TypedByReference should be treated as a normal valuetype for calling convention purposes #71675
Conversation
ghost
assigned 
|
@AaronRobinsonMSFT tagging you as this may have an impact on ref structures, and places we'd like to test. |
| <Project Sdk="Microsoft.NET.Sdk"> | ||
| <PropertyGroup> | ||
| <OutputType>Exe</OutputType> | ||
| <CLRTestPriority>0</CLRTestPriority> |
There was a problem hiding this comment.
Change to Pri1 before checkin, but leave as Pri0 for initial test pass to make sure that this works on all architectures and OSs.
| // Licensed to the .NET Foundation under one or more agreements. | ||
| // The .NET Foundation licenses this file to you under the MIT license. | ||
| using System; | ||
| using System.Runtime.CompilerServices; |
There was a problem hiding this comment.
I don't think this test is appropriate. The github number thing here makes sense with the JIT tests, but not with defined behaviors. I'd prefer to see this test added to an existing collection for TypedReference under src/tests/Loader/classloader or, since this is pure managed, in TypedReferenceTests.cs.
There was a problem hiding this comment.
Reasonable. I originally put the test here, as I guessed it was weird optimizer thing.... as it was only reproducing on Linux then I looked a bit deeper.
There was a problem hiding this comment.
Instead of moving this, I'd also be fine with clearly documenting in the test the state we are validating here.
The previous logic would cause the runtime to compute an incorrect size for a TypedByReference argument, and then only shuffle a single register of data instead of two.
Fixes #42732