I've decided to fix this in a very explicit way. The only "magic" part that's
left is how we decide that the AGR of the perspective item is outside the
scrolled frame (and I'm not sure myself how that works).
I didn't want to change what scroll clips we set on what items, because the
scroll clip really belongs on the perspective item, because that's the item
that needs to be clipped, and it should also be the item that should be
scrolled if it weren't for the fact that APZ wouldn't know that it should
apply the perspective transform before the APZ transform.
MozReview-Commit-ID: BBw8VPohQI4
Matrix4x4 remains a typedef for Matrix4x4Typed<UnknownUnits, UnknownUnits>.
No client code needed changing, except for forward-declarations of Matrix4x4
as a class (since it's now a typedef).
This change necessitates a few other header changes around the tree:
either places that we relying on FrameLayerBuilder.h to #include
ImageLayers.h for them, or places that were bootlegging headers from
ImageLayers.h.
This change means that we don't have to have the definition of
LayerManager visible when FrameLayerBuilder is declared (which would be
necessary for, e.g. the ctor/dtor of ClippedDisplayItem), which means we
can get rid of the Layers.h include here in a subsequent patch.
Having to include all of Layers.h just to get at the definition of
LayerUserData is inconvenient, especially as most of the interesting
things in Layers.h can be forward-declared. Let's move LayerUserData to
its own header, so clients can include a small header for that,
forward-declare anything else they need from Layers.h, and reduce header
bloat.
The bulk of this commit was generated with a script, executed at the top
level of a typical source code checkout. The only non-machine-generated
part was modifying MFBT's moz.build to reflect the new naming.
CLOSED TREE makes big refactorings like this a piece of cake.
# The main substitution.
find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
xargs perl -p -i -e '
s/nsRefPtr\.h/RefPtr\.h/g; # handle includes
s/nsRefPtr ?</RefPtr</g; # handle declarations and variables
'
# Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h.
perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h
# Handle nsRefPtr.h itself, a couple places that define constructors
# from nsRefPtr, and code generators specially. We do this here, rather
# than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename
# things like nsRefPtrHashtable.
perl -p -i -e 's/nsRefPtr/RefPtr/g' \
mfbt/nsRefPtr.h \
xpcom/glue/nsCOMPtr.h \
xpcom/base/OwningNonNull.h \
ipc/ipdl/ipdl/lower.py \
ipc/ipdl/ipdl/builtin.py \
dom/bindings/Codegen.py \
python/lldbutils/lldbutils/utils.py
# In our indiscriminate substitution above, we renamed
# nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up.
find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \
xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g'
if [ -d .git ]; then
git mv mfbt/nsRefPtr.h mfbt/RefPtr.h
else
hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h
fi
Make the FrameLayerBuilder remember for what region it has calculated
display item visibility, then recompute the visibility whenever the
dirty region it is passed to DrawPaintedLayer changes.
This means that the caller does not have to know the entire dirty region
that will be drawn for the transaction, but we can still optimise cases
where it knows some of the dirty region in advance.
This fixes a regression where MultiTiledContentClient's low-res display
port would not be painted if a smaller region of its high-res buffer had
already been painted that transaction, since the FrameLayerBuilder
had decided that most of the larger low-res region was invisible.
Bug 1176077 introduced the parameter aDirtyRegion to
DrawPaintedLayerCallback, which allows the callback to recompute the
visibility of all items to be painted in that transaction in a single
go. However, this parameter can not always be determined correctly
when using RotatedBuffer, and using an incorrect value was causing
graphical glitches.
Make the parameter optional, and on null values do not perform the
optimisation. Pass null from ClientPaintedLayer, which uses
RotatedBuffer and was causing problems, but continue to pass the
correct value from other Layer implementations. This optimisation was
most important for tiled layers using progressive paint, so this is
okay.
FrameLayerManager::RecomputeItemsVisibility() was being called on every
call to FrameLayerBuilder::DrawPaintedLayer(), each time for the region
to be painted by that paint call. This is inefficient when progressive
paint is enabled. Change it so that we compute the visibility of all the
layer's items within the total region to be painted, but only on the
first paint after the display list has been modified.
Get rid of EnumerateEntries by inlining those enumerate functions. Also
move gPaintedDisplayItemLayerUserData, gColorLayerUserData, etc. to the
front of the file since LayerManagerData::Dump() references to one of
them.
The bulk of this commit was generated by running:
run-clang-tidy.py \
-checks='-*,llvm-namespace-comment' \
-header-filter=^/.../mozilla-central/.* \
-fix
This adds support for class="reftest-opaque-layer" and for
reftest-assigned-layer="some-layer-name" to the reftest harness.
From reftest/README.txt:
Opaque Layer Tests: class="reftest-opaque-layer"
================================================
If an element should be assigned to a PaintedLayer that's opaque, set the class
"reftest-opaque-layer" on it. This checks whether the layer is opaque during
the last paint of the test, and it works whether your test is an invalidation
test or not. In order to pass the test, the element has to have a primary
frame, and that frame's display items must all be assigned to a single painted
layer and no other layers, so it can't be used on elements that create stacking
contexts (active or inactive).
Layerization Tests: reftest-assigned-layer="layer-name"
=======================================================
If two elements should be assigned to the same PaintedLayer, choose any string
value as the layer name and set the attribute reftest-assigned-layer="yourname"
on both elements. Reftest will check whether all elements with the same
reftest-assigned-layer value share the same layer. It will also test whether
elements with different reftest-assigned-layer values are assigned to different
layers.
The same restrictions as with class="reftest-opaque-layer" apply: All elements
must have a primary frame, and that frame's display items must all be assigned
to the same PaintedLayer and no other layers. If these requirements are not
met, the test will fail.
