Make sure WebRootComponentManager is initialized for dynamic JS roots

[ilonatommy]

Fixes #64159.

Source of the issue:

_webRootComponentManager is not defined in

aspnetcore/src/Components/Server/src/Circuits/RemoteRenderer.cs

Line 324 in 10ef000

return remoteComponentState.ParentComponentState != null ?

How it typically works :

_webRootComponentManager gets created when CircuitHost goes through

aspnetcore/src/Components/Server/src/Circuits/CircuitHost.cs

Line 882 in 10ef000

var webRootComponentManager = Renderer.GetOrCreateWebRootComponentManager();

That is the normal flow for prerendered/SSR roots: the client sends an UpdateRootComponents message, CircuitHost materialises the operations, the manager is created, and every root component added in that batch is registered inside the manager’s dictionary.

How it works with dynamic JS root:

• The root cames from JavaScript via Blazor.rootComponents.add(...). That call hits CircuitJSComponentInterop.AddRootComponent, which simply calls base.AddRootComponent(...) on the renderer:
  

  aspnetcore/src/Components/Web.JS/src/Rendering/JSRootComponents.ts

  Line 27 in 6819df1

  const componentId = await getRequiredManager().invokeMethodAsync<number>('AddRootComponent', componentIdentifier, containerIdentifier);

  
  

  aspnetcore/src/Components/Server/src/Circuits/CircuitJSComponentInterop.cs

  Line 26 in 10ef000

  var id = base.AddRootComponent(identifier, domElementSelector);

  
  (getRequiredManager is not the manager we are interested in, it's only a set of interop methods available for a given component, see

  aspnetcore/src/Components/Web.JS/src/Rendering/WebRendererInteropMethods.ts

  Line 78 in 6819df1

  function getInteropMethods(rendererId: number): DotNet.DotNetObject {

  )
• This bypasses the whole CircuitHost.PerformRootComponentOperations pipeline, so _webRootComponentManager never gets initialised and no entry is recorded for the new component.
• When the renderer later constructs RemoteComponentState for that JS-added root, RemoteComponentState.GetComponentKey() invokes RemoteRenderer.GetMarkerKey(...). Because the manager had never been created on this code path, _webRootComponentManager was still null and the null-forgiving access ! resulted in the NullReferenceException.

Fix:

• Make sure the manager is initialized.

[Copilot]

Pull Request Overview

This PR fixes a bug where dynamically added JavaScript root components with persistent state were causing circuit errors because they were entering the GetMarkerKey method without the _webRootComponentManager being initialized. The fix ensures the manager is created if it doesn't exist when getting the marker key.

Key changes:

• Modified GetMarkerKey method in both RemoteRenderer.cs and WebAssemblyRenderer.cs to lazily initialize _webRootComponentManager if null
• Added test coverage with a new E2E test to verify JS-added persistent state root components work correctly

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/Components/Server/src/Circuits/RemoteRenderer.cs	Fixed GetMarkerKey to call GetOrCreateWebRootComponentManager() when manager is null
src/Components/WebAssembly/WebAssembly/src/Rendering/WebAssemblyRenderer.cs	Fixed GetMarkerKey to call GetOrCreateWebRootComponentManager() when manager is null
src/Components/test/E2ETest/Tests/StatePersistenceTest.cs	Added new test case for JS-added persistent state root components
src/Components/test/testassets/BasicTestApp/wwwroot/persistent-state-js-root-component.html	Added test HTML page for JS root component scenario
src/Components/test/testassets/Components.TestServer/RazorComponentEndpointsStartup.cs	Registered CounterWithPersistentState component for JavaScript testing
src/Components/test/testassets/Components.TestServer/RazorComponents/Pages/PersistentState/CounterWithPersistentState.razor	Added test component with persistent state for testing

[ilonatommy]

The tests:
Circuit_CanShutDownAndReInitializeMultipleTimes, DotNetObjectReference_CannotBeUsed_AfterCircuitShutsDown_AndANewCircuitIsInitialized, StateIsProvidedEveryTimeACircuitGetsCreated fail not because of the framework changes but because of test project changes to RazorComponentEndpointsStartup.cs:

 + options.RootComponents.RegisterForJavaScript<TestContentPackage.PersistentComponents.ComponentWithPersistentState>("dynamic-js-root-counter");

The failure happens on reinitialization after shutdown. The reinitialization process doesn't properly handle the case where JS components with persistent state are registered but not yet added to the circuit. The circuit seems to get "stuck" during reinitialization - it connects (WebSocket shows connected):

[2025-11-24T12:43:08Z] [Info] http://127.0.0.1:51676/subdir/_framework/blazor.web.crlaspqfcs.js 0:59500 "[2025-11-24T12:43:08.420Z] Information: WebSocket connected to ws://127.0.0.1:51676/subdir/_blazor?id=shfXstthdREtTim5LfjW8g."

the HTML shows the component markup, but the components never become interactive (<span id="is-interactive-1">False</span> never changes to True). This suggests the circuit initialization hangs or fails silently somewhere.. (?)

The question is: how adding RegisterForJavaScript (calling enableJSRootComponents on client etc) influences the way we detect interactivity?

[ilonatommy]

The question is: how adding RegisterForJavaScript (calling enableJSRootComponents on client etc) influences the way we detect interactivity?

Calling enableJSRootComponents throws when called more than once which breaks the flow:

aspnetcore/src/Components/Web.JS/src/Rendering/JSRootComponents.ts

Line 126 in da2e7e1

throw new Error('Dynamic root components have already been enabled.');

@javiercn, I'm planning to remove that exception and replace it with no-op.

Edit:
the comment above the exception "This will only happen in very nonstandard cases where someone has multiple hosts." makes me think no-op is not a good solution. I will file a separate issue (#64523) and make a workaround for tests to have this PR fix the original problem only.

[javiercn]

Looks great!