fix: allow eval filtering with EvalCase objects

py/src/braintrust/framework.py

@@ -9,7 +9,7 @@
 import traceback
 import warnings
 from collections import defaultdict
-from collections.abc import Awaitable, Callable, Coroutine, Iterable, Iterator, Sequence
+from collections.abc import Awaitable, Callable, Coroutine, Iterable, Iterator, Mapping, Sequence
 from concurrent.futures import ThreadPoolExecutor
 from contextlib import contextmanager
 from multiprocessing import cpu_count
@@ -1154,7 +1154,10 @@ def parse_filters(filters: list[str]) -> list[Filter]:
 def evaluate_filter(object, filter: Filter):
     key = object
     for p in filter.path:
-        key = key.get(p)
+        if isinstance(key, Mapping):
+            key = key.get(p)
+        else:
+            key = getattr(key, p, None)
         if key is None:
             return False
     return filter.pattern.match(serialize_json_with_plain_string(key)) is not None

py/src/braintrust/test_framework.py

@@ -1,4 +1,5 @@
 import importlib.util
+import re
 from typing import List
 from unittest.mock import MagicMock
 
@@ -11,6 +12,9 @@
     EvalHooks,
     EvalResultWithSummary,
     Evaluator,
+    Filter,
+    evaluate_filter,
+    parse_filters,
     run_evaluator,
 )
 from .score import Score, Scorer
@@ -626,3 +630,77 @@ async def test_run_evaluator_empty_dataset_warns(capsys):
     captured = capsys.readouterr()
     assert "Warning" in captured.err
     assert "empty" in captured.err.lower()
+
+
+class TestEvaluateFilter:
+    """Regression tests for https://github.com/braintrustdata/braintrust-sdk-python/issues/207."""
+
+    @pytest.mark.parametrize(
+        "datum",
+        [
+            {"input": "hello", "metadata": {"name": "foo"}},
+            EvalCase(input="hello", metadata={"name": "foo"}),
+        ],
+        ids=["dict", "evalcase"],
+    )
+    def test_evaluate_filter_match(self, datum):
+        f = Filter(path=["metadata", "name"], pattern=re.compile("foo"))
+        assert evaluate_filter(datum, f) is True
+
+    @pytest.mark.parametrize(
+        "datum",
+        [
+            {"input": "hello", "metadata": {"name": "bar"}},
+            EvalCase(input="hello", metadata={"name": "bar"}),
+        ],
+        ids=["dict", "evalcase"],
+    )
+    def test_evaluate_filter_no_match(self, datum):
+        f = Filter(path=["metadata", "name"], pattern=re.compile("foo"))
+        assert evaluate_filter(datum, f) is False
+
+    @pytest.mark.parametrize(
+        "datum",
+        [
+            {"input": "hello"},
+            EvalCase(input="hello"),
+        ],
+        ids=["dict", "evalcase"],
+    )
+    def test_evaluate_filter_missing_key(self, datum):
+        f = Filter(path=["metadata", "name"], pattern=re.compile("foo"))
+        assert evaluate_filter(datum, f) is False
+
+    def test_evaluate_filter_nested_metadata(self):
+        datum = EvalCase(input="hello", metadata={"priority": "P0", "owner": "alice"})
+        f = Filter(path=["metadata", "priority"], pattern=re.compile("^P0$"))
+        assert evaluate_filter(datum, f) is True
+
+    def test_evaluate_filter_input_field(self):
+        datum = EvalCase(input={"text": "hello world"}, metadata={"name": "foo"})
+        f = Filter(path=["input", "text"], pattern=re.compile("hello"))
+        assert evaluate_filter(datum, f) is True
+
+    @pytest.mark.asyncio
+    async def test_run_evaluator_with_filter_and_evalcase(self):
+        data = [
+            EvalCase(input="hello", metadata={"name": "foo"}),
+            EvalCase(input="world", metadata={"name": "bar"}),
+        ]
+
+        evaluator = Evaluator(
+            project_name="test-project",
+            eval_name="test-filter-evalcase",
+            data=data,
+            task=lambda x: x,
+            scores=[],
+            experiment_name=None,
+            metadata=None,
+        )
+
+        filters = parse_filters(["metadata.name=foo"])
+        result = await run_evaluator(experiment=None, evaluator=evaluator, position=None, filters=filters)
+
+        # Only the "foo" case should pass the filter
+        assert len(result.results) == 1
+        assert result.results[0].input == "hello"