Coverage remained the same at 39.85% when pulling 4912fa5 on craigcitro:cleanup_tox into c431900 on google:master.

@craigcitro the first example uses the [testing] extras from the PyPI version (rather than the local checkout).

I think I would probably just punt on the DRY and inline the list into tox.ini -- anybody who is likely to care will be running tox anyhow.

The Travis 3.4 failure is likely due to hash order stuff: pass sort_keys=True to json.dumps() to normalize, or parse them back to dicts to compare.

@tseaver i had thought the same thing about dropping the pip line, but @dhermes said it failed when he tried. that leaves me utterly confused. 😉

i'll fix the sort_keys thing as soon as i get a sec.

My take on a better fix is in #10.