From a conversion through 2to3:

<       default_value=unicode("", "utf-8"),
---
>       default_value=str("", "utf-8"),

The Python 2 unicode constructor takes an optional second parameter which is the codec to use to convert when the first parameter is non-unicode.

2to3 should check the parameters on uses of unicode() and if there is a second parameter and the first is explicitly b"" bytes it should turn it into
  default_value=b"whatever".decode(second_param)

if the first is valid utf-8 and the second is "utf-8" (or its other spellings) it should leave it as is and simply become:
  default_value="thing passed to unicode() that was already utf-8"

This is not a bug, str accepts the encoding argument in Python 3. And in contrast to the decode method it works with arbitrary byte-like objects (i.e. array.array).

Correct, my characterization above was wrong (I shouldn't write these up without the interpreter right in front of me). What is wrong with the conversion is:

unicode("", "utf-8") in python 2.x should become either str(b"", "utf-8") or, better, just "" in Python 3.x.  The better version could be done if the codec and value can be represented in the encoding of the output 3.x source code file as is but that optimization is not critical.

In order for str() to take a second arg (the codec) the first cannot be a unicode string already:

>>> str("foo", "utf-8")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: decoding str is not supported

Just add the "b" prefix to literal string argument of unicode() in Python 2.