria=[], tags={"safety"})
+
+
+def test_unconstrained_criterion_preserves_default_filter() -> None:
+    assert SeedDatasetFilter(criteria=[SeedDatasetMetadata()]).criteria == SeedDatasetFilter().criteria