gecko-dev/layout/base/UnitTransforms.h
Botond Ballo a9d2b5cbba Bug 1288210 - Simplify AlignFixedAndStickyLayers by not considering the local transform. r=kats,mstange
There shouldn't be a local transform because we create container layers
for fixed layers, so any transform would be on a descendant layer instead.

MozReview-Commit-ID: Kmya9vHZx1n

--HG--
extra : rebase_source : 88f31ae44e7baeff546624ba0e058d33ba6dc72f
2016-09-28 13:33:38 -04:00

295 lines
14 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* 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/. */
#ifndef MOZ_UNIT_TRANSFORMS_H_
#define MOZ_UNIT_TRANSFORMS_H_
#include "Units.h"
#include "mozilla/gfx/Matrix.h"
#include "mozilla/Maybe.h"
#include "nsRegion.h"
namespace mozilla {
// Convenience functions for converting an entity from one strongly-typed
// coordinate system to another without changing the values it stores (this
// can be thought of as a cast).
// To use these functions, you must provide a justification for each use!
// Feel free to add more justifications to PixelCastJustification, along with
// a comment that explains under what circumstances it is appropriate to use.
enum class PixelCastJustification : uint8_t {
// For the root layer, Screen Pixel = Parent Layer Pixel.
ScreenIsParentLayerForRoot,
// On the layout side, Screen Pixel = LayoutDevice at the outer-window level.
LayoutDeviceIsScreenForBounds,
// For the root layer, Render Target Pixel = Parent Layer Pixel.
RenderTargetIsParentLayerForRoot,
// For the root composition size we want to view it as layer pixels in any layer
ParentLayerToLayerForRootComposition,
// The Layer coordinate space for one layer is the ParentLayer coordinate
// space for its children
MovingDownToChildren,
// The transform that is usually used to convert between two coordinate
// systems is not available (for example, because the object that stores it
// is being destroyed), so fall back to the identity.
TransformNotAvailable,
// When an OS event is initially constructed, its reference point is
// technically in screen pixels, as it has not yet accounted for any
// asynchronous transforms. This justification is for viewing the initial
// reference point as a screen point. The reverse is useful when synthetically
// created WidgetEvents need to be converted back to InputData.
LayoutDeviceIsScreenForUntransformedEvent,
// Similar to LayoutDeviceIsScreenForUntransformedEvent, PBrowser handles
// some widget/tab dimension information as the OS does -- in screen units.
LayoutDeviceIsScreenForTabDims,
// A combination of LayoutDeviceIsScreenForBounds and
// ScreenIsParentLayerForRoot, which is how we're using it.
LayoutDeviceIsParentLayerForRCDRSF,
// Used to treat the product of AsyncTransformComponentMatrix objects
// as an AsyncTransformMatrix. See the definitions of these matrices in
// LayersTypes.h for details.
MultipleAsyncTransforms,
// We have reason to believe a layer doesn't have a local transform.
// Should only be used if we've already checked or asserted this.
NoTransformOnLayer
};
template <class TargetUnits, class SourceUnits>
gfx::CoordTyped<TargetUnits> ViewAs(const gfx::CoordTyped<SourceUnits>& aCoord, PixelCastJustification) {
return gfx::CoordTyped<TargetUnits>(aCoord.value);
}
template <class TargetUnits, class SourceUnits>
gfx::SizeTyped<TargetUnits> ViewAs(const gfx::SizeTyped<SourceUnits>& aSize, PixelCastJustification) {
return gfx::SizeTyped<TargetUnits>(aSize.width, aSize.height);
}
template <class TargetUnits, class SourceUnits>
gfx::IntSizeTyped<TargetUnits> ViewAs(const gfx::IntSizeTyped<SourceUnits>& aSize, PixelCastJustification) {
return gfx::IntSizeTyped<TargetUnits>(aSize.width, aSize.height);
}
template <class TargetUnits, class SourceUnits>
gfx::PointTyped<TargetUnits> ViewAs(const gfx::PointTyped<SourceUnits>& aPoint, PixelCastJustification) {
return gfx::PointTyped<TargetUnits>(aPoint.x, aPoint.y);
}
template <class TargetUnits, class SourceUnits>
gfx::IntPointTyped<TargetUnits> ViewAs(const gfx::IntPointTyped<SourceUnits>& aPoint, PixelCastJustification) {
return gfx::IntPointTyped<TargetUnits>(aPoint.x, aPoint.y);
}
template <class TargetUnits, class SourceUnits>
gfx::RectTyped<TargetUnits> ViewAs(const gfx::RectTyped<SourceUnits>& aRect, PixelCastJustification) {
return gfx::RectTyped<TargetUnits>(aRect.x, aRect.y, aRect.width, aRect.height);
}
template <class TargetUnits, class SourceUnits>
gfx::IntRectTyped<TargetUnits> ViewAs(const gfx::IntRectTyped<SourceUnits>& aRect, PixelCastJustification) {
return gfx::IntRectTyped<TargetUnits>(aRect.x, aRect.y, aRect.width, aRect.height);
}
template <class TargetUnits, class SourceUnits>
gfx::MarginTyped<TargetUnits> ViewAs(const gfx::MarginTyped<SourceUnits>& aMargin, PixelCastJustification) {
return gfx::MarginTyped<TargetUnits>(aMargin.top, aMargin.right, aMargin.bottom, aMargin.left);
}
template <class TargetUnits, class SourceUnits>
gfx::IntMarginTyped<TargetUnits> ViewAs(const gfx::IntMarginTyped<SourceUnits>& aMargin, PixelCastJustification) {
return gfx::IntMarginTyped<TargetUnits>(aMargin.top, aMargin.right, aMargin.bottom, aMargin.left);
}
template <class TargetUnits, class SourceUnits>
gfx::IntRegionTyped<TargetUnits> ViewAs(const gfx::IntRegionTyped<SourceUnits>& aRegion, PixelCastJustification) {
return gfx::IntRegionTyped<TargetUnits>::FromUnknownRegion(aRegion.ToUnknownRegion());
}
template <class NewTargetUnits, class OldTargetUnits, class SourceUnits>
gfx::ScaleFactor<SourceUnits, NewTargetUnits> ViewTargetAs(
const gfx::ScaleFactor<SourceUnits, OldTargetUnits>& aScaleFactor,
PixelCastJustification) {
return gfx::ScaleFactor<SourceUnits, NewTargetUnits>(aScaleFactor.scale);
}
// Unlike the other functions in this category, this function takes the
// target matrix type, rather than its source and target unit types, as
// the explicit template argument, so an example invocation is:
// ViewAs<ScreenToLayerMatrix4x4>(otherTypedMatrix, justification)
// The reason is that if it took the source and target unit types as two
// template arguments, there may be some confusion as to which is the
// source and which is the target.
template <class TargetMatrix, class SourceMatrixSourceUnits, class SourceMatrixTargetUnits>
TargetMatrix ViewAs(
const gfx::Matrix4x4Typed<SourceMatrixSourceUnits, SourceMatrixTargetUnits>& aMatrix,
PixelCastJustification) {
return TargetMatrix::FromUnknownMatrix(aMatrix.ToUnknownMatrix());
}
// Convenience functions for casting untyped entities to typed entities.
// Using these functions does not require a justification, but once we convert
// all code to use strongly typed units they should not be needed any longer.
template <class TargetUnits>
gfx::PointTyped<TargetUnits> ViewAs(const gfxPoint& aPoint) {
return gfx::PointTyped<TargetUnits>(aPoint.x, aPoint.y);
}
template <class TargetUnits>
gfx::PointTyped<TargetUnits> ViewAs(const gfx::Point& aPoint) {
return gfx::PointTyped<TargetUnits>(aPoint.x, aPoint.y);
}
template <class TargetUnits>
gfx::RectTyped<TargetUnits> ViewAs(const gfx::Rect& aRect) {
return gfx::RectTyped<TargetUnits>(aRect.x, aRect.y, aRect.width, aRect.height);
}
template <class TargetUnits>
gfx::IntSizeTyped<TargetUnits> ViewAs(const nsIntSize& aSize) {
return gfx::IntSizeTyped<TargetUnits>(aSize.width, aSize.height);
}
template <class TargetUnits>
gfx::IntPointTyped<TargetUnits> ViewAs(const nsIntPoint& aPoint) {
return gfx::IntPointTyped<TargetUnits>(aPoint.x, aPoint.y);
}
template <class TargetUnits>
gfx::IntRectTyped<TargetUnits> ViewAs(const nsIntRect& aRect) {
return gfx::IntRectTyped<TargetUnits>(aRect.x, aRect.y, aRect.width, aRect.height);
}
template <class TargetUnits>
gfx::IntRegionTyped<TargetUnits> ViewAs(const nsIntRegion& aRegion) {
return gfx::IntRegionTyped<TargetUnits>::FromUnknownRegion(aRegion);
}
// Unlike the other functions in this category, this function takes the
// target matrix type, rather than its source and target unit types, as
// the template argument, so an example invocation is:
// ViewAs<ScreenToLayerMatrix4x4>(untypedMatrix)
// The reason is that if it took the source and target unit types as two
// template arguments, there may be some confusion as to which is the
// source and which is the target.
template <class TypedMatrix>
TypedMatrix ViewAs(const gfx::Matrix4x4& aMatrix) {
return TypedMatrix::FromUnknownMatrix(aMatrix);
}
// Convenience functions for transforming an entity from one strongly-typed
// coordinate system to another using the provided transformation matrix.
template <typename TargetUnits, typename SourceUnits>
static gfx::PointTyped<TargetUnits>
TransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::PointTyped<SourceUnits>& aPoint)
{
return aTransform.TransformPoint(aPoint);
}
template <typename TargetUnits, typename SourceUnits>
static gfx::IntPointTyped<TargetUnits>
TransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::IntPointTyped<SourceUnits>& aPoint)
{
return RoundedToInt(TransformBy(aTransform, gfx::PointTyped<SourceUnits>(aPoint)));
}
template <typename TargetUnits, typename SourceUnits>
static gfx::RectTyped<TargetUnits>
TransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::RectTyped<SourceUnits>& aRect)
{
return aTransform.TransformBounds(aRect);
}
template <typename TargetUnits, typename SourceUnits>
static gfx::IntRectTyped<TargetUnits>
TransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::IntRectTyped<SourceUnits>& aRect)
{
return RoundedToInt(TransformBy(aTransform, gfx::RectTyped<SourceUnits>(aRect)));
}
template <typename TargetUnits, typename SourceUnits>
static gfx::IntRegionTyped<TargetUnits>
TransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::IntRegionTyped<SourceUnits>& aRegion)
{
return ViewAs<TargetUnits>(aRegion.ToUnknownRegion().Transform(
aTransform.ToUnknownMatrix()));
}
// Transform |aVector|, which is anchored at |aAnchor|, by the given transform
// matrix, yielding a point in |TargetUnits|.
// The anchor is necessary because with 3D tranforms, the location of the
// vector can affect the result of the transform.
template <typename TargetUnits, typename SourceUnits>
static gfx::PointTyped<TargetUnits>
TransformVector(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::PointTyped<SourceUnits>& aVector,
const gfx::PointTyped<SourceUnits>& aAnchor)
{
gfx::PointTyped<TargetUnits> transformedStart = TransformBy(aTransform, aAnchor);
gfx::PointTyped<TargetUnits> transformedEnd = TransformBy(aTransform, aAnchor + aVector);
return transformedEnd - transformedStart;
}
// UntransformBy() and UntransformVector() are like TransformBy() and
// TransformVector(), respectively, but are intended for cases where
// the transformation matrix is the inverse of a 3D projection. When
// using such transforms, the resulting Point4D is only meaningful
// if it has a positive w-coordinate. To handle this, these functions
// return a Maybe object which contains a value if and only if the
// result is meaningful
template <typename TargetUnits, typename SourceUnits>
static Maybe<gfx::PointTyped<TargetUnits>>
UntransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::PointTyped<SourceUnits>& aPoint)
{
gfx::Point4DTyped<TargetUnits> point = aTransform.ProjectPoint(aPoint);
if (!point.HasPositiveWCoord()) {
return Nothing();
}
return Some(point.As2DPoint());
}
template <typename TargetUnits, typename SourceUnits>
static Maybe<gfx::IntPointTyped<TargetUnits>>
UntransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::IntPointTyped<SourceUnits>& aPoint)
{
gfx::PointTyped<SourceUnits> p = aPoint;
gfx::Point4DTyped<TargetUnits> point = aTransform.ProjectPoint(p);
if (!point.HasPositiveWCoord()) {
return Nothing();
}
return Some(RoundedToInt(point.As2DPoint()));
}
// The versions of UntransformBy() that take a rectangle also take a clip,
// which represents the bounds within which the target must fall. The
// result of the transform is intersected with this clip, and is considered
// meaningful if the intersection is not empty.
template <typename TargetUnits, typename SourceUnits>
static Maybe<gfx::RectTyped<TargetUnits>>
UntransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::RectTyped<SourceUnits>& aRect,
const gfx::RectTyped<TargetUnits>& aClip)
{
gfx::RectTyped<TargetUnits> rect = aTransform.ProjectRectBounds(aRect, aClip);
if (rect.IsEmpty()) {
return Nothing();
}
return Some(rect);
}
template <typename TargetUnits, typename SourceUnits>
static Maybe<gfx::IntRectTyped<TargetUnits>>
UntransformBy(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::IntRectTyped<SourceUnits>& aRect,
const gfx::IntRectTyped<TargetUnits>& aClip)
{
gfx::RectTyped<TargetUnits> rect = aTransform.ProjectRectBounds(aRect, aClip);
if (rect.IsEmpty()) {
return Nothing();
}
return Some(RoundedToInt(rect));
}
template <typename TargetUnits, typename SourceUnits>
static Maybe<gfx::PointTyped<TargetUnits>>
UntransformVector(const gfx::Matrix4x4Typed<SourceUnits, TargetUnits>& aTransform,
const gfx::PointTyped<SourceUnits>& aVector,
const gfx::PointTyped<SourceUnits>& aAnchor)
{
gfx::Point4DTyped<TargetUnits> projectedAnchor = aTransform.ProjectPoint(aAnchor);
gfx::Point4DTyped<TargetUnits> projectedTarget = aTransform.ProjectPoint(aAnchor + aVector);
if (!projectedAnchor.HasPositiveWCoord() || !projectedTarget.HasPositiveWCoord()){
return Nothing();
}
return Some(projectedTarget.As2DPoint() - projectedAnchor.As2DPoint());
}
} // namespace mozilla
#endif