Handles the impeller unittests line in
https://github.com/flutter/flutter/issues/161456
This PR is focused on the unit tests in Impeller. There are still some
uses in other non-test areas which will be handled in a separate PR.
There can only be one handler that is shared between the views.
Move event redispatch matcher back into FlKeyboardManager - if a
redispatched event was moved between views due to a focus change it
would not be matched against.
Instead of doing the redispatching from inside FlKeyboardManager, return
the event asynchronously to the code that provided the event in the
first case. That code has all the context on how to do the redispatch -
this will be more complicated in a multi-view case.
The existing technique of offsetting a smaller texture is very vunerable
to bugs in the renderer. Rather than this approach, we can allocate a
new offscreen that is full sized and then blit a smaller region. To
reduce the allocation costs, we can also set up a transients cache which
will reuse this texture. In total, this should be more performant than
the existing partial repaint (due to lack of continual re-allocation) at
the cost of higher peak memory usage.
Fixes https://github.com/flutter/flutter/issues/140877
Fixes https://github.com/flutter/flutter/issues/160588
Fixes https://github.com/flutter/flutter/issues/156113
Fixes SkRSXform task in https://github.com/flutter/flutter/issues/161456
Removes (nearly) all uses of Skia SkRSXform object from DisplayList and
replaces it with a new Impeller RSTransform geometry object.
There are remaining uses in:
- Skia adapter code which needs to convert them back to SkRSXform when
using the Skia backend
- dl_rendering_tests which is waiting for a major conversion effort
- ?Fuchsia? code has an SkCanvas spy adapter used in its embedder code
(not DisplayList related)
- web_ui/skwasm
Address the first item in
https://github.com/flutter/flutter/issues/161456 (Unit tests in the
display_list/ directory)
Some new `DlPath::Make<Shape>` factories were added to make test writing
simpler.
`DlPath` is now bi-directional! You can construct one from either an
`SkPath` or an `impeller::Path` and it will auto-convert to the other as
needed. This allows unit tests with custom paths to rely on
`impeller::Path` for path construction instead of `SkPath` (as long as
only simple move/line/quad/curve verbs are needed).
`RoundRect` now normalizes the argument rect in all constructors to
match Flutter expectations and `SkRRect` legacy behavior. This behavior
was already being enforced in `ui.rrect` but the unit tests we have to
verify the behavior are written against the `RoundRect` object itself so
this was the simplest way to make the unit tests work right, while
ensuring that we maintain correct behavior for `ui` objects. Ideally
these issues would be tested at the `ui` native interface instead of as
unit tests on our internal objects and we should be allowed to decide
how we want our internal APIs to behave with regard to this concept.
Skia inverted path types are no longer allowed in `DlPath` and all use
of them should be eliminated in the engine (except to test if they crash
when used in a debug unit test)
A couple of unit tests for `DlOpSpy` and Impeller's interop package were
migrated here along for the ride even though this PR was focused
primarily on `display_list/` unit tests.
Adds an opt in flag for the Android surface control based swapchain.
This is off by default, but added for the sake of bug reports and
investigation of the functionality.
Follow-up to https://github.com/flutter/flutter/pull/160937
(https://github.com/flutter/flutter/issues/160933).
This more or less mirrors what we did already for `onSurfaceCreated` to
match `onSurfaceAvailable`.
With this approach, the master branch should immediately start seeing
better behavior in terms of being able to use the `onSurfaceDestroyed`
callback effectively (before the surface has been released). For
example, for a plugin that already uses `onSurfaceDestroyed`, [i.e
`camerax`](97ce56a68e/packages/camera/camera_android_camerax/android/src/main/java/io/flutter/plugins/camerax/PreviewHostApiImpl.java):
```java
@Override
public void onSurfaceDestroyed() {
// Invalidate the SurfaceRequest so that CameraX knows to to make a new request
// for a surface.
request.invalidate();
}
```
... the request is now invalidated _before_ the surface has been
destroyed, which is what (I believe) we wanted?
---
Folks that want to publish package updates that _definitely_ use the
correct timing will have to wait for the next stable.
/cc @hasali19 would be great to get your input here.
Ensures that the onscreen command buffer is blocked via the render ready
semaphore, instead of just the layout transition. I can confirm this
fixes the rendering artifacts on the Samsung a50 - which I suspect are
the same as the issues these other devices are having. Essentially,
without the semaphore protecting the onscreen pass we can end up writing
to the color attachment while it is still being read.
* https://github.com/flutter/flutter/issues/160522
* https://github.com/flutter/flutter/issues/160370
Closes https://github.com/flutter/flutter/issues/160933.
The timing of this callback gives our users (and plugin authors) a
chance to stop using the `Surface` before it becomes invalid, allowing
us to fix https://github.com/flutter/flutter/issues/156488 - we no
longer need to do shenanigans on storing and restoring state because
`ExoPlayer` can now handle it out of the box; see
https://github.com/flutter/flutter/issues/160933#issuecomment-2564092567.
It's unfortunate we have to go through a bit of churn on the callback
API, but realistically this _is_ the feedback we were looking for when
originally creating it - it just took longer than expected due to the
long release cycle.
/cc @hasali19, @xxoo, @camsim99
Fixes https://github.com/flutter/flutter/issues/160480
Both The Impeller and Skia variant of the OES texture rendering use the
same shared code path, so the Impeller code must match the weird 1x1
texture behavior of Skia. In addition, we have to add back the
TiledTextureContents support, since we need to render a texture with a
transform. I had previously tested this but neglected to force the
SurfaceTexture path, so I only tested the ImageReader path which does
not use a transform.
Adds files from flutter/flaux which contain modifications for the engine
structure. The history for engine/ has been edited. Please see
flutter/engine for the original PRs.
The RGB565 format documentation in embedder.h incorrectly stated that the red component uses the least significant bits. Unit tests in embedder/testdefs/embedder_unittests.cc demonstrate this is incorrect, showing:
- Red test (0xF800): Uses bits [15:11]
- Green test (0x07E0): Uses bits [10:5]
- Blue test (0x001F): Uses bits [4:0]
This commit fixes the documentation to correctly reflect the actual bit layout:
- Red uses 5 MSBs [15:11]
- Green uses 6 middle bits [10:5]
- Blue uses 5 LSBs [4:0]
Also fixes the example bit extraction code to use correct masks, matching the test expectations.
Impact: This change helps prevent potential developer confusion about RGB565 bit ordering and ensures the documentation matches the actual implementation as verified by the test suite.
The old path doesn't exist.
`dart format` stumbles over this non-existent include.
`analysis_options.yaml` files that just imported something non-existent were deleted.
I am surprised that this never caused any other issues. Is this all dead code that isn't actually analyzed?
This PR limits the search depth, because we don't want to enable this workaround for AdMob banner, which has a WKWebView in the depth of 7. See the previous PR for more context: https://github.com/flutter/engine/pull/57168
I was able to confirm that this returns YES for the 3P plugin, and NO for AdMob.
*List which issues are fixed by this PR. You must list at least one issue.*
https://github.com/flutter/flutter/issues/158961
*If you had to change anything in the [flutter/tests] repo, include a link to the migration guide as per the [breaking change policy].*
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
Continuing the migration of engine code to the new geometry classes. Only DlRTree and DlRegion are converted in this pass, plus a small amount of associated code.
The original workaround ([PR](https://github.com/flutter/engine/pull/56804)) works for the official web view plugin, but it doesn't work for a third party plugin `flutter_inappwebview` ([issue](https://github.com/pichillilorenzo/flutter_inappwebview)). Upon discussion with the author of that plugin, it turns out that their platform view is not a WKWebView, but rather a wrapper of WKWebView.
This PR performs a DFS search of the view hierarchy, and enable the workaround as long as there's a WKWebView inside.
TODO: pending sample project:
I am quite positive that it should work, but **I haven't tried it since I don't have a sample project yet**. I have requested a sample project with them so I can verify the solution.
*List which issues are fixed by this PR. You must list at least one issue.*
https://github.com/pichillilorenzo/flutter_inappwebview/issues/2415
*If you had to change anything in the [flutter/tests] repo, include a link to the migration guide as per the [breaking change policy].*
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
Eliminates some cases where `FlutterViewController` was relying on `FlutterEngine` internals:
* `[FlutterEngine shell]`
* `[FlutterEngine platformView]`
* `[FlutterEngine iosPlatformView]`
Instead, `FlutterEngine` now exposes:
* `installFirstFrameCallback:`
* `enableSemantics:withFlags:`
* `notifyViewCreated`
* `notifyViewDestroyed`
* `waitForFirstFrameSync:callback:`
Also fixes a couple cases where we were relying on transitive header includes:
* `FlutterAppController` relied on `FlutterViewController_Internal.h` for `sendDeepLinkToFramework:completionHandler:`
This is a refactoring followup to https://github.com/flutter/engine/pull/57099 that introduces no semantic changes.
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
`GetLastError` returns an unsigned 32 bit integer that was being
implicitly cast to an int for the std::variant<..., int>. This was
causing my build to fail with:
```
../../flutter/shell/platform/windows/platform_handler.cc(178,12): error: no viable conversion from returned value of type 'DWORD' (aka 'unsigned long') to function return type 'std::variant<std::wstring, int>' (aka 'variant<basic_string<wchar_t, char_traits<wchar_t>, allocator<wchar_t>>, int>')
178 | return ::GetLastError();
| ^~~~~~~~~~~~~~~~
../../../../../../../Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.42.34433/include\variant(923,7): note: candidate constructor (the implicit copy constructor) not viable: no known conversion from 'DWORD' (aka 'unsigned long') to 'const variant<basic_string<wchar_t>, int> &' for 1st argument
923 | class variant : private _SMF_control<_Variant_destroy_layer<_Types...>, _Types...> { // discriminated union
| ^~~~~~~
```
Commands:
```
./flutter/tools/gn --runtime-mode release --no-rbe
ninja -C .\out\host_release windows gen_snapshot flutter/build/archives:windows_flutter
```
Explicitly casting `::GetLastError` to an int fixes this issue.
I'm running on Windows 11 (Version 10.0.26100 Build 26100) with VS 2022
Community Edition.
@loic-sharma
Co-authored-by: Eric Seidel <eric@shorebird.dev>
Collection of changes to DlVertices::Builder and the stopwatch visualizer.
At a high level:
* improve performance of the stopwatch visualizer by pre-allocating storage (and sharing it across both visualizers), lookup up font once, and cache the debug frame rate used. Updates to use Dl types instead of SkTypes.
* Change DlVerticesBuilder to allow storing the bounds and use that in the visualizer, since we already know them. Make FML_CHECKS into dchecks, as the dart:ui vertices will already bounds check correctly - so these should only be necessary for debugging engine changes.
`FlutterEngine` at the `_shell` unique_ptr ivar it owns have different lifetimes. `_shell` is initialised transitively from `runWithEntrypoint`, and reset in `[FlutterEngine destroyContext]`, which is called transitively from `[FlutterviewController dealloc]` via `[FlutterEngine notifyViewControllerDeallocated]`.
As such, all uses of `_shell` should be checked either via an assertion, in cases we know the shell should be non-null, or via a runtime null check in cases where it's expected that it may be null.
Specifically, this guards against a crash that can occur if we get a CoreAnimation transaction commit callback for an inflight frame just as we're shutting down the app (or removing the FlutterView in an add-to-app scenario).
Example stack trace:
```
0 Flutter 0x11b28 -[FlutterEngine platformView] + 53 (weak_ptr.h:53)
1 Flutter 0x11994 -[FlutterEngine updateViewportMetrics:] + 186 (ref_ptr.h:186)
2 Flutter 0x1f854 -[FlutterViewController updateViewportMetricsIfNeeded] + 427 (vector:427)
3 Flutter 0x1f9b8 -[FlutterViewController viewDidLayoutSubviews] + 1411 (FlutterViewController.mm:1411)
4 UIKitCore 0x8c864 -[UIView(CALayerDelegate) layoutSublayersOfLayer:] + 2376
5 QuartzCore 0x1fa0c CA::Layer::layout_if_needed(CA::Transaction*) + 516
6 QuartzCore 0x1ae84c CA::Context::commit_transaction(CA::Transaction*, double, double*) + 516
7 QuartzCore 0x2888 CA::Transaction::commit() + 648
```
Issue: https://github.com/flutter/flutter/issues/98735
Issue: https://github.com/flutter/flutter/issues/159639
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
Migrate the use of fuchsia.io open functions to the new open3 replacements (which also requires transitioning from OpenFlags -> Flags). Likewise, we update all uses of the SDK VFS library to use the new set of flags.
This work is being done as part of the ongoing io2 migration in https://fxbug.dev/378924259
[C++, Objective-C, Java style guides]: https://github.com/flutter/engine/blob/main/CONTRIBUTING.md#style
