Bisected to 081e02da202cd7394d4fdc2b47f0a6b4fb1957b7

I've been staring a bit at this today, and I think I've come to the conclusion that the change above exposed a mismatch in what the local coordinate space is (i.e where the 'transform' is applied) during paint and hit-test.
Trying to figure out in which order 'transform' and the object-bounding-box transform should apply, I believe I landed at that we have it backwards (based on [1].) Doing it in that order has the nice benefit of making the transformations work out "by accident"... (applying in that order should make little visible difference in example at hand, and at also appears to be what Gecko does - although they seem to exhibit some weird behavior in some cases.)
I'd welcome an additional interpretation of [1], but it's consistent with <*Gradient> and <pattern>.

[1] https://drafts.csswg.org/css-transforms/#svg-user-coordinate-space

If you're saying that the object-bounding-box transform should apply on top of the gradient/pattern/transform then I would agree. It's always confusing to talk about this, so another way of saying it is that the gradient/pattern/transform is "closer" to the element than the object bounding box transform.

If M_transform is the "additional" transform and M_bbox is the bbox transform, then we'd do:

M = M_bbox * M_transform [1]

(which is what we do for gradients), but currently the order is the other way around (for clipPath.)

[1] Or in AffineTransform parlance I think it'd be:

AffineTransform t = OBBTransform();
t.Multiply(AdditionalTransform());

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/2a25370e4ddbaa7a291dc09dbda2241d7dc51ce6

commit 2a25370e4ddbaa7a291dc09dbda2241d7dc51ce6
Author: Fredrik Söderquist <fs@opera.com>
Date: Fri Aug 24 22:42:48 2018

Align HitTestClippedOutByClipPath coordinate space with paint

In the paint code (ClipPathClipper), the reference box used is the one
computed by LocalReferenceBox(), and the coordinate space is adjusted
for that. In the hit-testing code however, the reference box is adjusted
instead, leading to incorrect results when the reference box is
subjected to an additional transform (a 'transform' on a <clipPath>.)

Instead, set up the reference box and coordinate space in the hit-
testing code in the same way as for paint, by translating to the correct
local coordinate space before performing the actual hit-test.

Bug:  876390 
Cq-Include-Trybots: luci.chromium.try:linux_layout_tests_slimming_paint_v2;master.tryserver.blink:linux_trusty_blink_rel
Change-Id: I60acd616964a942893f0e256766aca728d14b494
Reviewed-on: https://chromium-review.googlesource.com/1188302
Reviewed-by: Stephen Chenney <schenney@chromium.org>
Commit-Queue: Fredrik Söderquist <fs@opera.com>
Cr-Commit-Position: refs/heads/master@{#586028}
[add] https://crrev.com/2a25370e4ddbaa7a291dc09dbda2241d7dc51ce6/third_party/WebKit/LayoutTests/css3/masking/clip-path-hittest-reference-obb-w-transform.html
[modify] https://crrev.com/2a25370e4ddbaa7a291dc09dbda2241d7dc51ce6/third_party/blink/renderer/core/paint/paint_layer.cc