Problem

pybabel extract aborts the whole run when a keyword call receives a byte string literal, e.g. a source file containing _(b'foo') (as produced by some tools converting legacy code):

  File "babel/messages/extract.py", line 619, in extract_python
    messages.append(''.join(buf))
TypeError: sequence item 0: expected str instance, bytes found

Root cause: _parse_python_string() is annotated -> str | None, but for a byte string literal the ast.Constant branch returns the raw bytes value, which later reaches ''.join(buf).

As agreed in #1190, byte string literals are not translatable messages, so they should be skipped with a warning rather than crashing the extraction.

Fix

In _parse_python_string(), check the constant's type: a bytes value now emits a UserWarning naming the literal and returns None, which the extractor already handles as "nothing extracted" (the same way non-constant arguments such as _(hello.there) are skipped).

>>> from io import BytesIO
>>> from babel.messages import extract
>>> buf = BytesIO(b"msg1 = _(b'foo')\nmsg2 = _('bar')\n")
>>> list(extract.extract('python', buf, extract.DEFAULT_KEYWORDS, []))
...UserWarning: Byte string literal b'foo' is not a translatable message and was not extracted.
[(2, 'bar', [], None)]

Tests

test_extract_byte_string_literal_warns_and_skips — red on master (reproduces the exact TypeError from the issue), green with this change. Full suite: 31 failed, 6277 passed with the fix vs 32 failed, 6276 passed on master — the only delta is the new test; the remaining failures are the pre-existing CLDR-data-drift cases (this checkout runs against release locale data).

Fixes #1190.

Disclosure: this pull request was prepared with the assistance of an AI coding agent (analysis, patch and tests), and was reviewed and submitted by a human contributor.