Only 9 tests failing!
```
bindgen_test_layout_FontFamilyList
bindgen_test_layout_StyleAnimation
bindgen_test_layout_nsFont
bindgen_test_layout_nsStyleBackground
bindgen_test_layout_nsStyleDisplay
bindgen_test_layout_nsStyleFont
bindgen_test_layout_nsStyleGridTemplate
bindgen_test_layout_nsStylePosition
bindgen_test_layout_nsStyleSVG
```
The dependencies are:
* FontFamilyList makes nsFont fail which makes nsStyleFont fail.
* nsStyleGridTemplate makes nsStylePosition fail
* StyleAnimation makes nsStyleDisplay fail
nsStyleGridTemplate and nsStyleSVG failures seem to be bitfield-related.
Except the bitfield-related values and those deps, I don't know what
might cause the other failures, we'll have to check carefully C++
member alignments.
This still fails some tests (20 of 130 IIRC).
Most of them are derived from others, and those who arent I think it's
because of:
1. A vtable: Since we don't generate functions I think we don't do the
vtable calculation.
2. Premature opaque structs: It's the case for nsDataHashtable, where
we generate a 0 byte struct, since we calculate it using sizeof() and
the type is not complete without the type parameters.
3. Some other subtlety? I think we miss a case of enum-in-union or
something like that, that is explicitely disabled in bindgen since the C
standard forbids it. I think it can be enabled without too much problem.
Anyways, I'll leave it for tomorrow! \o/
We *do* generate bindings for them, and they compile fine, but the
expected size for them is not correct.
My guess is that C++ types without members have at least 1 byte to be
addressable.
I'll test that later, but for now we don't care about them...
Generalize the style structs
This allows geckolib to pass gecko style structs and have the style system write to them directly, provided we implement all the traits.
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10155)
<!-- Reviewable:end -->
This changes headless operation to strictly be a runtime option, rather
than a compile-time one. Note that the old headless version still relied
on a display server to support WebGL, while it now requires one all the
time.
Fixes#8573
Canvas cleanup
This PR cleans up the layerization infrastructure for canvas, which was unused, and removes unused dependencies.
It also takes in account my recent username change to update angle's dependency (offscreen_gl_context requires extra work due to webrender depending on it).
r? @jdm
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10150)
<!-- Reviewable:end -->
Support navigation keys
Rather useful.
If most people have these keys on their keyboard, I'd prefer to remove the backspace navigation handler. I've never used it on purpose, but it gets hit often by accident when an input widget isn't focused (either due to a misclick or debug build lag).
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10122)
<!-- Reviewable:end -->
array, and consolidate the multitude of display list messages into one.
Improves display list building performance by 50%-100% for small display
lists.
This pulls in https://github.com/servo/ipc-channel/pull/52 , and
especially 8e2357604f7af8869b489b9682a2cf8b58177637, which fixes another
likely cause of intermittent failures on GNU/Linux.
