Function _looks_like_binary

src/command_system/input_processing.py:229–260

Return True if the first ``_BINARY_SNIFF_BYTES`` bytes contain a NUL AND the file is not BOM-prefixed text. Defense-in-depth check for files whose extension isn't in ``_AT_MENTION_BINARY_EXTENSIONS`` — e.g. a misnamed ``.txt`` that's actually a tarball, or any extension we haven't e

(path: str)

def _looks_like_binary(path: str) -> bool:
    """Return True if the first ``_BINARY_SNIFF_BYTES`` bytes contain a NUL
    AND the file is not BOM-prefixed text.

    Defense-in-depth check for files whose extension isn't in
    ``_AT_MENTION_BINARY_EXTENSIONS`` — e.g. a misnamed ``.txt`` that's
    actually a tarball, or any extension we haven't enumerated. NUL bytes
    do not appear in well-formed utf-8 text, so their presence in a sniff
    window is a high-precision signal.

    Exception: UTF-16 / UTF-32 text files contain NUL bytes for every
    ASCII char and would trip the naive NUL sniffer. A BOM prefix
    (``\\xff\\xfe`` LE / ``\\xfe\\xff`` BE / UTF-32 variants) tells us
    the file is real text that just needs a different decoder; we return
    False so the text-read branch can decode it properly via
    ``_read_text_with_encoding``. Without that paired fix this BOM
    short-circuit would let UTF-16 mojibake flow into a system-reminder
    (the exact failure mode the binary branch was added to prevent).

    Conservative on read errors (treats unreadable files as "not binary"
    so the existing text-read path can surface the OSError its caller
    already handles).
    """
    try:
        with open(path, "rb") as fh:
            chunk = fh.read(_BINARY_SNIFF_BYTES)
    except OSError:
        return False
    if _detect_bom_encoding(chunk) is not None:
        # Encoded text — handled by the text branch with the right codec.
        return False
    return b"\x00" in chunk


# If a decoded file has more than this fraction of U+FFFD replacement