[MichalStrehovsky]

Part of a series of pull requests that enable compiling, linking, and running a Console.WriteLine hello world on WASI/native AOT.

Summary

Resolve NativeAOT Wasm interface targets and call them indirectly with the actual method signature, rather than using the arbitrary-signature RhpInterfaceDispatch stub.

ILC reports CORINFO_VIRTUALCALL_LDVIRTFTN, retains concrete or generic dispatch-cell lookups, and requests null-instance checks. RyuJIT passes the receiver and cell to CORINFO_HELP_INTERFACELOOKUP_FOR_SLOT; Wasm maps that to the existing fixed-signature RhpResolveInterfaceMethod rather than the unsupported fast helper.

The existing generic-virtual-method path is retained. The importer also enables NativeAOT Wasm's existing ReadyToRun resolver for class-virtual ldvirtftn, covering delegates bound to class virtual methods. Non-Wasm NativeAOT stub selection and the CoreCLR Wasm fallback are preserved.

Validation

• Checked Wasm JIT, native x64 JIT, ILC, and crossgen2 builds - passed.
• Local verify-interface-codegen.ps1 - 6/6 Wasm cases compiled and passed Wasm validation; no case imported RhpInterfaceDispatch, and the concrete/generic call cases imported RhpResolveInterfaceMethod.
• The same script's 6/6 Windows x64 compilation controls passed.

The cases cover concrete interface calls, shared generic methods/types, interface delegates, and class-virtual delegates. These are codegen checks, not interface-dispatch execution tests.

These results are from the combined local bringup tree, including prerequisite NativeAOT/Wasm enablement, not a fresh build of this isolated main-based branch. The probe script is a local fixture, not a test added by this PR.

Note

This PR description was generated by GitHub Copilot.

[dotnet-policy-service]

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

[azure-pipelines]

Azure Pipelines:
Successfully started running 5 pipeline(s).
11 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟡 Changes recommended

Unsupported generic lookups can crash compilation, and standalone virtual function-pointer lookups lack required receiver null checks.

3 open findings

• Handle failed generic interface lookup before helper construction · New
• Add null check for standalone interface ldvirtftn · New
• Add receiver null check for class-virtual ldvirtftn · New

What changed in this PR

Advances NativeAOT/WASI support by resolving interface targets before calling them with their actual signatures.

Changes:

• Selects Wasm-specific interface lookup behavior and the fixed-signature resolver.
• Adds importer support for interface lookups and enables class-virtual function-pointer resolution.

Validation here was limited to source inspection; no builds or tests were run.

File	Description
src/​coreclr/​tools/​aot/​ILCompiler.RyuJit/​JitInterface/​CorInfoImpl.RyuJit.cs	Selects Wasm dispatch lookups, resolver, and null-check requirements.
src/​coreclr/​jit/​importer.cpp	Imports Wasm interface lookups and enables NativeAOT class-virtual resolution.

🧠 Review effort: Balanced

[MichalStrehovsky]

@dotnet/jit-contrib could someone have a look? this is blocking progress for native AOT WASM. We cannot start testing anything in the CI if interface dispatch doesn't work and native AOT WASM already got broken twice since I started working on it.

[MichalStrehovsky]

@dotnet/jit-contrib could someone have a look? this is blocking progress for native AOT WASM. We cannot start testing anything in the CI if interface dispatch doesn't work and native AOT WASM already got broken twice since I started working on it.

@adamperlin @EgorBo could you have a look please? I thought I can wait for Jakob to be back but he's still not back and this is getting really blocking.

[EgorBo]

@MichalStrehovsky it looks like we need to apply this patch:

    bool IsGenericVirtual(Compiler* compiler) const
    {
        return (gtCallType == CT_INDIRECT && (gtControlExpr->IsHelperCall(CORINFO_HELP_VIRTUAL_FUNC_PTR) ||
                                              gtControlExpr->IsHelperCall(CORINFO_HELP_GVMLOOKUP_FOR_SLOT) ||
+                                             gtControlExpr->IsHelperCall(CORINFO_HELP_INTERFACELOOKUP_FOR_SLOT)
#ifdef FEATURE_READYTORUN
                                              || gtControlExpr->IsHelperCall(CORINFO_HELP_READYTORUN_VIRTUAL_FUNC_PTR)
#endif
                                                  ));
    }

for divirtualizations to work.
Other than that LGTM I think. CORINFO_HELP_INTERFACELOOKUP_FOR_SLOT might also need some VN treatment for CSE, but not sure (perf only, not correctness).

[MichalStrehovsky]

Thanks for the reviews! I ran some additional validation - enabled this on Windows and ran a decent subset of src/tests tree with it. This should work, we can do perf fixes later.