Currently, `GfxViewportElement` hides DOM blocks outside the viewport using `display: none` to optimize performance. However, this approach presents two issues:
1. Even when hidden, all top-level blocks still undergo frequent CSS transform updates during viewport panning and zooming.
2. Hidden blocks cannot access DOM layout information, preventing `TurboRenderer` from updating the complete canvas bitmap.
To address this, this PR introduces a refactoring that divides all top-level edgeless blocks into two states: `idle` and `active`. The improvements are as follows:
1. Blocks outside the viewport are set to the `idle` state, meaning they no longer update their DOM during viewport panning or zooming. Only `active` blocks within the viewport are updated frame by frame.
2. For `idle` blocks, the hiding method switches from `display: none` to `visibility: hidden`, ensuring their layout information remains accessible to `TurboRenderer`.
[Screen Recording 2025-03-07 at 3.23.56 PM.mov <span class="graphite__hidden">(uploaded via Graphite)</span> <img class="graphite__hidden" src="https://app.graphite.dev/api/v1/graphite/video/thumbnail/lEGcysB4lFTEbCwZ8jMv/4bac640b-f5b6-4b0b-904d-5899f96cf375.mov" />](https://app.graphite.dev/media/video/lEGcysB4lFTEbCwZ8jMv/4bac640b-f5b6-4b0b-904d-5899f96cf375.mov)
While this minimizes DOM updates, it introduces a trade-off: `idle` blocks retain an outdated layout state. Since their positions are updated using a lazy update strategy, their layout state remains frozen at the moment they were last moved out of the viewport:
To resolve this, the PR serializes and stores the viewport field of the block at that moment on the `idle` block itself. This allows the correct layout, positioned in the model coordinate system, to be restored from the stored data.
The primary purpose of this PR appears to be:
1. Simplifying the change notification API by removing the redundant value parameter from callbacks
2. Improving the reactive system's handling of specialized types (Text and Boxed) in flat data
3. Adding better test coverage for text handling in the flat data model
Close [BS-1869](https://linear.app/affine-design/issue/BS-1869/[bug]-android-chrome-%E8%BE%93%E5%85%A5%E9%94%99%E8%AF%AF)
## Problem
On Android devices, keyboard events do not properly capture key information, causing the backspace key and other keyboard functionalities to malfunction. This is due to the specific behavior of Android platform, as discussed in:
- https://stackoverflow.com/a/68188679
- https://stackoverflow.com/a/66724830
## Solution
1. Added special handling for Android platform in `KeyboardControl` class by using `beforeInput` event instead of `keyDown` event
2. Implemented `androidBindKeymapPatch` function to handle special key events on Android platform
3. Updated event handling logic in related components, including:
- CodeBlock
- ListKeymap
- ParagraphKeymap
- PageKeyboardManager
## Changes
- Added `androidBindKeymapPatch` function for handling key events on Android platform
- Modified `KeyboardControl.bindHotkey` method to add `beforeInput` event handling for Android
- Unified event object access using `ctx.get('defaultState').event` instead of `keyboardState.raw`
- Updated key event handling logic in multiple components
## Before
https://github.com/user-attachments/assets/e8602de4-d584-4adf-816f-369f38312022
## After
https://github.com/user-attachments/assets/f9e1680e-28ff-4d52-bdab-7683cdcb6f82
The new code should be more efficient as it:
- Avoids unnecessary iterations when objects aren't empty
- Has clearer path management
- Reduces redundant object traversals
The main changes in this PR involve replacing the deprecated `BlockServiceWatcher` with the new `LifeCycleWatcher` across multiple files. Here's a detailed breakdown:
1. **Core Architectural Change:**
- Removed `BlockServiceWatcher` class completely (deleted file)
- Migrated to `LifeCycleWatcher` as the new standard for watching component lifecycle events
2. **Key Changes in Implementation:**
- Changed from using `blockService.specSlots` events to using `view.viewUpdated` events
- Replaced `flavour` static property with `key` static property
- Updated event handling to use more specific payload type checking
3. **Major File Changes:**
- Modified multiple block components:
- Embed synced doc block
- Frame preview
- Edgeless root spec
- AI-related components (code, image, paragraph, etc.)
- Quick search service
- Edgeless clipboard
4. **Pattern of Changes:**
The migration follows a consistent pattern:
```typescript
// Old pattern
class SomeWatcher extends BlockServiceWatcher {
static override readonly flavour = 'some:flavour';
mounted() {
this.blockService.specSlots.viewConnected.on(...)
}
}
// New pattern
class SomeWatcher extends LifeCycleWatcher {
static override key = 'some-watcher';
mounted() {
const { view } = this.std;
view.viewUpdated.on(payload => {
if (payload.type !== 'block' || payload.method !== 'add') return;
// Handle event
});
}
}
```
5. **Benefits:**
- More explicit and type-safe event handling
- Cleaner architecture by removing deprecated code
- More consistent approach to lifecycle management
- Better separation of concerns
This appears to be a significant architectural improvement that modernizes the codebase by removing deprecated patterns and standardizing on a more robust lifecycle management system.
1. **Major Architectural Change: Schema Management**
- Moved from `workspace.schema` to `store.schema` throughout the codebase
- Removed schema property from Workspace and Doc interfaces
- Added `BlockSchemaExtension` pattern across multiple block types
2. **Block Schema Extensions Added**
- Added new `BlockSchemaExtension` to numerous block types including:
- DataView, Surface, Attachment, Bookmark, Code
- Database, Divider, EdgelessText, Embed blocks (Figma, Github, HTML, etc.)
- Frame, Image, Latex, List, Note, Paragraph
- Root, Surface Reference, Table blocks
3. **Import/Export System Updates**
- Updated import functions to accept `schema` parameter:
- `importHTMLToDoc`
- `importHTMLZip`
- `importMarkdownToDoc`
- `importMarkdownZip`
- `importNotionZip`
- Modified export functions to use new schema pattern
4. **Test Infrastructure Updates**
- Updated test files to use new schema extensions
- Modified test document creation to include schema extensions
- Removed direct schema registration in favor of extensions
5. **Service Layer Changes**
- Updated various services to use `getAFFiNEWorkspaceSchema()`
- Modified transformer initialization to use document schema
- Updated collection initialization patterns
6. **Version Management**
- Removed version-related properties and methods from:
- `WorkspaceMetaImpl`
- `TestMeta`
- `DocImpl`
- Removed `blockVersions` and `workspaceVersion/pageVersion`
7. **Store and Extension Updates**
- Added new store extensions and adapters
- Updated store initialization patterns
- Added new schema-related functionality in store extension
This PR represents a significant architectural shift in how schemas are managed, moving from a workspace-centric to a store-centric approach, while introducing a more extensible block schema system through `BlockSchemaExtension`. The changes touch multiple layers of the application including core functionality, services, testing infrastructure, and import/export capabilities.
