#566 RoundTripScanner incorrectly scans comments with CRLF line endings

[Sven Stegemann]

When loading a file using RoundTripParser on Windows, RoundTripScanner.scan_next_token() will mistakenly scan the first CRLF of each comment as "\r\n" instead of "\n". On dumping the parsed data to a file Python will replace the "\n" with a full CRLF, resulting in an additional CR character per comment.

[Sven Stegemann]

To reproduce the bug:

def test_example(tmp_path: Path):
    import ruamel.yaml
    input_file = Path("example.yaml")
    yaml = ruamel.yaml.YAML()

    correctly_parsed = yaml.load(open(input_file, "r"))
    incorrectly_parsed = yaml.load(input_file)

    output_path_correct = tmp_path / "correct.yaml"
    output_path_incorrect = tmp_path / "incorrect.yaml"

    yaml.dump(correctly_parsed, output_path_correct)
    yaml.dump(incorrectly_parsed, output_path_incorrect)

    # This assertion fails, because incorrect.yaml contains additional CR characters. 
    assert output_path_correct.read_text() == output_path_incorrect.read_text()

By passing an text stream (mode="r") to yaml.load, every "\r\n" gets translated to "\n", which ruamel will parse correctly. When passing a Path object instead ruamel will open a binary stream (mode="rb"), encounter the bug described above and write the file using a text stream which translates "\n" to "\r\n" resulting in an additional "\r" in the first line of every comment.

[Sven Stegemann]

example.yaml contained a typo. Here is the corrected version