From 7cf121002aa1c4cefef3ae6a274d14fe14dba8ed Mon Sep 17 00:00:00 2001
From: simpleqt <simpleqt@users.noreply.github.com>
Date: Tue, 6 Oct 2026 18:38:39 +0800
Subject: [PATCH] Warn and skip byte string literals in the Python extractor

`_(b'foo')` crashed `pybabel extract` with
`TypeError: sequence item 0: expected str instance, bytes found`
because `_parse_python_string` returned the raw `bytes` value of an
`ast.Constant`, violating its `str | None` annotation, and the value
was later fed to `''.join(buf)`.

Byte string literals are not translatable messages, so following the
behavior agreed in #1190 they are now skipped with a UserWarning instead
of aborting the whole extraction run.

Fixes #1190.
---
 babel/messages/extract.py             | 13 ++++++++++++-
 tests/messages/test_extract_python.py | 16 ++++++++++++++++
 2 files changed, 28 insertions(+), 1 deletion(-)

diff --git a/babel/messages/extract.py b/babel/messages/extract.py
index 6fad84304..433c85959 100644
--- a/babel/messages/extract.py
+++ b/babel/messages/extract.py
@@ -710,7 +710,18 @@ 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
+            value = body.value
+            if isinstance(value, bytes):
+                # Byte string literals (e.g. `_(b'foo')`) are not translatable
+                # messages; warn and leave them unextracted instead of letting
+                # the bytes value crash the string buffer later on.
+                warnings.warn(
+                    f"Byte string literal {value!r} is not a translatable message and was not extracted.",
+                    UserWarning,
+                    stacklevel=2,
+                )
+                return None
+            return value
         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)
diff --git a/tests/messages/test_extract_python.py b/tests/messages/test_extract_python.py
index be1cf970f..6828d1405 100644
--- a/tests/messages/test_extract_python.py
+++ b/tests/messages/test_extract_python.py
@@ -505,3 +505,19 @@ def test_nested_messages():
     assert messages[6][3] == []
     assert messages[7][2] == 'Armin'
     assert messages[7][3] == []
+
+
+def test_extract_byte_string_literal_warns_and_skips():
+    # `_(b'foo')` used to crash the extractor with a TypeError from
+    # `''.join(buf)` because the bytes value leaked through
+    # `_parse_python_string`. Byte literals are not translatable messages, so
+    # they are now skipped with a warning (see #1190).
+    buf = BytesIO(b"""\
+msg1 = _(b'foo')
+msg2 = _('bar')
+""")
+    with pytest.warns(UserWarning, match="Byte string literal"):
+        messages = list(extract.extract("python", buf, extract.DEFAULT_KEYWORDS, []))
+    assert messages == [
+        (2, 'bar', [], None),
+    ]