mirror of
https://github.com/servo/servo.git
synced 2025-06-21 15:49:04 +01:00
Instead annotate all flows with their owning ScrollRoots. When processing the display list items into a flattened display list, we add PushScrollRoot and PopScrollRoot to signal when scrolling regions start and end. It is possible for content from different scrolling regions to intersect and when they do, the stack of scrolling regions is duplicated. When these duplicated scrolling regions stacks reach WebRender, it will scroll them in tandem. The PushScrollRoot and PopScrollRoot items are currently represented as StackingContexts in WebRender, but eventually these will be replaced with special WebRender display items. Fixes #13529. Fixed #13298.
1223 lines
44 KiB
Rust
1223 lines
44 KiB
Rust
/* This Source Code Form is subject to the terms of the Mozilla Public
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
|
|
|
//! Servo heavily uses display lists, which are retained-mode lists of painting commands to
|
|
//! perform. Using a list instead of painting elements in immediate mode allows transforms, hit
|
|
//! testing, and invalidation to be performed using the same primitives as painting. It also allows
|
|
//! Servo to aggressively cull invisible and out-of-bounds painting elements, to reduce overdraw.
|
|
//! Finally, display lists allow tiles to be farmed out onto multiple CPUs and painted in parallel
|
|
//! (although this benefit does not apply to GPU-based painting).
|
|
//!
|
|
//! Display items describe relatively high-level drawing operations (for example, entire borders
|
|
//! and shadows instead of lines and blur operations), to reduce the amount of allocation required.
|
|
//! They are therefore not exactly analogous to constructs like Skia pictures, which consist of
|
|
//! low-level drawing primitives.
|
|
|
|
use app_units::Au;
|
|
use euclid::{Matrix4D, Point2D, Rect, Size2D};
|
|
use euclid::num::{One, Zero};
|
|
use euclid::rect::TypedRect;
|
|
use euclid::side_offsets::SideOffsets2D;
|
|
use gfx_traits::{ScrollPolicy, ScrollRootId, StackingContextId};
|
|
use gfx_traits::print_tree::PrintTree;
|
|
use ipc_channel::ipc::IpcSharedMemory;
|
|
use msg::constellation_msg::PipelineId;
|
|
use net_traits::image::base::{Image, PixelFormat};
|
|
use range::Range;
|
|
use std::cmp::{self, Ordering};
|
|
use std::collections::HashMap;
|
|
use std::fmt;
|
|
use std::sync::Arc;
|
|
use style::computed_values::{border_style, filter, image_rendering, mix_blend_mode};
|
|
use style_traits::cursor::Cursor;
|
|
use text::TextRun;
|
|
use text::glyph::ByteIndex;
|
|
use util::geometry::{self, max_rect};
|
|
use webrender_traits::{self, ColorF, GradientStop, WebGLContextId};
|
|
|
|
pub use style::dom::OpaqueNode;
|
|
|
|
/// The factor that we multiply the blur radius by in order to inflate the boundaries of display
|
|
/// items that involve a blur. This ensures that the display item boundaries include all the ink.
|
|
pub static BLUR_INFLATION_FACTOR: i32 = 3;
|
|
|
|
#[derive(HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct DisplayList {
|
|
pub list: Vec<DisplayItem>,
|
|
}
|
|
|
|
impl DisplayList {
|
|
// Return all nodes containing the point of interest, bottommost first, and
|
|
// respecting the `pointer-events` CSS property.
|
|
pub fn hit_test(&self,
|
|
translated_point: &Point2D<Au>,
|
|
client_point: &Point2D<Au>,
|
|
scroll_offsets: &ScrollOffsetMap)
|
|
-> Vec<DisplayItemMetadata> {
|
|
let mut result = Vec::new();
|
|
let mut traversal = DisplayListTraversal::new(self);
|
|
self.hit_test_contents(&mut traversal,
|
|
translated_point,
|
|
client_point,
|
|
scroll_offsets,
|
|
&mut result);
|
|
result
|
|
}
|
|
|
|
pub fn hit_test_contents<'a>(&self,
|
|
traversal: &mut DisplayListTraversal<'a>,
|
|
translated_point: &Point2D<Au>,
|
|
client_point: &Point2D<Au>,
|
|
scroll_offsets: &ScrollOffsetMap,
|
|
result: &mut Vec<DisplayItemMetadata>) {
|
|
while let Some(item) = traversal.next() {
|
|
match item {
|
|
&DisplayItem::PushStackingContext(ref stacking_context_item) => {
|
|
self.hit_test_stacking_context(traversal,
|
|
&stacking_context_item.stacking_context,
|
|
translated_point,
|
|
client_point,
|
|
scroll_offsets,
|
|
result);
|
|
}
|
|
&DisplayItem::PushScrollRoot(ref item) => {
|
|
self.hit_test_scroll_root(traversal,
|
|
&item.scroll_root,
|
|
*translated_point,
|
|
client_point,
|
|
scroll_offsets,
|
|
result);
|
|
}
|
|
&DisplayItem::PopStackingContext(_) | &DisplayItem::PopScrollRoot(_) => return,
|
|
_ => {
|
|
if let Some(meta) = item.hit_test(*translated_point) {
|
|
result.push(meta);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
fn hit_test_scroll_root<'a>(&self,
|
|
traversal: &mut DisplayListTraversal<'a>,
|
|
scroll_root: &ScrollRoot,
|
|
mut translated_point: Point2D<Au>,
|
|
client_point: &Point2D<Au>,
|
|
scroll_offsets: &ScrollOffsetMap,
|
|
result: &mut Vec<DisplayItemMetadata>) {
|
|
// Adjust the translated point to account for the scroll offset if
|
|
// necessary. This can only happen when WebRender is in use.
|
|
//
|
|
// We don't perform this adjustment on the root stacking context because
|
|
// the DOM-side code has already translated the point for us (e.g. in
|
|
// `Window::hit_test_query()`) by now.
|
|
if let Some(scroll_offset) = scroll_offsets.get(&scroll_root.id) {
|
|
translated_point.x -= Au::from_f32_px(scroll_offset.x);
|
|
translated_point.y -= Au::from_f32_px(scroll_offset.y);
|
|
}
|
|
self.hit_test_contents(traversal, &translated_point, client_point, scroll_offsets, result);
|
|
}
|
|
|
|
fn hit_test_stacking_context<'a>(&self,
|
|
traversal: &mut DisplayListTraversal<'a>,
|
|
stacking_context: &StackingContext,
|
|
translated_point: &Point2D<Au>,
|
|
client_point: &Point2D<Au>,
|
|
scroll_offsets: &ScrollOffsetMap,
|
|
result: &mut Vec<DisplayItemMetadata>) {
|
|
// Convert the parent translated point into stacking context local transform space if the
|
|
// stacking context isn't fixed. If it's fixed, we need to use the client point anyway.
|
|
debug_assert!(stacking_context.context_type == StackingContextType::Real);
|
|
let is_fixed = stacking_context.scroll_policy == ScrollPolicy::FixedPosition;
|
|
let translated_point = if is_fixed {
|
|
*client_point
|
|
} else {
|
|
let point = *translated_point - stacking_context.bounds.origin;
|
|
let inv_transform = stacking_context.transform.inverse().unwrap();
|
|
let frac_point = inv_transform.transform_point(&Point2D::new(point.x.to_f32_px(),
|
|
point.y.to_f32_px()));
|
|
Point2D::new(Au::from_f32_px(frac_point.x), Au::from_f32_px(frac_point.y))
|
|
};
|
|
|
|
self.hit_test_contents(traversal, &translated_point, client_point, scroll_offsets, result);
|
|
}
|
|
|
|
pub fn print(&self) {
|
|
let mut print_tree = PrintTree::new("Display List".to_owned());
|
|
self.print_with_tree(&mut print_tree);
|
|
}
|
|
|
|
pub fn print_with_tree(&self, print_tree: &mut PrintTree) {
|
|
print_tree.new_level("Items".to_owned());
|
|
for item in &self.list {
|
|
print_tree.add_item(format!("{:?} StackingContext: {:?} ScrollRoot: {:?}",
|
|
item,
|
|
item.base().stacking_context_id,
|
|
item.scroll_root_id()));
|
|
}
|
|
print_tree.end_level();
|
|
}
|
|
}
|
|
|
|
pub struct DisplayListTraversal<'a> {
|
|
pub display_list: &'a DisplayList,
|
|
pub next_item_index: usize,
|
|
pub first_item_index: usize,
|
|
pub last_item_index: usize,
|
|
}
|
|
|
|
impl<'a> DisplayListTraversal<'a> {
|
|
pub fn new(display_list: &'a DisplayList) -> DisplayListTraversal {
|
|
DisplayListTraversal {
|
|
display_list: display_list,
|
|
next_item_index: 0,
|
|
first_item_index: 0,
|
|
last_item_index: display_list.list.len(),
|
|
}
|
|
}
|
|
|
|
pub fn new_partial(display_list: &'a DisplayList,
|
|
stacking_context_id: StackingContextId,
|
|
start: usize,
|
|
end: usize)
|
|
-> DisplayListTraversal {
|
|
debug_assert!(start <= end);
|
|
debug_assert!(display_list.list.len() > start);
|
|
debug_assert!(display_list.list.len() > end);
|
|
|
|
let stacking_context_start = display_list.list[0..start].iter().rposition(|item|
|
|
match item {
|
|
&DisplayItem::PushStackingContext(ref item) =>
|
|
item.stacking_context.id == stacking_context_id,
|
|
_ => false,
|
|
}).unwrap_or(start);
|
|
debug_assert!(stacking_context_start <= start);
|
|
|
|
DisplayListTraversal {
|
|
display_list: display_list,
|
|
next_item_index: stacking_context_start,
|
|
first_item_index: start,
|
|
last_item_index: end + 1,
|
|
}
|
|
}
|
|
|
|
pub fn previous_item_id(&self) -> usize {
|
|
self.next_item_index - 1
|
|
}
|
|
|
|
pub fn skip_to_end_of_stacking_context(&mut self, id: StackingContextId) {
|
|
self.next_item_index = self.display_list.list[self.next_item_index..].iter()
|
|
.position(|item| {
|
|
match item {
|
|
&DisplayItem::PopStackingContext(ref item) => item.stacking_context_id == id,
|
|
_ => false
|
|
}
|
|
}).unwrap_or(self.display_list.list.len());
|
|
debug_assert!(self.next_item_index < self.last_item_index);
|
|
}
|
|
}
|
|
|
|
impl<'a> Iterator for DisplayListTraversal<'a> {
|
|
type Item = &'a DisplayItem;
|
|
|
|
fn next(&mut self) -> Option<&'a DisplayItem> {
|
|
while self.next_item_index < self.last_item_index {
|
|
debug_assert!(self.next_item_index <= self.last_item_index);
|
|
|
|
let reached_first_item = self.next_item_index >= self.first_item_index;
|
|
let item = &self.display_list.list[self.next_item_index];
|
|
|
|
self.next_item_index += 1;
|
|
|
|
if reached_first_item {
|
|
return Some(item)
|
|
}
|
|
|
|
// Before we reach the starting item, we only emit stacking context boundaries. This
|
|
// is to ensure that we properly position items when we are processing a display list
|
|
// slice that is relative to a certain stacking context.
|
|
match item {
|
|
&DisplayItem::PushStackingContext(_) |
|
|
&DisplayItem::PopStackingContext(_) => return Some(item),
|
|
_ => {}
|
|
}
|
|
}
|
|
|
|
None
|
|
}
|
|
}
|
|
|
|
/// Display list sections that make up a stacking context. Each section here refers
|
|
/// to the steps in CSS 2.1 Appendix E.
|
|
///
|
|
#[derive(Clone, Copy, Debug, Deserialize, Eq, HeapSizeOf, Ord, PartialEq, PartialOrd, RustcEncodable, Serialize)]
|
|
pub enum DisplayListSection {
|
|
BackgroundAndBorders,
|
|
BlockBackgroundsAndBorders,
|
|
Content,
|
|
Outlines,
|
|
}
|
|
|
|
#[derive(Clone, Copy, Debug, Deserialize, Eq, HeapSizeOf, Ord, PartialEq, PartialOrd, RustcEncodable, Serialize)]
|
|
pub enum StackingContextType {
|
|
Real,
|
|
PseudoPositioned,
|
|
PseudoFloat,
|
|
PseudoScrollingArea,
|
|
}
|
|
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
/// Represents one CSS stacking context, which may or may not have a hardware layer.
|
|
pub struct StackingContext {
|
|
/// The ID of this StackingContext for uniquely identifying it.
|
|
pub id: StackingContextId,
|
|
|
|
/// The type of this StackingContext. Used for collecting and sorting.
|
|
pub context_type: StackingContextType,
|
|
|
|
/// The position and size of this stacking context.
|
|
pub bounds: Rect<Au>,
|
|
|
|
/// The overflow rect for this stacking context in its coordinate system.
|
|
pub overflow: Rect<Au>,
|
|
|
|
/// The `z-index` for this stacking context.
|
|
pub z_index: i32,
|
|
|
|
/// CSS filters to be applied to this stacking context (including opacity).
|
|
pub filters: filter::T,
|
|
|
|
/// The blend mode with which this stacking context blends with its backdrop.
|
|
pub blend_mode: mix_blend_mode::T,
|
|
|
|
/// A transform to be applied to this stacking context.
|
|
pub transform: Matrix4D<f32>,
|
|
|
|
/// The perspective matrix to be applied to children.
|
|
pub perspective: Matrix4D<f32>,
|
|
|
|
/// Whether this stacking context creates a new 3d rendering context.
|
|
pub establishes_3d_context: bool,
|
|
|
|
/// The scroll policy of this layer.
|
|
pub scroll_policy: ScrollPolicy,
|
|
|
|
/// Children of this StackingContext.
|
|
pub children: Vec<StackingContext>,
|
|
|
|
/// The id of the parent scrolling area that contains this StackingContext.
|
|
pub parent_scroll_id: ScrollRootId,
|
|
}
|
|
|
|
impl StackingContext {
|
|
/// Creates a new stacking context.
|
|
#[inline]
|
|
pub fn new(id: StackingContextId,
|
|
context_type: StackingContextType,
|
|
bounds: &Rect<Au>,
|
|
overflow: &Rect<Au>,
|
|
z_index: i32,
|
|
filters: filter::T,
|
|
blend_mode: mix_blend_mode::T,
|
|
transform: Matrix4D<f32>,
|
|
perspective: Matrix4D<f32>,
|
|
establishes_3d_context: bool,
|
|
scroll_policy: ScrollPolicy,
|
|
parent_scroll_id: ScrollRootId)
|
|
-> StackingContext {
|
|
StackingContext {
|
|
id: id,
|
|
context_type: context_type,
|
|
bounds: *bounds,
|
|
overflow: *overflow,
|
|
z_index: z_index,
|
|
filters: filters,
|
|
blend_mode: blend_mode,
|
|
transform: transform,
|
|
perspective: perspective,
|
|
establishes_3d_context: establishes_3d_context,
|
|
scroll_policy: scroll_policy,
|
|
children: Vec::new(),
|
|
parent_scroll_id: parent_scroll_id,
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
pub fn root() -> StackingContext {
|
|
StackingContext::new(StackingContextId::new(0),
|
|
StackingContextType::Real,
|
|
&Rect::zero(),
|
|
&Rect::zero(),
|
|
0,
|
|
filter::T::new(Vec::new()),
|
|
mix_blend_mode::T::normal,
|
|
Matrix4D::identity(),
|
|
Matrix4D::identity(),
|
|
true,
|
|
ScrollPolicy::Scrollable,
|
|
ScrollRootId::root())
|
|
}
|
|
|
|
pub fn add_child(&mut self, mut child: StackingContext) {
|
|
child.update_overflow_for_all_children();
|
|
self.children.push(child);
|
|
}
|
|
|
|
pub fn child_at_mut(&mut self, index: usize) -> &mut StackingContext {
|
|
&mut self.children[index]
|
|
}
|
|
|
|
pub fn children(&self) -> &[StackingContext] {
|
|
&self.children
|
|
}
|
|
|
|
fn update_overflow_for_all_children(&mut self) {
|
|
for child in self.children.iter() {
|
|
if self.context_type == StackingContextType::Real &&
|
|
child.context_type == StackingContextType::Real {
|
|
// This child might be transformed, so we need to take into account
|
|
// its transformed overflow rect too, but at the correct position.
|
|
let overflow = child.overflow_rect_in_parent_space();
|
|
self.overflow = self.overflow.union(&overflow);
|
|
}
|
|
}
|
|
}
|
|
|
|
fn overflow_rect_in_parent_space(&self) -> Rect<Au> {
|
|
// Transform this stacking context to get it into the same space as
|
|
// the parent stacking context.
|
|
//
|
|
// TODO: Take into account 3d transforms, even though it's a fairly
|
|
// uncommon case.
|
|
let origin_x = self.bounds.origin.x.to_f32_px();
|
|
let origin_y = self.bounds.origin.y.to_f32_px();
|
|
|
|
let transform = Matrix4D::identity().pre_translated(origin_x, origin_y, 0.0)
|
|
.pre_mul(&self.transform);
|
|
let transform_2d = transform.to_2d();
|
|
|
|
let overflow = geometry::au_rect_to_f32_rect(self.overflow);
|
|
let overflow = transform_2d.transform_rect(&overflow);
|
|
geometry::f32_rect_to_au_rect(overflow)
|
|
}
|
|
|
|
pub fn print_with_tree(&self, print_tree: &mut PrintTree) {
|
|
print_tree.new_level(format!("{:?}", self));
|
|
for kid in self.children() {
|
|
kid.print_with_tree(print_tree);
|
|
}
|
|
print_tree.end_level();
|
|
}
|
|
|
|
pub fn to_display_list_items(self) -> (DisplayItem, DisplayItem) {
|
|
let mut base_item = BaseDisplayItem::empty();
|
|
base_item.stacking_context_id = self.id;
|
|
|
|
let pop_item = DisplayItem::PopStackingContext(Box::new(
|
|
PopStackingContextItem {
|
|
base: base_item.clone(),
|
|
stacking_context_id: self.id,
|
|
}
|
|
));
|
|
|
|
let push_item = DisplayItem::PushStackingContext(Box::new(
|
|
PushStackingContextItem {
|
|
base: base_item,
|
|
stacking_context: self,
|
|
}
|
|
));
|
|
|
|
(push_item, pop_item)
|
|
}
|
|
}
|
|
|
|
impl Ord for StackingContext {
|
|
fn cmp(&self, other: &Self) -> Ordering {
|
|
if self.z_index != 0 || other.z_index != 0 {
|
|
return self.z_index.cmp(&other.z_index);
|
|
}
|
|
|
|
match (self.context_type, other.context_type) {
|
|
(StackingContextType::PseudoFloat, StackingContextType::PseudoFloat) => Ordering::Equal,
|
|
(StackingContextType::PseudoFloat, _) => Ordering::Less,
|
|
(_, StackingContextType::PseudoFloat) => Ordering::Greater,
|
|
(_, _) => Ordering::Equal,
|
|
}
|
|
}
|
|
}
|
|
|
|
impl PartialOrd for StackingContext {
|
|
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
|
|
Some(self.cmp(other))
|
|
}
|
|
}
|
|
|
|
impl Eq for StackingContext {}
|
|
impl PartialEq for StackingContext {
|
|
fn eq(&self, other: &Self) -> bool {
|
|
self.id == other.id
|
|
}
|
|
}
|
|
|
|
impl fmt::Debug for StackingContext {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
let type_string = if self.context_type == StackingContextType::Real {
|
|
"StackingContext"
|
|
} else {
|
|
"Pseudo-StackingContext"
|
|
};
|
|
|
|
write!(f, "{} at {:?} with overflow {:?}: {:?}",
|
|
type_string,
|
|
self.bounds,
|
|
self.overflow,
|
|
self.id)
|
|
}
|
|
}
|
|
|
|
/// Defines a stacking context.
|
|
#[derive(Clone, Debug, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct ScrollRoot {
|
|
/// The unique ID of this ScrollRoot.
|
|
pub id: ScrollRootId,
|
|
|
|
/// The unique ID of the parent of this ScrollRoot.
|
|
pub parent_id: ScrollRootId,
|
|
|
|
/// The position of this scroll root's frame in the parent stacking context.
|
|
pub clip: Rect<Au>,
|
|
|
|
/// The size of the contents that can be scrolled inside of the scroll root.
|
|
pub size: Size2D<Au>,
|
|
}
|
|
|
|
impl ScrollRoot {
|
|
pub fn to_push(&self) -> DisplayItem {
|
|
DisplayItem::PushScrollRoot(box PushScrollRootItem {
|
|
base: BaseDisplayItem::empty(),
|
|
scroll_root: self.clone(),
|
|
})
|
|
}
|
|
}
|
|
|
|
|
|
/// One drawing command in the list.
|
|
#[derive(Clone, Deserialize, HeapSizeOf, Serialize)]
|
|
pub enum DisplayItem {
|
|
SolidColor(Box<SolidColorDisplayItem>),
|
|
Text(Box<TextDisplayItem>),
|
|
Image(Box<ImageDisplayItem>),
|
|
WebGL(Box<WebGLDisplayItem>),
|
|
Border(Box<BorderDisplayItem>),
|
|
Gradient(Box<GradientDisplayItem>),
|
|
Line(Box<LineDisplayItem>),
|
|
BoxShadow(Box<BoxShadowDisplayItem>),
|
|
Iframe(Box<IframeDisplayItem>),
|
|
PushStackingContext(Box<PushStackingContextItem>),
|
|
PopStackingContext(Box<PopStackingContextItem>),
|
|
PushScrollRoot(Box<PushScrollRootItem>),
|
|
PopScrollRoot(Box<BaseDisplayItem>),
|
|
}
|
|
|
|
/// Information common to all display items.
|
|
#[derive(Clone, Deserialize, HeapSizeOf, Serialize)]
|
|
pub struct BaseDisplayItem {
|
|
/// The boundaries of the display item, in layer coordinates.
|
|
pub bounds: Rect<Au>,
|
|
|
|
/// Metadata attached to this display item.
|
|
pub metadata: DisplayItemMetadata,
|
|
|
|
/// The region to clip to.
|
|
pub clip: ClippingRegion,
|
|
|
|
/// The section of the display list that this item belongs to.
|
|
pub section: DisplayListSection,
|
|
|
|
/// The id of the stacking context this item belongs to.
|
|
pub stacking_context_id: StackingContextId,
|
|
|
|
/// The id of the scroll root this item belongs to.
|
|
pub scroll_root_id: ScrollRootId,
|
|
}
|
|
|
|
impl BaseDisplayItem {
|
|
#[inline(always)]
|
|
pub fn new(bounds: &Rect<Au>,
|
|
metadata: DisplayItemMetadata,
|
|
clip: &ClippingRegion,
|
|
section: DisplayListSection,
|
|
stacking_context_id: StackingContextId,
|
|
scroll_root_id: ScrollRootId)
|
|
-> BaseDisplayItem {
|
|
// Detect useless clipping regions here and optimize them to `ClippingRegion::max()`.
|
|
// The painting backend may want to optimize out clipping regions and this makes it easier
|
|
// for it to do so.
|
|
BaseDisplayItem {
|
|
bounds: *bounds,
|
|
metadata: metadata,
|
|
clip: if clip.does_not_clip_rect(&bounds) {
|
|
ClippingRegion::max()
|
|
} else {
|
|
(*clip).clone()
|
|
},
|
|
section: section,
|
|
stacking_context_id: stacking_context_id,
|
|
scroll_root_id: scroll_root_id,
|
|
}
|
|
}
|
|
|
|
#[inline(always)]
|
|
pub fn empty() -> BaseDisplayItem {
|
|
BaseDisplayItem {
|
|
bounds: TypedRect::zero(),
|
|
metadata: DisplayItemMetadata {
|
|
node: OpaqueNode(0),
|
|
pointing: None,
|
|
},
|
|
clip: ClippingRegion::max(),
|
|
section: DisplayListSection::Content,
|
|
stacking_context_id: StackingContextId::root(),
|
|
scroll_root_id: ScrollRootId::root(),
|
|
}
|
|
}
|
|
}
|
|
|
|
/// A clipping region for a display item. Currently, this can describe rectangles, rounded
|
|
/// rectangles (for `border-radius`), or arbitrary intersections of the two. Arbitrary transforms
|
|
/// are not supported because those are handled by the higher-level `StackingContext` abstraction.
|
|
#[derive(Clone, PartialEq, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct ClippingRegion {
|
|
/// The main rectangular region. This does not include any corners.
|
|
pub main: Rect<Au>,
|
|
/// Any complex regions.
|
|
///
|
|
/// TODO(pcwalton): Atomically reference count these? Not sure if it's worth the trouble.
|
|
/// Measure and follow up.
|
|
pub complex: Vec<ComplexClippingRegion>,
|
|
}
|
|
|
|
/// A complex clipping region. These don't as easily admit arbitrary intersection operations, so
|
|
/// they're stored in a list over to the side. Currently a complex clipping region is just a
|
|
/// rounded rectangle, but the CSS WGs will probably make us throw more stuff in here eventually.
|
|
#[derive(Clone, PartialEq, Debug, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct ComplexClippingRegion {
|
|
/// The boundaries of the rectangle.
|
|
pub rect: Rect<Au>,
|
|
/// Border radii of this rectangle.
|
|
pub radii: BorderRadii<Au>,
|
|
}
|
|
|
|
impl ClippingRegion {
|
|
/// Returns an empty clipping region that, if set, will result in no pixels being visible.
|
|
#[inline]
|
|
pub fn empty() -> ClippingRegion {
|
|
ClippingRegion {
|
|
main: Rect::zero(),
|
|
complex: Vec::new(),
|
|
}
|
|
}
|
|
|
|
/// Returns an all-encompassing clipping region that clips no pixels out.
|
|
#[inline]
|
|
pub fn max() -> ClippingRegion {
|
|
ClippingRegion {
|
|
main: max_rect(),
|
|
complex: Vec::new(),
|
|
}
|
|
}
|
|
|
|
/// Returns a clipping region that represents the given rectangle.
|
|
#[inline]
|
|
pub fn from_rect(rect: &Rect<Au>) -> ClippingRegion {
|
|
ClippingRegion {
|
|
main: *rect,
|
|
complex: Vec::new(),
|
|
}
|
|
}
|
|
|
|
/// Mutates this clipping region to intersect with the given rectangle.
|
|
///
|
|
/// TODO(pcwalton): This could more eagerly eliminate complex clipping regions, at the cost of
|
|
/// complexity.
|
|
#[inline]
|
|
pub fn intersect_rect(&mut self, rect: &Rect<Au>) {
|
|
self.main = self.main.intersection(rect).unwrap_or(Rect::zero())
|
|
}
|
|
|
|
/// Returns true if this clipping region might be nonempty. This can return false positives,
|
|
/// but never false negatives.
|
|
#[inline]
|
|
pub fn might_be_nonempty(&self) -> bool {
|
|
!self.main.is_empty()
|
|
}
|
|
|
|
/// Returns true if this clipping region might contain the given point and false otherwise.
|
|
/// This is a quick, not a precise, test; it can yield false positives.
|
|
#[inline]
|
|
pub fn might_intersect_point(&self, point: &Point2D<Au>) -> bool {
|
|
self.main.contains(point) &&
|
|
self.complex.iter().all(|complex| complex.rect.contains(point))
|
|
}
|
|
|
|
/// Returns true if this clipping region might intersect the given rectangle and false
|
|
/// otherwise. This is a quick, not a precise, test; it can yield false positives.
|
|
#[inline]
|
|
pub fn might_intersect_rect(&self, rect: &Rect<Au>) -> bool {
|
|
self.main.intersects(rect) &&
|
|
self.complex.iter().all(|complex| complex.rect.intersects(rect))
|
|
}
|
|
|
|
/// Returns true if this clipping region completely surrounds the given rect.
|
|
#[inline]
|
|
pub fn does_not_clip_rect(&self, rect: &Rect<Au>) -> bool {
|
|
self.main.contains(&rect.origin) && self.main.contains(&rect.bottom_right()) &&
|
|
self.complex.iter().all(|complex| {
|
|
complex.rect.contains(&rect.origin) && complex.rect.contains(&rect.bottom_right())
|
|
})
|
|
}
|
|
|
|
/// Returns a bounding rect that surrounds this entire clipping region.
|
|
#[inline]
|
|
pub fn bounding_rect(&self) -> Rect<Au> {
|
|
let mut rect = self.main;
|
|
for complex in &*self.complex {
|
|
rect = rect.union(&complex.rect)
|
|
}
|
|
rect
|
|
}
|
|
|
|
/// Intersects this clipping region with the given rounded rectangle.
|
|
#[inline]
|
|
pub fn intersect_with_rounded_rect(&mut self, rect: &Rect<Au>, radii: &BorderRadii<Au>) {
|
|
let new_complex_region = ComplexClippingRegion {
|
|
rect: *rect,
|
|
radii: *radii,
|
|
};
|
|
|
|
// FIXME(pcwalton): This is O(n²) worst case for disjoint clipping regions. Is that OK?
|
|
// They're slow anyway…
|
|
//
|
|
// Possibly relevant if we want to do better:
|
|
//
|
|
// http://www.inrg.csie.ntu.edu.tw/algorithm2014/presentation/D&C%20Lee-84.pdf
|
|
for existing_complex_region in &mut self.complex {
|
|
if existing_complex_region.completely_encloses(&new_complex_region) {
|
|
*existing_complex_region = new_complex_region;
|
|
return
|
|
}
|
|
if new_complex_region.completely_encloses(existing_complex_region) {
|
|
return
|
|
}
|
|
}
|
|
|
|
self.complex.push(ComplexClippingRegion {
|
|
rect: *rect,
|
|
radii: *radii,
|
|
});
|
|
}
|
|
|
|
/// Translates this clipping region by the given vector.
|
|
#[inline]
|
|
pub fn translate(&self, delta: &Point2D<Au>) -> ClippingRegion {
|
|
ClippingRegion {
|
|
main: self.main.translate(delta),
|
|
complex: self.complex.iter().map(|complex| {
|
|
ComplexClippingRegion {
|
|
rect: complex.rect.translate(delta),
|
|
radii: complex.radii,
|
|
}
|
|
}).collect(),
|
|
}
|
|
}
|
|
|
|
#[inline]
|
|
pub fn is_max(&self) -> bool {
|
|
self.main == max_rect() && self.complex.is_empty()
|
|
}
|
|
}
|
|
|
|
impl fmt::Debug for ClippingRegion {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
if *self == ClippingRegion::max() {
|
|
write!(f, "ClippingRegion::Max")
|
|
} else if *self == ClippingRegion::empty() {
|
|
write!(f, "ClippingRegion::Empty")
|
|
} else if self.main == max_rect() {
|
|
write!(f, "ClippingRegion(Complex={:?})", self.complex)
|
|
} else {
|
|
write!(f, "ClippingRegion(Rect={:?}, Complex={:?})", self.main, self.complex)
|
|
}
|
|
}
|
|
}
|
|
|
|
impl ComplexClippingRegion {
|
|
// TODO(pcwalton): This could be more aggressive by considering points that touch the inside of
|
|
// the border radius ellipse.
|
|
fn completely_encloses(&self, other: &ComplexClippingRegion) -> bool {
|
|
let left = cmp::max(self.radii.top_left.width, self.radii.bottom_left.width);
|
|
let top = cmp::max(self.radii.top_left.height, self.radii.top_right.height);
|
|
let right = cmp::max(self.radii.top_right.width, self.radii.bottom_right.width);
|
|
let bottom = cmp::max(self.radii.bottom_left.height, self.radii.bottom_right.height);
|
|
let interior = Rect::new(Point2D::new(self.rect.origin.x + left, self.rect.origin.y + top),
|
|
Size2D::new(self.rect.size.width - left - right,
|
|
self.rect.size.height - top - bottom));
|
|
interior.origin.x <= other.rect.origin.x && interior.origin.y <= other.rect.origin.y &&
|
|
interior.max_x() >= other.rect.max_x() && interior.max_y() >= other.rect.max_y()
|
|
}
|
|
}
|
|
|
|
/// Metadata attached to each display item. This is useful for performing auxiliary threads with
|
|
/// the display list involving hit testing: finding the originating DOM node and determining the
|
|
/// cursor to use when the element is hovered over.
|
|
#[derive(Clone, Copy, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct DisplayItemMetadata {
|
|
/// The DOM node from which this display item originated.
|
|
pub node: OpaqueNode,
|
|
/// The value of the `cursor` property when the mouse hovers over this display item. If `None`,
|
|
/// this display item is ineligible for pointer events (`pointer-events: none`).
|
|
pub pointing: Option<Cursor>,
|
|
}
|
|
|
|
/// Paints a solid color.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct SolidColorDisplayItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The color.
|
|
pub color: ColorF,
|
|
}
|
|
|
|
/// Paints text.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct TextDisplayItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The text run.
|
|
#[ignore_heap_size_of = "Because it is non-owning"]
|
|
pub text_run: Arc<TextRun>,
|
|
|
|
/// The range of text within the text run.
|
|
pub range: Range<ByteIndex>,
|
|
|
|
/// The color of the text.
|
|
pub text_color: ColorF,
|
|
|
|
/// The position of the start of the baseline of this text.
|
|
pub baseline_origin: Point2D<Au>,
|
|
|
|
/// The orientation of the text: upright or sideways left/right.
|
|
pub orientation: TextOrientation,
|
|
|
|
/// The blur radius for this text. If zero, this text is not blurred.
|
|
pub blur_radius: Au,
|
|
}
|
|
|
|
#[derive(Clone, Eq, PartialEq, HeapSizeOf, Deserialize, Serialize)]
|
|
pub enum TextOrientation {
|
|
Upright,
|
|
SidewaysLeft,
|
|
SidewaysRight,
|
|
}
|
|
|
|
/// Paints an image.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct ImageDisplayItem {
|
|
pub base: BaseDisplayItem,
|
|
|
|
pub webrender_image: WebRenderImageInfo,
|
|
|
|
#[ignore_heap_size_of = "Because it is non-owning"]
|
|
pub image_data: Option<Arc<IpcSharedMemory>>,
|
|
|
|
/// The dimensions to which the image display item should be stretched. If this is smaller than
|
|
/// the bounds of this display item, then the image will be repeated in the appropriate
|
|
/// direction to tile the entire bounds.
|
|
pub stretch_size: Size2D<Au>,
|
|
|
|
/// The amount of space to add to the right and bottom part of each tile, when the image
|
|
/// is tiled.
|
|
pub tile_spacing: Size2D<Au>,
|
|
|
|
/// The algorithm we should use to stretch the image. See `image_rendering` in CSS-IMAGES-3 §
|
|
/// 5.3.
|
|
pub image_rendering: image_rendering::T,
|
|
}
|
|
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct WebGLDisplayItem {
|
|
pub base: BaseDisplayItem,
|
|
#[ignore_heap_size_of = "Defined in webrender_traits"]
|
|
pub context_id: WebGLContextId,
|
|
}
|
|
|
|
|
|
/// Paints an iframe.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct IframeDisplayItem {
|
|
pub base: BaseDisplayItem,
|
|
pub iframe: PipelineId,
|
|
}
|
|
|
|
/// Paints a gradient.
|
|
#[derive(Clone, Deserialize, HeapSizeOf, Serialize)]
|
|
pub struct GradientDisplayItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The start point of the gradient (computed during display list construction).
|
|
pub start_point: Point2D<Au>,
|
|
|
|
/// The end point of the gradient (computed during display list construction).
|
|
pub end_point: Point2D<Au>,
|
|
|
|
/// A list of color stops.
|
|
pub stops: Vec<GradientStop>,
|
|
}
|
|
|
|
/// Paints a border.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct BorderDisplayItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// Border widths.
|
|
pub border_widths: SideOffsets2D<Au>,
|
|
|
|
/// Border colors.
|
|
pub color: SideOffsets2D<ColorF>,
|
|
|
|
/// Border styles.
|
|
pub style: SideOffsets2D<border_style::T>,
|
|
|
|
/// Border radii.
|
|
///
|
|
/// TODO(pcwalton): Elliptical radii.
|
|
pub radius: BorderRadii<Au>,
|
|
}
|
|
|
|
/// Information about the border radii.
|
|
///
|
|
/// TODO(pcwalton): Elliptical radii.
|
|
#[derive(Clone, PartialEq, Debug, Copy, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct BorderRadii<T> {
|
|
pub top_left: Size2D<T>,
|
|
pub top_right: Size2D<T>,
|
|
pub bottom_right: Size2D<T>,
|
|
pub bottom_left: Size2D<T>,
|
|
}
|
|
|
|
impl<T> Default for BorderRadii<T> where T: Default, T: Clone {
|
|
fn default() -> Self {
|
|
let top_left = Size2D::new(Default::default(),
|
|
Default::default());
|
|
let top_right = Size2D::new(Default::default(),
|
|
Default::default());
|
|
let bottom_left = Size2D::new(Default::default(),
|
|
Default::default());
|
|
let bottom_right = Size2D::new(Default::default(),
|
|
Default::default());
|
|
BorderRadii { top_left: top_left,
|
|
top_right: top_right,
|
|
bottom_left: bottom_left,
|
|
bottom_right: bottom_right }
|
|
}
|
|
}
|
|
|
|
impl BorderRadii<Au> {
|
|
// Scale the border radii by the specified factor
|
|
pub fn scale_by(&self, s: f32) -> BorderRadii<Au> {
|
|
BorderRadii { top_left: BorderRadii::scale_corner_by(self.top_left, s),
|
|
top_right: BorderRadii::scale_corner_by(self.top_right, s),
|
|
bottom_left: BorderRadii::scale_corner_by(self.bottom_left, s),
|
|
bottom_right: BorderRadii::scale_corner_by(self.bottom_right, s) }
|
|
}
|
|
|
|
// Scale the border corner radius by the specified factor
|
|
pub fn scale_corner_by(corner: Size2D<Au>, s: f32) -> Size2D<Au> {
|
|
Size2D::new(corner.width.scale_by(s), corner.height.scale_by(s))
|
|
}
|
|
}
|
|
|
|
impl<T> BorderRadii<T> where T: PartialEq + Zero {
|
|
/// Returns true if all the radii are zero.
|
|
pub fn is_square(&self) -> bool {
|
|
let zero = Zero::zero();
|
|
self.top_left == zero && self.top_right == zero && self.bottom_right == zero &&
|
|
self.bottom_left == zero
|
|
}
|
|
}
|
|
|
|
impl<T> BorderRadii<T> where T: PartialEq + Zero + Clone {
|
|
/// Returns a set of border radii that all have the given value.
|
|
pub fn all_same(value: T) -> BorderRadii<T> {
|
|
BorderRadii {
|
|
top_left: Size2D::new(value.clone(), value.clone()),
|
|
top_right: Size2D::new(value.clone(), value.clone()),
|
|
bottom_right: Size2D::new(value.clone(), value.clone()),
|
|
bottom_left: Size2D::new(value.clone(), value.clone()),
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Paints a line segment.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct LineDisplayItem {
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The line segment color.
|
|
pub color: ColorF,
|
|
|
|
/// The line segment style.
|
|
pub style: border_style::T
|
|
}
|
|
|
|
/// Paints a box shadow per CSS-BACKGROUNDS.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct BoxShadowDisplayItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The dimensions of the box that we're placing a shadow around.
|
|
pub box_bounds: Rect<Au>,
|
|
|
|
/// The offset of this shadow from the box.
|
|
pub offset: Point2D<Au>,
|
|
|
|
/// The color of this shadow.
|
|
pub color: ColorF,
|
|
|
|
/// The blur radius for this shadow.
|
|
pub blur_radius: Au,
|
|
|
|
/// The spread radius of this shadow.
|
|
pub spread_radius: Au,
|
|
|
|
/// The border radius of this shadow.
|
|
///
|
|
/// TODO(pcwalton): Elliptical radii; different radii for each corner.
|
|
pub border_radius: Au,
|
|
|
|
/// How we should clip the result.
|
|
pub clip_mode: BoxShadowClipMode,
|
|
}
|
|
|
|
/// Defines a stacking context.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct PushStackingContextItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
pub stacking_context: StackingContext,
|
|
}
|
|
|
|
/// Defines a stacking context.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct PopStackingContextItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
pub stacking_context_id: StackingContextId,
|
|
}
|
|
|
|
/// Starts a group of items inside a particular scroll root.
|
|
#[derive(Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct PushScrollRootItem {
|
|
/// Fields common to all display items.
|
|
pub base: BaseDisplayItem,
|
|
|
|
/// The scroll root that this item starts.
|
|
pub scroll_root: ScrollRoot,
|
|
}
|
|
|
|
/// How a box shadow should be clipped.
|
|
#[derive(Clone, Copy, Debug, PartialEq, HeapSizeOf, Deserialize, Serialize)]
|
|
pub enum BoxShadowClipMode {
|
|
/// No special clipping should occur. This is used for (shadowed) text decorations.
|
|
None,
|
|
/// The area inside `box_bounds` should be clipped out. Corresponds to the normal CSS
|
|
/// `box-shadow`.
|
|
Outset,
|
|
/// The area outside `box_bounds` should be clipped out. Corresponds to the `inset` flag on CSS
|
|
/// `box-shadow`.
|
|
Inset,
|
|
}
|
|
|
|
impl DisplayItem {
|
|
pub fn base(&self) -> &BaseDisplayItem {
|
|
match *self {
|
|
DisplayItem::SolidColor(ref solid_color) => &solid_color.base,
|
|
DisplayItem::Text(ref text) => &text.base,
|
|
DisplayItem::Image(ref image_item) => &image_item.base,
|
|
DisplayItem::WebGL(ref webgl_item) => &webgl_item.base,
|
|
DisplayItem::Border(ref border) => &border.base,
|
|
DisplayItem::Gradient(ref gradient) => &gradient.base,
|
|
DisplayItem::Line(ref line) => &line.base,
|
|
DisplayItem::BoxShadow(ref box_shadow) => &box_shadow.base,
|
|
DisplayItem::Iframe(ref iframe) => &iframe.base,
|
|
DisplayItem::PushStackingContext(ref stacking_context) => &stacking_context.base,
|
|
DisplayItem::PopStackingContext(ref item) => &item.base,
|
|
DisplayItem::PushScrollRoot(ref item) => &item.base,
|
|
DisplayItem::PopScrollRoot(ref base) => &base,
|
|
}
|
|
}
|
|
|
|
pub fn scroll_root_id(&self) -> ScrollRootId {
|
|
self.base().scroll_root_id
|
|
}
|
|
|
|
pub fn stacking_context_id(&self) -> StackingContextId {
|
|
self.base().stacking_context_id
|
|
}
|
|
|
|
pub fn section(&self) -> DisplayListSection {
|
|
self.base().section
|
|
}
|
|
|
|
pub fn bounds(&self) -> Rect<Au> {
|
|
self.base().bounds
|
|
}
|
|
|
|
pub fn debug_with_level(&self, level: u32) {
|
|
let mut indent = String::new();
|
|
for _ in 0..level {
|
|
indent.push_str("| ")
|
|
}
|
|
println!("{}+ {:?}", indent, self);
|
|
}
|
|
|
|
fn hit_test(&self, point: Point2D<Au>) -> Option<DisplayItemMetadata> {
|
|
// TODO(pcwalton): Use a precise algorithm here. This will allow us to properly hit
|
|
// test elements with `border-radius`, for example.
|
|
let base_item = self.base();
|
|
|
|
if !base_item.clip.might_intersect_point(&point) {
|
|
// Clipped out.
|
|
return None;
|
|
}
|
|
if !self.bounds().contains(&point) {
|
|
// Can't possibly hit.
|
|
return None;
|
|
}
|
|
if base_item.metadata.pointing.is_none() {
|
|
// `pointer-events` is `none`. Ignore this item.
|
|
return None;
|
|
}
|
|
|
|
match *self {
|
|
DisplayItem::Border(ref border) => {
|
|
// If the point is inside the border, it didn't hit the border!
|
|
let interior_rect =
|
|
Rect::new(
|
|
Point2D::new(border.base.bounds.origin.x +
|
|
border.border_widths.left,
|
|
border.base.bounds.origin.y +
|
|
border.border_widths.top),
|
|
Size2D::new(border.base.bounds.size.width -
|
|
(border.border_widths.left +
|
|
border.border_widths.right),
|
|
border.base.bounds.size.height -
|
|
(border.border_widths.top +
|
|
border.border_widths.bottom)));
|
|
if interior_rect.contains(&point) {
|
|
return None;
|
|
}
|
|
}
|
|
DisplayItem::BoxShadow(_) => {
|
|
// Box shadows can never be hit.
|
|
return None;
|
|
}
|
|
_ => {}
|
|
}
|
|
|
|
Some(base_item.metadata)
|
|
}
|
|
}
|
|
|
|
impl fmt::Debug for DisplayItem {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
if let DisplayItem::PushStackingContext(ref item) = *self {
|
|
return write!(f, "PushStackingContext({:?})", item.stacking_context);
|
|
}
|
|
|
|
if let DisplayItem::PopStackingContext(ref item) = *self {
|
|
return write!(f, "PopStackingContext({:?}", item.stacking_context_id);
|
|
}
|
|
|
|
if let DisplayItem::PushScrollRoot(ref item) = *self {
|
|
return write!(f, "PushScrollRoot({:?}", item.scroll_root);
|
|
}
|
|
|
|
if let DisplayItem::PopScrollRoot(_) = *self {
|
|
return write!(f, "PopScrollRoot");
|
|
}
|
|
|
|
write!(f, "{} @ {:?} {:?}",
|
|
match *self {
|
|
DisplayItem::SolidColor(ref solid_color) =>
|
|
format!("SolidColor rgba({}, {}, {}, {})",
|
|
solid_color.color.r,
|
|
solid_color.color.g,
|
|
solid_color.color.b,
|
|
solid_color.color.a),
|
|
DisplayItem::Text(_) => "Text".to_owned(),
|
|
DisplayItem::Image(_) => "Image".to_owned(),
|
|
DisplayItem::WebGL(_) => "WebGL".to_owned(),
|
|
DisplayItem::Border(_) => "Border".to_owned(),
|
|
DisplayItem::Gradient(_) => "Gradient".to_owned(),
|
|
DisplayItem::Line(_) => "Line".to_owned(),
|
|
DisplayItem::BoxShadow(_) => "BoxShadow".to_owned(),
|
|
DisplayItem::Iframe(_) => "Iframe".to_owned(),
|
|
DisplayItem::PushStackingContext(_) |
|
|
DisplayItem::PopStackingContext(_) |
|
|
DisplayItem::PushScrollRoot(_) |
|
|
DisplayItem::PopScrollRoot(_) => "".to_owned(),
|
|
},
|
|
self.bounds(),
|
|
self.base().clip
|
|
)
|
|
}
|
|
}
|
|
|
|
#[derive(Copy, Clone, HeapSizeOf, Deserialize, Serialize)]
|
|
pub struct WebRenderImageInfo {
|
|
pub width: u32,
|
|
pub height: u32,
|
|
pub format: PixelFormat,
|
|
#[ignore_heap_size_of = "WebRender traits type, and tiny"]
|
|
pub key: Option<webrender_traits::ImageKey>,
|
|
}
|
|
|
|
impl WebRenderImageInfo {
|
|
#[inline]
|
|
pub fn from_image(image: &Image) -> WebRenderImageInfo {
|
|
WebRenderImageInfo {
|
|
width: image.width,
|
|
height: image.height,
|
|
format: image.format,
|
|
key: image.id,
|
|
}
|
|
}
|
|
}
|
|
|
|
/// The type of the scroll offset list. This is only populated if WebRender is in use.
|
|
pub type ScrollOffsetMap = HashMap<ScrollRootId, Point2D<f32>>;
|
|
|
|
|
|
pub trait SimpleMatrixDetection {
|
|
fn is_identity_or_simple_translation(&self) -> bool;
|
|
}
|
|
|
|
impl SimpleMatrixDetection for Matrix4D<f32> {
|
|
#[inline]
|
|
fn is_identity_or_simple_translation(&self) -> bool {
|
|
let (_0, _1) = (Zero::zero(), One::one());
|
|
self.m11 == _1 && self.m12 == _0 && self.m13 == _0 && self.m14 == _0 &&
|
|
self.m21 == _0 && self.m22 == _1 && self.m23 == _0 && self.m24 == _0 &&
|
|
self.m31 == _0 && self.m32 == _0 && self.m33 == _1 && self.m34 == _0 &&
|
|
self.m44 == _1
|
|
}
|
|
}
|