bug(material/dialog): exit animation "jumps" (shrinks, then snaps back) when close() is called inside a requestAnimationFrame callback — common in zoneless apps
維護者通常 1 天內回覆
還沒有人認領這個 Issue。
評估
- 難度
- 3/5
- 預估耗時
- 1-2 天
- 新手友好度
- 67/100
- Issue 類型
- 缺陷
- 描述清晰度
- 描述清楚
- 活躍度
- 活躍
- 技術堆疊
- angular, typescript
- 領域
- frontend
研究方向
Start with MatDialogContainer._startExitAnimation() in the dialog container source and trace how it changes the open and closing classes. Reproduce the issue with the linked StackBlitz by closing from a requestAnimationFrame callback, then check the dialog animation tests for existing exit behavior. Done means the exit transition moves directly from open to closing without a frame in the class-less state.
由索引模型根據 Issue 內容生成。
描述
Is this a regression?
- Yes, this behavior used to work in the previous version
The previous version in which this bug was not present was
No response
Description
MatDialogContainer._startExitAnimation() removes mdc-dialog--open synchronously, but adds mdc-dialog--closing in a requestAnimationFrame callback:
this._hostElement.classList.remove(OPEN_CLASS);
if (this._animationsEnabled) {
this._hostElement.style.setProperty(TRANSITION_DURATION_PROPERTY, `${this._exitAnimationDuration}ms`);
this._requestAnimationFrame(() => this._hostElement.classList.add(CLOSING_CLASS));
...
}
If close() is called from a normal task (click handler, setTimeout, XHR callback), the rAF runs before the next paint, so the dialog goes straight from --open to --closing.
If close() is called from inside a rAF callback, the nested rAF is deferred to the next frame. One frame is painted with neither class. In that state the base styles apply: .mat-mdc-dialog-surface { transform: scale(0.8) } and .mat-mdc-dialog-inner-container { opacity: 0 }, both with a transition. The surface starts shrinking, then --closing arrives (transform: none) and it snaps back to full size before fading out.
In a zoneless app this happens without any explicit rAF in user code. The zoneless scheduler runs change detection via scheduleCallbackWithRafRace (setTimeout vs rAF, whichever fires first). Effects run during that tick (runEffect → runEffectsInView → ApplicationRef.tick → ChangeDetectionSchedulerImpl.tick), so a dialog that closes itself from an effect() is closed inside a rAF whenever the rAF wins the race. This is a common pattern when the close is driven by state, e.g. "close when the save request succeeded".
Because it depends on the race, it is intermittent. Measured on the reproduction below, by hooking classList.add/remove and recording the rAF frame timestamp of each change:
How close() is called |
Class-less frame painted | Result |
|---|---|---|
Inside a requestAnimationFrame callback |
10 / 10 | ❌ always jumps |
From an effect() (zoneless, real mouse clicks) |
2 / 4 | ⚠️ intermittent |
| Directly in the click handler | 0 / 10 | ✅ clean fade-out |
Example log for the rAF case: remove(mdc-dialog--open) happens in frame 92399.7, and add(mdc-dialog--closing) happens in frame 92424.8. In a real app, closing from an effect() after an HTTP response showed the glitch in 2 out of 6 attempts.
Not a regression as far as we can tell. The nested requestAnimationFrame was added in #26971 (v16), but it only becomes visible when close() runs inside a rAF callback, which zoneless change detection makes common.
Reproduction
StackBlitz link: https://stackblitz.com/edit/mat-dialog-close-in-raf?file=src%2Fmain.ts
The dialog component from the StackBlitz (the app is bootstrapped with provideZonelessChangeDetection()):
@Component({
imports: [MatButton, MatDialogActions, MatDialogContent, MatDialogTitle],
template: `
<h2 mat-dialog-title>Example dialog</h2>
<mat-dialog-content>Pick a way to close this dialog.</mat-dialog-content>
<mat-dialog-actions>
<button mat-button (click)="closeInRaf()">Close inside rAF</button>
<button mat-button (click)="closeRequested.set(true)">Close via effect</button>
<button mat-button (click)="dialogRef.close()">Close directly</button>
</mat-dialog-actions>
`,
})
export class ExampleDialog {
protected dialogRef = inject(MatDialogRef<ExampleDialog>);
protected closeRequested = signal(false);
constructor() {
effect(() => {
if (this.closeRequested()) {
this.dialogRef.close();
}
});
}
closeInRaf(): void {
requestAnimationFrame(() => this.dialogRef.close());
}
}
Steps to reproduce:
- Click "Open dialog", then "Close inside rAF". The surface visibly shrinks, jumps back to full size, then fades out. Every time.
- Click "Open dialog", then "Close via effect". Repeat several times. Some closes show the same jump.
- Click "Open dialog", then "Close directly". The dialog always fades out cleanly.
Slowing animations down in the DevTools Animations panel makes the jump easy to see.
Expected Behavior
The dialog transitions directly from the open state to the closing state, no matter which task or callback close() is called from.
Actual Behavior
When close() runs inside a rAF callback, one frame is rendered without mdc-dialog--open or mdc-dialog--closing. The surface starts animating toward scale(0.8) and then snaps back to transform: none when --closing is applied.
Related issues
- angular/angular#57528 (canonical: consequences of macrotask scheduling). This is a related consequence of zoneless scheduling, but the intermediate state here is created by
MatDialogContaineritself:--openis removed in one frame and--closingis added in the next. Angular's scheduler only determines whereclose()runs. The fix can stay local to the dialog container and needs no change to the scheduler. - Same class of zoneless timing bugs in other components: #29174 (virtual scroll flicker, fixed in #31316 by keeping the DOM write and the render in the same tick), #33837 (drag-drop blink), #33851 (select exit fade).
Possible fix
Either of these removes the class-less frame:
- Add
CLOSING_CLASSsynchronously in_startExitAnimation(). The rAF from #26971 is needed for the opening animation, because a freshly attached element has no previous computed style to transition from. On close, the element has already been rendered in the open state, so a synchronous class change starts the transition correctly. The duration custom property set in the same task is picked up as well, because transitions use the after-change style. - Keep the rAF, but move the
OPEN_CLASSremoval into the same callback:this._requestAnimationFrame(() => { this._hostElement.classList.remove(OPEN_CLASS); this._hostElement.classList.add(CLOSING_CLASS); });
Environment
- Angular: 22.2.4 (zoneless)
- CDK/Material: 22.2.6. The same code is present on main.
- Browser(s): Chromium-based
- Operating System: macOS
- 主要語言
- TypeScript
- 星號
- 25k
- 分支
- 6.9k
- 平均合併
- 2 天 7 小時
- 30 天內合併 PR
- 83
環境準備
- 沒有 Dockerfile 或 Docker Compose 檔案
- 沒有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
angular/components 的其他 Issue
-
bug(mat-slider): on Firefox, slider thumb is not aligned with cursor可能已有人在做 @Devyalamaddi 於 4 天前認領。 未關閉area: material/slider P4
難度 2/5 1-3 小時 新手友好度 62/100
angular/components#34094 · 1 則留言 ·
維護者通常 1 天內回覆
-
bug(material/sort): hover hint arrow on an unsorted header shows the active sort direction, not the resulting direction after click.可能已有人在做 @abdul-karim-mia 於 2 天前認領。 未關閉area: material/sort P3
難度 2/5 1-3 小時 新手友好度 70/100
angular/components#33933 ·
維護者通常 1 天內回覆
-
bug(mat-date-picker): month shown twice in date picker可能已有人在做 @PreethiPantangi 於 54 天前認領。 未關閉area: material/datepicker P4
難度 2/5 1-3 小時 新手友好度 72/100
angular/components#33763 · 2 則留言 · 3 個 reaction ·
維護者通常 1 天內回覆
-
Bug in table.scss footer cell height: unset instead of footer row height unset可能已有人在做 @minwookshin 於 41 天前認領。 未關閉area: material/table P4
難度 1/5 1 小時以內 新手友好度 88/100
angular/components#33709 · 1 則留言 ·
維護者通常 1 天內回覆
-
bug(date-adapter): Misleading JSDoc for createDate returns type可能已有人在做 @BHUVANSH855 於 171 天前認領。 未關閉area: material/core P3
難度 1/5 1 小時以內 新手友好度 78/100
angular/components#33059 · 2 則留言 ·
維護者通常 1 天內回覆
查看 angular/components 的全部 Issue
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 78/100
vercel-labs/ui#540 ·
維護者通常 1 天內回覆
-
accepted ai-dev ai-implementing
難度 2/5 1-3 小時 新手友好度 74/100
維護者通常 1 天內回覆
-
normalizePropName strips `#` characters from Figma property names instead of only the `#id` suffix未關閉
難度 2/5 1-3 小時 新手友好度 72/100
figma/code-connect#419 ·
-
難度 2/5 1-3 小時 新手友好度 78/100
-
Negative values skip k/M abbreviation and medium decimal format on line, stacked-area, and tooltip未關閉
難度 2/5 1-3 小時 新手友好度 84/100
britecharts/britecharts#1255 ·
維護者通常 1 天內回覆