Implement actions for Automation API.

chrome/test/data/extensions/api_test/automation/sanity_check/test.js

Index: chrome/test/data/extensions/api_test/automation/sanity_check/test.js
diff --git a/chrome/test/data/extensions/api_test/automation/sanity_check/test.js b/chrome/test/data/extensions/api_test/automation/sanity_check/test.js
index 3cf6d289d6d869d1348e1aad1a2d765e3b093d78..be3eadf045d2789516f2ec1bad019f73947aeee9 100644
--- a/chrome/test/data/extensions/api_test/automation/sanity_check/test.js
+++ b/chrome/test/data/extensions/api_test/automation/sanity_check/test.js
@@ -94,6 +94,22 @@ var allTests = [
         });
       });
     });
+  },
+  function testSimpleAction() {
+    function makeAssertions(tree) {
+      var okButton = tree.root.firstChild().firstChild();

[aboxhall, 2014/04/01 17:49:51]

This suggests to me that we should spec out getting an automation node from a
DOM node. This would also be generally useful anyway, and should be fairly
straightforward to implement, I think.

[David Tseng, 2014/04/01 21:27:15]

I was looking for the equivalent of 'getElementById' here. But, yes, we need
some way of getting an automation node based on various criteria; let's spec it.

[aboxhall, 2014/04/01 21:39:52]

Let's chat about this. I spent some time chatting to adamk@ and it should be
possible to pass a querySelector over to Blink and pass an a11y node ID back,
which would do most of what we want. We possibly also want some kind of search
by text API?

+      okButton.addEventListener('focus', function() {
+        chrome.test.succeed();
+      }, true);
+      okButton.focus();
+    }
+    chrome.tabs.create({url: 'test.html'});

[aboxhall, 2014/04/01 17:49:51]

Can we use chrome.tabs.update to be consistent with above and avoid creating a
new tab unnecessarily?

[David Tseng, 2014/04/01 21:27:15]

I'd prefer to leave this as is.

I didn't use update here because I didn't want to specify an existing tab id. I
don't think it matters if we create a new tab right? I'm also not sure about
state left over from previous tests here.

The other aspect I should note is that the tree (before the load_complete event)
is in a "unactionable" state. That is, the subtree under root actually changes a
few times (allocating new nodes with new ids). So, trying to focus an item
before load_complete would fail since the ids potentially changed by the time we
send that action over to the target renderer.

[aboxhall, 2014/04/01 21:39:52]

I see, makes sense.

+    chrome.automation.getTree(function(tree) {
+      tree.root.addEventListener('load_complete',
+          function() {
+            makeAssertions(tree);
+          }, true);
+    });
   }
 ];