Chrome remote control multipage tests: Add interactions.

tools/perf/page_sets/top_25_usage.json

Index: tools/perf/page_sets/top_25_usage.json
diff --git a/tools/perf/page_sets/top_25_usage.json b/tools/perf/page_sets/top_25_usage.json
new file mode 100644
index 0000000000000000000000000000000000000000..4ebf6102281d0acafe78f5ee61c37fcb4b692f1e
--- /dev/null
+++ b/tools/perf/page_sets/top_25_usage.json
@@ -0,0 +1,28 @@
+{
+    "description": "Pages for executing tests which not only load the pages, but also do some user actions.",
+    "credentials_path": "../data/credentials.json",
+    "archive_path": "../data/top_25_usage.wpr",
+    "pages": [
+        {
+            "name": "Facebook",
+            "url": "http://www.facebook.com/barackobama",
+            "why": "top social,Public profile",
+            "credentials": "facebook",
+            "scroll_is_infinite": true,
+            "interactions": [

[tonyg, 2012/11/06 00:28:20]

What do you think about adding this to top_25.json rather than creating a new
set? Am I missing a strong reason to have a new one?

[nduca, 2012/11/06 05:03:27]

Agreed, I think this is a must have--- top25 should be shared.

Not loving the verbosity of having to specify every different interaction with
an object. What about

  interactions: ["click_link", "scroll"]
  clickable_links: ["Women for Obama", "About", "Home"]?

If i read this right, you're supposed to do each one in sequence, right? E.g.
click women, then wait, then click about, then wait, then click home? If thats
the case, then you really may want a better name. Like "click_sequence", where
you have this instead:

interactions: ["scroll", "click_sequence"],
click_sequence: [
  {action: "click_link", link_text: "foo"},
  {action: "wait_for_navigation_to_complete"},
}

which brings up the point, should you make the action between each click
explicit?

finally, yno use a selector instead of link text? E.g a[text=blah]?

[marja, 2012/11/06 16:01:21]

Merged this with the top_25.json. This means that when I update the .wpr file in
the repository, the scroll benchmark results won't be comparable to the previous
version (because the pages will contain different data).

Yes, it's a sequence, so, "About" is a link on the "Women for Obama" page.

I'd have just one sequence of interactions, where each interaction can be e.g.,
click or scroll. So we could define a test which clicks a link, scrolls the
page, clicks another link, etc. I wouldn't put "stress_memory" into the name,
since that action sequence is generic, and after executing the actions, we can
measure multiple things, not only memory.

Your comments had a couple of ideas how to note down the action sequence and how
to add the completion condition, the latest patch set has:
"interaction_sequence": [
    {
        "action": "click_link",
        "selector": "a[href=\\\"http://www.facebook.com/WomenforObama\\\"]",
	"action_done": "navigation_complete"
    }, ...
}

Selectors... you meant document.querySelector(), right? Hmm, I can't pass
a[text="foobar"] to it, since "text" is not an attribute of the "a".

So this is a bit ugly:
"selector": "a[href=\\\"http://www.facebook.com/WomenforObama\\\"]"

[marja, 2012/11/06 16:36:07]

I simplified this to
"selector": "a[href=\"http://www.facebook.com/WomenforObama\"]"
and also eliminated this unnecessary round trip of passing the interaction as a
json string instead of an object and converting it back on the js side.

+                {
+                    "type": "clickLink",
+                    "link_text": "Women for Obama"
+                },
+                {
+                    "type": "clickLink",
+                    "link_text": "About"
+                },
+                {
+                    "type": "clickLink",
+                    "link_text": "Home"
+                }
+            ]
+        }
+    ]
+}