Make PicturePile pile.

Make PicturePile pile.

This follows a similar policy as Android browser:

- If an invalidate only intersects the base plus at most one non-base SkPicture, then it creates a new SkPicture of that size.
- If an invalidate intersects two or more non-base SkPictures, then a new SkPicture is created at the top of the pile, sized to the union of the invalidate plus the bounds of all the intersecting pictures.
- Whenever new picture's area fully contains an existing picture, that old picture is destroyed.
- If an SkPicture's area would be >70% of the base, the pile is
  destroyed and the base SkPicture is recreated.

During the invalidate pass, invalidated pictures are represented as
Pictures with a size but no recording.  Then all the blank pictures
are filled in at the end of the Update().

BUG=163429
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=170917

[aelias_OOO_until_Jul13, 2012-12-04 03:23:59]

Hi Vangelis,

I haven't tested this logic fully works yet, but it's not obviously more
broken-looking than the existing impl-side painting.

[enne (OOO), 2012-12-04 04:11:09]

lgtm

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.cc
File cc/picture_pile.cc (right):

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.cc#newcode11
cc/picture_pile.cc:11: const float kResetThreshold = 0.7f;
Can you add a comment for what this constant is for?

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.cc#newcode26
cc/picture_pile.cc:26: pile_.clear();
Ah, quite right. I guess if you're resizing and have to rerecord, then you might
as well just make a new pile.

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.cc#newcode73
cc/picture_pile.cc:73: > kResetThreshold)
style nit: I'm pretty sure operators go on the previous line in Chromium.

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.h
File cc/picture_pile.h (right):

https://codereview.chromium.org/11299324/diff/1/cc/picture_pile.h#newcode48
cc/picture_pile.h:48: typedef std::list<scoped_refptr<Picture> > Pile;
Just curious, why list instead of vector?

[aelias_OOO_until_Jul13, 2012-12-04 04:44:34]

I realized I had oversimplified Android browser's piling logic.  I introduced a
new line "if (overlaps.size() >= kMaxOverlapping) {" in my last patchset
(notwithstanding misleading description "fix style nits", wrote the wrong thing
there).  I updated the commit description to describe the change in behavior.

I chose a list because the erase() operation is cheaper on them.

[enne (OOO), 2012-12-04 05:13:35]

Still lgtm.

[nduca, 2012-12-04 05:25:35]

piles of lgtm from me too

[Vangelis Kokkevis, 2012-12-04 07:05:41]

lgtm as well!

[commit-bot: I haz the power, 2012-12-04 08:23:43]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/aelias@chromium.org/11299324/9001

[commit-bot: I haz the power, 2012-12-04 09:51:21]

Change committed as 170917