Default Http3Support to false for PublishTrimmed

[MichalStrehovsky]

EDIT: per discussion below, disabling for PublishTrimmed in general, not just PublishAot.

Since HTTP3 doesn't work with native AOT (or single file for that matter) without extra gestures outside Linux (#73290), we should probably require an extra gesture to include the code too. The extra code is sizeable - I've seen 15% savings from turning Http3Support off.

There are multiple ways to go about this:

1. Do nothing (close this PR)
2. Disable Http3Support on native AOT outside Linux
3. Disable Http3Support on native AOT everywhere

1 is a possibility. People could discover this switch in docs. In practice, very few people will discover it.
I think 2 is an option because the delivery mechanism for QUIC outside Linux is our own build of msquic.dll (AFAIK) that we don't include in native AOT apps and user needs to manually copy. AFAIK on Linux we rely on an OS-provided msquic. I could be wrong. If I'm wrong, we can just cross out option 2.
3 is nice and consistent. We'll need to consider marking this a breaking change (does this fit definition of a breaking change?)

Cc @dotnet/ncl for throughts/review.

[Copilot]

Pull Request Overview

This PR disables HTTP/3 support by default in Native AOT builds to reduce application size and avoid unsupported configurations.

• Adds an MSBuild property Http3Support defaulting to false
• Inserts a runtime host configuration option to turn off HTTP/3 in SocketsHttpHandler

Comments suppressed due to low confidence (2)

src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets:42

• Add a brief comment above this property explaining why HTTP/3 support is disabled by default in Native AOT (e.g., to save size and avoid unsupported scenarios) for future maintainers.

    <Http3Support Condition="'$(Http3Support)' == ''">false</Http3Support>

src/coreclr/nativeaot/BuildIntegration/Microsoft.NETCore.Native.targets:51

• Consider adding a unit or integration test to verify that HTTP/3 support is indeed disabled by default for Native AOT applications and that enabling Http3Support toggles this behavior as expected.

    <RuntimeHostConfigurationOption Include="System.Net.SocketsHttpHandler.Http3Support"

[jkotas]

AFAIK on Linux we rely on an OS-provided msquic.

Nit: We rely on Microsoft-provided distro-specific msquic binary. You have to add Microsoft Linux package feed for given distro and install msquic from there: https://learn.microsoft.com/en-us/dotnet/fundamentals/networking/quic/quic-overview#platform-dependencies .

[jkotas]

This makes sense to me. I think we may want to go even further and disabled it by default for all trimmed apps, not just NAOT.

does this fit definition of a breaking change?

I think so.

[ManickaP]

I'm perfectly fine with disabling this on native AOT.

[stephentoub]

Will we follow this up to update Kestrel to also trim out its HTTP/3 support by default when trimming? I'm wondering if we need a more general name than "System.Net.SocketsHttpHandler.Http3Support".

[MichalStrehovsky]

I'm perfectly fine with disabling this on native AOT.

Any opinions about disabling this in trimmed apps in general like Jan suggested?

Will we follow this up to update Kestrel to also trim out its HTTP/3 support by default when trimming? I'm wondering if we need a more general name than "System.Net.SocketsHttpHandler.Http3Support".

This is currently doc'd (https://learn.microsoft.com/en-us/dotnet/core/deploying/trimming/trimming-options) as specific to System.Net.Http. i don't know what size savings we would be talking about for Kestrel and whether we'd ever need to control it independently. Cc @davidfowl for opinion.

[MichalStrehovsky]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[ManickaP]

Any opinions about disabling this in trimmed apps in general like Jan suggested?

Is trimming automatic when you just specify RID when publishing (or any other way it's on automatically without user input)? If so, than I'm rather against as it would be too easy for the customers to unintentionally turn off H/3.

[jkoritzinsky]

Trimming only happens (for desktop platforms) when PublishTrimmed or PublishAot is set to true.

Mobile platforms trim by default, but I don't think we have an msquic story there anyway.

[MichalStrehovsky]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[ManickaP]

Trimming only happens (for desktop platforms) when PublishTrimmed or PublishAot is set to true.

Then I'm fine with "disabling this in trimmed apps in general" as well.

[MichalStrehovsky]

Trimming only happens (for desktop platforms) when PublishTrimmed or PublishAot is set to true.

Then I'm fine with "disabling this in trimmed apps in general" as well.

Okay, pushed out a commit that moves this default under PublishTrimmed.

Does the test update look reasonable too?

[ManickaP]

[MichalStrehovsky]

/ba-g helix deadletters on wasm