[page_load_metrics] Add support for chrome:// urls and add extensions observer

chrome/browser/page_load_metrics/observers/webui_page_load_metrics_observer.cc

+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "chrome/browser/page_load_metrics/observers/webui_page_load_metrics_observer.h"
+
+#include "chrome/browser/page_load_metrics/page_load_metrics_util.h"
+#include "chrome/common/url_constants.h"
+#include "content/public/browser/navigation_handle.h"
+
+namespace internal {
+
+const char kHistogramExtensionsParseDuration[] =
+    "PageLoad.Clients.WebUI.Extensions.ParseTiming."
+    "ParseDuration";
+
+}  // namespace internal
+WebUIPageLoadMetricsObserver::WebUIPageLoadMetricsObserver() {}
+
+page_load_metrics::TrackingPolicy WebUIPageLoadMetricsObserver::ShouldTrackLoad(
+    content::NavigationHandle* navigation_handle) {
+  const GURL& url = navigation_handle->GetURL();
+  if (url.SchemeIs("chrome")) {

[Devlin, 2016/09/20 22:59:15]

kChromeUIScheme

+    if (url.path_piece().find(chrome::kChromeUIExtensionsHost) == 1) {

[Devlin, 2016/09/20 22:59:15]

more information than you ever wanted:
In the current settings page, we have two frames - the top frame
(chrome://chrome/extensions) and the content frame (chrome://extensions-frame). 
The former isn't very exciting; the latter has all the interesting stuff.  If we
can limit ourselves to tracking the extensions-frame, that's probably better.
In the new (MD) world, we enter a more sane state, where there is a single frame
- chrome://extensions.

Can we change this to track those two instances, rather than the top frame of
the settings page?

[Charlie Harrison, 2016/09/21 00:32:48]

Hmm this is tricky with the current architecture, which only tracks metrics for
a main frame navigation. FCP tracks the min of {first text paint, first image
paint}. If necessary we can probably log first meaningful paint which tracks the
most significant layout.

[Devlin, 2016/09/21 00:47:22]

This would still be useful for MD extensions, then, but I'd be a little dubious
of relying on it for the current world.  You can see for yourself how little
content there is when you remove the big iframe in chrome://extensions through
devtools - it's basically just the sidebar.

FirstMeaningfulPaint sounds interesting, and might be worth trying out.  I don't
mind being the guinea pig. :)

[Charlie Harrison, 2016/09/26 20:05:57]

I did an experiment and it still seems like (most of the time) FMP is triggered
at first paint of the shell, even when I add ~7 extensions which makes the frame
fill my entire screen.

This seems like an error in FMP, which I thought would take this into account.
How would you like to proceed with this patch? I can cc ksakamoto@ to the bug
and explain this failure.

[Devlin, 2016/09/26 20:39:54]

If this turns out to be very complicated, could we do this to only record for MD
extensions for now, and then pursue metrics for more complicated cases (like the
uber page) separately?  It's a shame it wouldn't land in parallel since it means
we'd be lacking comparison data, but it would be a good starting point.

[Charlie Harrison, 2016/09/26 21:38:19]

SG. I will amend the patch to only log for MD.

+      tracked_page_ = EXTENSIONS;
+    }
+  }
+  return tracked_page_ == NONE ? page_load_metrics::IGNORE
+                               : page_load_metrics::TRACK;
+}
+
+void WebUIPageLoadMetricsObserver::OnParseStop(
+    const page_load_metrics::PageLoadTiming& timing,
+    const page_load_metrics::PageLoadExtraInfo& info) {
+  if (!WasStartedInForegroundOptionalEventInForeground(timing.parse_stop,
+                                                       info)) {
+    return;
+  }
+
+  switch (tracked_page_) {
+    case EXTENSIONS:
+      PAGE_LOAD_HISTOGRAM(
+          internal::kHistogramExtensionsParseDuration,

[Devlin, 2016/09/20 22:59:15]

It would be useful to have this broken apart by MD vs non-MD* - see e.g. how we
do this in extensions_ui.cc today:
https://cs.chromium.org/chromium/src/chrome/browser/ui/webui/extensions/exten...

*In theory, Finch grants us this breakdown, but that's only useful when we have
equivalent populations, and I'd love to be able to look at this for performance
data prior to rollout.

[Charlie Harrison, 2016/09/21 00:32:48]

ACK. This is a good idea.

+          timing.parse_stop.value() - timing.parse_start.value());
+      break;
+    case NONE:
+      NOTREACHED();
+      break;
+  }
+}
+
+void WebUIPageLoadMetricsObserver::OnFirstContentfulPaint(
+    const page_load_metrics::PageLoadTiming& timing,
+    const page_load_metrics::PageLoadExtraInfo& info) {
+  if (!WasStartedInForegroundOptionalEventInForeground(
+          timing.first_contentful_paint, info)) {
+    return;
+  }
+
+  switch (tracked_page_) {
+    case EXTENSIONS:
+      PAGE_LOAD_HISTOGRAM(
+          "PageLoad.Clients.WebUI.Extensions.PaintTiming."
+          "NavigationToFirstContentfulPaint",
+          timing.first_contentful_paint.value());
+      break;
+    case NONE:
+      NOTREACHED();
+      break;
+  }
+}