gecko-dev/servo/components/style/stylist.rs

1560 lines
64 KiB
Rust
Raw Normal View History

/* 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/. */
//! Selector matching.
use {Atom, LocalName, Namespace};
use applicable_declarations::{ApplicableDeclarationBlock, ApplicableDeclarationList};
use bit_vec::BitVec;
use context::QuirksMode;
use data::ComputedStyle;
use dom::TElement;
use element_state::ElementState;
use error_reporting::create_error_reporter;
use font_metrics::FontMetricsProvider;
#[cfg(feature = "gecko")]
use gecko_bindings::structs::{nsIAtom, StyleRuleInclusion};
use invalidation::element::invalidation_map::InvalidationMap;
use invalidation::media_queries::EffectiveMediaQueryResults;
use media_queries::Device;
use properties::{self, CascadeFlags, ComputedValues};
use properties::{AnimationRules, PropertyDeclarationBlock};
#[cfg(feature = "servo")]
use properties::INHERIT_ALL;
use rule_tree::{CascadeLevel, RuleTree, StrongRuleNode, StyleSource};
use selector_map::{SelectorMap, SelectorMapEntry};
use selector_parser::{SelectorImpl, PseudoElement};
use selectors::attr::NamespaceConstraint;
use selectors::bloom::BloomFilter;
use selectors::matching::{ElementSelectorFlags, matches_selector, MatchingContext, MatchingMode};
use selectors::matching::AFFECTED_BY_PRESENTATIONAL_HINTS;
use selectors::parser::{AncestorHashes, Combinator, Component, Selector, SelectorAndHashes};
use selectors::parser::{SelectorIter, SelectorMethods};
use selectors::visitor::SelectorVisitor;
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
use shared_lock::{Locked, SharedRwLockReadGuard, StylesheetGuards};
use sink::Push;
use smallvec::VecLike;
use std::fmt::Debug;
#[cfg(feature = "servo")]
use std::marker::PhantomData;
use style_traits::viewport::ViewportConstraints;
use stylearc::Arc;
#[cfg(feature = "gecko")]
use stylesheets::{CounterStyleRule, FontFaceRule};
use stylesheets::{CssRule, StyleRule};
use stylesheets::{Stylesheet, Origin, UserAgentStylesheets};
use stylesheets::keyframes_rule::KeyframesAnimation;
use stylesheets::viewport_rule::{self, MaybeNew, ViewportRule};
use thread_state;
pub use ::fnv::FnvHashMap;
/// This structure holds all the selectors and device characteristics
/// for a given document. The selectors are converted into `Rule`s
/// (defined in rust-selectors), and introduced in a `SelectorMap`
/// depending on the pseudo-element (see `PerPseudoElementSelectorMap`),
/// and stylesheet origin (see the fields of `PerPseudoElementSelectorMap`).
///
/// This structure is effectively created once per pipeline, in the
/// LayoutThread corresponding to that pipeline.
#[cfg_attr(feature = "servo", derive(HeapSizeOf))]
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
pub struct Stylist {
/// Device that the stylist is currently evaluating against.
///
/// This field deserves a bigger comment due to the different use that Gecko
/// and Servo give to it (that we should eventually unify).
///
/// With Gecko, the device is never changed. Gecko manually tracks whether
/// the device data should be reconstructed, and "resets" the state of the
/// device.
///
/// On Servo, on the other hand, the device is a really cheap representation
/// that is recreated each time some constraint changes and calling
/// `set_device`.
device: Device,
/// Viewport constraints based on the current device.
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
viewport_constraints: Option<ViewportConstraints>,
/// Effective media query results cached from the last rebuild.
effective_media_query_results: EffectiveMediaQueryResults,
/// If true, the quirks-mode stylesheet is applied.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "defined in selectors")]
quirks_mode: QuirksMode,
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
/// If true, the device has changed, and the stylist needs to be updated.
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
is_device_dirty: bool,
/// If true, the stylist is in a cleared state (e.g. just-constructed, or
/// had clear() called on it with no following rebuild()).
is_cleared: bool,
/// The current selector maps, after evaluating media
/// rules against the current device.
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
element_map: PerPseudoElementSelectorMap,
/// The rule tree, that stores the results of selector matching.
rule_tree: RuleTree,
/// The selector maps corresponding to a given pseudo-element
/// (depending on the implementation)
pseudos_map: FnvHashMap<PseudoElement, PerPseudoElementSelectorMap>,
/// A map with all the animations indexed by name.
animations: FnvHashMap<Atom, KeyframesAnimation>,
/// Applicable declarations for a given non-eagerly cascaded pseudo-element.
/// These are eagerly computed once, and then used to resolve the new
/// computed values on the fly on layout.
///
/// FIXME(emilio): Use the rule tree!
precomputed_pseudo_element_decls: FnvHashMap<PseudoElement, Vec<ApplicableDeclarationBlock>>,
/// A monotonically increasing counter to represent the order on which a
/// style rule appears in a stylesheet, needed to sort them by source order.
rules_source_order: u32,
/// The invalidation map for this document.
invalidation_map: InvalidationMap,
/// The attribute local names that appear in attribute selectors. Used
/// to avoid taking element snapshots when an irrelevant attribute changes.
/// (We don't bother storing the namespace, since namespaced attributes
/// are rare.)
///
/// FIXME(heycam): This doesn't really need to be a counting Bloom filter.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "just an array")]
attribute_dependencies: BloomFilter,
/// Whether `"style"` appears in an attribute selector. This is not common,
/// and by tracking this explicitly, we can avoid taking an element snapshot
/// in the common case of style=""` changing due to modifying
/// `element.style`. (We could track this in `attribute_dependencies`, like
/// all other attributes, but we should probably not risk incorrectly
/// returning `true` for `"style"` just due to a hash collision.)
style_attribute_dependency: bool,
/// The element state bits that are relied on by selectors. Like
/// `attribute_dependencies`, this is used to avoid taking element snapshots
/// when an irrelevant element state bit changes.
state_dependencies: ElementState,
/// The ids that appear in the rightmost complex selector of selectors (and
/// hence in our selector maps). Used to determine when sharing styles is
/// safe: we disallow style sharing for elements whose id matches this
/// filter, and hence might be in one of our selector maps.
///
/// FIXME(bz): This doesn't really need to be a counting Blooom filter.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "just an array")]
mapped_ids: BloomFilter,
/// Selectors that require explicit cache revalidation (i.e. which depend
/// on state that is not otherwise visible to the cache, like attributes or
/// tree-structural state like child index and pseudos).
#[cfg_attr(feature = "servo", ignore_heap_size_of = "Arc")]
selectors_for_cache_revalidation: SelectorMap<RevalidationSelectorAndHashes>,
/// The total number of selectors.
num_selectors: usize,
/// The total number of declarations.
num_declarations: usize,
/// The total number of times the stylist has been rebuilt.
num_rebuilds: usize,
}
/// This struct holds data which user of Stylist may want to extract
/// from stylesheets which can be done at the same time as updating.
#[cfg(feature = "gecko")]
pub struct ExtraStyleData<'a> {
/// A list of effective font-face rules and their origin.
pub font_faces: &'a mut Vec<(Arc<Locked<FontFaceRule>>, Origin)>,
/// A map of effective counter-style rules.
pub counter_styles: &'a mut FnvHashMap<Atom, Arc<Locked<CounterStyleRule>>>,
}
#[cfg(feature = "gecko")]
impl<'a> ExtraStyleData<'a> {
/// Clear the internal data.
fn clear(&mut self) {
self.font_faces.clear();
self.counter_styles.clear();
}
/// Add the given @font-face rule.
fn add_font_face(&mut self, rule: &Arc<Locked<FontFaceRule>>, origin: Origin) {
self.font_faces.push((rule.clone(), origin));
}
/// Add the given @counter-style rule.
fn add_counter_style(&mut self, guard: &SharedRwLockReadGuard,
rule: &Arc<Locked<CounterStyleRule>>) {
let name = rule.read_with(guard).mName.raw::<nsIAtom>().into();
self.counter_styles.insert(name, rule.clone());
}
}
#[allow(missing_docs)]
#[cfg(feature = "servo")]
pub struct ExtraStyleData<'a> {
pub marker: PhantomData<&'a usize>,
}
#[cfg(feature = "servo")]
impl<'a> ExtraStyleData<'a> {
fn clear(&mut self) {}
}
/// What cascade levels to include when styling elements.
#[derive(Copy, Clone, PartialEq)]
pub enum RuleInclusion {
/// Include rules for style sheets at all cascade levels. This is the
/// normal rule inclusion mode.
All,
/// Only include rules from UA and user level sheets. Used to implement
/// `getDefaultComputedStyle`.
DefaultOnly,
}
#[cfg(feature = "gecko")]
impl From<StyleRuleInclusion> for RuleInclusion {
fn from(value: StyleRuleInclusion) -> Self {
match value {
StyleRuleInclusion::All => RuleInclusion::All,
StyleRuleInclusion::DefaultOnly => RuleInclusion::DefaultOnly,
}
}
}
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
impl Stylist {
/// Construct a new `Stylist`, using given `Device` and `QuirksMode`.
/// If more members are added here, think about whether they should
/// be reset in clear().
#[inline]
pub fn new(device: Device, quirks_mode: QuirksMode) -> Self {
let mut stylist = Stylist {
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
viewport_constraints: None,
device: device,
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
is_device_dirty: true,
is_cleared: true,
quirks_mode: quirks_mode,
effective_media_query_results: EffectiveMediaQueryResults::new(),
element_map: PerPseudoElementSelectorMap::new(),
pseudos_map: Default::default(),
animations: Default::default(),
precomputed_pseudo_element_decls: Default::default(),
rules_source_order: 0,
rule_tree: RuleTree::new(),
invalidation_map: InvalidationMap::new(),
attribute_dependencies: BloomFilter::new(),
style_attribute_dependency: false,
state_dependencies: ElementState::empty(),
mapped_ids: BloomFilter::new(),
selectors_for_cache_revalidation: SelectorMap::new(),
num_selectors: 0,
num_declarations: 0,
num_rebuilds: 0,
};
SelectorImpl::each_eagerly_cascaded_pseudo_element(|pseudo| {
stylist.pseudos_map.insert(pseudo, PerPseudoElementSelectorMap::new());
});
// FIXME: Add iso-8859-9.css when the documents encoding is ISO-8859-8.
stylist
}
/// Returns the number of selectors.
pub fn num_selectors(&self) -> usize {
self.num_selectors
}
/// Returns the number of declarations.
pub fn num_declarations(&self) -> usize {
self.num_declarations
}
/// Returns the number of times the stylist has been rebuilt.
pub fn num_rebuilds(&self) -> usize {
self.num_rebuilds
}
/// Returns the number of revalidation_selectors.
pub fn num_revalidation_selectors(&self) -> usize {
self.selectors_for_cache_revalidation.len()
}
/// Gets a reference to the invalidation map.
pub fn invalidation_map(&self) -> &InvalidationMap {
&self.invalidation_map
}
/// Clear the stylist's state, effectively resetting it to more or less
/// the state Stylist::new creates.
///
/// We preserve the state of the following members:
/// device: Someone might have set this on us.
/// quirks_mode: Again, someone might have set this on us.
/// num_rebuilds: clear() followed by rebuild() should just increment this
///
/// We don't just use struct update syntax with Stylist::new(self.device)
/// beause for some of our members we can clear them instead of creating new
/// objects. This does cause unfortunate code duplication with
/// Stylist::new.
pub fn clear(&mut self) {
if self.is_cleared {
return
}
self.is_cleared = true;
self.effective_media_query_results.clear();
self.viewport_constraints = None;
// preserve current device
self.is_device_dirty = true;
// preserve current quirks_mode value
self.element_map = PerPseudoElementSelectorMap::new();
self.pseudos_map = Default::default();
self.animations.clear(); // Or set to Default::default()?
self.precomputed_pseudo_element_decls = Default::default();
self.rules_source_order = 0;
// We want to keep rule_tree around across stylist rebuilds.
self.invalidation_map.clear();
self.attribute_dependencies.clear();
self.style_attribute_dependency = false;
self.state_dependencies = ElementState::empty();
self.mapped_ids.clear();
self.selectors_for_cache_revalidation = SelectorMap::new();
self.num_selectors = 0;
self.num_declarations = 0;
// preserve num_rebuilds value, since it should stay across
// clear()/rebuild() cycles.
}
/// rebuild the stylist for the given document stylesheets, and optionally
/// with a set of user agent stylesheets.
///
/// This method resets all the style data each time the stylesheets change
/// (which is indicated by the `stylesheets_changed` parameter), or the
/// device is dirty, which means we need to re-evaluate media queries.
pub fn rebuild<'a, 'b, I>(&mut self,
doc_stylesheets: I,
guards: &StylesheetGuards,
ua_stylesheets: Option<&UserAgentStylesheets>,
stylesheets_changed: bool,
author_style_disabled: bool,
extra_data: &mut ExtraStyleData<'a>) -> bool
where I: Iterator<Item = &'b Arc<Stylesheet>> + Clone,
{
debug_assert!(!self.is_cleared || self.is_device_dirty);
self.is_cleared = false;
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
if !(self.is_device_dirty || stylesheets_changed) {
return false;
}
self.num_rebuilds += 1;
self.viewport_constraints = None;
if viewport_rule::enabled() {
// TODO(emilio): This doesn't look so efficient.
//
// Presumably when we properly implement this we can at least have a
// bit on the stylesheet that says whether it contains viewport
// rules to skip it entirely?
//
// Processing it with the rest of rules seems tricky since it
// overrides the viewport size which may change the evaluation of
// media queries (or may not? how are viewport units in media
// queries defined?)
let cascaded_rule = ViewportRule {
declarations: viewport_rule::Cascade::from_stylesheets(
doc_stylesheets.clone(), guards.author, &self.device
).finish()
};
self.viewport_constraints =
ViewportConstraints::maybe_new(&self.device,
&cascaded_rule,
self.quirks_mode)
}
if let Some(ref constraints) = self.viewport_constraints {
self.device.account_for_viewport_rule(constraints);
}
SelectorImpl::each_eagerly_cascaded_pseudo_element(|pseudo| {
self.pseudos_map.insert(pseudo, PerPseudoElementSelectorMap::new());
});
extra_data.clear();
if let Some(ua_stylesheets) = ua_stylesheets {
for stylesheet in &ua_stylesheets.user_or_user_agent_stylesheets {
self.add_stylesheet(&stylesheet, guards.ua_or_user, extra_data);
}
if self.quirks_mode != QuirksMode::NoQuirks {
self.add_stylesheet(&ua_stylesheets.quirks_mode_stylesheet,
guards.ua_or_user, extra_data);
}
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
}
// Only use author stylesheets if author styles are enabled.
let sheets_to_add = doc_stylesheets.filter(|s| {
!author_style_disabled || s.origin != Origin::Author
});
for ref stylesheet in sheets_to_add {
self.add_stylesheet(stylesheet, guards.author, extra_data);
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
}
SelectorImpl::each_precomputed_pseudo_element(|pseudo| {
if let Some(map) = self.pseudos_map.remove(&pseudo) {
let declarations = map.user_agent.get_universal_rules(CascadeLevel::UANormal);
self.precomputed_pseudo_element_decls.insert(pseudo, declarations);
}
});
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
self.is_device_dirty = false;
true
}
/// clear the stylist and then rebuild it. Chances are, you want to use
/// either clear() or rebuild(), with the latter done lazily, instead.
pub fn update<'a, 'b, I>(&mut self,
doc_stylesheets: I,
guards: &StylesheetGuards,
ua_stylesheets: Option<&UserAgentStylesheets>,
stylesheets_changed: bool,
author_style_disabled: bool,
extra_data: &mut ExtraStyleData<'a>) -> bool
where I: Iterator<Item = &'b Arc<Stylesheet>> + Clone,
{
debug_assert!(!self.is_cleared || self.is_device_dirty);
// We have to do a dirtiness check before clearing, because if
// we're not actually dirty we need to no-op here.
if !(self.is_device_dirty || stylesheets_changed) {
return false;
}
self.clear();
self.rebuild(doc_stylesheets, guards, ua_stylesheets, stylesheets_changed,
author_style_disabled, extra_data)
}
fn add_stylesheet<'a>(&mut self,
stylesheet: &Stylesheet,
guard: &SharedRwLockReadGuard,
_extra_data: &mut ExtraStyleData<'a>) {
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
if stylesheet.disabled() || !stylesheet.is_effective_for_device(&self.device, guard) {
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
return;
}
self.effective_media_query_results.saw_effective(stylesheet);
for rule in stylesheet.effective_rules(&self.device, guard) {
match *rule {
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
CssRule::Style(ref locked) => {
let style_rule = locked.read_with(&guard);
self.num_declarations += style_rule.block.read_with(&guard).len();
for selector_and_hashes in &style_rule.selectors.0 {
self.num_selectors += 1;
let map = if let Some(pseudo) = selector_and_hashes.selector.pseudo_element() {
self.pseudos_map
.entry(pseudo.canonical())
.or_insert_with(PerPseudoElementSelectorMap::new)
.borrow_for_origin(&stylesheet.origin)
} else {
self.element_map.borrow_for_origin(&stylesheet.origin)
};
map.insert(
Rule::new(selector_and_hashes.selector.clone(),
selector_and_hashes.hashes.clone(),
locked.clone(),
self.rules_source_order),
self.quirks_mode);
self.invalidation_map.note_selector(selector_and_hashes, self.quirks_mode);
if needs_revalidation(&selector_and_hashes.selector) {
self.selectors_for_cache_revalidation.insert(
RevalidationSelectorAndHashes::new(&selector_and_hashes),
self.quirks_mode);
}
selector_and_hashes.selector.visit(&mut AttributeAndStateDependencyVisitor {
attribute_dependencies: &mut self.attribute_dependencies,
style_attribute_dependency: &mut self.style_attribute_dependency,
state_dependencies: &mut self.state_dependencies,
});
selector_and_hashes.selector.visit(&mut MappedIdVisitor {
mapped_ids: &mut self.mapped_ids,
});
}
self.rules_source_order += 1;
}
CssRule::Import(ref lock) => {
let import_rule = lock.read_with(guard);
self.effective_media_query_results.saw_effective(import_rule);
// NOTE: effective_rules visits the inner stylesheet if
// appropriate.
}
CssRule::Media(ref lock) => {
let media_rule = lock.read_with(guard);
self.effective_media_query_results.saw_effective(media_rule);
}
CssRule::Keyframes(ref keyframes_rule) => {
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
let keyframes_rule = keyframes_rule.read_with(guard);
debug!("Found valid keyframes rule: {:?}", *keyframes_rule);
// Don't let a prefixed keyframes animation override a non-prefixed one.
let needs_insertion = keyframes_rule.vendor_prefix.is_none() ||
self.animations.get(keyframes_rule.name.as_atom()).map_or(true, |rule|
rule.vendor_prefix.is_some());
if needs_insertion {
let animation = KeyframesAnimation::from_keyframes(
&keyframes_rule.keyframes, keyframes_rule.vendor_prefix.clone(), guard);
debug!("Found valid keyframe animation: {:?}", animation);
self.animations.insert(keyframes_rule.name.as_atom().clone(), animation);
}
}
#[cfg(feature = "gecko")]
CssRule::FontFace(ref rule) => {
_extra_data.add_font_face(&rule, stylesheet.origin);
}
#[cfg(feature = "gecko")]
CssRule::CounterStyle(ref rule) => {
_extra_data.add_counter_style(guard, &rule);
}
// We don't care about any other rule.
_ => {}
}
}
}
/// Returns whether the given attribute might appear in an attribute
/// selector of some rule in the stylist.
pub fn might_have_attribute_dependency(&self,
local_name: &LocalName)
-> bool {
if *local_name == local_name!("style") {
self.style_attribute_dependency
} else {
self.attribute_dependencies.might_contain(local_name)
}
}
/// Returns whether the given ElementState bit is relied upon by a selector
/// of some rule in the stylist.
pub fn has_state_dependency(&self, state: ElementState) -> bool {
self.state_dependencies.intersects(state)
}
/// Computes the style for a given "precomputed" pseudo-element, taking the
/// universal rules and applying them.
///
/// If `inherit_all` is true, then all properties are inherited from the
/// parent; otherwise, non-inherited properties are reset to their initial
/// values. The flow constructor uses this flag when constructing anonymous
/// flows.
pub fn precomputed_values_for_pseudo(&self,
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
guards: &StylesheetGuards,
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
pseudo: &PseudoElement,
parent: Option<&Arc<ComputedValues>>,
cascade_flags: CascadeFlags,
font_metrics: &FontMetricsProvider)
-> ComputedStyle {
debug_assert!(pseudo.is_precomputed());
let rule_node = match self.precomputed_pseudo_element_decls.get(pseudo) {
Some(declarations) => {
// FIXME(emilio): When we've taken rid of the cascade we can just
// use into_iter.
self.rule_tree.insert_ordered_rules_with_important(
declarations.into_iter().map(|a| (a.source.clone(), a.level())),
guards)
}
None => self.rule_tree.root(),
};
// NOTE(emilio): We skip calculating the proper layout parent style
// here.
//
// It'd be fine to assert that this isn't called with a parent style
// where display contents is in effect, but in practice this is hard to
// do for stuff like :-moz-fieldset-content with a
// <fieldset style="display: contents">. That is, the computed value of
// display for the fieldset is "contents", even though it's not the used
// value, so we don't need to adjust in a different way anyway.
//
// In practice, I don't think any anonymous content can be a direct
// descendant of a display: contents element where display: contents is
// the actual used value, and the computed value of it would need
// blockification.
let computed =
properties::cascade(&self.device,
&rule_node,
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
guards,
parent.map(|p| &**p),
parent.map(|p| &**p),
None,
None,
&create_error_reporter(),
font_metrics,
cascade_flags,
self.quirks_mode);
ComputedStyle::new(rule_node, Arc::new(computed))
}
/// Returns the style for an anonymous box of the given type.
#[cfg(feature = "servo")]
pub fn style_for_anonymous(&self,
guards: &StylesheetGuards,
pseudo: &PseudoElement,
parent_style: &Arc<ComputedValues>)
-> Arc<ComputedValues> {
use font_metrics::ServoMetricsProvider;
// For most (but not all) pseudo-elements, we inherit all values from the parent.
let inherit_all = match *pseudo {
PseudoElement::ServoText |
PseudoElement::ServoInputText => false,
PseudoElement::ServoAnonymousBlock |
PseudoElement::ServoAnonymousTable |
PseudoElement::ServoAnonymousTableCell |
PseudoElement::ServoAnonymousTableRow |
PseudoElement::ServoAnonymousTableWrapper |
PseudoElement::ServoTableWrapper |
PseudoElement::ServoInlineBlockWrapper |
PseudoElement::ServoInlineAbsolute => true,
PseudoElement::Before |
PseudoElement::After |
PseudoElement::Selection |
PseudoElement::DetailsSummary |
PseudoElement::DetailsContent => {
unreachable!("That pseudo doesn't represent an anonymous box!")
}
};
let mut cascade_flags = CascadeFlags::empty();
if inherit_all {
cascade_flags.insert(INHERIT_ALL);
}
self.precomputed_values_for_pseudo(guards, &pseudo, Some(parent_style), cascade_flags,
&ServoMetricsProvider)
.values.unwrap()
}
/// Computes a pseudo-element style lazily during layout.
///
/// This can only be done for a certain set of pseudo-elements, like
/// :selection.
///
/// Check the documentation on lazy pseudo-elements in
/// docs/components/style.md
pub fn lazily_compute_pseudo_element_style<E>(&self,
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
guards: &StylesheetGuards,
element: &E,
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
pseudo: &PseudoElement,
rule_inclusion: RuleInclusion,
parent_style: &ComputedValues,
font_metrics: &FontMetricsProvider)
-> Option<ComputedStyle>
where E: TElement,
{
let rule_node =
match self.lazy_pseudo_rules(guards, element, pseudo, rule_inclusion) {
Some(rule_node) => rule_node,
None => return None
};
// Read the comment on `precomputed_values_for_pseudo` to see why it's
// difficult to assert that display: contents nodes never arrive here
// (tl;dr: It doesn't apply for replaced elements and such, but the
// computed value is still "contents").
// Bug 1364242: We need to add visited support for lazy pseudos
let computed =
properties::cascade(&self.device,
&rule_node,
guards,
Some(parent_style),
Some(parent_style),
None,
None,
&create_error_reporter(),
font_metrics,
CascadeFlags::empty(),
self.quirks_mode);
Some(ComputedStyle::new(rule_node, Arc::new(computed)))
}
/// Computes the rule node for a lazily-cascaded pseudo-element.
///
/// See the documentation on lazy pseudo-elements in
/// docs/components/style.md
pub fn lazy_pseudo_rules<E>(&self,
guards: &StylesheetGuards,
element: &E,
pseudo: &PseudoElement,
rule_inclusion: RuleInclusion)
-> Option<StrongRuleNode>
where E: TElement
{
let pseudo = pseudo.canonical();
debug_assert!(pseudo.is_lazy());
if self.pseudos_map.get(&pseudo).is_none() {
return None
}
// Apply the selector flags. We should be in sequential mode
// already, so we can directly apply the parent flags.
let mut set_selector_flags = |element: &E, flags: ElementSelectorFlags| {
if cfg!(feature = "servo") {
// Servo calls this function from the worker, but only for internal
// pseudos, so we should never generate selector flags here.
unreachable!("internal pseudo generated slow selector flags?");
}
// No need to bother setting the selector flags when we're computing
// default styles.
if rule_inclusion == RuleInclusion::DefaultOnly {
return;
}
// Gecko calls this from sequential mode, so we can directly apply
// the flags.
debug_assert!(thread_state::get() == thread_state::LAYOUT);
let self_flags = flags.for_self();
if !self_flags.is_empty() {
unsafe { element.set_selector_flags(self_flags); }
}
let parent_flags = flags.for_parent();
if !parent_flags.is_empty() {
if let Some(p) = element.parent_element() {
unsafe { p.set_selector_flags(parent_flags); }
}
}
};
// Bug 1364242: We need to add visited support for lazy pseudos
let mut declarations = ApplicableDeclarationList::new();
let mut matching_context =
MatchingContext::new(MatchingMode::ForStatelessPseudoElement,
None,
self.quirks_mode);
self.push_applicable_declarations(element,
Some(&pseudo),
None,
None,
AnimationRules(None, None),
rule_inclusion,
&mut declarations,
&mut matching_context,
&mut set_selector_flags);
servo: Merge #16637 - Fix Stylist::lazily_compute_pseudo_element_style to return None when it should (from bzbarsky:probing-lazy-pseudos); r=emilio We have a fast path to return None if we have no rules for the pseudo-element at all, but we should also return None if we have no _matching_ rules for it for the given originating element. This is relied on by consumers like Servo_ResolvePseudoStyle, which needs to be able to detect the "no styles for this pseudo-element for this originating element case", to support probing for whether work for a specific pseudo-element should be done at all. <!-- Please describe your changes on the following line: --> --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [ ] `./mach build -d` does not report any errors - [X] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Also, please make sure that "Allow edits from maintainers" checkbox is checked, so that we can help you if you get stuck somewhere along the way.--> <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 04aac0247aea29e361589a75954b4f769478dd02 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : d1fa90d8723576f4a0ade231fa81595521445be9
2017-04-27 23:31:20 +00:00
if declarations.is_empty() {
return None
}
let rule_node =
self.rule_tree.insert_ordered_rules_with_important(
declarations.into_iter().map(|a| a.order_and_level()),
guards);
if rule_node == self.rule_tree.root() {
None
} else {
Some(rule_node)
}
}
/// Set a given device, which may change the styles that apply to the
/// document.
///
/// This means that we may need to rebuild style data even if the
/// stylesheets haven't changed.
///
/// Also, the device that arrives here may need to take the viewport rules
/// into account.
///
/// TODO(emilio): Probably should be unified with `update`, right now I
/// don't think we take into account dynamic updates to viewport rules.
///
/// Probably worth to make the stylist own a single `Device`, and have a
/// `update_device` function?
///
/// feature = "servo" because gecko only has one device, and manually tracks
/// when the device is dirty.
///
/// FIXME(emilio): The semantics of the device for Servo and Gecko are
/// different enough we may want to unify them.
#[cfg(feature = "servo")]
pub fn set_device(&mut self,
mut device: Device,
guard: &SharedRwLockReadGuard,
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
stylesheets: &[Arc<Stylesheet>]) {
let cascaded_rule = ViewportRule {
declarations: viewport_rule::Cascade::from_stylesheets(stylesheets.iter(), guard, &device).finish(),
};
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
self.viewport_constraints =
ViewportConstraints::maybe_new(&device, &cascaded_rule, self.quirks_mode);
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
if let Some(ref constraints) = self.viewport_constraints {
device.account_for_viewport_rule(constraints);
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
}
self.device = device;
let features_changed = self.media_features_change_changed_style(
stylesheets.iter(),
guard
);
self.is_device_dirty |= features_changed;
}
/// Returns whether, given a media feature change, any previously-applicable
/// style has become non-applicable, or vice-versa.
pub fn media_features_change_changed_style<'a, I>(
&self,
stylesheets: I,
guard: &SharedRwLockReadGuard,
) -> bool
where I: Iterator<Item = &'a Arc<Stylesheet>>
{
use invalidation::media_queries::PotentiallyEffectiveMediaRules;
debug!("Stylist::media_features_change_changed_style");
for stylesheet in stylesheets {
let effective_now =
stylesheet.media.read_with(guard)
.evaluate(&self.device, self.quirks_mode);
let effective_then =
self.effective_media_query_results.was_effective(&**stylesheet);
if effective_now != effective_then {
debug!(" > Stylesheet changed -> {}, {}",
effective_then, effective_now);
return true
}
if !effective_now {
continue;
}
let mut iter =
stylesheet.iter_rules::<PotentiallyEffectiveMediaRules>(
&self.device,
guard
);
while let Some(rule) = iter.next() {
match *rule {
CssRule::Style(..) |
CssRule::Namespace(..) |
CssRule::FontFace(..) |
CssRule::CounterStyle(..) |
CssRule::Supports(..) |
CssRule::Keyframes(..) |
CssRule::Page(..) |
CssRule::Viewport(..) |
CssRule::Document(..) => {
// Not affected by device changes.
continue;
}
CssRule::Import(ref lock) => {
let import_rule = lock.read_with(guard);
let mq = import_rule.stylesheet.media.read_with(guard);
let effective_now =
mq.evaluate(&self.device, self.quirks_mode);
let effective_then =
self.effective_media_query_results.was_effective(import_rule);
if effective_now != effective_then {
debug!(" > @import rule changed {} -> {}",
effective_then, effective_now);
return true;
}
if !effective_now {
iter.skip_children();
}
}
CssRule::Media(ref lock) => {
let media_rule = lock.read_with(guard);
let mq = media_rule.media_queries.read_with(guard);
let effective_now =
mq.evaluate(&self.device, self.quirks_mode);
let effective_then =
self.effective_media_query_results.was_effective(media_rule);
if effective_now != effective_then {
debug!(" > @media rule changed {} -> {}",
effective_then, effective_now);
return true;
}
if !effective_now {
iter.skip_children();
}
}
}
}
}
return false;
}
/// Returns the viewport constraints that apply to this document because of
/// a @viewport rule.
pub fn viewport_constraints(&self) -> Option<&ViewportConstraints> {
self.viewport_constraints.as_ref()
}
/// Returns the Quirks Mode of the document.
pub fn quirks_mode(&self) -> QuirksMode {
self.quirks_mode
}
/// Sets the quirks mode of the document.
pub fn set_quirks_mode(&mut self, quirks_mode: QuirksMode) {
// FIXME(emilio): We don't seem to change the quirks mode dynamically
// during multiple layout passes, but this is totally bogus, in the
// sense that it's updated asynchronously.
//
// This should probably be an argument to `update`, and use the quirks
// mode info in the `SharedLayoutContext`.
self.quirks_mode = quirks_mode;
}
/// Returns the applicable CSS declarations for the given element by
/// treating us as an XBL stylesheet-only stylist.
pub fn push_applicable_declarations_as_xbl_only_stylist<E, V>(&self,
element: &E,
applicable_declarations: &mut V)
where E: TElement,
V: Push<ApplicableDeclarationBlock> + VecLike<ApplicableDeclarationBlock>,
{
let mut matching_context =
MatchingContext::new(MatchingMode::Normal, None, self.quirks_mode);
let mut dummy_flag_setter = |_: &E, _: ElementSelectorFlags| {};
self.element_map.author.get_all_matching_rules(element,
element,
applicable_declarations,
&mut matching_context,
self.quirks_mode,
&mut dummy_flag_setter,
CascadeLevel::XBL);
}
/// Returns the applicable CSS declarations for the given element.
///
/// This corresponds to `ElementRuleCollector` in WebKit.
///
/// The `StyleRelations` recorded in `MatchingContext` indicate hints about
/// which kind of rules have matched.
pub fn push_applicable_declarations<E, V, F>(
&self,
element: &E,
pseudo_element: Option<&PseudoElement>,
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
style_attribute: Option<&Arc<Locked<PropertyDeclarationBlock>>>,
smil_override: Option<&Arc<Locked<PropertyDeclarationBlock>>>,
animation_rules: AnimationRules,
rule_inclusion: RuleInclusion,
applicable_declarations: &mut V,
context: &mut MatchingContext,
flags_setter: &mut F)
where E: TElement,
V: Push<ApplicableDeclarationBlock> + VecLike<ApplicableDeclarationBlock> + Debug,
F: FnMut(&E, ElementSelectorFlags),
{
debug_assert!(!self.is_device_dirty);
// Gecko definitely has pseudo-elements with style attributes, like
// ::-moz-color-swatch.
debug_assert!(cfg!(feature = "gecko") ||
style_attribute.is_none() || pseudo_element.is_none(),
"Style attributes do not apply to pseudo-elements");
debug_assert!(pseudo_element.map_or(true, |p| !p.is_precomputed()));
let map = match pseudo_element {
Some(pseudo) => self.pseudos_map.get(pseudo).unwrap(),
None => &self.element_map,
};
let is_implemented_pseudo =
element.implemented_pseudo_element().is_some();
// NB: This causes use to rule has pseudo selectors based on the
// properties of the originating element (which is fine, given the
// find_first_from_right usage).
let rule_hash_target = if is_implemented_pseudo {
element.closest_non_native_anonymous_ancestor().unwrap()
} else {
*element
};
debug!("Determining if style is shareable: pseudo: {}",
pseudo_element.is_some());
let only_default_rules = rule_inclusion == RuleInclusion::DefaultOnly;
// Step 1: Normal user-agent rules.
map.user_agent.get_all_matching_rules(element,
&rule_hash_target,
applicable_declarations,
context,
self.quirks_mode,
flags_setter,
CascadeLevel::UANormal);
debug!("UA normal: {:?}", context.relations);
if pseudo_element.is_none() && !only_default_rules {
// Step 2: Presentational hints.
let length_before_preshints = applicable_declarations.len();
element.synthesize_presentational_hints_for_legacy_attributes(
context.visited_handling,
applicable_declarations
);
if applicable_declarations.len() != length_before_preshints {
if cfg!(debug_assertions) {
for declaration in &applicable_declarations[length_before_preshints..] {
assert_eq!(declaration.level(), CascadeLevel::PresHints);
}
}
// Note the existence of presentational attributes so that the
// style sharing cache can avoid re-querying them if they don't
// exist.
context.relations |= AFFECTED_BY_PRESENTATIONAL_HINTS;
}
debug!("preshints: {:?}", context.relations);
}
// NB: the following condition, although it may look somewhat
// inaccurate, would be equivalent to something like:
//
// element.matches_user_and_author_rules() ||
// (is_implemented_pseudo &&
// rule_hash_target.matches_user_and_author_rules())
//
// Which may be more what you would probably expect.
if rule_hash_target.matches_user_and_author_rules() {
// Step 3a: User normal rules.
map.user.get_all_matching_rules(element,
&rule_hash_target,
applicable_declarations,
context,
self.quirks_mode,
flags_setter,
CascadeLevel::UserNormal);
debug!("user normal: {:?}", context.relations);
} else {
debug!("skipping user rules");
}
// Step 3b: XBL rules.
let cut_off_inheritance =
rule_hash_target.get_declarations_from_xbl_bindings(applicable_declarations);
debug!("XBL: {:?}", context.relations);
if rule_hash_target.matches_user_and_author_rules() && !only_default_rules {
// Gecko skips author normal rules if cutting off inheritance.
// See nsStyleSet::FileRules().
if !cut_off_inheritance {
// Step 3c: Author normal rules.
map.author.get_all_matching_rules(element,
&rule_hash_target,
applicable_declarations,
context,
self.quirks_mode,
flags_setter,
CascadeLevel::AuthorNormal);
debug!("author normal: {:?}", context.relations);
} else {
debug!("skipping author normal rules due to cut off inheritance");
}
} else {
debug!("skipping author normal rules");
}
if !only_default_rules {
// Step 4: Normal style attributes.
if let Some(sa) = style_attribute {
Push::push(
applicable_declarations,
ApplicableDeclarationBlock::from_declarations(sa.clone(),
CascadeLevel::StyleAttributeNormal));
}
debug!("style attr: {:?}", context.relations);
// Step 5: SMIL override.
// Declarations from SVG SMIL animation elements.
if let Some(so) = smil_override {
Push::push(
applicable_declarations,
ApplicableDeclarationBlock::from_declarations(so.clone(),
CascadeLevel::SMILOverride));
}
debug!("SMIL: {:?}", context.relations);
// Step 6: Animations.
// The animations sheet (CSS animations, script-generated animations,
// and CSS transitions that are no longer tied to CSS markup)
if let Some(anim) = animation_rules.0 {
Push::push(
applicable_declarations,
ApplicableDeclarationBlock::from_declarations(anim.clone(),
CascadeLevel::Animations));
}
debug!("animation: {:?}", context.relations);
} else {
debug!("skipping style attr and SMIL & animation rules");
}
//
// Steps 7-10 correspond to !important rules, and are handled during
// rule tree insertion.
//
if !only_default_rules {
// Step 11: Transitions.
// The transitions sheet (CSS transitions that are tied to CSS markup)
if let Some(anim) = animation_rules.1 {
Push::push(
applicable_declarations,
ApplicableDeclarationBlock::from_declarations(anim.clone(),
CascadeLevel::Transitions));
}
debug!("transition: {:?}", context.relations);
} else {
debug!("skipping transition rules");
}
debug!("push_applicable_declarations: shareable: {:?}", context.relations);
}
/// Given an id, returns whether there might be any rules for that id in any
/// of our rule maps.
#[inline]
pub fn may_have_rules_for_id(&self, id: &Atom) -> bool {
self.mapped_ids.might_contain(id)
}
/// Return whether the device is dirty, that is, whether the screen size or
/// media type have changed (for now).
#[inline]
servo: Merge #8039 - Move Stylesheet loading and ownership from the layout task into HTML elements (from tschneidereit:script-owns-stylesheets); r=jdm Stylesheets for `HTMLLinkElement`s are now loaded by the resource task, triggered by the element in question. Stylesheets are owned by the elements they're associated with, which can be `HTMLStyleElement`, `HTMLLinkElement`, and `HTMLMetaElement` (for `<meta name="viewport">). Additionally, the quirks mode stylesheet (just as the user and user agent stylesheets a couple of commits ago), is implemented as a lazy static, loaded once per process and shared between all documents. This all has various nice consequences: - Stylesheet loading becomes a non-blocking operation. - Stylesheets are removed when the element they're associated with is removed from the document. - It'll be possible to implement the CSSOM APIs that require direct access to the stylesheets (i.e., ~ all of them). - Various subtle correctness issues are fixed. One piece of interesting follow-up work would be to move parsing of external stylesheets to the resource task, too. Right now, it happens in the link element once loading is complete, so blocks the script task. Moving it to the resource task would probably be fairly straight-forward as it doesn't require access to any external state. Depends on #7979 because without that loading stylesheets asynchronously breaks lots of content. Source-Repo: https://github.com/servo/servo Source-Revision: 7ff3a17524e0e703e3ac279441729c185444be24
2015-11-07 19:41:54 +00:00
pub fn is_device_dirty(&self) -> bool {
self.is_device_dirty
}
/// Returns the map of registered `@keyframes` animations.
#[inline]
pub fn animations(&self) -> &FnvHashMap<Atom, KeyframesAnimation> {
&self.animations
}
/// Returns the rule root node.
#[inline]
pub fn rule_tree_root(&self) -> StrongRuleNode {
self.rule_tree.root()
}
/// Computes the match results of a given element against the set of
/// revalidation selectors.
pub fn match_revalidation_selectors<E, F>(&self,
element: &E,
bloom: Option<&BloomFilter>,
flags_setter: &mut F)
-> BitVec
where E: TElement,
servo: Merge #16641 - Bug 1360399: Don't deduplicate revalidation selectors. r=bholley (from emilio:dupedupedupedupe); r=bholley It's unfortunate, but it's a correctness issue. I was looking at the expectations update here: * https://hg.mozilla.org/integration/autoland/rev/659cddddd434 And investigating it I realised that it's wrong to coalesce selectors like that, because we keep the bloom filter flags. So in the test cases disabled, we have a selector that looks like this: ``` msub > :not(:first-child), msup > :not(:first-child), msubsup > :not(:first-child), mmultiscripts > :not(:first-child) { -moz-script-level: +1; -moz-math-display: inline; } ``` And an element that looks like this: ``` <msubsup><mi></mi><mi></mi></msubsup> ``` We're only inserting the first selector msub > :not(:first-child) into the set, so when we're going to match the <mi> elements we fast-reject it in both cases due to the bloom filter, so they share style. I can't see an easy way to fix this keeping the deduplication. If we keep it, we need to remove the bloom filter optimization, which means that we'd trash the cache for every first-child in the document (the :not(:first-child) effectively becomes a global rule). MozReview-Commit-ID: 9VPkmdj9zDg Signed-off-by: Emilio Cobos Álvarez <emilio@crisal.io> Source-Repo: https://github.com/servo/servo Source-Revision: eb975ab890b21c6a5c55716151c0c5eb144cb0d9 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 050be3042ec6e90bed7cbb9fdb24bcb10b5a20cc
2017-04-28 00:59:25 +00:00
F: FnMut(&E, ElementSelectorFlags),
{
// NB: `MatchingMode` doesn't really matter, given we don't share style
// between pseudos.
let mut matching_context =
MatchingContext::new(MatchingMode::Normal, bloom, self.quirks_mode);
// Note that, by the time we're revalidating, we're guaranteed that the
// candidate and the entry have the same id, classes, and local name.
// This means we're guaranteed to get the same rulehash buckets for all
// the lookups, which means that the bitvecs are comparable. We verify
// this in the caller by asserting that the bitvecs are same-length.
let mut results = BitVec::new();
self.selectors_for_cache_revalidation.lookup(
*element, self.quirks_mode, &mut |selector_and_hashes| {
results.push(matches_selector(&selector_and_hashes.selector,
selector_and_hashes.selector_offset,
&selector_and_hashes.hashes,
element,
&mut matching_context,
flags_setter));
true
}
);
results
}
/// Computes styles for a given declaration with parent_style.
pub fn compute_for_declarations(&self,
guards: &StylesheetGuards,
parent_style: &Arc<ComputedValues>,
declarations: Arc<Locked<PropertyDeclarationBlock>>)
-> Arc<ComputedValues> {
use font_metrics::get_metrics_provider_for_product;
let v = vec![
ApplicableDeclarationBlock::from_declarations(declarations.clone(),
CascadeLevel::StyleAttributeNormal)
];
let rule_node =
self.rule_tree.insert_ordered_rules(v.into_iter().map(|a| a.order_and_level()));
// This currently ignores visited styles. It appears to be used for
// font styles in <canvas> via Servo_StyleSet_ResolveForDeclarations.
// It is unclear if visited styles are meaningful for this case.
let metrics = get_metrics_provider_for_product();
Arc::new(properties::cascade(&self.device,
&rule_node,
guards,
Some(parent_style),
Some(parent_style),
None,
None,
&create_error_reporter(),
&metrics,
CascadeFlags::empty(),
self.quirks_mode))
}
/// Accessor for a shared reference to the device.
pub fn device(&self) -> &Device {
&self.device
}
/// Accessor for a mutable reference to the device.
pub fn device_mut(&mut self) -> &mut Device {
&mut self.device
}
/// Accessor for a shared reference to the rule tree.
pub fn rule_tree(&self) -> &RuleTree {
&self.rule_tree
}
}
impl Drop for Stylist {
fn drop(&mut self) {
// This is the last chance to GC the rule tree. If we have dropped all
// strong rule node references before the Stylist is dropped, then this
// will cause the rule tree to be destroyed correctly. If we haven't
// dropped all strong rule node references before now, then we will
// leak them, since there will be no way to call gc() on the rule tree
// after this point.
//
// TODO(emilio): We can at least assert all the elements in the free
// list are indeed free.
unsafe { self.rule_tree.gc(); }
}
}
/// Visitor to collect names that appear in attribute selectors and any
/// dependencies on ElementState bits.
struct AttributeAndStateDependencyVisitor<'a> {
attribute_dependencies: &'a mut BloomFilter,
style_attribute_dependency: &'a mut bool,
state_dependencies: &'a mut ElementState,
}
impl<'a> SelectorVisitor for AttributeAndStateDependencyVisitor<'a> {
type Impl = SelectorImpl;
fn visit_attribute_selector(&mut self, _ns: &NamespaceConstraint<&Namespace>,
name: &LocalName, lower_name: &LocalName)
-> bool {
#[cfg(feature = "servo")]
let style_lower_name = local_name!("style");
#[cfg(feature = "gecko")]
let style_lower_name = atom!("style");
if *lower_name == style_lower_name {
*self.style_attribute_dependency = true;
} else {
self.attribute_dependencies.insert(&name);
self.attribute_dependencies.insert(&lower_name);
}
true
}
fn visit_simple_selector(&mut self, s: &Component<SelectorImpl>) -> bool {
if let Component::NonTSPseudoClass(ref p) = *s {
self.state_dependencies.insert(p.state_flag());
}
true
}
}
/// Visitor to collect ids that appear in the rightmost portion of selectors.
struct MappedIdVisitor<'a> {
mapped_ids: &'a mut BloomFilter,
}
impl<'a> SelectorVisitor for MappedIdVisitor<'a> {
type Impl = SelectorImpl;
/// We just want to insert all the ids we find into mapped_ids.
fn visit_simple_selector(&mut self, s: &Component<SelectorImpl>) -> bool {
if let Component::ID(ref id) = *s {
self.mapped_ids.insert(id);
}
true
}
/// We want to stop as soon as we've moved off the rightmost ComplexSelector
/// that is not a psedo-element. That can be detected by a
/// visit_complex_selector call with a combinator other than None and
/// PseudoElement. Importantly, this call happens before we visit any of
/// the simple selectors in that ComplexSelector.
fn visit_complex_selector(&mut self,
_: SelectorIter<SelectorImpl>,
combinator: Option<Combinator>) -> bool {
match combinator {
None | Some(Combinator::PseudoElement) => true,
_ => false,
}
}
}
/// SelectorMapEntry implementation for use in our revalidation selector map.
#[derive(Clone, Debug)]
struct RevalidationSelectorAndHashes {
selector: Selector<SelectorImpl>,
selector_offset: usize,
hashes: AncestorHashes,
}
impl RevalidationSelectorAndHashes {
fn new(selector_and_hashes: &SelectorAndHashes<SelectorImpl>) -> Self {
// We basically want to check whether the first combinator is a
// pseudo-element combinator. If it is, we want to use the offset one
// past it. Otherwise, our offset is 0.
let mut index = 0;
let mut iter = selector_and_hashes.selector.iter();
// First skip over the first ComplexSelector. We can't check what sort
// of combinator we have until we do that.
for _ in &mut iter {
index += 1; // Simple selector
}
let offset = match iter.next_sequence() {
Some(Combinator::PseudoElement) => index + 1, // +1 for the combinator
_ => 0
};
RevalidationSelectorAndHashes {
selector: selector_and_hashes.selector.clone(),
selector_offset: offset,
hashes: selector_and_hashes.hashes.clone(),
}
}
}
impl SelectorMapEntry for RevalidationSelectorAndHashes {
fn selector(&self) -> SelectorIter<SelectorImpl> {
self.selector.iter_from(self.selector_offset)
}
fn hashes(&self) -> &AncestorHashes {
&self.hashes
}
}
/// Visitor to determine whether a selector requires cache revalidation.
///
/// Note that we just check simple selectors and eagerly return when the first
/// need for revalidation is found, so we don't need to store state on the
/// visitor.
///
/// Also, note that it's important to check the whole selector, due to cousins
/// sharing arbitrarily deep in the DOM, not just the rightmost part of it
/// (unfortunately, though).
///
/// With cousin sharing, we not only need to care about selectors in stuff like
/// foo:first-child, but also about selectors like p:first-child foo, since the
/// two parents may have shared style, and in that case we can test cousins
/// whose matching depends on the selector up in the chain.
///
/// TODO(emilio): We can optimize when matching only siblings to only match the
/// rightmost selector until a descendant combinator is found, I guess, and in
/// general when we're sharing at depth `n`, to the `n + 1` sequences of
/// descendant combinators.
///
/// I don't think that in presence of the bloom filter it's worth it, though.
struct RevalidationVisitor;
impl SelectorVisitor for RevalidationVisitor {
type Impl = SelectorImpl;
fn visit_complex_selector(&mut self,
_: SelectorIter<SelectorImpl>,
combinator: Option<Combinator>) -> bool {
let is_sibling_combinator =
combinator.map_or(false, |c| c.is_sibling());
!is_sibling_combinator
}
/// Check whether sequence of simple selectors containing this simple
/// selector to be explicitly matched against both the style sharing cache
/// entry and the candidate.
///
/// We use this for selectors that can have different matching behavior
/// between siblings that are otherwise identical as far as the cache is
/// concerned.
fn visit_simple_selector(&mut self, s: &Component<SelectorImpl>) -> bool {
match *s {
Component::AttributeInNoNamespaceExists { .. } |
Component::AttributeInNoNamespace { .. } |
Component::AttributeOther(_) |
Component::Empty |
// FIXME(bz) We really only want to do this for some cases of id
// selectors. See
// https://bugzilla.mozilla.org/show_bug.cgi?id=1369611
Component::ID(_) |
Component::FirstChild |
Component::LastChild |
Component::OnlyChild |
Component::NthChild(..) |
Component::NthLastChild(..) |
Component::NthOfType(..) |
Component::NthLastOfType(..) |
Component::FirstOfType |
Component::LastOfType |
Component::OnlyOfType => {
false
},
servo: Merge #16641 - Bug 1360399: Don't deduplicate revalidation selectors. r=bholley (from emilio:dupedupedupedupe); r=bholley It's unfortunate, but it's a correctness issue. I was looking at the expectations update here: * https://hg.mozilla.org/integration/autoland/rev/659cddddd434 And investigating it I realised that it's wrong to coalesce selectors like that, because we keep the bloom filter flags. So in the test cases disabled, we have a selector that looks like this: ``` msub > :not(:first-child), msup > :not(:first-child), msubsup > :not(:first-child), mmultiscripts > :not(:first-child) { -moz-script-level: +1; -moz-math-display: inline; } ``` And an element that looks like this: ``` <msubsup><mi></mi><mi></mi></msubsup> ``` We're only inserting the first selector msub > :not(:first-child) into the set, so when we're going to match the <mi> elements we fast-reject it in both cases due to the bloom filter, so they share style. I can't see an easy way to fix this keeping the deduplication. If we keep it, we need to remove the bloom filter optimization, which means that we'd trash the cache for every first-child in the document (the :not(:first-child) effectively becomes a global rule). MozReview-Commit-ID: 9VPkmdj9zDg Signed-off-by: Emilio Cobos Álvarez <emilio@crisal.io> Source-Repo: https://github.com/servo/servo Source-Revision: eb975ab890b21c6a5c55716151c0c5eb144cb0d9 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 050be3042ec6e90bed7cbb9fdb24bcb10b5a20cc
2017-04-28 00:59:25 +00:00
Component::NonTSPseudoClass(ref p) => {
!p.needs_cache_revalidation()
},
_ => {
true
}
}
}
}
/// Returns true if the given selector needs cache revalidation.
pub fn needs_revalidation(selector: &Selector<SelectorImpl>) -> bool {
let mut visitor = RevalidationVisitor;
!selector.visit(&mut visitor)
}
/// Map that contains the CSS rules for a specific PseudoElement
/// (or lack of PseudoElement).
#[cfg_attr(feature = "servo", derive(HeapSizeOf))]
#[derive(Debug)]
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
struct PerPseudoElementSelectorMap {
/// Rules from user agent stylesheets
user_agent: SelectorMap<Rule>,
/// Rules from author stylesheets
author: SelectorMap<Rule>,
/// Rules from user stylesheets
user: SelectorMap<Rule>,
}
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
impl PerPseudoElementSelectorMap {
#[inline]
servo: Merge #12515 - Make the style crate more concrete (from servo:concrete-style); r=bholley Background: The changes to Servo code to support Stylo began in the `selectors` crate with making pseudo-elements generic, defined be the user, so that different users (such as Servo and Gecko/Stylo) could have a different set of pseudo-elements supported and parsed. Adding a trait makes sense there since `selectors` is in its own repository and has others users (or at least [one](https://github.com/SimonSapin/kuchiki)). Then we kind of kept going with the same pattern and added a bunch of traits in the `style` crate to make everything generic, allowing Servo and Gecko/Stylo to do things differently. But we’ve also added a `gecko` Cargo feature to do conditional compilation, at first to enable or disable some CSS properties and values in the Mako templates. Since we’re doing conditional compilation anyway, it’s often easier and simpler to do it more (with `#[cfg(feature = "gecko")]` and `#[cfg(feature = "servo")]`) that to keep adding traits and making everything generic. When a type is generic, any method that we want to call on it needs to be part of some trait. ---- The first several commits move some code around, mostly from `geckolib` to `style` (with `#[cfg(feature = "gecko")]`) but otherwise don’t change much. The following commits remove some traits and many type parameters through the `style` crate, replacing them with pairs of conditionally-compiled API-compatible items (types, methods, …). Simplifying code is nice to make it more maintainable, but this is motivated by another change described in https://github.com/servo/servo/pull/12391#issuecomment-232183942. (Porting Servo for that change proved difficult because some code in the `style` crate was becoming generic over `String` vs `Atom`, and this PR will help make that concrete. That change, in turn, is motivated by removing geckolib’s `[replace]` override for string-cache, in order to enable using a single Cargo "workspace" in this repository.) r? @bholley --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [x] These changes do not require new tests because refactoring <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: 2d01d41a506bcbc7f26a2284b9f42390d6ef96ab --HG-- rename : servo/ports/geckolib/selector_impl.rs => servo/components/style/gecko_selector_impl.rs rename : servo/ports/geckolib/values.rs => servo/components/style/gecko_values.rs rename : servo/ports/geckolib/properties.mako.rs => servo/components/style/properties/gecko.mako.rs
2016-07-20 07:58:34 +00:00
fn new() -> Self {
PerPseudoElementSelectorMap {
user_agent: SelectorMap::new(),
author: SelectorMap::new(),
user: SelectorMap::new(),
}
}
#[inline]
fn borrow_for_origin(&mut self, origin: &Origin) -> &mut SelectorMap<Rule> {
match *origin {
Origin::UserAgent => &mut self.user_agent,
Origin::Author => &mut self.author,
Origin::User => &mut self.user,
}
}
}
/// A rule, that wraps a style rule, but represents a single selector of the
/// rule.
#[cfg_attr(feature = "servo", derive(HeapSizeOf))]
#[derive(Clone, Debug)]
pub struct Rule {
/// The selector this struct represents. We store this and the
/// any_{important,normal} booleans inline in the Rule to avoid
/// pointer-chasing when gathering applicable declarations, which
/// can ruin performance when there are a lot of rules.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "Arc")]
pub selector: Selector<SelectorImpl>,
/// The ancestor hashes associated with the selector.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "No heap data")]
pub hashes: AncestorHashes,
/// The source order this style rule appears in. Note that we only use
/// three bytes to store this value in ApplicableDeclarationsBlock, so
/// we could repurpose that storage here if we needed to.
pub source_order: u32,
/// The actual style rule.
#[cfg_attr(feature = "servo", ignore_heap_size_of = "Arc")]
servo: Merge #16014 - Per-process lock for CSSOM objects (from servo:style-ref); r=emilio <!-- Please describe your changes on the following line: --> Before this PR, every object reflected in CSSOM is in `Arc<RwLock<_>>` to enable safe (synchronized) mutable aliasing. Acquiring all these locks has significant cost during selector matching: * https://bugzilla.mozilla.org/show_bug.cgi?id=1311469 * https://bugzilla.mozilla.org/show_bug.cgi?id=1335941 * https://bugzilla.mozilla.org/show_bug.cgi?id=1339703 This PR introduce a mechanism to protect many objects with the same `RwLock` that only needs to be acquired once. In Stylo, there is one such lock per process (in a `lazy_static`), used for everything. I non-Stylo Servo, I originally intended to have one such lock per document (for author-origin stylesheets, and one per process for user-agent and user sytlesheets since they’re shared across documents, and never mutated anyway). However I failed to have the same document-specific (or pipeline-specific) `Arc` reachable from both `Document` nodes and `LayoutThread`. Recursively following callers lead me to include this `Arc` in `UnprivilegedPipelineContent`, but that needs to be serializable. So there is a second process-wide lock. This was previously #15998, closed accidentally. --- <!-- Thank you for contributing to Servo! Please replace each `[ ]` by `[X]` when the step is complete, and replace `__` with appropriate data: --> - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [ ] These changes fix #__ (github issue number if applicable). <!-- Either: --> - [ ] There are tests for these changes OR - [ ] These changes do not require tests because _____ <!-- Pull requests that do not address these steps are welcome, but they will require additional verification as part of the review process. --> Source-Repo: https://github.com/servo/servo Source-Revision: bb54f0a429de0e8b8861f8071b6cf82f73622664 --HG-- extra : subtree_source : https%3A//hg.mozilla.org/projects/converted-servo-linear extra : subtree_revision : 851230e57ac8775707df5f0f103be5feac81fc41
2017-03-19 21:31:19 +00:00
pub style_rule: Arc<Locked<StyleRule>>,
}
impl SelectorMapEntry for Rule {
fn selector(&self) -> SelectorIter<SelectorImpl> {
self.selector.iter()
}
fn hashes(&self) -> &AncestorHashes {
&self.hashes
}
}
impl Rule {
/// Returns the specificity of the rule.
pub fn specificity(&self) -> u32 {
self.selector.specificity()
}
/// Turns this rule into an `ApplicableDeclarationBlock` for the given
/// cascade level.
pub fn to_applicable_declaration_block(&self,
level: CascadeLevel)
-> ApplicableDeclarationBlock {
let source = StyleSource::Style(self.style_rule.clone());
ApplicableDeclarationBlock::new(source,
self.source_order,
level,
self.specificity())
}
/// Creates a new Rule.
pub fn new(selector: Selector<SelectorImpl>,
hashes: AncestorHashes,
style_rule: Arc<Locked<StyleRule>>,
source_order: u32)
-> Self
{
Rule {
selector: selector,
hashes: hashes,
style_rule: style_rule,
source_order: source_order,
}
}
}