fix browser test driver

This fixes http://code.google.com/p/google-caja/issues/detail?id=1590

Webdriver communication with firefox 17 is kind of unreliable, and this
unreliability seems to happen only when opening new firefox windows.
I tried digging into why this was happening, but haven't been able to
isolate the cause yet.

Instead, this patch changes our browser test driver to just re-use one
window.  I think the difference between "navigate window to url x" and
"open new window with url x" are not relevant to our test cases.
This is also faster, and improves our chances of success testing
other browsers.

In the case where a browser test fails, this still opens a new window,
so that the failing windows remain available for manual inspection.

I'm also upgrading webdriver to 2.26.0.

I'm also fixing an issue where stderr is closed prematurely.

I'm also adding support for RemoteWebDriver and ChromeDriver.
Chrome works but doesn't pass 100% yet.
HtmlUnit doesn't work, no idea why yet.
I haven't tried IE or iphone yet.

[felix8a, 2012-11-22 18:00:58]

New snapshot, includes all the changes to support Chrome driver and
RemoteWebDriver.

[kpreid2, 2012-11-22 18:35:33]

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
File tests/com/google/caja/plugin/BrowserTestCase.java (right):

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/BrowserTestCase.java:65: *  
<dt>caja.test.thihostname</dt>
typo

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/BrowserTestCase.java:170: String page = "http://" +
ThisHostName.value() + ":" + portNumber
I'd like this to ask localServer for the address — even if it uses the same
utility code in the end, it will more accurately express the dependencies.

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
File tests/com/google/caja/plugin/es53-test-client-uri-rewriting.js (right):

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/es53-test-client-uri-rewriting.js:32: var selfname
= location.hostname;
Can we abstract out the entire prefix including scheme, hostname, and port
number instead? (here and all other occurrences of selfname)

[felix8a, 2012-11-22 19:10:38]

updated snapshot

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
File tests/com/google/caja/plugin/BrowserTestCase.java (right):

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/BrowserTestCase.java:65: *  
<dt>caja.test.thihostname</dt>
On 2012/11/22 18:35:33, kpreid2 wrote:
> typo

Done.

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/BrowserTestCase.java:170: String page = "http://" +
ThisHostName.value() + ":" + portNumber
On 2012/11/22 18:35:33, kpreid2 wrote:
> I'd like this to ask localServer for the address — even if it uses the same
> utility code in the end, it will more accurately express the dependencies.

Done.

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
File tests/com/google/caja/plugin/es53-test-client-uri-rewriting.js (right):

https://codereview.appspot.com/6851095/diff/3001/tests/com/google/caja/plugin...
tests/com/google/caja/plugin/es53-test-client-uri-rewriting.js:32: var selfname
= location.hostname;
On 2012/11/22 18:35:33, kpreid2 wrote:
> Can we abstract out the entire prefix including scheme, hostname, and port
> number instead? (here and all other occurrences of selfname)

ok, done.  actually changing the scheme and port number is not going to work
without some other changes (which I'm planning on doing eventually, but not
right now).

[kpreid2, 2012-11-23 17:17:32]

LGTM

[felix8a, 2012-11-23 17:30:31]

@r5158