Unicode literal in BufferedStream._readFromBuffer causes failure in HTMLBinaryInputStream

[niklasl]

In html5lib/inputstream.py, unicode_literals is imported from __future__. This causes html5lib.inputstream.BufferedStream to misbehave, specifically the _readFromBuffer method, which ends with return "".join(rv). Due to this being a unicode literal, any read from after the first becomes a chunk of unicode instead of a chunk of bytes.

An example of the problem caused:

from urllib2 import Request, urlopen
from html5lib.inputstream import HTMLBinaryInputStream

req = Request(url='http://example.org/')
source = urlopen(req)
HTMLBinaryInputStream(source)

Causing:

Traceback (most recent call last):
  File "<stdin>", line 6, in <module>
  File ".../html5lib/inputstream.py", line 411, in __init__
    self.charEncoding = self.detectEncoding(parseMeta, chardet)
  File ".../html5lib/inputstream.py", line 448, in detectEncoding
    encoding = self.detectEncodingMeta()
  File ".../html5lib/inputstream.py", line 535, in detectEncodingMeta
    assert isinstance(buffer, bytes)
AssertionError

(That is, when HTMLBinaryInputStream is used with a file-like object (such as the result of urllib2.urlopen), it wraps it in a BufferedStream, which then fails (at line 535) with the assert isinstance(buffer, bytes).)

This can be fixed by using a byte literal in _readFromBuffer, instead, i.e. return b"".join(rv). (There are at least three places in inputstream.py where string literals are used like this: at lines 117, 318 and 348.)