iOS] Migrate VSyncClient to a pure Obj-C implementation (#186166)#186935
Conversation
Migrates `FlutterVSyncClient` from a hybrid C++/Obj-C implementation to a pure Objective-C implementation by removing C++ types from its interface and implementation. Previously, `FlutterVSyncClient` stored a C++ `flutter::VsyncWaiter::Callback` and used a temporary `std::unique_ptr<flutter::FrameTimingsRecorder>` to propagate vsync timestamps. This was redundant: * `FlutterVSyncClient` allocated a temporary `FrameTimingsRecorder` and recorded vsync times. * It passed this to a C++ lambda wrapper. * The lambda extracted the start/target times as seconds and passed them to the Obj-C block callback. * The temporary recorder was immediately discarded. * `VsyncWaiterIOS` converted the seconds back to `fml::TimePoint`. * `VsyncWaiter::FireCallback` allocated *another* `FrameTimingsRecorder` and recorded the same times again. We now: * manage the callback as pure Obj-C block that accepts raw `CFTimeInterval` (double seconds) directly from `CADisplayLink`. * Remove the redundant `FrameTimingsRecorder` and C++ lambda wrapper. * Rely solely on `VsyncWaiter::FireCallback` to handle the final `FrameTimingsRecorder` creation and recording, which it already does. No test changes because this introduces no semantic change and is covered by existing tests recently added in: flutter#186457 Issue: flutter#112232
|
It looks like this pull request may not have tests. Please make sure to add tests or get an explicit test exemption before merging. If you are not sure if you need tests, consider this rule of thumb: the purpose of a test is to make sure someone doesn't accidentally revert the fix. Ask yourself, is there anything in your PR that you feel it is important we not accidentally revert back to how it was before your fix? Reviewers: Read the Tree Hygiene page and make sure this patch meets those guidelines before LGTMing. If you believe this PR qualifies for a test exemption, contact "@test-exemption-reviewer" in the #hackers channel in Discord (don't just cc them here, they won't see it!). The test exemption team is a small volunteer group, so all reviewers should feel empowered to ask for tests, without delegating that responsibility entirely to the test exemption group. |
There was a problem hiding this comment.
Code Review
This pull request refactors VSync timing on iOS by transitioning from FML-based time points to CACurrentMediaTime and replacing C++ callbacks with Objective-C blocks in FlutterVSyncClient. It also updates VsyncWaiterIOS to align these timestamps with the engine's steady clock. Feedback was provided regarding a potential division-by-zero error and a misleading code comment in the duration calculation logic.
| # for example), consider adding them to the ":framework_common_swift_unittests" | ||
| # target. | ||
| # for example), consider adding them to the | ||
| # ":framework_common_swift_unittests" target. |
There was a problem hiding this comment.
Not an actual change but turns out we were past 80 cols which upset the format presubmit.
| "framework/Source/SemanticsObjectTest.mm", | ||
| "framework/Source/SemanticsObjectTestMocks.h", | ||
| "framework/Source/UIViewController_FlutterScreenAndSceneIfLoadedTest.mm", | ||
| "framework/Source/VsyncWaiterIOSTest.mm", |
There was a problem hiding this comment.
Yes, the filename is inconsistent with the implementation file, but the test is an Obj-C XCTest and the class under test is a C++ class, hence the difference in naming.
| CGFloat currentInset = delegate.physicalViewInsetBottom; | ||
| self.keyboardAnimationView.frame = CGRectMake(0, currentInset, 0, 0); | ||
| self.keyboardAnimationStartTime = fml::TimePoint::Now().ToEpochDelta().ToSecondsF(); | ||
| self.keyboardAnimationStartTime = CACurrentMediaTime(); |
There was a problem hiding this comment.
Note that this was adjusted since FlutterVSyncClient is now pure Obj-C and sticks to CoreAnimation APIs (along with the rest of the embedder now).
| VsyncWaiterIOS::SnapDuration(targetTime - startTime, max_refresh_rate_); | ||
|
|
||
| // Align target time to the C++ steady_clock used by fml::TimePoint. | ||
| fml::TimePoint target_time = start_time + fml::TimeDelta::FromSecondsF(duration); |
There was a problem hiding this comment.
The core engine deals in fml::TimePoints which are based on std::chrono::steady_clock which is backed by mach_continuous_time, whereas CoreAnimation uses mach_absolute_time.
Consistent with embedders on the embedder API, we adjust the epoch for start and target time here.
There was a problem hiding this comment.
The core engine deals in fml::TimePoints which are based on std::chrono::steady_clock which is backed by mach_continuous_time, whereas CoreAnimation uses mach_absolute_time.
This comment is helpful. can you add this info to the comment?
Added tests. |
hellohuanlin
left a comment
There was a problem hiding this comment.
should we rename this to .m file?
| VsyncWaiterIOS::SnapDuration(targetTime - startTime, max_refresh_rate_); | ||
|
|
||
| // Align target time to the C++ steady_clock used by fml::TimePoint. | ||
| fml::TimePoint target_time = start_time + fml::TimeDelta::FromSecondsF(duration); |
There was a problem hiding this comment.
The core engine deals in fml::TimePoints which are based on std::chrono::steady_clock which is backed by mach_continuous_time, whereas CoreAnimation uses mach_absolute_time.
This comment is helpful. can you add this info to the comment?
| return client_.refreshRate; | ||
| } | ||
|
|
||
| CFTimeInterval VsyncWaiterIOS::SnapDuration(CFTimeInterval duration, double max_refresh_rate) { |
There was a problem hiding this comment.
what's "snap duration"?
There was a problem hiding this comment.
See the comment on the method: it snaps the duration to a round number of Hz. Due to floating point error or tiny inconsistencies in timing the duration may be off by a few microseconds etc, so we compute the inverse (frequency in Hz) snap to the closes integer Hz (60, 120, etc. whatever the nearest int value of Hz is), then adjust the time to keep the frame timing from drifting)
We were doing this before but decided it'd be better to put it in a named method with a doc comment.
We can do better :) In my next PR, I'm renaming it to a .swift file. Will update the comment in that PR too of that's okay to avoid the reapproval dance this close to your end of day. |
Roll Flutter from e03b91f1fe34 to f3a4b9897834 (63 revisions) flutter/flutter@e03b91f...f3a4b98 2026-05-26 47866232+chunhtai@users.noreply.github.com Update batch release doc to reflect latest workflow (flutter/flutter#186979) 2026-05-26 engine-flutter-autoroll@skia.org Roll Skia from 0442274cc696 to 27a819894f7c (5 revisions) (flutter/flutter#187094) 2026-05-26 bkonyi@google.com [Tool Robustness] Gracefully handle asynchronous subprocess crashes and connection timeouts (flutter/flutter#186964) 2026-05-26 bkonyi@google.com [pubspec] Bump Dart SDK constraint to ^3.13.0 (flutter/flutter#186957) 2026-05-26 engine-flutter-autoroll@skia.org Roll Dart SDK from 7eb54169841d to 00e625453c43 (1 revision) (flutter/flutter#187086) 2026-05-26 bdero@google.com [Impeller] Retire Y-coord-scale plumbing by flipping GLES at the vertex stage (flutter/flutter#186556) 2026-05-26 engine-flutter-autoroll@skia.org Roll Skia from f4f294bdf98d to 0442274cc696 (2 revisions) (flutter/flutter#187079) 2026-05-26 kevmoo@users.noreply.github.com [flutter_tools] Fix version cache poisoning from git environment variables (flutter/flutter#186595) 2026-05-26 bkonyi@google.com [Tool] Handle DTD connection failures gracefully in widget-preview (flutter/flutter#186952) 2026-05-25 engine-flutter-autoroll@skia.org Roll Skia from 9d1adb5f2427 to f4f294bdf98d (1 revision) (flutter/flutter#187056) 2026-05-25 engine-flutter-autoroll@skia.org Roll Skia from 4dd78179e6ec to 9d1adb5f2427 (1 revision) (flutter/flutter#187048) 2026-05-25 engine-flutter-autoroll@skia.org Roll Skia from 1f26101197bf to 4dd78179e6ec (4 revisions) (flutter/flutter#187044) 2026-05-24 engine-flutter-autoroll@skia.org Roll Skia from bbe9ccc2bdbf to 1f26101197bf (1 revision) (flutter/flutter#187016) 2026-05-24 engine-flutter-autoroll@skia.org Roll Fuchsia Linux SDK from nsgcNDlZOuweOvy3Q... to Itd2Jq_ZIABH2rW7B... (flutter/flutter#187032) 2026-05-23 engine-flutter-autoroll@skia.org Roll Dart SDK from 7e0f28eb5315 to 7eb54169841d (1 revision) (flutter/flutter#187005) 2026-05-23 engine-flutter-autoroll@skia.org Roll Dart SDK from 90e55fa88456 to 7e0f28eb5315 (1 revision) (flutter/flutter#186990) 2026-05-23 engine-flutter-autoroll@skia.org Roll Fuchsia Linux SDK from 6T6BY9PTftoG3vP_1... to nsgcNDlZOuweOvy3Q... (flutter/flutter#186984) 2026-05-23 chris@bracken.jp iOS] Migrate VSyncClient to a pure Obj-C implementation (#186166) (flutter/flutter#186935) 2026-05-23 30870216+gaaclarke@users.noreply.github.com Disables embedder_tests.cm for fuchsia (flutter/flutter#186969) 2026-05-23 engine-flutter-autoroll@skia.org Roll Dart SDK from b8414c46f6c7 to 90e55fa88456 (2 revisions) (flutter/flutter#186977) 2026-05-22 engine-flutter-autoroll@skia.org Roll Skia from 6fdb013d1953 to bbe9ccc2bdbf (1 revision) (flutter/flutter#186980) 2026-05-22 mdebbar@google.com [web] Fix cutoff text in WebParagraph (flutter/flutter#186819) 2026-05-22 1961493+harryterkelsen@users.noreply.github.com fix(web): Removes the iterative downscaling hack (flutter/flutter#186914) 2026-05-22 30870216+gaaclarke@users.noreply.github.com opts the linux embedder into sdf rendering (flutter/flutter#186909) 2026-05-22 engine-flutter-autoroll@skia.org Roll Skia from dae8778ca40d to 6fdb013d1953 (5 revisions) (flutter/flutter#186970) 2026-05-22 dacoharkes@google.com Fix hooks inputs outputs rebuilt (flutter/flutter#186701) 2026-05-22 30870216+gaaclarke@users.noreply.github.com adds linux impeller integration test for external textures (flutter/flutter#186759) 2026-05-22 kevmoo@users.noreply.github.com fix(flutter_tools): defensively catch DWDS unregistered service extension errors (flutter/flutter#186896) 2026-05-22 bdero@google.com [Impeller] Add golden harness support to the renderer test layer (flutter/flutter#186735) 2026-05-22 mdebbar@google.com [web] Remove image codecs from canvaskit_chromium (flutter/flutter#178133) 2026-05-22 30870216+gaaclarke@users.noreply.github.com opts all macos into wide gamut (flutter/flutter#186277) 2026-05-22 engine-flutter-autoroll@skia.org Roll Skia from 356185490a75 to dae8778ca40d (9 revisions) (flutter/flutter#186949) 2026-05-22 1598289+lukemmtt@users.noreply.github.com Filter out SwiftPM schemes when fetching schemes (flutter/flutter#186006) 2026-05-22 engine-flutter-autoroll@skia.org Roll Packages from 3754d04 to 69cf959 (1 revision) (flutter/flutter#186950) 2026-05-22 engine-flutter-autoroll@skia.org Roll Dart SDK from eca46bec956d to b8414c46f6c7 (2 revisions) (flutter/flutter#186944) 2026-05-22 30870216+gaaclarke@users.noreply.github.com Saves a DeviceHolderVK with the CommandPoolVK (flutter/flutter#186749) 2026-05-22 engine-flutter-autoroll@skia.org Roll Dart SDK from e0d509fd676e to eca46bec956d (1 revision) (flutter/flutter#186922) 2026-05-22 bkonyi@google.com [ Tool ] Stop generating widget preview scaffold under $TMP (flutter/flutter#186476) 2026-05-21 737941+loic-sharma@users.noreply.github.com Fix typo in StretchingOverscrollIndicator docs (flutter/flutter#186897) 2026-05-21 engine-flutter-autoroll@skia.org Roll Dart SDK from 28c7cb5a8e8d to e0d509fd676e (1 revision) (flutter/flutter#186903) 2026-05-21 jason-simmons@users.noreply.github.com Fix some issues in the integration between EmbedderExternalViewEmbedder and Impeller (flutter/flutter#184905) 2026-05-21 jason-simmons@users.noreply.github.com Fix a potential buffer overflow in the animated PNG decoder when parsing malformed fdAT chunks (flutter/flutter#186700) 2026-05-21 engine-flutter-autoroll@skia.org Roll Skia from 2ff20950975d to 356185490a75 (5 revisions) (flutter/flutter#186892) 2026-05-21 flar@google.com Add primitive shadows to rendering benchmark (flutter/flutter#186779) 2026-05-21 15619084+vashworth@users.noreply.github.com Move prefetchSwiftPackages to be per platform (flutter/flutter#186468) 2026-05-21 okorohelijah@google.com Upgrade iOS version (flutter/flutter#186889) ...
Migrates
FlutterVSyncClientfrom a hybrid C++/Obj-C implementation to a pure Objective-C implementation by removing C++ types from its interface and implementation.Previously,
FlutterVSyncClientstored a C++flutter::VsyncWaiter::Callbackand used a temporarystd::unique_ptr<flutter::FrameTimingsRecorder>to propagate vsync timestamps. This was redundant:FlutterVSyncClientallocated a temporaryFrameTimingsRecorderand recorded vsync times.VsyncWaiterIOSconverted the seconds back tofml::TimePoint.VsyncWaiter::FireCallbackallocated anotherFrameTimingsRecorderand recorded the same times again.We now:
CFTimeInterval(double seconds) directly fromCADisplayLink.FrameTimingsRecorderand C++ lambda wrapper.VsyncWaiter::FireCallbackto handle the finalFrameTimingsRecordercreation and recording, which it already does.This is a re-land of #186166 which was reverted in #186266. That patch failed to correctly adjust the epoch of timestamps obtained from CoreAnimation (which uses
mach_absolute_timeunder the hood) to the epoch used by the core engine, which is built onfml::TimePoint, which usesstd::chrono::steady_clock, which usesmach_continuous_timeunder the hood, consistent with the C++ specification's of a continuous clock. While these two clocks start roughly synchronized at boot,mach_absolute_timepauses on device sleep, whereasmach_continuous_timedoes not.The iOS embedder now uses timings based on CoreAnimation throughout and translates to
fml::TimePointonly at the engine-embedder boundary. This is consistent with embedders based on the embedder API, such as macOS, which does this work in-[FlutterEngine viewControllerViewDidLoad]usingFlutterTimeConverter.Added tests for frame rate snapping/fallback. No other test changes because this introduces no semantic change and is covered by existing tests recently added in: #186457
Issue: #112232
Pre-launch Checklist
///).If you need help, consider asking for advice on the #hackers-new channel on Discord.
If this change needs to override an active code freeze, provide a comment explaining why. The code freeze workflow can be overridden by code reviewers. See pinned issues for any active code freezes with guidance.
Note: The Flutter team is currently trialing the use of Gemini Code Assist for GitHub. Comments from the
gemini-code-assistbot should not be taken as authoritative feedback from the Flutter team. If you find its comments useful you can update your code accordingly, but if you are unsure or disagree with the feedback, please feel free to wait for a Flutter team member's review for guidance on which automated comments should be addressed.