fix(extract): skip bytes literals in _parse_python_string (#1190)

babel/messages/extract.py

@@ -710,7 +710,19 @@ def _parse_python_string(value: str, encoding: str, future_flags: int) -> str |
     if isinstance(code, ast.Expression):
         body = code.body
         if isinstance(body, ast.Constant):
-            return body.value
+            if isinstance(body.value, str):
+                return body.value
+            if isinstance(body.value, bytes):
+                # A bytes literal used to propagate out of here and crash the
+                # extraction frontend (GH #1190). Warn rather than fail, and
+                # return None so the message is skipped.
+                warnings.warn(
+                    f"Bytes literal {value!r} passed to a gettext function; "
+                    "it will be skipped during message extraction. "
+                    "Use a str literal instead.",
+                    SyntaxWarning,
+                    stacklevel=2,
+                )
         if isinstance(body, ast.JoinedStr):  # f-string
             if all(isinstance(node, ast.Constant) for node in body.values):
                 return ''.join(node.value for node in body.values)

tests/messages/test_extract.py

@@ -180,3 +180,19 @@ def test_issue_1195_2():
         'NOTE: This should still be considered, even if',
         'the text is far away',
     ]
+
+
+def test_bytes_literal_warning():
+    buf = BytesIO(b"""
+t = _(b'hello')
+""")
+    with pytest.warns(SyntaxWarning, match="Bytes literal"):
+        messages = list(extract.extract('python', buf, extract.DEFAULT_KEYWORDS, [], {}))
+    assert len(messages) == 0
+
+
+# Note: a non-string constant such as _(42) is not covered here. It reaches the
+# tokenizer as a NUMBER token, and _parse_python_string() is only called for
+# STRING and FSTRING_START tokens, so no code path inspects it. Warning about it
+# would mean handling NUMBER inside a translator call, which is a wider change
+# than the bytes crash in GH #1190 that this fixes.