Use ast.literal_eval instead of eval in hellaswag_arabic_pfn

src/lighteval/tasks/multilingual/tasks/arabic.py

@@ -17,6 +17,7 @@
 paper:
 """
 
+import ast
 import random
 import re
 from string import ascii_uppercase
@@ -600,7 +601,7 @@ def copa_arabic_pfn(line, task_name: str = None):
 def hellaswag_arabic_pfn(line, task_name: str = None):
     ctx = re.sub(r"\[.*?\]", "", line["ctx"])  # Remove latin words within brackets
     endings = [
-        re.sub(r"\[.*?\]", "", e) for e in eval(line["endings"])
+        re.sub(r"\[.*?\]", "", e) for e in ast.literal_eval(line["endings"])
     ]  # endings is a string representation of a list
     answer_index = line["label"]
     instruction = "بناء على السياق التالي، اختر النهاية الصحيحة من الاقتراحات التالية"

tests/unit/tasks/test_arabic_tasks.py

@@ -0,0 +1,65 @@
+# MIT License
+
+# Copyright (c) 2024 The HuggingFace Team
+
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+import pytest
+
+from lighteval.tasks.multilingual.tasks.arabic import hellaswag_arabic_pfn
+
+
+def test_hellaswag_arabic_pfn_parses_endings():
+    line = {
+        "ctx": "sentence one [latin] and more",
+        "endings": "['ending one [x]', 'ending two']",
+        "label": 1,
+    }
+
+    doc = hellaswag_arabic_pfn(line, "test_task")
+
+    assert doc.choices == ["ending one ", "ending two"]
+    assert doc.gold_index == 1
+
+
+def test_hellaswag_arabic_pfn_rejects_non_literal_endings():
+    # endings comes from a hub dataset; anything that is not a plain literal
+    # must raise instead of being evaluated. literal_eval raises ValueError on
+    # non-literal nodes but SyntaxError on malformed input (e.g. a truncated
+    # field), so both count as rejection.
+    line = {
+        "ctx": "sentence",
+        "endings": "__import__('os').system('echo pwned')",
+        "label": 0,
+    }
+
+    with pytest.raises((ValueError, SyntaxError)):
+        hellaswag_arabic_pfn(line, "test_task")
+
+
+def test_hellaswag_arabic_pfn_does_not_execute_endings():
+    # pytest.raises alone would also pass if something ran before raising;
+    # the side effect pins down that nothing in the field executes at all
+    executed = []
+    line = {"ctx": "s", "endings": "[executed.append(1)]", "label": 0}
+
+    with pytest.raises((ValueError, SyntaxError)):
+        hellaswag_arabic_pfn(line, "test_task")
+
+    assert executed == []