The reason we didn't merge just the gpu and platform threads from the get go was a deadlock in Shell:OnPlatformViewCreated and Shell:OnPlatformViewDestroyed.
The deadlock was caused by the platform thread starting a thread-hopping flow that ends ends up with the gpu thread releasing a latch that the platform thread is waiting on just after starting the cross-thread dance.
If the platform and gpu threads are the same, that last task that is posted to the gpu thread will never get executed as the gpu/platform thread is blocked on a latch.
This works around the deadlock by having a special case in the code for the scenario where the gpu and platform threads are the same.
Fixes: flutter/flutter#23974
This will allow us to easily visualize the time the platform informed the engine of a vsync event, its arguments, and when the engine began its UI thread workload using this information.
Allow Flutter to automatically dump the skp that triggers new shader compilations. This is useful for writing custom ShaderWarmUp to reduce jank. By default, it's not enabled to reduce the overhead. This is only available in profile or debug build.
Later, we can add service protocol support to pull the skp from the client to the host. Currently, it works fine for Android-based devices (including our urgent internal clients) where we can `adb shell` into the cache directory.
The ICU data is packaged within libflutter.so, and the engine uses
dlsym(RTLD_DEFAULT) to look up the ICU data symbols in the libraries
loaded by the process.
This approach apparently fails on some Xiaomi devices. If RTLD_DEFAULT
can not find the symbols, then the platform will pass the path to the
libflutter.so library so the engine can use that to resolve the symbols.
Fixes https://github.com/flutter/flutter/issues/28208
Some clients (e.g., embedded devices) prefer generating persistent cache files for the specific device beforehand, and ship them as readonly files in OTA packages.
The current assumption is that the embedder will wait till the vsync event and
then fire the callback. However, some embedders have that information upfront.
Since the time point has already been specified by the embedder, there is no
reason to burden the embedder with having to setup a wait either.
The shell was already designed to cleanly shut down the VM but it couldnt
earlier as |Dart_Initialize| could never be called after a |Dart_Cleanup|. This
meant that shutting down an engine instance could not shut down the VM to save
memory because newly created engines in the process after that point couldn't
restart the VM. There can only be one VM running in a process at a time.
This patch separate the previous DartVM object into one that references a
running instance of the DartVM and a set of immutable dependencies that
components can reference even as the VM is shutting down.
Unit tests have been added to assert that non-overlapping engine launches use
difference VM instances.
If a frame is scheduled by the time the delayed animator notify idle
task runs, then we should not call |NotifyIdle|.
Also, add a trace event in the task, to make it more clear in traces
when a notify idle call is coming from here (as opposed to
|Animator::AwaitVSync|).
Prior to this the Android embedder code would extract the icudtl.dat asset out
of the APK and write it to local disk during the first startup of the app.
This change will make that work unnecessary and eliminate the risk of ICU
failures due to errors in the extraction process.
Behavior (visual) changes should be very minor. Things that are to be expected:
* A few things were not color managed correctly by the transform canvas (color emoji, some color filters). Those will be handled correctly with the tagged surfaces (although we're always transforming to sRGB, so nothing should change until we target a wider gamut).
* Image filtering will happen in the source color space, rather than the destination. Very minor.
* The transform canvas did caching of images in the destination color space. Now, the conversion happens at draw time. If there are performance issues, images can be pre-converted to the destination with makeColorSpace().
* Support real fonts in 'flutter test'
Change the order of font_managers to query in font_collection
so that dynamic_font_manager fonts will be resolved.
Tested with test case in `flutter/flutter` repo:
`packages/flutter/test/rendering/localized_fonts_test.dart`
Ensured:
- A font loaded with FontLoader will be used
- The default 'Ahem' font is still loaded by default
The test above still cannot be fixed because FontLoader and the
underlying mechanisms don't cover Locale-specific font loading
and therefore a CJK font-family won't be able to be loaded as needed
for that test.
Fixes#17700
* Format fixup
In Dart 2, runtime checked mode has been eliminated. Many of these type
checks have been moved to static compile-time checks, the remainder are
enforced at runtime, and are no longer optional.
Pushes the check up front so that we don't always land on the
application_kernel_list_asset error message, which may be misleading to
readers.
Also switches the empty check to only emit in debug builds, since this
is almost certainly not a runtime error but a programmer error.
Adds a check/error message for the case where running from kernel list,
but application_kernel_list_asset is left unset (or empty).
Adds a check/error message for the case where we fail to load the
application_kernel_list_asset specified in the settings.
The service protocol's ListViews method needs to return description data for
each engine in the process. Previously ListViews would queue a task to each
UI thread to gather this data. However, the UI thread might be blocked from
executing tasks (e.g. if the Dart isolate is paused), resulting in a deadlock.
This change provides a copy of the engine's description data to the
ServiceProtocol's global list of engines, allowing ListViews to run without
accessing any UI threads.
Fixes https://github.com/flutter/flutter/issues/24400
Right now we do it whenever the platform views preview flag is on.
This is less efficient, filed
https://github.com/flutter/flutter/issues/24133 to only do this when
there's a platform view in the tree.
Moved the frame buffer specific logic from IOSGLContext to IOSGLRenderTarget.
use recording canvases for overlays
Support platform view overlays with gl rendering.
This also changes the overlay canvases (for both software and gl
rendering) be recording canvases, and only rasterize them after
finishing the paint traversal.
Handing a UIView refererence directly to the engine makes it challenging
for plugin authors to retain a controller for that UIView (e.g the
controller that talks over the platform channel) for as long as the
embedded view is needed.
We instead make the factory return a FlutterPlatformView which is a
wrapper around the UIView that the engine retains as long as the
platform view instance is needed. This allows plugin authors to keep
their control logic in the FlutterPlatformView and know that the engine
is responsible for retaining the reference.
When we visit a PlatformViewLayer during the paint traversal it replaces
the PaintContext's canvas with a new one that is painted ontop of the
embedded view.
We need to make sure that operations applied by parent layers are also
applied to the new canvas.
To achieve this we collect all the canvases in a SkNWayCanvas and use
this canvas by non leaf nodes. Leaf nodes still paint only to the "current"
canvas.
This PR moves the overlay canvas creation from the paint phase to the
preroll phase, collects them into a SkNWayCanvas and set it in
PaintContext.
To keep this PR focused, I only used the internal_nodes_canvas in the
tranform_layer.
Will followup with a PR that changes all internal layers to use the
internal_nodes_canvas.
