Better Preserve3D support

[kvark]

Follow up to #1169
Includes #1207

r? @glennw @mrobinson

Fixes

• "preserve-3d" only affects children stacking contexts and contained items
• generated polygon coordinates of non-zero local bounds
• sorting order

Servo

The Servo PR is being worked on, but it's not required here, since it may safely continue using TransformStyle::Flat.

Future work

There is at least one feature on the horizon to be implemented before #739 can be truly closed.
The TransformStyle should be moved out of the stacking context and deserve it's own pushable item (similar to clip-scroll groups). This is required because an item without "preserve-3d" automatically becomes "flat" but does not establish a stacking context, which is tested by Servo.

Edit: apparently, Chrome disagrees here, so the current approach of WR might stay.
Create Bugstar issue-1362543.

---

This change is 

[glennw]

Reviewed 24 of 24 files at r1.
Review status: all files reviewed at latest revision, 3 unresolved discussions.

---

webrender/src/frame_builder.rs, line 90 at r1 (raw file):

fn make_polygon(sc: &StackingContext, node: &ClipScrollNode, anchor: usize)
                -> Polygon<f32, WorldPixel> {
    //TODO: only work with `sc.local_bounds` worth of space

Could you expand on this comment a bit, and why local_bounds.origin is not used?

---

webrender/src/frame_builder.rs, line 92 at r1 (raw file):

    //TODO: only work with `sc.local_bounds` worth of space
    let size = sc.local_bounds.bottom_right();
    let bounds = LayerRect::new(sc.reference_frame_offset, LayerSize::new(size.x, size.y));

Is size already a LayerSize?

---

wrench/reftests/split/reftest.list, line 3 at r1 (raw file):

== simple.yaml simple-ref.yaml
== order.yaml order-ref.yaml
#TODO: == nested.yaml nested-ref.yaml

Is this meant to be resolved before merge, or as a follow up?

---

Comments from Reviewable

[glennw]

@kvark Looks good - just a couple minor issues.

[kvark]

@glennw thanks for the review! Your concerns are hopefully addressed now. Please take a look at the last commit.

---

Review status: 11 of 21 files reviewed at latest revision, 3 unresolved discussions.

---

webrender/src/frame_builder.rs, line 90 at r1 (raw file):

Previously, glennw (Glenn Watson) wrote…

Could you expand on this comment a bit, and why local_bounds.origin is not used?

good idea!

---

webrender/src/frame_builder.rs, line 92 at r1 (raw file):

Previously, glennw (Glenn Watson) wrote…

Is size already a LayerSize?

unfortunately, it's LayerPoint, and euclid doesn't seem to provide convenience conversion between these

---

wrench/reftests/split/reftest.list, line 3 at r1 (raw file):

Previously, glennw (Glenn Watson) wrote…

Is this meant to be resolved before merge, or as a follow up?

nicely spotted! I worked on that test and encountered a logical problem with the code, related to handling the order of planes that geometrically match each other.

---

Comments from Reviewable

[glennw]

@bors-servo r+

[bors-servo]

📌 Commit 3262ea1 has been approved by glennw

[bors-servo]

⌛ Testing commit 3262ea1 with merge def8647...

[bors-servo]

☀️ Test successful - status-travis
Approved by: glennw
Pushing def8647 to master...

[staktrace]

Edit: apparently, Chrome disagrees here, so the current approach of WR might stay.
Create Bugstar issue-1362543.

@kvark, from the discussion on bug 1362543, it looks like Chrome was correct? Are there any other WR changes needed for preserve-3d, or can we close out #739?

[kvark]

@staktrace I want to get some real use cases going (outside of our artificial tests) before closing the issue.

[leeoniya]

http://keithclark.co.uk/labs/css-fps/
http://www.michaelbromley.co.uk/experiments/css-space-shooter/
https://ondras.zarovi.cz/demos/rubik/

;)