Line 402 in lib/python3.3/tokenize.py, contains the following line:

    if first.startswith(BOM_UTF8):

BOM_UTF8 is a bytes object. str.startswith does not accept bytes objects. I was able to use tokenize.tokenize only after making the following changes:

Change line 402 to the following:

    if first.startswith(BOM_UTF8.decode()):

Add these two lines at line 374:

        except AttributeError:
            line_string = line

Change line 485 to the following:

            try:
                line = line.decode(encoding)
            except AttributeError:
                pass

I do not know if these changes are correct as I have not fully tested this module after these changes, but it started working for me. This is the meat of my invokation of tokenize.tokenize:

import tokenize

with open('example.py') as file: # opening a file encoded as UTF-8
	for token in tokenize.tokenize(file.readline):
		print(token)

I am not suggesting that these changes are correct, but I do believe that the current implementation is incorrect. I am also unsure as to what other versions of Python are affected by this.

The docs could certainly be more explicit...currently they state that tokenize is *detecting* the encoding of the file, which *implies* but does not make explicit that the input must be binary, not text.

The doc problem will get fixed as part of the fix to issue 12486, so I'm closing this as a duplicate.  If you want to help out with a patch review and doc patch suggestions on that issue, that would be great.