The following commits added the necessary tracking to ensure we always end
a profiler marker when we start one, and this just plugs them in.
MozReview-Commit-ID: C5VXpnlH2QA
--HG--
extra : rebase_source : 832c8abdef45e0a9bbb71bbf105fe4a76d0eb60d
This commit adds instrumentation and some asserting to track a "paint group",
which is essentially the first PaintThread::AsyncPaintContents until
PaintThread::AsyncEndLayerTransaction.
I didn't add an AsyncBeginLayerTransaction and use that to start a "paint group"
as I think it makes sense to not have a paint marker if we don't do
any painting.
MozReview-Commit-ID: AlSsUUF5ZOH
--HG--
extra : rebase_source : 84d9feaf5aac39fb9ce32984efda704fa2a68838
I believe this is the only case where we miss calling EndLayerTransaction,
but still had queued async paints. This wasn't an issue before, because
the following transaction would then do the synchronization. But I'd
like to use EndLayerTransaction for a profiler mark, so we should call
it unconditionally.
MozReview-Commit-ID: 9rUXBzCZaLO
--HG--
extra : rebase_source : ba360f2451190cf38745121d4dc32f4ed8d1bfe2
And remove unreachable code after MOZ_CRASH_UNSAFE_OOL().
MOZ_CRASH_UNSAFE_OOL causes data collection because crash strings are annotated to crash-stats and are publicly visible. Firefox data stewards must do data review on usages of this macro. However, all the crash strings this patch collects with MOZ_CRASH_UNSAFE_OOL are already collected with NS_RUNTIMEABORT.
MozReview-Commit-ID: IHmJfuxXSqw
--HG--
extra : rebase_source : 031f30934b58a7b87f960e57179641d44aefe5c5
extra : source : fe9f638a56a53c8721eecc4273dcc074c988546e
And remove unreachable code after MOZ_CRASH().
MozReview-Commit-ID: 6ShBtPRKYlF
--HG--
extra : rebase_source : 0fe45a59411bda663828336e2686707b550144ae
extra : source : 8473fd7333d2abe1ea1cc176510c292a5b34df45
Even if we cannot reuse the back buffer, we still use it to initialize the new back buffer.
The correct condition for detecting changes to buffer mode should only care if there is a
buffer, not if we decided to reuse it.
If the buffer mode has changed to component alpha, we need to invalidate and repaint everything
because we won't be able to copy over content to the new buffer correctly.
I believe this is a regression from the refactoring that created this function from
RotatedContentBuffer::BeginPaint.
MozReview-Commit-ID: H9G7GxqekLt
--HG--
extra : rebase_source : afb3321694ce3f1b54ad4dce28b4a7c7c7b47904
extra : amend_source : fdef1994514fcc9b9f2e952767a785841e9ddda0
extra : histedit_source : 3e8324255f516094c242ab3ec0688d2285c85a99
Add a new TextureClientData type, AndroidNativeWindowTextureData,
backed by a SurfaceTexture in single-buffer mode. It uses the
NativeWindow API, which provides producer-side access to the buffer.
This provides a DrawTarget which can be used to paint directly in to
the SurfaceTexture, which can then be composited using a SurfaceTextureHost.
Due to API restrictions it is not possible to read from a NativeWindow
while the corresponding SurfaceTexture has ownership of the
buffer. TiledContentClient now handles that by painting the additional
region that it cannot copy from the front buffer, if required.
MozReview-Commit-ID: 1NZq6MQqwFq
--HG--
extra : rebase_source : 9d1db721d4892f3df033d43127489a85421e8863
This patch was generated automatically by the "modeline.py" script, available
here: https://github.com/amccreight/moz-source-tools/blob/master/modeline.py
For every file that is modified in this patch, the changes are as follows:
(1) The patch changes the file to use the exact C++ mode lines from the
Mozilla coding style guide, available here:
https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Coding_Style#Mode_Line
(2) The patch deletes any blank lines between the mode line & the MPL
boilerplate comment.
(3) If the file previously had the mode lines and MPL boilerplate in a
single contiguous C++ comment, then the patch splits them into
separate C++ comments, to match the boilerplate in the coding style.
MozReview-Commit-ID: 77D61xpSmIl
--HG--
extra : rebase_source : c6162fa3cf539a07177a19838324bf368faa162b
This patch was automatically generated. I found the files to be fixed in this
patch with the following command:
grep -r "// \* This Source Code" gfx
...and then I modified each of these files with the following script
(where $1 is the filename to be modified):
####
old="// \* This Source Code Form is subject to the terms of the Mozilla Public"
new=" \* This Source Code Form is subject to the terms of the Mozilla Public"
sed -i s%"^$old"%"$new"% $1
####
MozReview-Commit-ID: Ihx0EAOkT2g
--HG--
extra : rebase_source : 4dc97f234495e3ec1f4904e91d43d00fd3810626
This patch was automatically generated. I found the files to be fixed in this
patch with the following command:
grep -r "^\* This Source Code" gfx
...and then I modified each of these files with the following script
(where $1 is the filename to be modified):
###
line1="\* This Source Code Form is subject to the terms of the Mozilla Public"
line2="\* License, v\. 2\.0\. If a copy of the MPL was not distributed with this"
line3="\* file, You can obtain one at http://mozilla\.org/MPL/2\.0/\. \*/"
# Insert 1 space at beginning:
sed -i s%"^$line1"%" $line1"% $1
sed -i s%"^$line2"%" $line2"% $1
sed -i s%"^$line3"%" $line3"% $1
###
MozReview-Commit-ID: HXBMrfnhlVr
--HG--
extra : rebase_source : de4c78563711f8366e2978c5199a5041875fbe38
This patch doesn't modify the mode lines at all -- it just swaps their order,
and makes each one its own C++ comment, separate from the MPL boilerplate
comment.
MozReview-Commit-ID: BEZJVj2sMuK
--HG--
extra : rebase_source : 9e0946c8a8d0b67c11b5932b9d1e51e0e6e8ebef
These items don't paint anything and are handled separately, so we don't need
to call CreateWebRenderCommands for them. We could just implement
CreateWebRenderCommands for the item and have it return true unconditionally,
but it's still an extra virtual call. Plus there's existing code in
WebRenderCommandBuilder that's conditioned specifically for event regions so
I prefer keeping all the conditioned code together.
MozReview-Commit-ID: 5kTuv7qgiKU
--HG--
extra : rebase_source : 8bd1a3fdf6711d06ff7523a25324da3a170215a3
This changes the serialization format a little bit.
We now have an index at the end of the blob. This
is currently used to store a list of the used font keys.
In the future we'll add rects and can use it for invalidation.
This extends the fix in bug 1373802 to account for extra levels of
display item nesting. If those extra intermediate display items don't
push any clips we still want to pick up the ClipAndScroll from the
enclosing ancestor that has it.
MozReview-Commit-ID: AmxRz4fBKnX
--HG--
extra : rebase_source : ae12a9f797ab201169ff199d0d42a29df51ee1cb
We already support cases where we have scrolling clips applied to fixed
items. However if we had to build nested clips inside those items, then
those nested clips would not properly inherit from the scrolling clips.
This patch addresses that case.
MozReview-Commit-ID: CWp1x0EsyaP
--HG--
extra : rebase_source : f8c80ace2da39edebaabd5339670a68038a18489
The window buttons are drawn as part of the AddWindowOverlayWebRenderCommands
function which is invoked in the full-transaction codepath. It should be possible
to have the empty transaction codepath simply update the image (without building
a full WR display list) and do a recomposite. That would be more performant but
it requires some plumbing to build and ship across a IpcResourceUpdateQueue on
empty transactions.
MozReview-Commit-ID: 2Mrb0wELD6E
--HG--
extra : rebase_source : 9a94c32f94403050835bf3445176f4fe2c1579fa
This is functionally unrelated to the bug but I noticed it while fiddling with
the code, and the lines affected are kind of intertwined with the next patch so
I'm just doing the code removal as part of this bug.
MozReview-Commit-ID: CwmluhyCdbR
--HG--
extra : rebase_source : 0a86ba7299f63587b77c44fadc804e34ada5a474
We want to always use Servo animation backend on the compositor.
However, Android doesn't support Stylo now, so add a defined flag for it.
MozReview-Commit-ID: 63MnTBnq6yv
--HG--
extra : rebase_source : dce46737f81e4e217e9fd67b6bec722994dca6a2
Use the new added FFI, Servo_ComposeAnimationSegment, to compose an
animation segment from Servo backend on the compositor.
MozReview-Commit-ID: LNgpCSIlDl9
--HG--
extra : rebase_source : 5b5c145fae877b4f4b01ea54259737dc9dad2951
We will use Servo backend on the compositor, so implement this for opacity.
MozReview-Commit-ID: BeWR2nBSbjb
--HG--
extra : rebase_source : eb5db3cf04640a83f13857984e792a949f26bcc7
My original understanding of the API was flawed, and so while I had
position:sticky working in some cases it didn't work properly in a lot
of other cases. This patch corrects the usage of the API to match what
WR is expecting and fixes a lot of test cases.
MozReview-Commit-ID: AdMux19Fk9U
--HG--
extra : rebase_source : e7f970a710b2079501a91eeeac8c292b603210dc
Most of this change is just fiddling with function signatures so that they take
a LayerManager* instead of a Layer* (or in some cases, both). This allows
the WebRender codepaths to pass a WebRenderLayerManager* instead of having to
produce a Layer* which it doesn't have.
MozReview-Commit-ID: Fb0C8OUVDin
--HG--
extra : rebase_source : e4c3324cfa20c295db85d5c09df8d8d77865bb6a
Instead of unconditionally pushing and popping clips per display item,
this patch changes things so that for each recursive display list, we
create an ItemClips struct. We push this onto the stack when we enter
the display list, and pop it off at the end. For each display item, we
check to see if the clips would actually change compared to the previous
display item, and only do the pop/repush in that case.
MozReview-Commit-ID: J0MCc2V9hWT
--HG--
extra : rebase_source : 8617cfaa7391457867f01c1b619cb871a21bf3f5
This makes ScrollingLayersHelper a non-RAII type class, and instead adds
methods to notify it of when we start processing a new transaction or a
new display item within the transaction. This patch has no functional
changes, it's non-obvious refactoring.
MozReview-Commit-ID: GEZzCGbVqB1
--HG--
extra : rebase_source : dd4fab7f34da7c5465ba474db670cf583e8dadf4
Storing the per-item clip state in a struct like this will allow us to
easily compare the desired clip state across items, so we can avoid
doing unnecessary work when going from one item to the next. This patch
has no functional changes, it's just refactoring.
MozReview-Commit-ID: GX2FX4YDusO
--HG--
extra : rebase_source : 06d7bae5cbae99d2987881f26f7ebd26965c1799
Instead just keep a ref to it as a member variable. No functional
change.
MozReview-Commit-ID: 9jSBdZRIGuV
--HG--
extra : rebase_source : cd1b2f500e9a833f7d42bce53bdaec2118e0d4c3
Instead of unconditionally pushing and popping clips per display item,
this patch changes things so that for each recursive display list, we
create an ItemClips struct. We push this onto the stack when we enter
the display list, and pop it off at the end. For each display item, we
check to see if the clips would actually change compared to the previous
display item, and only do the pop/repush in that case.
MozReview-Commit-ID: GadIp2J8TrA
--HG--
extra : rebase_source : ba64c6b4659b8e51cab19b807088b9a50d71b85a
This makes ScrollingLayersHelper a non-RAII type class, and instead adds
methods to notify it of when we start processing a new transaction or a
new display item within the transaction. This patch has no functional
changes, it's non-obvious refactoring.
MozReview-Commit-ID: 3yq9sPiHMge
--HG--
extra : rebase_source : 286423f56de59211e320f015cb1004a1e98332b8
Storing the per-item clip state in a struct like this will allow us to
easily compare the desired clip state across items, so we can avoid
doing unnecessary work when going from one item to the next. This patch
has no functional changes, it's just refactoring.
MozReview-Commit-ID: 49B6hmsWZ4V
--HG--
extra : rebase_source : 8ac4bbf039f81bc2d26e92924ed041fa3d18e5ba
Instead just keep a ref to it as a member variable. No functional
change.
MozReview-Commit-ID: 5fccUlSifsA
--HG--
extra : rebase_source : a14672f926c383354db76e553ae23613fe4a432a
One of the sticky-pos tests was only passing because of two wrongs that
cancelled each other out in the old code. Specifically, instead of
defining a nested clip with the sticky clip as an ancestor, the clip was being
defined with the root ASR as an ancestor. Both resulted in the nested
clip not scrolling with the actual scrolling scrollframe and so the test
was passing.
The new code changes things so that the nested clip is defined with the
actual scrolling scrollframe as the ancestor, causing the reftest to fail.
Fixing the clip ancestry is not hard but it reveals other problems so
so I'm deferring that to a follow-up bug.
MozReview-Commit-ID: DldAKi1AP4l
--HG--
extra : rebase_source : a448e06fd26fff21bbc4a6f50e04dbbdb427298c
This handles some cases where a nested display item's clip chain
implicitly extends from the wrapper item's clip chain.
MozReview-Commit-ID: DmghxOWi81K
--HG--
extra : rebase_source : 8ec95df64fed247650baf8f5e0c868d1934aa6bc
Bug 1409442 is tracking a change that will allow scroll layers to have
multiple ancestors. Without that, there are cases we cannot properly
handle, and so we need to ignore a clip in those scenarios. This patch
makes sure we do that instead of crashing.
MozReview-Commit-ID: 7AU4uyzT6if
--HG--
extra : rebase_source : d483c2a7ecff5cd488a53fa5e6b6da55cc3a1e29
When display items (such as mask items) push an out-of-band clip, we
can't use clip ids from, or update clip ids into, the cache. We also
need to ensure we take these out-of-band clips into account when
determining the parent for a new clip we are going to define.
MozReview-Commit-ID: GcUI2Hf6SLB
--HG--
extra : rebase_source : 512be5a6998b9c1d6db9d8c550231f0883d721c9
This code is more straightforward in its recursion than the old code,
and provides a relatively clean way to explicitly pass the desired
parent when defining a new clip or scroll layer.
Refer to the documentation in the patch for more details.
Note that this patch provides the basic recursive algorithm to define
the clips and scroll layers, although it omits some of the complicating
edge cases which will be added in later patches. The new code is not
invoked from anywhere until all the edge case handling has been done.
MozReview-Commit-ID: 7z51Kd7LlPU
--HG--
extra : rebase_source : 491efac9282a7379f7977a1bc0205ac70e356c3c
The APIs now allow providing the parent clip or scroll info explicitly
instead of having to push it on the stack. For now we just pass
Nothing() to preserve the existing behaviour, so this change is a
functinoal no-op.
MozReview-Commit-ID: dtNamN595
--HG--
extra : rebase_source : 3b6bd03b919bd31cd89e3f82283cb962f8f6abc5
From the crashes associated with bug 1389021, we know that some
compositor thread IPDL owners are not being cleaned up properly. We do
not know which protocols are causing the problem, so we temporarily will
annotate the logs with the ownership status. This should be limited to
under a dozen instances of the protocols.
This change will be backed out after a few builds are produced with it
and we see the first crash reports with the relevant data.
From the crashes associated with bug 1389021, we know that some
compositor thread IPDL owners are not being cleaned up properly. We do
not know which protocols are causing the problem, so we temporarily will
annotate the logs with the ownership status. This should be limited to
under a dozen instances of the protocols.
This change will be backed out after a few builds are produced with it
and we see the first crash reports with the relevant data.
mInAsyncPaint is only needed because the code using CreateBuffer expects the
buffer to be locked when it is returned. This isn't necessary because we can
now lock in BeginPaint.
MozReview-Commit-ID: 8WazvBKMbvb
--HG--
extra : rebase_source : ec59ca27e88257bd8e6d384685bee8227f545b31
Now that BeginPaintBuffer and BeginPaint are both implemented by content client,
I think it makes sense to just merge them into one method. This simplifies the
interface clients have to use.
MozReview-Commit-ID: 8bjH6WcpZS9
--HG--
extra : rebase_source : c2aad6cd9d85f7dd09652ee253683f03825e95c6
This commit does some more cleanup on the content client class hierarchy.
Some methods were virtual or instance methods when they didn't need to be.
MozReview-Commit-ID: 2y2D3zYtYvM
--HG--
extra : rebase_source : ebd137e4a1ee8eba10467e987bd2d91ea44ca5a2
CapturedPaintState::mTarget is actually a dual draw target to the black and
white draw targets, so this code will have the white draw target cleared black
and then white. This isn't incorrect, it's just wasteful.
MozReview-Commit-ID: ItgiSmegPK6
--HG--
extra : rebase_source : 9c3c88f463087a75383f1940ddeef6bb68ec35c4
UpdateDestinationFrom actually only uses a rect, so it's best to not be
misleading and force the clients of it to do the conversion.
MozReview-Commit-ID: EsUv3apqnku
--HG--
extra : rebase_source : 6ee0f7454ac2ea21618a62a9ab906857384d0fdf
This organizes the rotated buffer declarations and adds some documentation.
MozReview-Commit-ID: 5r24RiUMM6o
--HG--
extra : rebase_source : 28045e2d6b19db1f31796b7119717d417ee2dd34
This could be a part of the previous commit, but I omitted it to keep the size
down. Just some updates to comments and documentation.
MozReview-Commit-ID: HkCOSIQhBNv
--HG--
extra : rebase_source : 665b6bbe18dcd409809d33e1be37a56c0dc304d7
This commit is the guts of the refactoring.
This commit removes rotated content buffer, and moves the necessary
functionality to content client. All content clients now do not inherit from
rotated content buffer, but contain a refptr to a rotated buffer.
For the basic content client, this is a DrawTargetRotatedBuffer.
For the remote content client, this is a RemoteRotatedBuffer.
The double buffered content client contains an additional RemoteRotatedBuffer
for the front buffer.
Each derived class of content client only needs to implement a method to
create its own rotated buffer. The rest is handled by the base content
client. The remote content clients still override some additional methods
to sync buffers and do IPC.
MozReview-Commit-ID: B9vcSi8UYhm
--HG--
extra : rebase_source : a727d5548b0167eb12676837b173355eb0dd38a5
This commit splits off the part of BeginPaint that makes the decision about
whether to keep the buffer, its surface type, its content type, and the regions
to invalidate or paint.
MozReview-Commit-ID: JcPlv8GiRpA
--HG--
extra : rebase_source : 73c2de46ac9e11437e6a52aadd1da56028a7d3ec
This commit adds a method to rotated buffer to perform the adjusting to the new
destination rect, and potential unrotating that may need to happen. The goal of
this is to simplify and prepare BeginPaint to be moved to content client.
MozReview-Commit-ID: B4qeZqLjORR
--HG--
extra : rebase_source : 8ee7fb96ce8f1207b73027bb1d171dabd5b35788
Miscellaneous cleanup, PaintState doesn't need to know this as no one uses
this information outside of the content client classes.
MozReview-Commit-ID: 6K8LxEPvp4V
--HG--
extra : rebase_source : 931f7726d2f8f1f5ab32b5cf8f9030aa39a64bff
Miscellaneous cleanup, ContentClientSingleBuffered doesn't need to track if
the front and back buffer differ.
MozReview-Commit-ID: IvAw0b92jsh
--HG--
extra : rebase_source : c6b2d91666338f1d222f783b7a14e9d3b0ad6802
This commit refactors part of BeginPaint to treat the new and old back buffer's
as rotated buffers. The goal of this is to simplify and prepare this function to
be moved out of rotated content buffer.
MozReview-Commit-ID: 5iHeMbCWfpo
--HG--
extra : rebase_source : 232b699ca5d906e4784338914ebddd510730c379
This adds a new implementation of rotated buffer, which is backed by draw
targets. This will be the rotated buffer that basic content clients use.
MozReview-Commit-ID: Knn79yEFnxN
--HG--
extra : rebase_source : 78201bbb31c76e2b8131d912600b55500ba4cbdb
The front buffer for a double buffered content client is really just another
rotated buffer, so it can simplify the code to use the class we just added.
The goal is to have the back and the front buffer using remote rotated buffers,
but this is a good first step.
Note: The front buffer is represented as a Maybe<RemoteRotatedBuffer> in this
commit, but in the future it will be a RefPtr. That can't be done yet, because
rotated buffer can't implement refcounting in addition to compositable client.
MozReview-Commit-ID: Czk3otkf1pb
--HG--
extra : rebase_source : b7cc399643295bb63efa86739f3dae29d3375ae6
This adds a new implementation of rotated buffer, which is backed by texture
clients. This will be the rotated buffer that remote content clients use.
MozReview-Commit-ID: 3Y776uk5mFG
--HG--
extra : rebase_source : 40acce3f4150c5f6f51bf73f0b3c7e974ed8fd72
Similar to the previous patch, more functionality can be moved to rotated
buffer and out of rotated content client.
MozReview-Commit-ID: FNqfonyBBq9
--HG--
extra : rebase_source : fce8c6212321e6e25dc414c517a28e1e3c747c25
UpdateDestinationFrom is a useful method for any rotated buffer. This will be
needed when content clients are no longer rotated buffers.
Note: EnsureBuffer and EnsureBufferOnWhite were moved out of
UpdateDestinationFrom because I'd like to kill those methods, and so the calls
were moved to all users of UpdateDestinationFrom.
MozReview-Commit-ID: 2e3HhUsZ6iw
--HG--
extra : rebase_source : dd8f5a52808924a64ea0eb17359b6b749426b64e
This class doesn't seem to be necessary and only complicates the class
hierarchy.
MozReview-Commit-ID: 6dRXce1vB9T
--HG--
extra : rebase_source : 71591b44c68f9c8b73c7aadb6be245caaec0fbb5
Some css filter codes only used in layers-full mode. Since layers-full
already removed, we can also remove those unused codes.
MozReview-Commit-ID: 8YrfOfuXHNt
--HG--
extra : rebase_source : e9475d9100fbc2e5b301833f31fde1db458117aa
2017-10-20 16:10:03 +08:00
Matt Woodrow ext:(%2C%20Miko%20Mynttinen%20%3Cmikokm%40gmail.com%3E%2C%20Timothy%20Nikkel%20%3Ctnikkel%40gmail.com%3E)
It appears that in this case just waiting for the APZ flush to complete was
insufficient, probably because the APZ flush and touch inputs go via different
IPC channels and therefore do not provide ordering guarantees. So instead of
waiting for the APZ flush, let's just wait until the touchstart is received
in the content process and use that to resume the test.
MozReview-Commit-ID: AcPRhox1Xkg
--HG--
extra : rebase_source : be6653c7835d9a61bdc215266367adcc8adf8f0b
The two happening at the same time can lead to the APZ autoscroll being
cancelled due to APZ receiving a main-thread scroll offset update.
To achieve this:
- The content process assumes APZ is handling the autoscroll until
told otherwise.
- If the parent process knows APZ won't handle an autoscroll, it
tells the content process via its response to the Autoscroll:Start
message. This covers all cases where APZ doesn't handle the
autoscroll, except the case where APZCTreeManager itself rejects
the autoscroll and it lives in the compositor process rather than
the parent process.
- If APZCTreeManager rejects an autoscroll and it lives in the
compositor process, it sends an 'autoscroll-rejected-by-apz' message
to the content process.
MozReview-Commit-ID: L62v4COai6W
--HG--
extra : rebase_source : bc4c6417e77461634263defb88e67ed5036c454e