cc: Update picture pile on tiles to prevent memory leak

cc: Update picture pile on tiles to prevent memory leak

A given PicturePileImpl corresponds to a given main thread frame.  It should go
out of scope when that frame is no longer being used.  Currently, as Tile holds
a ref on PicturePileImpl, it won't go out of scope if the Tile persists, which
could happen if the Tile never gets invalidated.  To prevent this, the
PicturePileImpl on each Tile is updated to the newest version of the
PicturePileImpl when a tree is activated.  This will allow old PicturePileImpls
to be destroyed on a more regular basis.

Raster jobs in flight hold their own ref to a per-thead clone of the
PicturePileImpl, so this will not affect them.  When the job finishes, the
clone will also be destroyed.

R=reveman@chromium.org
BUG=171003
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=178106

[reveman, 2013-01-20 19:20:29]

Would it help to drop pile refs from tiles as raster completes? I'm guessing no
but it's not clear to me why that is.

https://codereview.chromium.org/11953005/diff/1/cc/picture_layer_tiling.cc
File cc/picture_layer_tiling.cc (right):

https://codereview.chromium.org/11953005/diff/1/cc/picture_layer_tiling.cc#ne...
cc/picture_layer_tiling.cc:318: void
PictureLayerTiling::MoveTilePriorities(WhichTree src_tree,
Should we update the name of this function to reflect that it's doing more than
moving priorities? or maybe add a separate UpdatePile function.

[reveman, 2013-01-20 22:28:38]

On 2013/01/20 19:20:29, David Reveman wrote:
> Would it help to drop pile refs from tiles as raster completes? I'm guessing
no
> but it's not clear to me why that is.

please ignore my last comment. we obviously can't drop the pile ref when raster
completes as we might have to re-raster the tile as a result of freeing its
resource from memory pressure.

[enne (OOO), 2013-01-20 22:43:52]

On 2013/01/20 22:28:38, David Reveman wrote:
> On 2013/01/20 19:20:29, David Reveman wrote:
> > Would it help to drop pile refs from tiles as raster completes? I'm guessing
> no
> > but it's not clear to me why that is.
> 
> please ignore my last comment. we obviously can't drop the pile ref when
raster
> completes as we might have to re-raster the tile as a result of freeing its
> resource from memory pressure.

If I'm reading the RasterWorkerPool correctly, the ref being held onto  in the
PostRasterTaskAndReply (namely the OnRasterTaskCompleted holds a ref).  If the
original pile (not clone) has been destroyed at that point, then this is the
only ref holding onto the pile.  If another rastering is needed, asking the Tile
for its picture pile and getting a clone will clone its new picture pile.

[reveman, 2013-01-20 23:57:06]

lgtm

[commit-bot: I haz the power, 2013-01-22 18:32:01]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/enne@chromium.org/11953005/7001

[commit-bot: I haz the power, 2013-01-22 21:38:58]

Change committed as 178106