servo/components/layout/traversal.rs
Emilio Cobos Álvarez 3af774bd75
Rewrite the style sharing candidate cache.
The style candidate cache had regressed a few times (see #12534), and my
intuition is that being able to disable all style sharing with a single rule in
the page is really unfortunate.

This commit redesigns the style sharing cache in order to be a optimistic cache,
but then reject candidates if they match different sibling-affecting selectors
in the page, for example.

So far the numbers have improved, but not so much as I'd wanted (~10%/20% of
non-incremental restyling time in general). The current implementation is really
dumb though (we recompute and re-match a lot of stuff), so we should be able to
optimise it quite a bit.

I have different ideas for improving it (that may or may not work), apart of the
low-hanging fruit like don't re-matching candidates all the time but I have to
measure the real impact.

Also, I need to verify it against try.
2016-08-17 14:16:16 -07:00

234 lines
9.1 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/. */
//! Traversals over the DOM and flow trees, running the layout computations.
use construct::FlowConstructor;
use context::{LayoutContext, SharedLayoutContext};
use display_list_builder::DisplayListBuildState;
use flow::{CAN_BE_FRAGMENTED, Flow, ImmutableFlowUtils, PostorderFlowTraversal};
use flow::{PreorderFlowTraversal, self};
use gfx::display_list::OpaqueNode;
use script_layout_interface::restyle_damage::{BUBBLE_ISIZES, REFLOW, REFLOW_OUT_OF_FLOW, REPAINT, RestyleDamage};
use script_layout_interface::wrapper_traits::{LayoutNode, ThreadSafeLayoutNode};
use std::mem;
use style::context::{LocalStyleContext, SharedStyleContext, StyleContext};
use style::dom::TNode;
use style::selector_impl::ServoSelectorImpl;
use style::traversal::RestyleResult;
use style::traversal::{DomTraversalContext, remove_from_bloom_filter, recalc_style_at};
use util::opts;
use wrapper::{LayoutNodeLayoutData, ThreadSafeLayoutNodeHelpers};
pub struct RecalcStyleAndConstructFlows<'lc> {
context: LayoutContext<'lc>,
root: OpaqueNode,
}
impl<'lc, N> DomTraversalContext<N> for RecalcStyleAndConstructFlows<'lc>
where N: LayoutNode + TNode,
N::ConcreteElement: ::selectors::Element<Impl=ServoSelectorImpl>
{
type SharedContext = SharedLayoutContext;
#[allow(unsafe_code)]
fn new<'a>(shared: &'a Self::SharedContext, root: OpaqueNode) -> Self {
// FIXME(bholley): This transmutation from &'a to &'lc is very unfortunate, but I haven't
// found a way to avoid it despite spending several days on it (and consulting Manishearth,
// brson, and nmatsakis).
//
// The crux of the problem is that parameterizing DomTraversalContext on the lifetime of
// the SharedContext doesn't work for a variety of reasons [1]. However, the code in
// parallel.rs needs to be able to use the DomTraversalContext trait (or something similar)
// to stack-allocate a struct (a generalized LayoutContext<'a>) that holds a borrowed
// SharedContext, which means that the struct needs to be parameterized on a lifetime.
// Given the aforementioned constraint, the only way to accomplish this is to avoid
// propagating the borrow lifetime from the struct to the trait, but that means that the
// new() method on the trait cannot require the lifetime of its argument to match the
// lifetime of the Self object it creates.
//
// This could be solved with an associated type with an unbound lifetime parameter, but
// that would require higher-kinded types, which don't exist yet and probably aren't coming
// for a while.
//
// So we transmute. :-( This is safe because the DomTravesalContext is stack-allocated on
// the worker thread while processing a WorkUnit, whereas the borrowed SharedContext is
// live for the entire duration of the restyle. This really could _almost_ compile: all
// we'd need to do is change the signature to to |new<'a: 'lc>|, and everything would
// work great. But we can't do that, because that would cause a mismatch with the signature
// in the trait we're implementing, and we can't mention 'lc in that trait at all for the
// reasons described above.
//
// [1] For example, the WorkQueue type needs to be parameterized on the concrete type of
// DomTraversalContext::SharedContext, and the WorkQueue lifetime is similar to that of the
// LayoutThread, generally much longer than that of a given SharedLayoutContext borrow.
let shared_lc: &'lc SharedLayoutContext = unsafe { mem::transmute(shared) };
RecalcStyleAndConstructFlows {
context: LayoutContext::new(shared_lc),
root: root,
}
}
fn process_preorder(&self, node: N) -> RestyleResult {
// FIXME(pcwalton): Stop allocating here. Ideally this should just be
// done by the HTML parser.
node.initialize_data();
recalc_style_at(&self.context, self.root, node)
}
fn process_postorder(&self, node: N) {
construct_flows_at(&self.context, self.root, node);
}
fn local_context(&self) -> &LocalStyleContext {
self.context.local_context()
}
}
/// A bottom-up, parallelizable traversal.
pub trait PostorderNodeMutTraversal<ConcreteThreadSafeLayoutNode: ThreadSafeLayoutNode> {
/// The operation to perform. Return true to continue or false to stop.
fn process(&mut self, node: &ConcreteThreadSafeLayoutNode);
}
/// The flow construction traversal, which builds flows for styled nodes.
#[inline]
#[allow(unsafe_code)]
fn construct_flows_at<'a, N: LayoutNode>(context: &'a LayoutContext<'a>, root: OpaqueNode, node: N) {
// Construct flows for this node.
{
let tnode = node.to_threadsafe();
// Always reconstruct if incremental layout is turned off.
let nonincremental_layout = opts::get().nonincremental_layout;
if nonincremental_layout || node.is_dirty() || node.has_dirty_descendants() {
let mut flow_constructor = FlowConstructor::new(context);
if nonincremental_layout || !flow_constructor.repair_if_possible(&tnode) {
flow_constructor.process(&tnode);
debug!("Constructed flow for {:x}: {:x}",
tnode.debug_id(),
tnode.flow_debug_id());
}
}
// Reset the layout damage in this node. It's been propagated to the
// flow by the flow constructor.
tnode.set_restyle_damage(RestyleDamage::empty());
}
unsafe {
node.set_changed(false);
node.set_dirty(false);
node.set_dirty_descendants(false);
}
remove_from_bloom_filter(context, root, node);
}
/// The bubble-inline-sizes traversal, the first part of layout computation. This computes
/// preferred and intrinsic inline-sizes and bubbles them up the tree.
pub struct BubbleISizes<'a> {
pub layout_context: &'a LayoutContext<'a>,
}
impl<'a> PostorderFlowTraversal for BubbleISizes<'a> {
#[inline]
fn process(&self, flow: &mut Flow) {
flow.bubble_inline_sizes();
flow::mut_base(flow).restyle_damage.remove(BUBBLE_ISIZES);
}
#[inline]
fn should_process(&self, flow: &mut Flow) -> bool {
flow::base(flow).restyle_damage.contains(BUBBLE_ISIZES)
}
}
/// The assign-inline-sizes traversal. In Gecko this corresponds to `Reflow`.
#[derive(Copy, Clone)]
pub struct AssignISizes<'a> {
pub shared_context: &'a SharedStyleContext,
}
impl<'a> PreorderFlowTraversal for AssignISizes<'a> {
#[inline]
fn process(&self, flow: &mut Flow) {
flow.assign_inline_sizes(self.shared_context);
}
#[inline]
fn should_process(&self, flow: &mut Flow) -> bool {
flow::base(flow).restyle_damage.intersects(REFLOW_OUT_OF_FLOW | REFLOW)
}
}
/// The assign-block-sizes-and-store-overflow traversal, the last (and most expensive) part of
/// layout computation. Determines the final block-sizes for all layout objects and computes
/// positions. In Gecko this corresponds to `Reflow`.
#[derive(Copy, Clone)]
pub struct AssignBSizes<'a> {
pub layout_context: &'a LayoutContext<'a>,
}
impl<'a> PostorderFlowTraversal for AssignBSizes<'a> {
#[inline]
fn process(&self, flow: &mut Flow) {
// Can't do anything with anything that floats might flow through until we reach their
// inorder parent.
//
// NB: We must return without resetting the restyle bits for these, as we haven't actually
// reflowed anything!
if flow.floats_might_flow_through() {
return
}
flow.assign_block_size(self.layout_context);
}
#[inline]
fn should_process(&self, flow: &mut Flow) -> bool {
let base = flow::base(flow);
base.restyle_damage.intersects(REFLOW_OUT_OF_FLOW | REFLOW) &&
// The fragmentation countainer is responsible for calling Flow::fragment recursively
!base.flags.contains(CAN_BE_FRAGMENTED)
}
}
#[derive(Copy, Clone)]
pub struct ComputeAbsolutePositions<'a> {
pub layout_context: &'a LayoutContext<'a>,
}
impl<'a> PreorderFlowTraversal for ComputeAbsolutePositions<'a> {
#[inline]
fn process(&self, flow: &mut Flow) {
flow.compute_absolute_position(self.layout_context);
}
}
pub struct BuildDisplayList<'a> {
pub state: DisplayListBuildState<'a>,
}
impl<'a> BuildDisplayList<'a> {
#[inline]
pub fn traverse(&mut self, flow: &mut Flow) {
if self.should_process() {
self.state.push_stacking_context_id(flow::base(flow).stacking_context_id);
flow.build_display_list(&mut self.state);
flow::mut_base(flow).restyle_damage.remove(REPAINT);
self.state.pop_stacking_context_id();
}
for kid in flow::child_iter_mut(flow) {
self.traverse(kid);
}
}
#[inline]
fn should_process(&self) -> bool {
true
}
}