Fix console viewport test failures with scroll anchoring.

third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html

Index: third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html
diff --git a/third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html b/third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html
index 2d7a9f52cbe51534aa0c1ffc6ca005de7db5c21e..ce80f8e579fb50c5987b9c04adbe7ed7d158d7c1 100644
--- a/third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html
+++ b/third_party/WebKit/LayoutTests/inspector/console/console-viewport-stick-to-bottom.html
@@ -88,13 +88,13 @@ function test()
 
         function testViewportMutationsShouldPreserveStickToBottom(next)
         {
-            viewport._contentElement.children[1].innerText = "More than 2 lines: foo\n\nbar";
+            viewport._contentElement.lastChild.innerText = "More than 2 lines: foo\n\nbar";

[lushnikov, 2016/09/12 20:18:39]

how does scroll anchoring picks a node to anchor?
Consider the following scenario:

1. log window object in console
2. log multiple messages after window object: for (var i = 0; i < 100; ++i)
console.log(i);
3. expand window object

Expected: scroll didn't change

[skobes, 2016/09/12 20:22:37]

Roughly, it will pick the first visible element.  If the window object is inside
the viewport when it is expanded, the scroll position won't change.

[lushnikov, 2016/09/13 00:09:21]

I still don't understand how the change helps to make the test pass. Does scroll
anchoring pick the first non-modified node?

[luoe, 2016/09/13 01:30:38]

skobes@, on second thought, I'm a bit uncertain, too.

When the first visible message is the anchor and an unseen message above it
grows, the content moves down and anchoring *would jump down with it.  However,
SANACLAP should suppress the snap because the content height changes, so it
won't jump.  Then, we are no longer at the bottom and the test should pass.

Is this the case?

[skobes, 2016/09/13 20:25:04]

What I observe is that when we set innerText for the second time (on line 97),
the content element has 13 children (log messages) but the first seven of these
are above the viewport.  So, we anchor inside
viewport._contentElement.children[7], which moves down if a preceding message is
modified.

I'm not sure exactly why the first seven messages are above the viewport at this
point.

[skobes, 2016/09/13 20:25:04]

SANACLAP doesn't trigger because no CSS properties have changed, except on the
gap top and bottom gap elements which are not ancestors of the anchor node.

             dumpAndContinue(onMessagesDumped);
 
             function onMessagesDumped()
             {
                 viewport.setStickToBottom(false);
-                viewport._contentElement.children[1].innerText = "More than 3 lines: foo\n\n\nbar";
+                viewport._contentElement.lastChild.innerText = "More than 3 lines: foo\n\n\nbar";
                 dumpAndContinue(next);
             }
         },