[lewing]

Object stack allocation skips every R2R array allocation because CORINFO_HELP_READYTORUN_NEWARR_1 carries no type argument. On WebAssembly, crossgen2 allocates arrays with the explicit-type helpers (CORINFO_HELP_NEWARR_1_DIRECT(type, length)) as JIT-compiled code does, so those can take the normal path. This removes the blanket R2R bail-out (and the now-unused m_isR2R); READYTORUN_NEWARR_1 stays excluded, so native R2R codegen is unchanged.

On wasm composite R2R, a non-escaping new int[4] ×1e7 goes from 194 ms to 27 ms, and the 9 array cases in ObjectStackAllocationTests now pass. This is the explicit-type subset of #116651, without its dependency on #116808.

SuperPMI: diffs

Missing SuperPMI contexts and test exclusions

A few R2R contexts replay with missing data (43 of 286,981 on linux-x64 libraries.crossgen2): shared-generic (__Canon) array allocations now query the array layout before being rejected. Crossgen of 9 framework assemblies with matched base/diff JITs produced zero stack-allocated arrays either way.

No test exclusion can be lifted: ObjectStackAllocationTests is still CrossGenTest=false because native R2R uses READYTORUN_NEWARR_1 (#109314), browser CoreCLR runtime tests run on the interpreter (#131321), and out-of-process tests don't run on WASI.

Note

This PR description was drafted with GitHub Copilot.

[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.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[lewing]

The JIT_ro failure in the browser-wasm R2R lane (TypeIntrinsics_ro, pMT->Validate() assert) is the known issue #135966, not this change. I've broadened that issue's match pattern to cover the _ro variant.

Checked locally under the Helix environment for this lane (checked browser-wasm runtime, node 23, the Helix payload):

• main without this change fails TypeIntrinsics_ro at the same rate as with it (4/15 vs 5/15), with the same assert. Whether _r or _ro fails depends on the run, which is why main currently shows JIT_r failing and this PR JIT_ro.
• With JIT: treat throwing trees as GC safe points #135970 applied, both tests pass on both bases (48/48).
• This change doesn't alter the compiled TypeIntrinsics_r/_ro images.

Note

This comment was drafted with GitHub Copilot.

[EgorBo]

If you don't mind, let's wait for @AndyAyersMS on this one, perhaps there was a reason R2R was not supported here.

[lewing]

If you don't mind, let's wait for @AndyAyersMS on this one, perhaps there was a reason R2R was not supported here.

Sounds good, nothing here is blocking and Andy's full fix is a superset.