I have a project (ldaplib) I am working on that needs to do ASN1-BER decoding and encoding. The function I have for the decoding portion is slightly more complex but neither are all that complicated. I would like to get some feedback on the overall approach as well as the code/exceptions/inline comments.

The decoding specifically uses a named tuple defined much higher in the code as a return type which is then used in many places throughout the code:

BER = namedtuple('BER', 'cls pc tag value')

The encoding portion:

def ber_encode(ber_cls, ber_pc, ber_tag, message):
    """ Encode a message into ASN1-BER (TLV) bytes.
    The ber_cls, ber_pc, and ber_tag should be numbers, the message can be
    bytes, str, or int. """
    if type(message) not in (str, int, bytes):
        raise TypeError("Requires str, int, or bytes object.")

    encoded = (ber_cls + ber_pc + ber_tag).to_bytes(1, byteorder='big')

    if type(message) is int:
        bytelength = (message.bit_length() // 8) + 1
        message = message.to_bytes(bytelength, byteorder='big')
    elif type(message) is str:
        message = message.encode(encoding='ascii')

    if len(message) == 0:
        return encoded + int(0).to_bytes(1, byteorder='big')

    # Short form or Long form?
    length = len(message)
    if length < 0x80:
        encoded += length.to_bytes(1, byteorder='big')
        bytelength = (length.bit_length() // 8) + 1
        encoded += (0x80 + bytelength).to_bytes(1, byteorder='big')
        encoded += length.to_bytes(bytelength, byteorder='big')

    # Add the message
    encoded += message

    return encoded

The decoding portion:

def ber_decode(message, result=None):
    """ Decode a message from ASN1-BER (TLV) bytes into a list of byte values.
    Values that contain embedded types will need to call this function to
    break the objects into their individual type/value components. """
    if type(message) is not bytes:
        raise TypeError("Requires bytes object.")

    if not result:
        result = []

    if len(message) == 0:
        raise Exception("No message")

    # Use masks to isolate the binary parts
    ber_type = message[0]  # NOQA
    ber_cls = ber_type & 0b11000000
    ber_pc  = ber_type & 0b00100000
    ber_tag = ber_type & 0b00011111

    ber_len = message[1]

    # Assume that the ber_value is offset 2 or is blank
    offset = 2

    # Only the definite form of length encoding is used.
    if ber_len > 0x80:
        nbytes = ber_len & 0x7f
        ber_len = int.from_bytes(
                message[offset:offset + nbytes], byteorder='big')
        offset = offset + nbytes
        ber_value = message[offset: offset + ber_len]
        ber_value = message[offset: offset + ber_len]

    # Append the decoded message part.
    result.append(BER(ber_cls, ber_pc, ber_tag, ber_value))
    message = message[ber_len + offset:]

    # If the message had only one type then we are done, if not recurse.
    if len(message) == 0:
        return result
        return ber_decode(message, result)

2 Answers 2


Some ideas:

String concatenation

The method you're using for string concatenation (successively using the += operator) is the slowest method in Python for constructing strings. Consider instead creating a list of strings and concatenating them at the end using return ''.join(created_strings). Because your BER encoder is likely to be used with a protocol, speed may be important.

Checking of inputs

Your cls, pc and tag elements are not checked on input to the ber_encode routine and would allow constructs such as ber_encode(0xc3,7,3,"bongo") which gets coded with a tag of 0xcd. That isn't necessarily wrong but it's not what I might have expected.

Orthogonality of ber_encode and ber_decode

It's not unreasonable for a user of your functions to assume that the two complementary functions can each be fed the output of the other, but that's not the case for these functions. While ber_decode will happily digest the output of ber_encode, what ber_decode produces is a named tuple rather than an output that's compatible with ber_encode's input requirements. Again, this isn't necessarily wrong, but it is a potential impediment to users of your functions.

Error checking

Generally, the code does a pretty good job of validating inputs and throwing appropriate errors, but there is at least one case which is accepted that should probably throw an error. Specifically,


results in a BER value which is 129 bytes long. It's probably worth double checking ITU X.690 section which says that an encoded length value of 11111111b shall not be used.



is decoded as three BER encodings, the first of which is 128 bytes long (which is suspect), the second is 97 bytes long (which is OK) and the third is 27 bytes long, which may or may not be OK, but shouldn't a truncated message throw an exception?

Test vectors

You may simply have omitted them from the posted code, but I'd highly recommend including test vectors with the code. Python has a number of ways to do testing including doctest and unittest. Not only do they help you make sure you've considered many different types of input (both good and bad), but they also serve as documentation for users of the functions.

  • \$\begingroup\$ Thanks for the review. These two functions are actually internal so I should add the _ at the beginning. I have full test coverage just not posted. :) \$\endgroup\$
    – clutton
    Apr 19, 2014 at 20:53
  • \$\begingroup\$ On the error checking for ber_cls, ber_pc, ber_tag. I should validate and raise, nice catch, I will add to my tests. \$\endgroup\$
    – clutton
    Apr 19, 2014 at 21:14
  • \$\begingroup\$ I did a little testing, it seems that doing encoded = b''.join((encoded, length.to_bytes(1, byteorder='big'))) in the various places does not really have a positive impact. There could be another way to address it by building up a list but from initial tests it is actually slower by about 3.8%. I would guess this is because I am not building up many byte strings. Adding the validation is a larger hit to performance at about 6.3% slower. Removing all validation (it is really for internal library use anyway) makes it 8% faster. \$\endgroup\$
    – clutton
    Apr 20, 2014 at 2:10
  • \$\begingroup\$ @clutton: good for you for actually measuring it. It only makes a difference if it's measurable on your computer with your data and if you can assure that you don't need the validation, it's perfectly reasonable to omit it. \$\endgroup\$
    – Edward
    Apr 20, 2014 at 2:15
  • \$\begingroup\$ One question on your comment regarding "encoded length value of 11111111b shall not be used." For my purpose (ldap) in your opinion should I be concerned about an LDAP server sending invalid data? This kind of goes along with the validation needs... this is actually my first time trying to write a library that would deal with potentially many different producers of inbound data. \$\endgroup\$
    – clutton
    Apr 20, 2014 at 2:20

In ber_encode: the line bytelength = (length.bit_length() // 8) + 1 counts 1 byte too many if bit_length is an exact multiple of 8. (e.g. 8 bits → 2 bytes?!) So this line should be:

bytelength = ((length.bit_length()-1) // 8) + 1

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.