chromium paint graphic

src hird_partylink enderercorepaintREADME.md

src hird_partylink endererplatformgraphicspaintREADME.md

---

renderer/core/paint

The code in this directory converts the LayoutObject tree into an efficient rendering format for the compositor (a list of cc::Layers containing display item lists, and associated cc::PropertyTrees). For a high level overview, see the Overview section.

For information about how the display list and paint property trees are implemented, see the platform paint README file.

This code is owned by the rendering team.

Contents

• Glossaries
• Stacked elements and stacking contexts
• Other glossaries
• Overview
• Current compositing algorithm (CompositeBeforePaint)
• New compositing algorithm (CompositeAfterPaint)
• Comparison of the current and new compositing algorithms
• PrePaint
• Paint invalidation
• Building paint property trees
• Paint
• Paint result caching
• Empty paint phase optimization
• Hit test information recording
• Scrollbar painting
• PaintNG

Glossaries

Stacked elements and stacking contexts

This chapter is basically a clarification of CSS 2.1 appendix E. Elaborate description of Stacking Contexts.

Note: we use ‘element’ instead of ‘object’ in this chapter to keep consistency with the spec. We use ‘object’ in other places in this document.

According to the documentation, we can have the following types of elements that are treated in different ways during painting:

• Stacked objects: objects that are z-ordered in stacking contexts, including:

  • Stacking contexts: elements with non-auto z-indices or other properties that affect stacking e.g. transform, opacity, blend-mode.

  • Replaced normal-flow stacking elements: replaced elements that do not have non-auto z-index but are stacking contexts for elements below them. Right now the only example is SVG <foreignObject>. The difference between these elements and regular stacking contexts is that they paint in the foreground phase of the painting algorithm (as opposed to the positioned descendants phase).

  • Elements that are not real stacking contexts but are treated as stacking contexts but don't manage other stacked elements. Their z-ordering are managed by real stacking contexts. They are positioned elements with z-index: auto (E.2.8 in the documentation).

    They must be managed by the enclosing stacking context as stacked elements because z-index:auto and z-index:0 are considered equal for stacking context sorting and they may interleave by DOM order.

    The difference of a stacked element of this type from a real stacking context is that it doesn't manage z-ordering of stacked descendants. These descendants are managed by the parent stacking context of this stacked element.

  “Stacked element” is not defined as a formal term in the documentation, but we found it convenient to use this term to refer to any elements participating z-index ordering in stacking contexts.

  A stacked element is represented by a PaintLayerStackingNode associated with a PaintLayer. It‘s painted as self-painting PaintLayers by PaintLayerPainter by executing all of the steps of the painting algorithm explained in the documentation for the element. When painting a stacked element of the second type, we don’t paint its stacked descendants which are managed by the parent stacking context.

• Non-stacked pseudo stacking contexts: elements that are not stacked, but paint their descendants (excluding any stacked contents) as if they created stacking contexts. This includes

  • inline blocks, inline tables, inline-level replaced elements (E.2.7.2.1.4 in the documentation)
  • non-positioned floating elements (E.2.5 in the documentation)
  • flex items
  • grid items
  • custom scrollbar parts

  They are painted by ObjectPainter::paintAllPhasesAtomically() which executes all of the steps of the painting algorithm explained in the documentation, except ignores any descendants which are positioned or have non-auto z-index (which is achieved by skipping descendants with self-painting layers).

• Other normal elements.

Other glossaries

• Paint container: the parent of an object for painting, as defined by CSS2.1 spec for painting. For regular objects, this is the parent in the DOM. For stacked objects, it's the containing stacking context-inducing object.

• Paint container chain: the chain of paint ancestors between an element and the root of the page.

• Compositing container: an implementation detail of Blink, which uses PaintLayers to represent some layout objects. It is the ancestor along the paint ancestor chain which has a PaintLayer. Implemented in PaintLayer::compositingContainer(). Think of it as skipping intermediate normal objects and going directly to the containing stacked object.

• Compositing container chain: same as paint chain, but for compositing container.

• Paint invalidation container: the nearest object on the compositing container chain which is composited. CompositeAfterPaint doesn't have this concept.

• Visual rect: the bounding box of all pixels that will be painted by a display item client. It's in the space of the containing transform property node (see Building paint property trees).

• Isolation nodes/boundary: In certain situations, it is possible to put in place a barrier that isolates a subtree from being affected by its ancestors. This barrier is called an isolation boundary and is implemented in the property trees as isolation nodes that serve as roots for any descendant property nodes. Currently, the contain: paint css property establishes an isolation boundary.

Overview

The primary responsibility of this directory is to convert the outputs from layout (the LayoutObject tree) to the inputs of the compositor (the cc::Layer list, which contains display items, and the associated cc::PropertyNodes).

This process is done in the following document lifecycle phases:

• Compositing update (kInCompositingUpdate, kCompositingInputsClean)
  • Decides layerization (GraphicsLayers).
  • This is only needed for the current compositing algorithm and will go away with CompositeAfterPaint.
• PrePaint (kInPrePaint)
  • Paint invalidation which invalidates display items which need to be painted.
  • Builds paint property trees.
• Paint (kInPaint)
  • Walks the LayoutObject tree and creates a display item list.
  • Groups the display list into paint chunks which share the same property tree state.
  • Commits the results to the compositor.
  • CompositeAfterPaint will decide layerization at this point.
  • Passes the paint chunks to the compositor in a cc::Layer list.
  • Converts the blink property tree nodes into cc property tree nodes.

Compositing decisions are currently made before paint (see Current compositing algorithm) but there is an in-progress refactoring to make compositing decisions after paint (see CompositeAfterPaint). The most recent step towards CompositeAfterPaint was a project called BlinkGenPropertyTrees which uses the compositing decisions from the current compositor (PaintLayerCompositor, which produces GraphicsLayers) with the new CompositeAfterPaint compositor (PaintArtifactCompositor). This is done by a step at the end of paint which collects all painted GraphicsLayers as a list of GraphicsLayerDisplayItems. Additionaly, ForeignLayerDisplayItems are used for cc::Layers managed outside blink (e.g., video layers, plugin layers) and are treated as opaque composited content by the PaintArtifactCompositor. This approach starts using much of the new PaintArtifactCompositor logic (e.g., converting blink property trees to cc property trees) without changing how compositing decisions are made.

Debugging blink objects has information about dumping the paint and compositing datastructures for debugging.

Current compositing algorithm (CompositeBeforePaint)

The current compositing system chooses which LayoutObjects paint into their own composited backing texture. This is called “having a compositing trigger”. These textures correspond to GraphicsLayers. There are also additional GraphicsLayers which represent property tree-related effects.

All elements which do not have a compositing trigger paint into the texture of the nearest LayoutObjectwith a compositing trigger on its compositing container chain (except for squashed layers; see below). For historical, practical and implementation detail reasons, only LayoutObjects with PaintLayers can have a compositing trigger. See crbug.com/370604 for a bug tracking this limitation, which is often referred to as the fundamental compositing bug.

The various compositing triggers are listed in compositing_reasons.h and fall in to several categories:

1. Direct reasons due to CSS style (see CompositingReason::kComboAllDirectStyleDeterminedReasons)
2. Direct reasons due to other conditions (see CompositingReason::kComboAllDirectNonStyleDeterminedReasons)
3. Composited scrolling-dependent reasons (see CompositingReason::kComboAllCompositedScrollingDeterminedReasons)
4. Composited descendant-dependent reasons (see CompositingReason::kComboCompositedDescendants)
5. Overlap-dependent reasons (See CompositingReasons::kComboSquashableReasons)

The triggers have no effect unless PaintLayerCompositor::CanBeComposited returns true.

Category (1) always triggers compositing of a LayoutObject based on its own style. Category (2) triggers based on the LayoutObject's style, its DOM ancestors, and whether it is a certain kind of frame root. Category (3) triggers based on whether composited scrolling applies to the LayoutObject, or the LayoutObject moves relative to a composited scroller (position: fixed or position: sticky). Category (4) triggers if there are any stacking descendants of the LayoutObject that end up composited. Category 5 triggers if the LayoutObject paints after and overlaps (or may overlap) another composited layer.

Note that composited scrolling is special. Several ways it is special:

• Composited descendants do not necessarily cause composited scrolling of an ancestor.
• The presence of LCD text prevents composited scrolling in the absence of other overriding triggers.
• Local frame roots always use composited scrolling if they have overflow.
• Non-local frame roots use composited scrolling if they have overflow and any composited descendants.
• Composited scrolling is indicated by a bit on PaintLayerScrollableArea, not a direct compositing reason. This bit is then transformed into a compositing reason from category (3) during the CompositingRequirementsUpdater

Note that overlap triggers have two special behaviors:

• Any LayoutObject which may overlap a LayoutObject that uses composited scrolling or a transform animation, paints after it, and scrolls with respect to it, receives an overlap trigger. In some cases this trigger is too aggressive.
• Inline CSS transform is treated as if it was a transform animation. (This is a heuristic to speed up the compositing step but leads to more composited layers.)

The sequence of work during the DocumentLifecycle to compute these triggers is as follows:

• kInStyleRecalc: compute (1) and most of (4) by calling CompositingReasonFinder::PotentialCompositingReasonsFromStyle and caching the result on PaintLayer, accessible via PaintLayer::PotentialCompositingReasonsFromStyle. Dirty bits in StyleDifference determine whether this has to be re-computed on a particular lifecycle update.
• kInCompositingUpdate: compute (2) CompositingInputsUpdater. Also set the composited scrolling bit on PaintLayerScrollableArea if applicable.
• kCompositingInputsClean: compute (3), the rest of (4), and (5), in CompositingRequirementsUpdater

The flow of data from the LayoutObject tree to the cc::Layer list and cc property trees is described below:

from layout
  |
  v
+------------------------------+
| LayoutObject/PaintLayer tree |-----------+
+------------------------------+           |
  |                                        |
  | PaintLayerCompositor::UpdateIfNeeded() |
  |   CompositingInputsUpdater::Update()   |
  |   CompositingLayerAssigner::Assign()   |
  |   GraphicsLayerUpdater::Update()       | PrePaintTreeWalk::Walk()
  |   GraphicsLayerTreeBuilder::Rebuild()  |   PaintPropertyTreeBuider::UpdatePropertiesForSelf()
  v                                        |
+--------------------+                   +------------------+
| GraphicsLayer tree |<------------------|  Property trees  |
+--------------------+                   +------------------+
      |                                    |              |
      |<-----------------------------------+              |
      | LocalFrameView::PaintTree()                       |
      |   LocalFrameView::PaintGraphicsLayerRecursively() |
      |     GraphicsLayer::Paint()                        |
      |       CompositedLayerMapping::PaintContents()     |
      |         PaintLayerPainter::PaintLayerContents()   |
      |           ObjectPainter::Paint()                  |
      v                                                   |
    +---------------------------------+                   |
    | DisplayItemList/PaintChunk list |                   |
    +---------------------------------+                   |
      |                                                   |
      |<--------------------------------------------------+
      | PaintChunksToCcLayer::Convert()                   |
      v                                                   |
+--------------------------------------------------+      |
| GraphicsLayerDisplayItem/ForeignLayerDisplayItem |      |
+--------------------------------------------------+      |
  |                                                       |
  |    LocalFrameView::PushPaintArtifactToCompositor()    |
  |         PaintArtifactCompositor::Update()             |
  +--------------------+       +--------------------------+
                       |       |
                       v       v
        +----------------+  +-----------------------+
        | cc::Layer list |  |   cc property trees   |
        +----------------+  +-----------------------+
                |              |
  +-------------+--------------+
  | to compositor
  v

Debugging blink objects has information about dumping these paint and compositing datastructures for debugging.

New compositing algorithm (CompositeAfterPaint)

This is a new mode under development. In this mode, layerization decisions are made after paint.

The process starts with pre-paint to generate property trees. During paint, each generated display item will be associated with a property tree state. Adjacent display items having the same property tree state will be grouped as PaintChunk. The list of paint chunks then will be processed by PaintArtifactCompositor for layerization. Property nodes that will be composited are converted into cc property nodes, while non-composited property nodes are converted into meta display items by PaintChunksToCcLayer.

from layout
  |
  v
+------------------------------+
| LayoutObject/PaintLayer tree |
+------------------------------+
  |     |
  |     | PrePaintTreeWalk::Walk()
  |     |   PaintPropertyTreeBuider::UpdatePropertiesForSelf()
  |     v
  |   +--------------------------------+
  |<--|         Property trees         |
  |   +--------------------------------+
  |                                  |
  | LocalFrameView::PaintTree()      |
  |   FramePainter::Paint()          |
  |     PaintLayerPainter::Paint()   |
  |       ObjectPainter::Paint()     |
  v                                  |
+---------------------------------+  |
| DisplayItemList/PaintChunk list |  |
+---------------------------------+  |
  |                                  |
  |<---------------------------------+
  | LocalFrameView::PushPaintArtifactToCompositor()
  |   PaintArtifactCompositor::Update()
  |
  +---+---------------------------------+
  |   v                                 |
  | +----------------------+            |
  | | Chunk list for layer |            |
  | +----------------------+            |
  |   |                                 |
  |   | PaintChunksToCcLayer::Convert() |
  v   v                                 v
+----------------+ +-----------------------+
| cc::Layer list | |   cc property trees   |
+----------------+ +-----------------------+
  |                  |
  +------------------+
  | to compositor
  v

Debugging blink objects has information about dumping these paint and compositing datastructures for debugging.

Comparison of the current and new compositing algorithms

The current compositing design is an incremental step towards the new CompositeAfterPaint design and was launched as BlinkGenPropertyTrees. The design before BlinkGenPropertyTrees is not described in this document.

	Current (CompositeBeforePaint)	New (CompositeAfterPaint)
REF::CompositeAfterPaintEnabled	False	True
Layerization	PaintLayerCompositor, CompositedLayerMapping	PaintArtifactCompositor
PaintController	One per GraphicsLayer	One per LocalFrameView

PrePaint

PrePaintTreeWalk

During the InPrePaint document lifecycle state, this class is called to walk the whole layout tree, beginning from the root FrameView, and across frame boundaries. This is an in-order tree traversal which is important for efficiently computing DOM-order hierarchy such as the parent containing block.

The PrePaint walk has two primary goals: paint invalidation and building paint property trees.

Paint invalidation

PaintInvalidator

Paint invalidator marks anything that need to be painted differently from the original cached painting.

During the document lifecycle stages prior to PrePaint, objects are marked for needing paint invalidation checking if needed by style change, layout change, compositing change, etc. In PrePaint stage, we traverse the layout tree in pre-order, crossing frame boundaries, for marked subtrees and objects and invalidate display item clients that will generate different display items.

At the beginning of the PrePaint tree walk, a root PaintInvalidatorContext is created for the root LayoutView. During the tree walk, one PaintInvalidatorContext is created for each visited object based on the PaintInvalidatorContext passed from the parent object. It tracks the following information to provide O(1) complexity access to them if possible:

• Paint invalidation container (Slimming Paint v1 only): As described by the definitions in Other glossaries, the paint invalidation container for stacked objects can differ from normal objects, we have to track both separately. Here is an example:

  <div style="overflow: scroll">
      <div id=A style="position: absolute"></div>
      <div id=B></div>
  </div>
  

  If the scroller is composited (for high-DPI screens for example), it is the paint invalidation container for div B, but not A.

• Painting layer: the layer which will initiate painting of the current object. It's the same value as LayoutObject::PaintingLayer().

PaintInvalidator initializes PaintInvalidatorContext for the current object, then calls LayoutObject::InvalidatePaint() which calls the object's paint invalidator (e.g. BoxPaintInvalidator) to complete paint invalidation of the object.

Paint invalidation of text

Text is painted by InlineTextBoxPainter using InlineTextBox as display item client. Text backgrounds and masks are painted by InlineTextFlowPainter using InlineFlowBox as display item client. We should invalidate these display item clients when their painting will change.

LayoutInlines and LayoutTexts are marked for full paint invalidation if needed when new style is set on them. During paint invalidation, we invalidate the InlineFlowBoxs directly contained by the LayoutInline in LayoutInline::InvalidateDisplayItemClients() and InlineTextBoxs contained by the LayoutText in LayoutText::InvalidateDisplayItemClients(). We don't need to traverse into the subtree of InlineFlowBoxs in LayoutInline::InvalidateDisplayItemClients() because the descendant InlineFlowBoxs and InlineTextBoxs will be handled by their owning LayoutInlines and LayoutTexts, respectively, when changed style is propagated.

Specialty of ::first-line

::first-line pseudo style dynamically applies to all InlineBox's in the first line in the block having ::first-line style. The actual applied style is computed from the ::first-line style and other applicable styles.

If the first line contains any LayoutInline, we compute the style from the ::first-line style and the style of the LayoutInline and apply the computed style to the first line part of the LayoutInline. In Blink's style implementation, the combined first line style of LayoutInline is identified with kPseudoIdFirstLineInherited.

The normal paint invalidation of texts doesn't work for first line because:

• ComputedStyle::VisualInvalidationDiff() can't detect first line style changes;
• The normal paint invalidation is based on whole LayoutObject's, not aware of the first line.

We have a special path for first line style change: the style system informs the layout system when the computed first-line style changes through LayoutObject::FirstLineStyleDidChange(). When this happens, we invalidate all InlineBoxes in the first line.

Building paint property trees

PaintPropertyTreeBuilder

This class is responsible for building property trees (see platform/paint/README.md for information about what property trees are).

Each PaintLayer's LayoutObject has one or more FragmentData objects (see below for more on fragments). Every FragmentData has an ObjectPaintProperties object if any property nodes are induced by it. For example, if the object has a transform, its ObjectPaintProperties::Transform() field points at the TransformPaintPropertyNode representing that transform.

The NeedsPaintPropertyUpdate, SubtreeNeedsPaintPropertyUpdate and DescendantNeedsPaintPropertyUpdate dirty bits on LayoutObject control how much of the layout tree is traversed during each PrePaintTreeWalk.

Additionally, some dirty bits are cleared at an isolation boundary. For example if the paint property tree topology has changed by adding or removing nodes for an element, we typically force a subtree walk for all descendants since the descendant nodes may now refer to new parent nodes. However, at an isolation boundary, we can reason that none of the descendants of an isolation element would be affected, since the highest node that the paint property nodes of an isolation element's subtree can reference are the isolation nodes established at this element itself.

Implementation note: the isolation boundary is achieved using alias nodes, which are nodes that are put in place on an isolated element for clip, transform, and effect trees. These nodes do not themselves contribute to any painted output, but serve as parents to the subtree nodes. The alias nodes and isolation nodes are synonymous and are used interchangeably. Also note that these nodes are placed as children of the regular nodes of the element. This means that the element itself is not isolated against ancestor mutations; it only isolates the element's subtree.

Example tree:

                        +----------------------+
                        | 1. Root LayoutObject |
                        +----------------------+
                          /                  
           +-----------------+            +-----------------+
           | 2. LayoutObject |            | 3. LayoutObject |
           +-----------------+            +-----------------+
             /                              /             
  +-----------------+          +-----------------+    +-----------------+
  | 4. LayoutObject |          | 5. LayoutObject |    | 6. LayoutObject |
  +-----------------+          +-----------------+    +-----------------+
                                 /             
                   +-----------------+     +-----------------+
                   | 7. LayoutObject |     | 8. LayoutObject |
                   +-----------------+     +-----------------+

Suppose that element 3's style changes to include a transform (e.g. transform: translateX(10px)).

Typically, here is the order of the walk (depth first) and updates:

• Root element 1 is visited since some descendant needs updates
• Element 2 is visited since it is one of the descendants, but it doesn't need updates.
• Element 4 is skipped since the above step didn't need to recurse.
• Element 3 is visited since it's a descendant of the root element, and its property trees are updated to include a new transform. This causes a flag to be flipped that all subtree nodes need an update.
• Elements are then visited in depth order: 5, 7, 8, 6. Elements 5 and 6 reparent their transform nodes to point to the transform node of element 3. Elements 7 and 8 are visited and updated but no changes occur.

Now suppose that element 5 has “contain: paint” style, which establishes an isolation boundary. The walk changes in the following way:

• Root element 1 is visited since some descendant needs updates
• Element 2 is visited since it is one of the descendants, but it doesn't need updates.
• Element 4 is skipped since the above step didn't need to recurse.
• Element 3 is visited since it's a descendant of the root element, and its property trees are updated to include a new transform. This causes a flag to be flipped that all subtree nodes need an update.
• Element 5 is visited and updated by reparenting the transform nodes. However, now the element is an isolation boundary so elements 7 and 8 are not visited (i.e. the forced subtree update flag is ignored).
• Element 6 is visited as before and is updated to reparent the transform node.

Note that there are subtleties when deciding whether we can skip the subtree walk. Specifically, not all subtree walks can be stopped at an isolation boundary. For more information, see PaintPropertyTreeBuilder and its use of IsolationPiercing vs IsolationBlocked subtree update reasons.

Fragments

In the absence of multicolumn/pagination, there is a 1:1 correspondence between LayoutObjects and FragmentData. If there is multicolumn/pagination, there may be more FragmentDatas. If a LayoutObject has a property node, each of its fragments will have one. The parent of a fragment‘s property node is the property node that belongs to the ancestor LayoutObject which is part of the same column. For example, if there are 3 columns and both a parent and child LayoutObject have a transform, there will be 3 FragmentData objects for the parent, 3 for the child, each FragmentData will have its own TransformPaintPropertyNode, and the child’s ith fragment‘s transform will point to the ith parent’s transform.

Each FragmentData receives its own ClipPaintPropertyNode. They also store a unique PaintOffset,PaginationOffset and LocalBordreBoxProperties object.

See LayoutMultiColumnFlowThread.h for a much more detail about multicolumn/pagination.

Paint

Paint walks the LayoutObject tree in paint-order and produces a list of display items. This is implemented using static painter classes (e.g., BlockPainter) and appends display items to a PaintController. During this treewalk, the current property tree state is maintained (see: PaintController::UpdateCurrentPaintChunkProperties). The PaintController segments the display item list into PaintChunks which are sequential display items that share a common property tree state.

With the current compositing algorithm, the paint-order LayoutObject treewalk is initiated by GraphicsLayers, and each GraphicsLayer contains a PaintController. In the new compositing approach, CompositeAfterPaint, there is only one PaintController for the entire LocalFrameView.

Paint result caching

PaintController holds the previous painting result as a cache of display items. If some painter would generate results same as those of the previous painting, we'll skip the painting and reuse the display items from cache.

Display item caching

When a painter would create a DrawingDisplayItem exactly the same as the display item created in the previous painting, we'll reuse the previous one instead of repainting it.

Subsequence caching

When possible, we create a scoped SubsequenceRecorder in PaintLayerPainter::PaintContents() to record all display items generated in the scope as a “subsequence”. Before painting a layer, if we are sure that the layer will generate exactly the same display items as the previous paint, we'll get the whole subsequence from the cache instead of repainting them.

There are many conditions affecting whether we need to generate subsequence for a PaintLayer and whether we can use cached subsequence for a PaintLayer. See ShouldCreateSubsequence() and shouldRepaintSubsequence() in PaintLayerPainter.cpp for the conditions.

Empty paint phase optimization

During painting, we walk the layout tree multiple times for multiple paint phases. Sometimes a layer contain nothing needing a certain paint phase and we can skip tree walk for such empty phases. Now we have optimized PaintPhaseDescendantOutlinesOnly and PaintPhaseFloat for empty paint phases.

During paint invalidation, we set the containing self-painting layer's NeedsPaintPhaseXXX flag if the object has something needing to be painted in the paint phase.

During painting, we check the flag before painting a paint phase and skip the tree walk if the flag is not set.

When layer structure changes, and we are not invalidate paint of the changed subtree, we need to manually update the NeedsPaintPhaseXXX flags. For example, if an object changes style and creates a self-painting-layer, we copy the flags from its containing self-painting layer to this layer, assuming that this layer needs all paint phases that its container self-painting layer needs.

Hit test information recording

Hit testing is done in paint-order, and to preserve this information the paint system is re-used to record hit test information when painting the background. This information is then used in the compositor to implement cc-side hit testing. Hit test information is recorded even if there is no painted content.

We record different types of hit test information in the following data structures:

1. Paint chunk bounds

   The bounds of the current paint chunk are expanded to ensure the bounds contain the hit testable area.

2. HitTestData::touch_action_rects

   Used for touch action rects which are areas of the page that allow certain gesture effects, as well as areas of the page that disallow touch events due to blocking touch event handlers.

3. HitTestData::scroll_translation and HitTestData::scroll_hit_test_rect

   Used to create non-fast scrollable regions to prevent compositor scrolling of non-composited scrollers, plugins with blocking scroll event handlers, and resize handles.

   If scroll_translation is not null, this is also used for CompositeAfterPaint to force a special cc::Layer that is marked as being scrollable when composited scrolling is needed for the scroller.

Scrollbar painting

For now in pre-CompositeAfterPaint, we have distinct paths for composited scrollbars and non-composited scrollbars. For a composited scrollbar, PaintArtifactCompositor creates a GraphicsLayer, then ScrollingCoordinator creates the cc scrollbar layer which is set as the content layer of the GraphicsLayer. For a non-composited scrollbar, ScrollableAreaPainter paints the scrollbar into various drawing display items.

In CompositeAfterPaint, during painting, for a non-custom scrollbar we create a ScrollbarDisplayItem which contains a cc::Scrollbar and other information that are needed to actually paint the scrollbar into a paint record or to create a cc scrollbar layer. During PaintArtifactCompositor update, we decide whether to composite the scrollbar and, if not composited, actually paint the scrollbar as a paint record, otherwise create a cc scrollbar layer of type cc::SolidColorScrollbarLayer, cc::PaintedScrollbarLayer or cc::PaintedOverlayScrollbarLayer depending on the type of the scrollbar.

In CompositeAfterPaint, custom scrollbars are still painted into drawing display items directly.

PaintNG

LayoutNG is a project that will change how Layout generates geometry/style information for painting. Instead of modifying LayoutObjects, LayoutNG will generate an NGFragment tree.

NGPaintFragments are:

• immutable
• all coordinates are physical. See layout_box_model_object.h.
• instead of Location(), NGFragment has Offset(), a physical offset from parent fragment.

The goal is for PaintNG to eventually paint from NGFragment tree, and not see LayoutObjects at all. Until this goal is reached, LegacyPaint, and NGPaint will coexist.

When a particular LayoutObject subclass fully migrates to NG, its LayoutObject geometry information might no longer be updated(*), and its painter needs to be rewritten to paint NGFragments. For example, see how BlockPainter is being rewritten as NGBoxFragmentPainter.

---

<!---
The live version of this document can be viewed at:
https://chromium.googlesource.com/chromium/src/+/master/third_party/blink/renderer/platform/graphics/paint/README.md
-->

# Platform paint code

This directory contains the implementation of display lists and display
list-based painting, except for code which requires knowledge of `core/`
concepts, such as DOM elements and layout objects.

For information about how the display list and paint property trees are
generated, see [the core paint README file](../../../core/paint/README.md).

This code is owned by the [rendering team](https://www.chromium.org/teams/rendering).

[TOC]

## Paint artifact

The CompositeAfterPaint [paint artifact](paint_artifact.h) consists of a list of
display items in paint order (ideally mostly or all drawings), partitioned into
*paint chunks* which define certain *paint properties* which affect how the
content should be drawn or composited.

## Paint properties

Paint properties define characteristics of how a paint chunk should be drawn,
such as the transform it should be drawn with. To enable efficient updates,
a chunk's paint properties are described hierarchically. For instance, each
chunk is associated with a transform node, whose matrix should be multiplied by
its ancestor transform nodes in order to compute the final transformation matrix
to the screen.

See [`ObjectPaintProperties`](../../../core/paint/object_paint_properties.h) for
description of all paint properties that we create for a `LayoutObject`.

Paint properties are represented by four paint property trees (transform, clip,
effect and scroll) each of which contains corresponding type of
[paint property nodes](paint_property_node.h). Each paint property node has a
pointer to the parent node. The parent node pointers link the paint property
nodes in a tree.

### Transforms

Each paint chunk is associated with a [transform node](transform_paint_property_node.h),
which defines the coordinate space in which the content should be painted.

Each transform node has:

* a 4x4 [`TransformationMatrix`](../../transforms/transformation_matrix.h)
* a 3-dimensional transform origin, which defines the origin relative to which
the transformation matrix should be applied (e.g. a rotation applied with some
transform origin will rotate the plane about that point)
* a boolean indicating whether the transform should be projected into the plane
of its parent (i.e., whether the total transform inherited from its parent
should be flattened before this node's transform is applied and propagated to
children)
* an integer rendering context ID; content whose transform nodes share a
rendering context ID should sort together
* other fields, see [the header file](transform_paint_property_node.h)

*** note
The painting system may create transform nodes which don't affect the position
of points in the xy-plane, but which have an apparent effect only when
multiplied with other transformation matrices. In particular, a transform node
may be created to establish a perspective matrix for descendant transforms in
order to create the illusion of depth.
***

Note that, even though CSS does not permit it in the DOM, the transform tree can
have nodes whose children do not flatten their inherited transform and
participate in no 3D rendering context. For example, not flattening is necessary
to preserve the 3D character of the perspective transform, but this does not
imply any 3D sorting.

### Clips

Each paint chunk is associated with a [clip node](clip_paint_property_node.h),
which defines the raster region that will be applied on the canvas when
the chunk is rastered.

Each clip node has:

* A float rect with (optionally) rounded corner radius.
* An optional clip path if the clip is a clip path.
* An associated transform node, which the clip rect is based on.

The raster region defined by a node is the rounded rect and/or clip path
transformed to the root space, intersects with the raster region defined by its
parent clip node (if not root).

### Effects

Each paint chunk is associated with an [effect node](effect_paint_property_node.h),
which defines the effect (opacity, transfer mode, filter, mask, etc.) that
should be applied to the content before or as it is composited into the content
below.

Each effect node has:

* effects, including opacity, transfer mode, filter, mask, etc.
* an optional associated clip node which clips the output of the effect when
composited into the current backdrop.
* an associated transform node which defines the geometry space of some
geometry-related effects (e.g. some filters).

The hierarchy in the *effect tree* defines the dependencies between
rasterization of different contents.

One can imagine each effect node as corresponding roughly to a bitmap that is
drawn before being composited into another bitmap, though for implementation
reasons this may not be how it is actually implemented.

### Scrolling

Each paint chunk is associated with a [scroll node](scroll_paint_property_node.h)
which defines information about how a subtree scrolls so threads other than the
main thread can perform scrolling. Scroll information includes:

* Which directions, if any, are scrollable by the user.
* A reference to a [transform node](transform_paint_property_node.h) which contains
a 2d scroll offset.
* The extent that can be scrolled. For example, an overflow clip of size 7x9
with scrolling contents of size 7x13 can scroll 4px vertically and none
horizontally.

To ensure geometry operations are simple and only deal with transforms, the
scroll offset is stored as a 2d transform in the transform tree.

## Display items

A display item is the smallest unit of a display list in Blink. Each display
item is identified by an ID consisting of:

* an opaque pointer to the *display item client* that produced it
* a type (from the `DisplayItem::Type` enum)

In practice, display item clients are generally subclasses of `LayoutObject`,
but can be other Blink objects which get painted, such as inline boxes and drag
images.

*** note
It is illegal for there to be two display items with the same ID in a display
item list, except for display items that are marked uncacheable
(see [DisplayItemCacheSkipper](display_item_cache_skipper.h)).
***

Generally, clients of this code should use stack-allocated recorder classes to
emit display items to a `PaintController` (using `GraphicsContext`).

### Standalone display items

#### [DrawingDisplayItem](drawing_display_item.h)

Holds a `PaintRecord` which contains the paint operations required to draw some
atom of content.

#### [ForeignLayerDisplayItem](foreign_layer_display_item.h)

Draws an atom of content, but using a `cc::Layer` produced by some agent outside
of the normal Blink paint system (for example, a plugin). Since they always map
to a `cc::Layer`, they are always the only display item in their paint chunk,
and are ineligible for squashing with other layers.

#### [ScrollHitTestDisplayItem](scroll_hit_test_display_item.h)

Placeholder for creating a cc::Layer for scrolling in paint order. Hit testing
in the compositor requires both property trees (scroll nodes) and a scrollable
`cc::layer` in paint order. This should be associated with the scroll
translation paint property node as well as any overflow clip nodes.

## Paint controller

Callers use `GraphicsContext` (via its drawing methods, and its
`paintController()` accessor) and scoped recorder classes, which emit items into
a `PaintController`.

`PaintController` is responsible for producing the paint artifact. It contains
the *current* paint artifact, and *new* display items and paint chunks, which
are added as content is painted.

Painters should call `PaintController::UseCachedItemIfPossible()` or
`PaintController::UseCachedSubsequenceIfPossible()` and if the function returns
`true`, existing display items that are still valid in the *current* paint artifact
will be reused and the painter should skip real painting of the item or subsequence.

When the new display items have been populated, clients call
`commitNewDisplayItems`, which replaces the previous artifact with the new data,
producing a new paint artifact.

At this point, the paint artifact is ready to be drawn or composited.

### Paint result caching and invalidation

See [Display item caching](../../../core/paint/README.md#paint-result-caching)
and [Paint invalidation](../../../core/paint/README.md#paint-invalidation) for
more details about how caching and invalidation are handled in blink core
module using `PaintController` API.

## Paint artifact compositor

[`PaintArtifactCompositor`](../compositing/paint_artifact_compositor.h) is
responsible for consuming the `PaintArtifact` produced by the `PaintController`,
and converting it into a form suitable for the compositor to consume.

At present, `PaintArtifactCompositor` creates a cc layer tree, with one layer
for each paint chunk. In the future, it is expected that we will use heuristics
to combine paint chunks into a smaller number of layers.

The owner of the `PaintArtifactCompositor` (e.g. `WebView`) can then attach its
root layer to the overall layer hierarchy to be displayed to the user.

In the future we would like to explore moving to a single shared property tree
representation across both cc and
Blink. See [Web Page Geometries](https://goo.gl/MwVIto) for more.

## Raster invalidation

This is to mark which parts of the composited layers need to be re-rasterized to
reflect changes of painting, by comparing the current paint artifact against the
previous paint artifact. It's the last step of painting.

It's done in two levels:

* Paint chunk level [`RasterInvalidator`](raster_invalidator.h): matches each
paint chunk in the current paint artifact against the corresponding paint
chunk in the previous paint artifact, by matching their ids. There are
following cases:

* A new paint chunk doesn't match any old paint chunk (appearing): The bounds
of the new paint chunk in the composited layer will be fully raster
invalidated.

* An old paint chunk doesn't match any new paint chunk (disappearing): The
bounds of the old paint chunk in the composited layer will be fully raster
invalidated.

* A new paint chunk matches an old paint chunk:

* The new paint chunk is moved backward (reordering): this may expose other
chunks that was previously covered by it: Both of the old bounds and the
new bounds will be fully raster invalidated.

* Paint properties of the paint chunk changed:

* If only clip changed, the difference between the old bounds and
the new bounds will be raster invalidated (i.e. do incremental
invalidation).

* Otherwise, both of the old bounds and the new bounds will be fully
raster invalidated.

* Otherwise, check for changed display items within the paint chunk.

* Display item level [`DisplayItemRasterInvalidator`](display_item_raster_invalidator.h):
This is executed when a new chunk matches an old chunk in-order and paint
properties didn't change. The algorithm checks changed display items within a
paint chunk.

* Similar to the paint chunk level, the visual rects (mapped to the space of
the composited layer) of appearing, disappearing, reordering display items
are fully raster invalidated.

* If a new paint chunk in-order matches an old paint chunk, if the display
item client has been [paint invalidated](../../../core/paint/README.md#paint-invalidation),
we will do full raster invalidation (which invalidates the old visual rect
and the new visual rect in the composted layer) or incremental raster
invalidation (which invalidates the difference between the old visual rect
and the new visual rect) according to the paint invalidation reason.

## Geometry routines

The [`GeometryMapper`](geometry_mapper.h) is responsible for efficiently computing
visual and transformed rects of display items in the coordinate space of ancestor
[`PropertyTreeState`](property_tree_state.h)s.

The transformed rect of a display item in an ancestor `PropertyTreeState` is
that rect, multiplied by the transforms between the display item's
`PropertyTreeState` and the ancestors, then flattened into 2D.

The visual rect of a display item in an ancestor `PropertyTreeState` is the
intersection of all of the intermediate clips (transformed in to the ancestor
state), with the display item's transformed rect.

---