From my limited experience, the code that needs the most work is that involving reading text files -- the whole unicode vs bytes thing.

For the rest, the 2to3 tool does the vast majority of the job.