| ActiveLayerTracker.cpp |
This tracks the state of a frame that may need active layers due to
ongoing content changes or style changes that indicate animation.
When no changes of *any* kind are detected after 75-100ms we remove this
object. Because we only track all kinds of activity with a single
nsExpirationTracker, it's possible a frame might remain active somewhat
spuriously if different kinds of changes kept happening, but that almost
certainly doesn't matter.
|
15792 |
| ActiveLayerTracker.h |
This class receives various notifications about style changes and content
changes that affect layerization decisions, and implements the heuristics
that drive those decisions. It manages per-frame state to support those
heuristics.
|
3822 |
| BorderCache.h |
mozilla_BorderCache_h_ |
1773 |
| BorderConsts.h |
mozilla_BorderConsts_h_ |
807 |
| crashtests |
|
|
| DashedCornerFinder.cpp |
|
12697 |
| DashedCornerFinder.h |
|
9710 |
| DisplayItemClip.cpp |
|
17980 |
| DisplayItemClip.h |
An DisplayItemClip represents the intersection of an optional rectangle
with a list of rounded rectangles (which is often empty), all in appunits.
It can represent everything CSS clipping can do to an element (except for
SVG clip-path), including no clipping at all.
|
7251 |
| DisplayItemClipChain.cpp |
static |
3088 |
| DisplayItemClipChain.h |
A DisplayItemClipChain is a linked list of DisplayItemClips where each clip
is associated with an active scrolled root that describes what the clip
moves with.
We use a chain instead of just one intersected clip due to async scrolling:
A clip that moves along with a display item can be fused to the item's
contents when drawing the layer contents, but all other clips in the chain
need to be kept separate so that they can be applied at composition time,
after any async scroll offsets have been applied.
The clip chain is created during display list construction by the builder's
DisplayListClipState.
The clip chain order is determined by the active scrolled root order.
For every DisplayItemClipChain object |clipChain|, the following holds:
!clipChain->mParent ||
ActiveScrolledRoot::IsAncestor(clipChain->mParent->mASR, clipChain->mASR).
The clip chain can skip over active scrolled roots. That just means that
there is no clip that moves with the skipped ASR in this chain.
|
3742 |
| DisplayListClipState.cpp |
|
5740 |
| DisplayListClipState.h |
All clip coordinates are in appunits relative to the reference frame
for the display item we're building.
|
10475 |
| DottedCornerFinder.cpp |
|
16963 |
| DottedCornerFinder.h |
|
14086 |
| HitTestInfo.cpp |
|
2151 |
| HitTestInfo.h |
A helper class that manages compositor hit testing information.
|
1547 |
| MatrixStack.h |
MatrixStack stores a stack of matrices and keeps track of the accumulated
transform matrix.
|
1556 |
| moz.build |
|
1712 |
| nsCSSRendering.cpp |
utility functions for drawing borders and backgrounds |
197022 |
| nsCSSRendering.h |
utility functions for drawing borders and backgrounds |
40392 |
| nsCSSRenderingBorders.cpp |
nsCSSRendering::PaintBorder
nsCSSRendering::PaintOutline
-> DrawBorders
DrawBorders
-> Ability to use specialized approach?
|- Draw using specialized function
|- separate corners?
|- dashed side mask
|
-> can border be drawn in 1 pass? (e.g., solid border same color all
around)
|- DrawBorderSides with all 4 sides
-> more than 1 pass?
|- for each corner
|- clip to DoCornerClipSubPath
|- for each side adjacent to corner
|- clip to GetSideClipSubPath
|- DrawBorderSides with one side
|- for each side
|- GetSideClipWithoutCornersRect
|- DrawDashedOrDottedSide || DrawBorderSides with one side
|
139542 |
| nsCSSRenderingBorders.h |
Helper class that handles border rendering.
aDrawTarget -- the DrawTarget to which the border should be rendered
outsideRect -- the rectangle on the outer edge of the border
For any parameter where an array of side values is passed in,
they are in top, right, bottom, left order.
borderStyles -- one border style enum per side
borderWidths -- one border width per side
borderRadii -- a RectCornerRadii struct describing the w/h for each rounded
corner. If the corner doesn't have a border radius, 0,0 should be given for
it. borderColors -- one nscolor per side
skipSides -- a bit mask specifying which sides, if any, to skip
backgroundColor -- the background color of the element.
Used in calculating colors for 2-tone borders, such as inset and outset
gapRect - a rectangle that should be clipped out to leave a gap in a border,
or nullptr if none.
|
12897 |
| nsCSSRenderingGradients.cpp |
utility functions for drawing borders and backgrounds |
57298 |
| nsCSSRenderingGradients.h |
Prepare a nsCSSGradientRenderer for a gradient for an element.
aIntrinsicSize - the size of the source gradient.
|
7804 |
| nsDisplayItemTypes.h |
It's useful to be able to dynamically check the type of certain items.
Every subclass of nsDisplayItem must have a new type added here for the
purposes of easy comparison and matching of items in different display lists.
|
2193 |
| nsDisplayItemTypesList.h |
|
6908 |
| nsDisplayList.cpp |
structures that represent things to be painted (ordered in z-order),
used during painting and hit testing
|
328545 |
| nsDisplayList.h |
structures that represent things to be painted (ordered in z-order),
used during painting and hit testing
|
243780 |
| nsDisplayListArenaTypes.h |
a list of all types that can be allocated in the display list's nsPresArena,
for preprocessing |
718 |
| nsDisplayListInvalidation.cpp |
|
4594 |
| nsDisplayListInvalidation.h |
This stores the geometry of an nsDisplayItem, and the area
that will be affected when painting the item.
It is used to retain information about display items so they
can be compared against new display items in the next paint.
|
7312 |
| nsImageRenderer.cpp |
utility code for drawing images as CSS borders, backgrounds, and shapes. |
41094 |
| nsImageRenderer.h |
This is a small wrapper class to encapsulate image drawing that can draw an
StyleImage image, which may internally be a real image, a sub image, or a CSS
gradient, etc...
@note Always call the member functions in the order of PrepareImage(),
SetSize(), and Draw*().
|
12097 |
| RetainedDisplayListBuilder.cpp |
Code for doing display list building for a modified subset of the window,
and then merging it into the existing display list (for the full window).
The approach primarily hinges on the observation that the 'true' ordering
of display items is represented by a DAG (only items that intersect in 2d
space have a defined ordering). Our display list is just one of a many
possible linear representations of this ordering.
Each time a frame changes (gets a new ComputedStyle, or has a size/position
change), we schedule a paint (as we do currently), but also reord the frame
that changed.
When the next paint occurs we union the overflow areas (in screen space) of
the changed frames, and compute a rect/region that contains all changed
items. We then build a display list just for this subset of the screen and
merge it into the display list from last paint.
Any items that exist in one list and not the other must not have a defined
ordering in the DAG, since they need to intersect to have an ordering and
we would have built both in the new list if they intersected. Given that, we
can align items that appear in both lists, and any items that appear between
matched items can be inserted into the merged list in any order.
Frames that are a stacking context, containing blocks for position:fixed
descendants, and don't have any continuations (see
CanStoreDisplayListBuildingRect) trigger recursion into the algorithm with
separate retaining decisions made.
RDL defines the concept of an AnimatedGeometryRoot (AGR), the nearest
ancestor frame which can be moved asynchronously on the compositor thread.
These are currently nsDisplayItems which return true from CanMoveAsync
(animated nsDisplayTransform and nsDisplayStickyPosition) and
ActiveScrolledRoots.
For each context that we run the retaining algorithm, there can only be
mutations to one AnimatedGeometryRoot. This is because we are unable to
reason about intersections of items that might then move relative to each
other without RDL running again. If there are mutations to multiple
AnimatedGeometryRoots, then we bail out and rebuild all the items in the
context.
Otherwise, when mutations are restricted to a single AGR, we pre-process the
old display list and mark the frames for all existing (unmodified!) items
that belong to a different AGR and ensure that we rebuild those items for
correct sorting with the modified ones.
|
64886 |
| RetainedDisplayListBuilder.h |
RetainedDisplayListData contains frame invalidation information.
Currently this is implemented as a map of frame pointers to flags.
|
9040 |
| RetainedDisplayListHelpers.h |
|
5552 |
| TransformClipNode.h |
TransformClipNode stores a transformation matrix and a post-transform
clip rect.
They can be used to transform and clip a display item inside a flattened
nsDisplayTransform to the coordinate space of that nsDisplayTransform.
|
4143 |
| WindowRenderer.cpp |
StartFrameTimeRecording, together with StopFrameTimeRecording
enable recording of frame intervals.
To allow concurrent consumers, a cyclic array is used which serves all
consumers, practically stateless with regard to consumers.
To save resources, the buffer is allocated on first call to
StartFrameTimeRecording and recording is paused if no consumer which called
StartFrameTimeRecording is able to get valid results (because the cyclic
buffer was overwritten since that call).
To determine availability of the data upon StopFrameTimeRecording:
- mRecording.mNextIndex increases on each RecordFrame, and never resets.
- Cyclic buffer position is realized as mNextIndex % bufferSize.
- StartFrameTimeRecording returns mNextIndex. When StopFrameTimeRecording is
called, the required start index is passed as an arg, and we're able to
calculate the required length. If this length is bigger than bufferSize, it
means data was overwritten. otherwise, we can return the entire sequence.
- To determine if we need to pause, mLatestStartIndex is updated to
mNextIndex on each call to StartFrameTimeRecording. If this index gets
overwritten, it means that all earlier start indices obtained via
StartFrameTimeRecording were also overwritten, hence, no point in
recording, so pause.
- mCurrentRunStartIndex indicates the oldest index of the recording after
which the recording was not paused. If StopFrameTimeRecording is invoked
with a start index older than this, it means that some frames were not
recorded, so data is invalid.
|
9305 |
| WindowRenderer.h |
Record (and return) frame-intervals and paint-times for frames which were
presented between calling StartFrameTimeRecording and
StopFrameTimeRecording.
- Uses a cyclic buffer and serves concurrent consumers, so if Stop is
called too late
(elements were overwritten since Start), result is considered invalid
and hence empty.)
- Buffer is capable of holding 10 seconds @ 60fps (or more if frames were
less frequent).
Can be changed (up to 1 hour) via pref:
toolkit.framesRecording.bufferSize.
- Note: the first frame-interval may be longer than expected because last
frame
might have been presented some time before calling
StartFrameTimeRecording.
|
9635 |