rebaseline_server: improve pixel diff reporting

rebaseline_server: improve pixel diff reporting
- fix differing pixels count (we weren't always doing this right)
- report number of differing pixels, as well as percentage
- report worst difference per RGB channel
- apply alpha mask to difference image (changed pixels = opaque, unchanged = transparent)
- add tooltips with further explanation of pixel diffs
(SkipBuildbotRuns)

R=mtklein@google.com

Committed: https://code.google.com/p/skia/source/detail?r=12366

[epoger, 2013-11-22 18:26:36]

Ready for review at patchset 1.

You can view this in action at http://172.24.179.183:7921/ ; compare that to the
existing one at http://108.170.217.246:10117/

[mtklein, 2013-11-22 18:47:34]

The new UI exceeds my wildest hopes.  Looks awesome.  Just one request below.

https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
File gm/rebaseline_server/imagediffdb.py (right):

https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
gm/rebaseline_server/imagediffdb.py:210: total_diff = sum(map(lambda
index,value: value * (index % VALUES_PER_BAND)**2,
I think this expression may have tipped over into the realm of
not-instantly-digestible.  Can you break this into smaller parts or perhaps
rewrite it with a loop?

[epoger, 2013-11-22 19:05:18]

https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
File gm/rebaseline_server/imagediffdb.py (right):

https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
gm/rebaseline_server/imagediffdb.py:210: total_diff = sum(map(lambda
index,value: value * (index % VALUES_PER_BAND)**2,
On 2013/11/22 18:47:34, mtklein wrote:
> I think this expression may have tipped over into the realm of
> not-instantly-digestible.  Can you break this into smaller parts or perhaps
> rewrite it with a loop?

I completely agree.  I was just trying to be one of the "Python cool kids" with
this nonsense.

Rewritten as a loop, much easier to understand IMO.  Hopefully performance is
similar.  Might even be better, because it doesn't create and throw away a
mapping every time.

[mtklein, 2013-11-22 19:14:03]

On 2013/11/22 19:05:18, epoger wrote:
>
https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
> File gm/rebaseline_server/imagediffdb.py (right):
> 
>
https://codereview.chromium.org/82823005/diff/1/gm/rebaseline_server/imagedif...
> gm/rebaseline_server/imagediffdb.py:210: total_diff = sum(map(lambda
> index,value: value * (index % VALUES_PER_BAND)**2,
> On 2013/11/22 18:47:34, mtklein wrote:
> > I think this expression may have tipped over into the realm of
> > not-instantly-digestible.  Can you break this into smaller parts or perhaps
> > rewrite it with a loop?
> 
> I completely agree.  I was just trying to be one of the "Python cool kids"
with
> this nonsense.
> 
> Rewritten as a loop, much easier to understand IMO.  Hopefully performance is
> similar.  Might even be better, because it doesn't create and throw away a
> mapping every time.

lgtm.  Thanks!

[epoger, 2013-11-22 19:26:22]

Committed patchset #2 manually as r12366 (presubmit successful).