Add browser WASM R2R test leg

[jkotas]

Goal

Maintain continuous browser-wasm ReadyToRun runtime-test coverage without blocking unrelated pull requests while known product failures remain.

Current status

• The original blocking R2R_CG2 leg was removed by Disable browser WASM R2R test leg #133294.
• Enable browser WASM R2R tests in manual pipeline #133308 provided a temporary manual census lane.
• A fresh current-main census was run at ef1850bdc9e7d142888247d9bb5df3b74a7ae341 in Azure DevOps build 1588226.
• Every failure from that census is fixed already or narrowly excluded against an active issue.
• Restore non-blocking browser WASM R2R runtime tests #133463 restores the default runtime pipeline leg as non-blocking and removes the now-redundant Enable browser WASM R2R tests in manual pipeline #133308 manual-pipeline plumbing.
• Promote the lane back to blocking after two consecutive current-main runs have zero failed Helix work items and zero failed test results in both published R2R runs, with no newly unclassified failures.

The restored lane remains visible in the default pipeline. While it is non-blocking, failures must still be inspected from Helix work-item status and both Azure DevOps test runs; the submission job result alone is not sufficient.

Final PR validation

The restored default runtime pipeline was validated on #133463 at commit 040c3069cc76d3261b7fedd7e1598757ae8f6f19 in Azure DevOps build 1588519:

• The R2R_CG2 browser wasm checked job completed succeededWithIssues, as intended for a non-blocking lane.
• Helix job 870b56d5-d077-41e1-9e9f-22ff41e50ddb completed with 0 failed work items.
• Test run 43844474 published 3813 total, 3328 passed, and 0 failed tests.

A preceding branch census exposed later failures that the fresh-main run's earlier aborts had masked. Those were also narrowed:

• Methodical_r1 / huge_b_r: generic high-rank array signature mismatch, tracked by [wasm][R2R] Generic multidimensional array access traps with null function #133307.
• Methodical_d1 / throwinfilter_d and Regression_6 / GitHub_4044: first-pass ExInfo collision, tracked by [wasm][R2R] b68872 collides with first-pass ExInfo #133468.
• The two >1,000-parameter merged assemblies require project-level R2R preflight exclusions because WebAssembly module construction fails before child dispatch.

Later PR-head census

Build 1589007 at 452dea3aba16d97244508b8a0fef4643de8dd26e exposed 13 additional or still-active test-result failures hidden behind completed Helix work items:

Failure	Handling
StructABI invalid f64.reinterpret_i64 operand	Out-of-process project R2R exclusion, #133502
more_tailcalls memory access out of bounds	Pre-existing limitation tracked by #131890; out-of-process project R2R exclusion
InterpreterTester.TestPInvoke uncaught missing-library failure	Guard only this subtest, #131811
badcodeinsidefinally_d/r returns null from GetObj	Type-level ActiveIssue, #133504
Runtime_101046 incorrect native small-integer sign extension	Out-of-process project R2R exclusion, #133506
github60486 duplicate stack frames	Method-level ActiveIssue, #133505
AssemblyOnDisk, ByRefLocals, GenericVirtualMethod, CollectibleStatics	Empty Assembly.Location treated as unavailable file-backed setup
Regression4	Empty Assembly.Location treated as unavailable file-backed setup
UnrollEqualsStartsWith preemptive-GC OBJECTREF assertion	Method-level ActiveIssue, #133219

These changes are included in #133463 at 67535696f88.

The final PR head 67535696f88a709f322896c0b608821659acb30a was validated in build 1589161:

• coreclr Pri0 Runtime Tests Run R2R_CG2 browser wasm checked completed successfully.
• Test run 43866592 reported 3815 total, 3340 passed, 475 not executed, 0 failed, and 0 unanalyzed.

Fresh main census

Build 1588226 produced two browser-wasm R2R test runs:

Run	Total	Passed	Failed	Not executed	Unanalyzed
R2R-CG2 browser wasm Checked no_tiered_compilation	2218	1798	7	413	7
R2R-CG2 browser wasm Checked	2218	1794	7	417	7

Helix reported 14 failed work items. The shared monitor surfaced only 10; JIT.opt, Regression_3, Regression_6, and Regression_PdbOnly_r_3 also failed and had to be found from the Helix job itself. Seven per-test failures were published even though their containing merged-runner work items did not appear in the monitor failure list.

Failed Helix work items

Work item	Exact failing test/signature	Tracking	Exclusion
Directed_do	tailcall/mutual_recursion, SIGSEGV / exit 139 after FifthMethod	#133464	Child project R2RIncompatible on wasm CoreCLR
Directed_ro	_Arrayscomplex3, null function or function signature mismatch	#133307	Method-level ActiveIssue
JIT.opt	NonVoidThrowHelper, OBJECTREF being accessed while thread is in preemptive GC mode	#133219 / fix #133269	Method-level ActiveIssue
Loader	ComplexHierarchyPositive, generic static-interface dispatch signature mismatch	#133466	Method-level ActiveIssue
Methodical_d1, Methodical_do, Methodical_r1, Methodical_ro	throwinfinally_50, RuntimeError: unreachable	#133465	Method-level ActiveIssue
Methodical_d2, Methodical_r2	lcs_gcref, multidimensional-array signature mismatch	#133307	Method-level ActiveIssue
Methodical_others	module construction fails: parameter count 4002	#132855 / fix #132865	Merged assembly R2RIncompatible on wasm CoreCLR; failure occurs before test dispatch
Regression_3	DevDiv_902271, GetGCStaticBase memory access out of bounds	#133467	Child IL project R2RIncompatible on wasm CoreCLR
Regression_6	b68872, first-pass ExInfo collision at exceptionhandling.cpp:4273	#133468	Child IL project R2RIncompatible on wasm CoreCLR
Regression_PdbOnly_r_3	module construction fails: parameter count 8194	#132855 / fix #132865	Merged assembly R2RIncompatible on wasm CoreCLR; failure occurs before test dispatch

Failed tests hidden from the work-item summary

Test	Exact signature	Tracking	Exclusion
Runtime_76273	AssemblyLoadContext.LoadFromAssemblyPath: path is empty	Applicability, not a product failure	Return successfully when Assembly.Location is empty; run the full scenario when files exist
GitHub_22888	Path.Combine: Assembly.Location produced a null directory	Applicability, not a product failure	Return successfully when Assembly.Location is empty; run the full scenario when files exist
InterpreterTester	InterpreterTest.TestMdArray, signature mismatch	#133307	Guard only the TestMdArray subtest; all other interpreter tests remain active
Runtime_101046	native small-integer return produces exit 101	#131811	Child IL project R2RIncompatible on wasm CoreCLR
bestfit, bestfit_1, bestfit-finalize	RuntimeError: unreachable after reaching steady state	#133469	Each child project R2RIncompatible on wasm CoreCLR

Revalidated non-failures

• b125091 passed in the fresh Regression_1 work item and remains unsuppressed. Assert failure: Popping a frame out of order in browser-wasm ReadyToRun runtime tests #133221 is retained as historical investigation context, not a current lane failure.
• pinnedlocal remains unsuppressed. The fresh JIT.opt work item terminated at NonVoidThrowHelper before providing fresh pinnedlocal evidence; recent main evidence showed it passing, so [wasm] R2R OSR pinnedlocal hits fFrameWasUnwound and cycles while dumping frames #133286 is not used for a suppression.
• Vector3Interop remains as fixed on main: forward P/Invoke executes, and only reverse P/Invoke is skipped for browser CoreCLR against [browser-wasm] CoreCLR runtime tests blocked on interop gaps after test-specific corerun enablement #131811.
• The [wasm][R2R] crossgen2 asserts 'wasmAbiType != CORINFO_WASM_TYPE_VOID' in LowerStoreSingleRegCallStruct, aborting the whole Regression_ro_2 work item #133224 ABI-classifier failure is fixed by merged [Wasm RyuJIT] Fix: Pass Call's struct type to Wasm ABI Classifier Instead of Dest Store Type #133281.
• The failures fixed by [wasm] Recognize R2R virtual IPs in IsManagedCode #133266 and the test-specific corerun work in [wasm] Build a test-specific corerun for nine more runtime tests #133367 are not re-suppressed.
• Generate ReadyToRun unboxing stubs for browser Wasm #133218 enabled shared structural Wasm R2R unboxing stubs but introduced the separate null-target-slot regression tracked by [ci-scan] Test failure: browser-wasm R2R interpreter-to-R2R thunk traps with null function or function signature mismatch #133491.

Open fixes and infrastructure dependencies

Issue	Fix PR	Current status
#132855, >1,000 wasm parameters	#132865	Open
#133219, cooperative GC mode after R2R catch resumption	#133269	Open
#133303, TypeMapApp compile/load failure	-	Open; Interop preflight remains narrowly excluded
#133307, generic MD-array signature mismatch	-	Open
#133491, shared Wasm R2R unboxing stub target/adapter dependencies	#133516	Draft is based on the earlier fallback approach; update to preserve shared U/UG/UM stubs with per-target M + I dependencies and the runtime availability guard
#131886, Runtime_79354 portable entry point	-	Open
Runtime-test call-helper generator alignment	#133419	Open; not required for this restoration
SDK crossgen2 resolution	#133413	Open with NO-MERGE; not treated as landed or ready

Non-blocking and blocking criteria

The default browser-wasm R2R lane is intentionally non-blocking while these active exclusions remain. It should continue to run on the same path triggers as the original lane and publish normal Helix/test-result evidence.

Promote it to blocking only after:

1. Two consecutive current-main runs report zero failed Helix work items.
2. Both the default and no_tiered_compilation test runs report zero failed tests, including failures hidden by merged-runner exit behavior.
3. Every exclusion added for this restoration has either been removed by a product fix or explicitly accepted as a longer-term platform limitation with its own tracker.
4. The shared Helix monitor can be re-enabled for the lane without turning known or hidden failures into unrelated PR noise.

Note

This tracking update was generated with GitHub Copilot.

[dotnet-policy-service]

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

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/crossgen-contrib
See info in area-owners.md if you want to be subscribed.

[lewing]

The helix queue monitor is delaying obvious visible reporting of failing legs until everything has completed which is causing some over aggressive merging on "yellow". That is what happened here but there are several other recent instances. I have a prospective Build Insights PR open and I'm engaging with the services side to try surface failures sooner on the pr status.

There are several open issues and prs in flight to fix the failures here and I will add tracking info once I've categorized them all. Given the rate of churn in the wasm r2r tests enabling this again is a high priority.