tools/metrics/histograms/histograms.xml

Index: tools/metrics/histograms/histograms.xml
diff --git a/tools/metrics/histograms/histograms.xml b/tools/metrics/histograms/histograms.xml
index f866bfbe3f88ba4da8303ea4274ed17bf365ff5e..535aa24b47b0f3559fbbd90155444d2158a08af3 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -2601,6 +2601,10 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="blink.ImageCodec" enum="ImageCodec">
+  <summary>Image codec inferred during decode.</summary>

[Noel Gordon, 2015/07/28 22:49:24]

Since this is for decoded images, could we call it blink.DecodedImageType?
(anything other but codec). 

[urvang, 2015/07/28 23:24:34]

Done.

[Noel Gordon, 2015/07/29 00:04:57]

Ok good.  You might need to re-title this CL, and I see you tried, thank you. 
Maybe "Add a histogram for Blink decoded image type".

[urvang, 2015/07/29 00:23:20]

Done.

+</histogram>
+
 <histogram name="Blink.MediaElement.Autoplay" enum="MediaElementAutoPlay">
   <owner>oysteine@chromium.org</owner>
   <summary>
@@ -59395,6 +59399,17 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </int>
 </enum>
 
+<enum name="ImageCodec" type="int">
+  <int value="0" label="kUnknownImage"/>

[Noel Gordon, 2015/07/28 22:49:24]

kImageUnknown?

[urvang, 2015/07/28 23:24:34]

Done.

+  <int value="1" label="kImageJPEG"/>
+  <int value="2" label="kImagePNG"/>
+  <int value="3" label="kImageGIF"/>
+  <int value="4" label="kImageWebP"/>
+  <int value="5" label="kImageICO"/>

[Noel Gordon, 2015/07/28 22:49:24]

Can blink discriminate between ICO and CUR types?  Their file types are decoded
with the same image decoder.  Should we just lump them together as kImageICO?

[urvang, 2015/07/28 23:24:34]

Ah my bad, I had already combined them into kImageICOorCUR in the Blink side
patch.
Did the same here now. Thanks!

[Noel Gordon, 2015/07/29 00:04:57]

I guess my questions were unclear, since I was asking them from two different
POV...

That fact that Blink can't discriminate b/w ICO/CUR is its problem, its
limitation.  I think it's unwise to reflect that limitation in this histogram
enum.

Here's what I'd do.  Define kImageICO only here.  If Blink decides in future
that it is important to discriminate, then we can add a kImageCUR entry here at
that time.  I'd describe why things are this way _briefly_ in the change
description.

On that point, the change description is a little empty for my tastes, but I
suppose we can fill it in with ~2 sentences now we know more details about this
change.

[urvang, 2015/07/29 00:23:20]

To be precise: Blink *can* descriminate between ICO and CUR by signature:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit...

But, as we need the width/height of the image while reporting the stats, we add
to the histogram inside the decoder. And because ICO and CUR have a common
decoder, we can't discriminate between the two at that point.

Given this, I'm OK with naming it either "ICOorCUR" or just "ICO". What do you
prefer?

+  <int value="6" label="kImageCUR"/>
+  <int value="7" label="kImageBMP"/>
+</enum>
+
 <enum name="IMECommitType" type="int">
   <obsolete>
     Deprecated 03/2015, and replaced by IMECommitType2.