| RFC-0166: One UI Stack | |
|---|---|
| Status | Accepted |
| Areas |
|
| Description | Clarifies plans and policy to move all Fuchsia Products to a single graphics and input stack. |
| Issues | |
| Gerrit change | |
| Authors | |
| Reviewers | |
| Date submitted (year-month-day) | 2022-05-04 |
| Date reviewed (year-month-day) | 2022-06-03 |
Summary
This RFC clarifies the plan of record with respect to legacy components and APIs used in the graphics, input and accessibility stacks (also called the "UI stack") in the Fuchsia platform. It outlines the planned order of migrations to get all Fuchsia product configurations using the same stack for graphics, input and accessibility. It also clarifies testing standards around these transitions. Specifically:
- The Root Presenter component is deprecated in favor of Scene Manager.
- The Input Pipeline library will run as part of Scene Manager on products with graphics.
- All contracts between UI components should be covered by platform integration tests to ensure consistent behavior across transitions.
- Scenic's GFX API is deprecated in favor of Flatland.
Motivation
The Fuchsia Platform currently contains multiple ways of
- Processing user input
- Creating a graphical scene
- Wiring up accessibility services
Different Fuchsia product configurations currently use different subsets of this functionality. This leads to confusion and difficulty debugging across different products.
This RFC seeks to clarify which of the components and APIs involved are deprecated, which are plan of record, and to ratify the planned order of the migrations. It provides additional context on the architecture discussed in RFC-0096: User input architecture and RFC-0147: View System.
Stakeholders
Who has a stake in whether this RFC is accepted? (This section is optional but encouraged.)
Facilitator:
cpu@google.com
Reviewers:
- Scenic, runtime integrations: emircan@google.com, jjosh@google.com, dworsham@google.com
- Accessibility: lucasradaelli@google.com
- Input: quiche@google.com, jaeheon@google.com
Consulted:
yeg@google.com
Socialization:
A detailed version of this plan was reviewed internally with Scenic, Input and Accessibility teams.
Glossary
- Scenic - Fuchsia platform component that handles graphical composition, focus management, and routing of graphical input. fuchsia.ui.scenic and fuchsia.ui.gfx.
- Flatland - Plan of record Scenic API. Found in fuchsia.ui.composition.
- Input Pipeline - Fuchsia library that handles routing and policy for input events. May run as part of Scene Manager or as a standalone component.
- Scene Manager - Fuchsia platform component that handles the setup and configuration of the Scenic scene, as well as starting the input pipeline. Currently missing some functionality, especially with respect to accessibility.
- Root Presenter - Legacy Fuchsia platform component that handles input routing and setup and configuration of the Scenic scene.
- Accessibility Manager - Fuchsia platform component that handles accessibility services.
Design
Legacy vs. desired state
In the legacy UI stack, Root Presenter is responsible for interpreting input events from the driver stack and for instantiating the scene graph using GFX. Root Presenter dispatches input events to Scenic, which sometimes forwards them to Accessibility Manager. Root Presenter sends commands to the Accessibility Manager to control magnification and color correction. Flutter and Chromium create views using GFX.
In the desired/future UI stack, Scene Manager is responsible for interpreting input events from the driver stack and for instantiating the scene graph using Flatland. The Input Pipeline is instantiated as part of the Scene Manager component and dispatches input events to Scenic. Scenic sometimes forwards input events to the Accessibility Manager. Scene Manager sends commands to the Accessibility Manager to control magnfication and color correction. Flutter and Chromium create views using Flatland. The Virtual Keyboard Controller is a separate component that is only responsible for passing messages about onscreen keyboard state.
What is changing
Deprecate Root Presenter
Scene management and input functionality will be removed from Root Presenter, which will be renamed to "Virtual Keyboard Manager" to reflect the remaining functionality. This removes a large amount of mostly-untested legacy code.
Any Root Presenter features not currently supported in Scene Manager (e.g. Accessibility Magnification) will be implemented in Scene Manager.
Integration tests
All features of the UI stack (graphics, different forms of user input, accessibility) will now require hermetic integration tests to ensure that the contracts between platform components remain stable.
GFX to Flatland
All references to and implementations of the GFX API will be removed from fuchsia.git. GFX is a legacy 3D API that is significantly more complex than is warranted by the use cases Fuchsia serves today. Removing it simplifies the system and removes a non-trivial chunk of poorly-understood API surface.
This also allows runtime implementations (Flutter, Chromium) to remove GFX support from their codebases.
Order of deprecation
To move all products to the same desired configuration, we must make changes to both the component topology (replacing Root Presenter with Scene Manager) as well as which API is used (GFX vs. Flatland). In order to provide solid test coverage for the graphics API transition, we will first align all products on the correct component topology. Once this is complete, we transition any products remaining on GFX to Flatland.
Implementation
In all phases migrations should be controlled with build flags to allow for easy manual testing and rollbacks.
Milestone 0 (completed): Flatland on Workstation only
The workstation configuration moved to use Flatland for Graphics in Q1 2022.
Milestone 1 (completed): Other products move to Input Pipeline
As of Q2 2022 all other Fuchsia products moved to use the Input Pipeline for input event routing (instead of Root Presenter). In products without Scene Manager it temporarily runs as a standalone component.
Milestone 2: Transition all products to Scene Manager
Currently Scene Manager lacks a number of features present in Root Presenter. The work to bring Scene Manager to feature parity with Root Presenter is tracked in https://fxbug.dev/42181047. Once this is completed and all features are covered by platform integration tests we will transition any remaining products to use Scene Manager + GFX.
After this transition, input and scene management code in Root Presenter will be removed. Root Presenter's virtual keyboard controller functionality will remain and the component will be renamed to reflect its reduced set of responsibilities.
Milestone 3: Transition remaining products to Flatland
There is additional work in Flatland to cover all features supported in GFX. This work is tracked in https://fxbug.dev/42175822. Once this work is completed and fully covered by platform integration tests we will transition remaining products to Flatland.
After this is landed and stable, the API and implementation for GFX may be removed. Note that this includes code in the Scenic component as well as integration code in multiple UI frameworks.
Performance
Milestone 3 could lead to performance improvements for some product configurations due to efficiencies in Flatland. This is expected when Flatland is able to delegate compositing to the display controller (i.e. without involving the GPU), which will be available on some device types. Performance should be monitored through each transition but is not expected to worsen.
Security and Privacy considerations
This RFC does not introduce any new security or privacy concerns not already covered by Flatland API reviews. No user data will be collected.
Testing
At each stage of the migration, we will gate transitions on full integration test coverage of the affected features. Because Root Presenter has historically lacked such coverage this should lead to an overall improvement in coverage for the UI stack.
Documentation
References to Root Presenter and GFX should be removed from documentation as milestones are completed.
Drawbacks, alternatives, and unknowns
This is a complex, multi-quarter migration which will take a substantial engineering effort to complete.
Do nothing
We could continue to maintain the legacy functionality for the lifespan of all products that use it. Historically this maintenance has been substantial due to continuous changes in the rest of the platform as well as a lack of comprehensive tests. Multiple UI stacks are also confusing, leading to uncertainty about what code was actually running on a given product. While the migration is costly, maintaining the status quo will require more effort over time with worse results and depressed engineering velocity.
Transition to Flatland first
This approach would require significant work to add Flatland support to Root Presenter. Because this code lacks comprehensive test coverage, this is risky and requires a large amount of throw-away work.