Support Firefox via geckodriver

[jmbowman]

Add support for using needle with recent versions of Firefox via geckodriver:

• Try rect from the W3C WebDriver spec before falling back to the Selenium location and size calls if necessary (geckodriver doesn't support the latter).
• Try directly getting a screenshot of the target element (supported for some drivers starting in selenium 2.46.1) before falling back to cropping a full-page screenshot if necessary. The fallback doesn't work correctly in geckodriver, so we can't just consistently use the legacy approach for all drivers.
• Update the tox configuration to simplify testing with Firefox and Chrome. Both pass all tests for me locally.
• Update the Travis configuration to also test Firefox/geckodriver with the latest selenium release. The selenium 2.x releases have a bug that often prevents Firefox from launching correctly, so I don't recommend using or testing that combination. The additional setup needed to run Firefox and geckodriver is skipped for PhantomJS test workers.

This seems like the minimum necessary step towards working, tested geckodriver support, but let me know if you want to split it up further into smaller PRs.

[jphalip]

Thanks a lot, this is great. Just a few comments:

• It seems that the changes attempt to run the tests with chromedriver, but that doesn't seem to be taken into consideration in the travis configuration file. Should that be added, and does chromedriver also need to be installed?
• Could you add a comment in the Travis conf about why tests aren't run with Firefox and Selenium 2.x?
• If it's not too much hassle then it'd be great if you could break up the changes into separate logical commits.

Thanks again!

[jmbowman]

I added support for testing chrome on local machines via tox, but didn't attempt to configure Travis to test it yet; I figured that would be a later PR.

Do you want a separate PR for each file? The changes to driver.py can't be verified without the changes to at least one of the other two files, although I could commit them separately now that I've tested them. And the test suite won't pass for Firefox either locally or in Travis without all of those changes in the driver module.

[jphalip]

Ok I see, no problem. I think it's ok to merge everything at once then. I'd just like to make sure chromedriver is covered for extra peace of mind. From a quick search, it seems it'd be fairly simple to install it on Travis: https://github.com/travis-ci/travis-ci/issues/272#issuecomment-283968954 Do you confirm?

[jmbowman]

Hmm, that doesn't seem to have actually used the updated .travis.yml file. Don't have time to follow up on it right now, but I'll look into it soon.

[jmbowman]

Not sure if it didn't like where I originally added the comment or it was just a force-push related hiccup, but looks like it worked this time. Let me know if you spot any other problems.

[jphalip]

Excellent, thanks again for your work!

[jphalip]

@jmbowman It looks like there is one failure after the merge. Do you know what that is? https://travis-ci.org/bfirsh/needle/jobs/214317627

[jmbowman]

It looks like the first test that tried to launch the browser failed, but all the later ones succeeded; maybe it just wasn't ready yet? I've seen some other Travis setups put in an explicit wait of 3 seconds to wait for xvfb to finish initializing; I thought the other subsequent setup tasks would take long enough for it to finish, but maybe not. I'll create a separate PR to both add that delay and add some retry logic with a timeout to get_web_driver().