[TrimmableTypeMap] Reflection-free TrimmableTypeMapType/ValueManager

[simonrozsival]

Summary

Adds the reflection-free TrimmableTypeMapValueManager and TrimmableTypeMapTypeManager, plus the opt-in wiring in JNIEnvInit (RuntimeFeature.TrimmableTypeMap) that selects them. Also includes the JavaConvert collection-factory refactor used by the trimmable path. NativeAOT default stays managed; trimmable is opt-in.

Status

All of the earlier prerequisites have now merged into main, and this PR has been rebased on latest main:

• [TrimmableTypeMap] Java.Interop bump and update to ValueManagers and TypeManagers #11799 — CoreCLR JavaMarshal split + Java.Interop bump (was the base; now merged)
• [TrimmableTypeMap] Add trimmable array proxy mapping #11753 array proxies · [TrimmableTypeMap] Fix inherited generic base typemap refs #11749 generic base · [TrimmableTypeMap] Skip unresolvable trimmable typemap peers #11751 unresolvable peers · [TrimmableTypeMap] Use non-generic JavaPeerProxy base for interface proxies (NativeAOT) #11769 interface proxies
• [TrimmableTypeMap] Honor AndroidApplicationJavaClass for the JCW base and manifest #11794 multidex/manifest base · [TrimmableTypeMap] Complete trimmable manifest generation parity #11796 manifest parity · [TrimmableTypeMap] Keep user AndroidJavaSource Java under R8 shrinking #11798 R8 keep

With those in main, this PR is no longer stacked or blocked — it contains only the value/type manager implementations on top of them.

[simonrozsival]

CoreCLRTrimmable device-test crash analysis (recording so we don't lose it)

Observed on the public dotnet-android pipeline, build 1487907 (run Mono.Android.NET_Tests-CoreCLRTrimmable, macOS > Tests > APKs 2).

Symptom

The step reported, after 52s:

Zero tests ran [+0/x0/?0] (52s 325ms)
Exit code: 1
  Error output: Error: Instrumentation did not report a resultsPath in the bundle.

"Zero tests ran" is misleading — it doesn't mean no tests executed, it means the test host never received a results file.

What actually happened

From the captured logcat artifact (net11.0-CoreCLRTrimmable/logcat-Mono.Android.NET_Tests-CoreCLRTrimmable.txt), the CoreCLRTrimmable app process (pid 4663) ran 263 tests before dying:

• 251 passed, 1 failed, 10 skipped
• App start 14:59:32 → crash 15:00:21 ≈ 49s of real test execution (so the 52s wasn't a hang/idle — it was doing work the whole time).

It crashed with a native SIGSEGV in the CoreCLR GC, mid-test:

06-30 15:00:21.301  4663  4686 I TestInstrumentation: [START] Xamarin.Android.NetTests.AndroidMessageHandlerTests.ServerCertificateCustomValidationCallback_ApprovesRequestWithInvalidCertificate
06-30 15:00:21.320  4663  4673 E DOTNET  : Fatal error. Got a SIGSEGV while executing native code.
06-30 15:00:21.324  4663  4673 F libc    : Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x4 in tid 4673 (HeapTaskDaemon), pid 4663 (droid.NET_Tests)
06-30 15:00:21.629  2657  2657 I Zygote  : Process 4663 exited due to signal 11 (Segmentation fault)
06-30 15:00:21.712  2674  2696 W ActivityManager: Crash of app Mono.Android.NET_Tests running instrumentation … TestInstrumentation

Fault addr 0x4 on the HeapTaskDaemon thread = a near-null deref inside the GC. It happened right after ServerCertificateCustomValidationCallback_ApproveRequest failed, during the next ServerCertificateCustomValidationCallback_* test.

Why it reported 0/0/0

Because the process died mid-run, the on-device instrumentation's final Finish() never executed, so no TRX was written and no resultsPath was reported → the host adapter threw "Instrumentation did not report a resultsPath in the bundle." and MTP discarded all 263 executed tests.

Follow-ups

1. The reporting gap (tests lost on crash) is fixed by [tests] Stream on-device test results to MTP as they finish #11833 — results now stream to MTP per-test, so a mid-run crash only loses the in-flight test and the crash itself is reported as a failure.
2. The ServerCertificateCustomValidationCallback_* crash matches Custom AndroidMessageHandler.ServerCertificateCustomValidationCallback that returns false may cause application crash while debugging. #8608 and is [Ignore]d in [tests] Stream on-device test results to MTP as they finish #11833 to unblock CoreCLR/Trimmable device runs; the native GC crash root cause still needs diagnosing separately.

Note: this manifested on PR #11802's validation run, but the crash is in the CoreCLR GC during an HTTP/TLS test and looks unrelated to the trimmable-typemap changes (the NativeAOT config in the same job passed).