[mono][wasm] Box gsharedvt Nullable<T> via a wrapper-free runtime helper

[pavelsavara]

What

Under minimal gsharedvt (llvmonly / WASM), boxing a gsharedvt Nullable<T> resolves a per-T Nullable<T>.Box out-sig wrapper (MONO_RGCTX_INFO_NULLABLE_CLASS_BOX). When corelib is AOT'd but the consuming assembly runs interpreted (the WasmTestOnChrome-MONO-ST configuration), that per-T wrapper is absent and the boxing call traps with function signature mismatch — e.g. boxing a Nullable<T> element through the non-generic IEnumerator.Current (new Queue(new List<Int128?> { ... })).

Fix

Emit a call to a new non-generic mono_helper_box_nullable icall — a thin wrapper over the existing mono_nullable_box, the same helper the interpreter uses (MINT_BOX_NULLABLE_PTR) — passing the value by address and the concrete Nullable<T> class from the rgctx. No per-T box wrapper is needed, so:

• the trap cannot occur, for corelib and value types defined in user assemblies, and
• the containing method stays fully AOT-compiled (nothing is dropped from the AOT image).

Only the llvm_only nullable-box paths in mini_emit_box are changed; other backends are untouched.

Change

• src/mono/mono/mini/jit-icalls.c / jit-icalls.h, src/mono/mono/metadata/jit-icall-reg.h, src/mono/mono/mini/mini-runtime.c: add and register the mono_helper_box_nullable(vbuf, klass) JIT icall.
• src/mono/mono/mini/method-to-ir.c: new mini_emit_nullable_box_helper; both llvm_only NULLABLE_CLASS_BOX sites now route through it.
• src/tests/Loader/classloader/generics/regressions/131537/: regression test that boxes Nullable<T> through Queue(ICollection). _AOT_InternalForceInterpretAssemblies reproduces the AOT-corelib + interpreted-assembly configuration on wasm-AOT lanes; it passes trivially elsewhere.

Fixes #131537.

Background

This supersedes two earlier approaches explored on this issue:

• The fixed-list rooting in [wasm][mono][AOT] Emit Nullable<T> gsharedvt box/unbox wrappers for corelib value types #131946 — only covered a hardcoded set of corelib value types, not Nullable<UserStruct>.
• @BrzVlad's suggestion to fail gsharedvt for box (mirroring unbox.any) — that cleared the trap but un-AOT'd whole methods, and CI surfaced a null function regression when an AOT'd caller reached a now-unemitted method (observed in the Stream.CopyToAsync / BrowserHttp async teardown on the MONO-ST Intrinsics lane).

Routing the nullable box through the wrapper-free icall keeps every method AOT-compiled, so it fixes the trap for all value types without that regression.

Testing

Validated on Release Mono WASM AOT (AOT'd corelib + force-interpreted app — confirmed System.Private.CoreLib is in the AOT'd set while the app assembly is interpreted): boxing Int128? / UInt128? / Half? / decimal? / Guid? / DateTime? / DateTimeOffset? / TimeSpan? through new Queue(List<T?>) round-trips cleanly — no function signature mismatch, no null function. Full library-test coverage (Number_AsCollectionElement_RoundTrip on WasmTestOnChrome-MONO) runs in CI.

Code size

Runtime binary: negligible (one small icall + a table entry). AOT'd app: roughly neutral — the per-T object(Nullable<T>) box out-sig wrappers are no longer needed for the gsharedvt path, offsetting the small per-call-site change, and it is much lighter than the rooting approach.

Note

This pull request was prepared with GitHub Copilot.

[azure-pipelines]

Azure Pipelines:
Successfully started running 4 pipeline(s).
12 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]

Pull request overview

This PR addresses a Mono WASM minimal-gsharedvt/AOT failure mode where boxing a gsharedvt variable valuetype (e.g., Nullable<T>) can require per-T wrappers that aren’t available when corelib is AOT’d but the consuming assembly is interpreted, leading to WebAssembly “function signature mismatch” traps. The change mirrors existing unbox.any behavior by bailing out of gsharedvt sharing for the problematic box case, and adds a regression test intended to reproduce the AOT-corelib + interpreted-app configuration.

Changes:

• Make MONO_CEE_BOX in method-to-ir.c trigger GSHAREDVT_FAILURE for gsharedvt variable-klass boxing under gsharedvt_min (mirrors MONO_CEE_UNBOX_ANY).
• Add a new loader/classloader regression test project and test case for boxing Nullable<T> via non-generic IEnumerator.Current.
• Configure the regression test project to force interpretation on wasm-AOT lanes via _AOT_InternalForceInterpretAssemblies.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/mono/mono/mini/method-to-ir.c	Adds minimal-gsharedvt guard for box on gsharedvt variable klass to force per-instantiation resolution.
src/tests/Loader/classloader/generics/regressions/131537/test131537.csproj	New regression test project with wasm-AOT interpreter-forcing configuration.
src/tests/Loader/classloader/generics/regressions/131537/test131537.cs	New regression test that boxes Nullable<T> elements through Queue(ICollection) / non-generic enumeration.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/tests/Loader/classloader/generics/regressions/131537/test131537.cs:48

• The test checks per-element equality inside the foreach loop, but it never asserts that the loop actually iterated over all expected elements. If enumeration stopped early, this could still pass. Add a final assertion on the number of items iterated.

        int i = 0;
        foreach (object boxed in queue)
        {
            Assert.Equal(source[i], (T)boxed);
            i++;

[pavelsavara]

/azp run runtime-wasm

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/tests/Loader/classloader/generics/regressions/131537/test131537.cs:33

• This test is unlikely to execute under the src/tests harness as-written: Loader/classloader tests in this area use the TestEntryPoint() + return-100 convention (see e.g. src/tests/Loader/classloader/generics/GenericMethods/method001.cs:33-52). With a void test method and only xUnit assertions, the regression may compile but never run on the intended lanes. Consider switching to a TestEntryPoint() that returns 100/101, and include at least one Nullable<T> where T is a struct defined in this test assembly (plus a null element) to cover the interpreted-assembly value-type case and HasValue=false boxing.

    [Fact]
    public static void BoxNullableThroughNonGenericEnumerator()
    {
        RoundTrip(new List<Int128?> { 1, 2, 3 });
        RoundTrip(new List<UInt128?> { 1, 2, 3 });

[pavelsavara]

/azp run runtime-wasm

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[azure-pipelines]

Azure Pipelines:
Successfully started running 4 pipeline(s).
12 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[BrzVlad]

Nice

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/mono/mono/mini/method-to-ir.c:3419

• mini_emit_nullable_box_helper manually promotes a vreg to a local and takes its address. This duplicates the existing EMIT_NEW_VARLOADA_VREG helper (already used elsewhere in method-to-ir.c) and makes the code harder to maintain.

	MonoInst *iargs [2], *addr, *var;

	var = get_vreg_to_inst (cfg, val->dreg);
	if (!var)
		var = mono_compile_create_var_for_vreg (cfg, m_class_get_byval_arg (klass), OP_LOCAL, val->dreg);

[pavelsavara]

/azp run runtime-wasm

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[pavelsavara]

/ba-g unrelated failures