Python2 to 3 Compatibility Pitfalls

This is a collection of surprising behavior changes important for writing python3 compatible code. Because of the presence of libfuture, these behaviors are also present when the codebase is run on python2.

Bytestrings

• bytes() cannot take an iterator. Though python3's bytes() handles this properly, libfuture's bytes() doesn't.
• bytes(n) where n is a number returns a bytestring of length n, all zeroes.
• Indexing into a bytestring returns an integer on python3 and a one-character string on python2. libfuture standardizes this to always return ints.
• Use the hexlify/unhexlify methods from binascii module to convert things to/from hexadecimal representation.

Strings

• Use the string methods in utils/helpers.py to determine types of strings being passed around. Prefer isbytestr/isunicode to isstring, unless it really doesn't matter.
• Functions that read in data (e,g, readlines, popen) will return str. In python3 str is basically a unicode string, and in python2 they're basically bytestrings. Use the helper methods to make sure you're ready for either case.
• Use the string literals u'foo' or b'foo' rather than 'foo' to avoid ambiguities, if you can.
• io.StringIO similarly now handles unicode rather than bytestrings.
• popen() with universal_newlines=True will return a bytestring no matter which version of python you use, which can be helpful.

Other types

• Use the is* methods in utils/helpers.py for checking types. Libfuture introduces python3 compatible replacement types for types like bytes, int, object etc. This is largely good but can be a weird headache if you are doing literal typechecking. E.g., a string could potentially be a newstring if it came from manticore, or str if it came from a different library. Most of the headaches have already been worked through.
• Python3 doesn't have a concept of a long. If you need to check for ints or longs, use isint. There's no reason to use long literals.

Python object model / metaclasses

• __hash__ needs to be re-implemented if your class overrides __eq__.
• Be careful with implementing custom __hasattr__ and __getattr__ methods. The resolution behaviors of these vary between python2 and 3 in confusing ways.
• Metaclass syntax and behavior is pretty significantly different in python3. Manticore doesn't use this much, but be aware of it.