Support null @rendermode

[halter73]

          > It looks we might need to change the runtime to provide a way to set the render mode to static server rendering.

Yes, I think you're right, and this is going to affect dotnet/aspnetcore#50920 (Auth with global interactivity) because it's exactly how we're most likely going to do it. cc @halter73

The reason it worked in my PR is probably that it didn't yet have the Razor compiler changes for @rendermode and hence was behaving as a regular attribute that gets elided if you pass null. We can change AddComponentRenderMode in the runtime to treat null as a no-op.

Originally posted by @SteveSandersonMS in dotnet/aspnetcore#51031 (comment)

[halter73]

@SteveSandersonMS @danroth27 @chsienki @jaredpar

I think we're going to need to update the razor compiler to allow a null @rendermode for .NET 8 GA if we're sometimes going to be using null in our .NET 8 project templates. Otherwise, users will see errors like the following warning when building a new project template unless we sprinkle in a deceitful !:

Components\App.razor(13,106): warning CS8600: Converting null literal or possible null value to non-nullable type.
Components\App.razor(17,105): warning CS8600: Converting null literal or possible null value to non-nullable type.

[chsienki]

Is that not just the underlying API needs the annotation applied to it to allow a nullable type?

[SteveSandersonMS]

Is that not just the underlying API needs the annotation applied to it to allow a nullable type?

Yeah I would have thought so. Not sure why a compiler change would be required, but if @halter73 knows it is, please clarify!

[halter73]

I've updated RenderTreeBuilder.AddComponentRenderMode(IComponentRenderMode renderMode) to RenderTreeBuilder.AddComponentRenderMode(IComponentRenderMode? renderMode) in the template update branch, but this does not silence the warnings from App.razor's <Routes @rendermode="null" /> (effectively).

I'm not 100% sure why this still leads to CS8600 warning, but could it because the code generator is still attempting to treat the IComponentRenderMode as non-nullable? For example:

razor/src/Compiler/Microsoft.AspNetCore.Razor.Language/src/Components/ComponentRenderModeDirectivePass.cs

Line 84 in 4258f88

Content = $"public override global::{ComponentsApi.IComponentRenderMode.FullTypeName} Mode => ModeImpl;"

Or does @rendermode's parameter type flow directly from the public runtime AddComponentRenderMode API? That seems unlikely, and is not what I'm seeing locally.

[chsienki]

@halter73 Ah you're right, its because we do the double assign to avoid side effects that this crops up. We can either make this nullable too (seems correct, if the underlying API is nullable) or not emit in the case of null.

[jaredpar]

Do we need to be concerned about ComponentsApi.IComponentRenderMode.FullTypeName representing say a struct type? Basically do we need have a bit of smarts or can we just append ? to the type name.

[halter73]

I'm pretty sure you can rely on it being a class, but I'll defer to @SteveSandersonMS.

[SteveSandersonMS]

Just appending ? to the type name seems fine. IComponentRenderMode itself will always be an interface, and the only implementations that the runtime cares about are classes.