Yes I think lots of cleanup needs to be done in that area. For example FillLayer::hasOpaqueImage() doesn't really mean whether the image is opaque, but whether the output pixel can depend on the existing pixels.

I think it is better to separate the criteria into logical groups:
hasOpaqueImage = image && image->canRender && image->imageSize && image->knownToBeOpaque
hasOpaqueTile = compositeMode == Clear || compositeMode == Copy || (hasOpaqueImage && compositeMode == SrcOver && blendMode == Normal)
hasFullFill = (repeatX == RepeatFill || repeatX == RoundFill) && (repeatY == RoundFill || repeatY == RoundFill)
isOpaque = hasOpaqueTile && hasFullFill

Only for illustrative purpose and may not be exhaustive. This thing takes serious mental effort to prove correctness. :/

Agreed. I'll generalize this bug title to reflect all the cleanup we need.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/e3ab046fcbe969bd1b38a0361f7d583998365c2c

commit e3ab046fcbe969bd1b38a0361f7d583998365c2c
Author: schenney <schenney@chromium.org>
Date: Mon Mar 21 16:02:40 2016

Consolidate background fill layer opaqueness into one method

Move all the logic for determining if a background layer is
opaque into FillLayer. Also add repeat round as a opaque
tiling mode.

R=trchen@chromium.org
BUG= 593437 

Review URL: https://codereview.chromium.org/1787733003

Cr-Commit-Position: refs/heads/master@{#382301}

[modify] https://crrev.com/e3ab046fcbe969bd1b38a0361f7d583998365c2c/third_party/WebKit/Source/core/paint/BoxDecorationData.cpp
[modify] https://crrev.com/e3ab046fcbe969bd1b38a0361f7d583998365c2c/third_party/WebKit/Source/core/paint/BoxPainter.cpp
[modify] https://crrev.com/e3ab046fcbe969bd1b38a0361f7d583998365c2c/third_party/WebKit/Source/core/style/FillLayer.cpp
[modify] https://crrev.com/e3ab046fcbe969bd1b38a0361f7d583998365c2c/third_party/WebKit/Source/core/style/FillLayer.h