[MONO][Interp] Tracking runtime test failures with Mono interpreter

[fanyang-mono]

Here are the issues:

Probably worth investigating in NET6 - most of these are probably intentionally bad IL or other IL edge case behavior (and could be triaged to lower priority pretty quickly) - but they might uncover interpreter issues:

• [x] #54374 [MONO][Interp] function call tests failed with SEGV with interpreter
• [x] #54559 [MONO][Interp] HardwareIntrinsics tests failed with SEGV with interpreter

Also warrants a deeper investigation in .NET6 - the failures here are not well understood:

• [ ] #54399 [MONO][Interp] Loader/classloader/* tests failed with interpreter (Rationale: default interface methods - probably not purely an interpreter issue)
• [x] #54389 [MONO][Interp] JIT/Methodical/eh/basics/throwinfilter_* failed with interpreter (Rationale: for browser-wasm we're relying on the interpreter for exception filter code to work correctly)
• [ ] #54358 [MONO][Interp] JIT/Performance/CodeQuality/Burgers/Burgers/Burgers.sh failed with interpreter (rationale: burgers is benchmark code - don't expect it to fail)

Other issues:

• [ ] #46622 Mono interpreter supports a maximum of 32k local space for a method
• [ ] #37955 Fix initlocals handling with mono interpreter
• [ ] #54359 [MONO][Interp] JIT/CodeGenBringUpTests/Localloc_* failed with interpreter
• [ ] #54373 [MONO][Interp] JIT/Directed/* failed with Error 101 with interpreter
• [x] #54375 [MONO][Interp] JIT/IL_Conformance/Old/Conformance_Base/div_r8/div_r8.sh failed with unhandled exception with interpreter
• [x] #54376 [MONO][Interp] JIT/IL_Conformance/* tests failed return the correct return code with interpreter
• [ ] #54359 [MONO][Interp] JIT/CodeGenBringUpTests/Localloc_* failed with interpreter
• [ ] #54388 [MONO][Interp] JIT/Methodical/* test failed with unhandled exception with interpreter
• [ ] #54392 [MONO][Interp] JIT/Regression/* failed return the correct return code with interpreter
• [ ] #54393 [MONO][Interp] JIT/Regression/* failed with unhandled exception with interpreter
• [ ] #92727

Lower priority:

• [x] #54371 [MONO][Interp] Assertion at /__w/1/s/src/mono/mono/mini/interp/transform.c:6813, condition `td->clause_indexes [in_offset] != -1' not met (Rationale: intentionally malformed IL)
• [x] #54372 [MONO][Interp] transform.c: Unimplemented opcode: ee at 0x0 (Rationale: illegal opcodes are unlikely in well-formed assemblies)
• [x] #54401 [MONO][Interp] baseservices/varargs/varargsupport* failed with interpreter (Rationale: varargs are rare)
• [x] #54396 [MONO][Interp] Assertion: should not be reached at /__w/1/s/src/mono/mono/mini/interp/interp.c:3285 with interpreter (Rationale: malformed IL test)
• [x] #54395 [MONO][Interp] Assertion at /__w/1/s/src/mono/mono/mini/interp/transform.c:4527, condition `td->sp >= td->stack' not met with interpreter (Rationale: bad IL)
• [x] #54560 [MONO][Interp]classloader tests failed with SEGV with interpreter (rationale: bad IL)
• [x] #54381 [MONO][Interp] Assertion: should not be reached at /__w/1/s/src/mono/mono/mini/interp/transform.c:5325 with interpreter (rationale: uncommon conversion only with IL)
• [ ] #54391 [MONO][Interp] Assertion: should not be reached at /__w/1/s/src/mono/mono/mini/interp/transform.c:5794 with interpreter (rationale: uncommon conversion only with IL)
• [x] #54394 [MONO][Interp] Assertion: should not be reached at /__w/1/s/src/mono/mono/mini/interp/transform.c:6158 with interpreter (rationale: uncommon conversion only with IL)
• [ ] https://github.com/dotnet/runtime/issues/92733

[ghost]

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

Issue Details

---

Comparing to Android x64 with JIT, here are the additional 93 tests failing to run on Android x64 with interpreter - [ ] #54358 [MONO][Android][Interp] JIT/Performance/CodeQuality/Burgers/Burgers/Burgers.sh failed on Android x64 with interpreter - [ ] #54359 [MONO][Android][Interp] JIT/CodeGenBringUpTests/Localloc_* failed on Android x64 with interpreter - [ ] #54371 [MONO][Android][Interp] Assertion at /__w/1/s/src/mono/mono/mini/interp/transform.c:6813, condition `td->clause_indexes [in_offset] != -1' not met - [ ] #54372 [MONO][Android][Interp] transform.c: Unimplemented opcode: ee at 0x0 - [ ] #54373 [MONO][Android][Interp] JIT/Directed/* failed with Error 101 on Android x64 with interpreter - [ ] #54374 [MONO][Android][Interp] JIT/Directed/tailcall/tailcall/tailcall.sh failed with SEGV on Android x64 with interpreter - [ ] #54375 [MONO][Android][Interp] JIT/IL_Conformance/Old/Conformance_Base/div_r8/div_r8.sh failed with unhandled exception on Android x64 with interpreter - [ ] #54376 [MONO][Android][Interp] JIT/IL_Conformance/* tests failed return the correct return code on Android x64 with interpreter - [ ] #54381 [MONO][Android][Interp] Assertion: should not be reached at /__w/1/s/src/mono/mono/mini/interp/transform.c:5325 - [ ] #54359 [MONO][Android][Interp] JIT/CodeGenBringUpTests/Localloc_* failed on Android x64 with interpreter

Author:	fanyang-mono
Assignees:	fanyang-mono
Labels:	`area-Codegen-Interpreter-mono`, `os-android`
Milestone:	-

[BrzVlad]

The title is misleading since these issues have nothing to do with android. These tests were already disabled on our CI runs on interpreter, however they were not triaged in detail. When it comes to android, it would have been more interesting to see whether there are any additional issues compared to desktop interpreter, which normally shouldn't be.

[fanyang-mono]

The title is misleading since these issues have nothing to do with android. These tests were already disabled on our CI runs on interpreter, however they were not triaged in detail. When it comes to android, it would have been more interesting to see whether there are any additional issues compared to desktop interpreter, which normally shouldn't be.

I see. Let me clean the issues up.

[fanyang-mono]

These are existing issues since .NET 5. None of them is currently block our customer. We will revisit them in .NET 7.

[lambdageek]

I went through and reorganized the list into 4 lists.

The first group "Probably worth investigating in NET6" are tests that I think a quick investigation will reprioritize for .NET 7. I suspect most of these are failing because the test is some bad IL, and the interpreter is generally not designed to gracefully exit for patently broken code.

The second group "Also warrants a deeper investigation in .NET6" Seem like they might be real high-impact issues.

The third group "Other issues" - particularly the JIT ones - I can't really tell from looking at them how big the impact is. I don't think they're intentionally bad IL, but I also can't tell from a superficial scanning of them if they're trying unlikely edge cases or something that is likely to be common

The fourth group are all the cases where the IL is intentionally bad and we're unlikely to see this from normal C# code.

[fanyang-mono]

Moving to 8.0.0