Fix compareTypesForEquality

[jkotas]

Fixes#96876

[ghost]

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

Issue Details

---

Fixes #96876

Author:	jkotas
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[jkotas]

This improves the original implementation in addition to make it correct. For example, this can prove that KeyValuePair<__Canon, int> and KeyValuePair<int, __Canon> are never equal that the original implementation was not able to.

[EgorBo]

@MihuBot

[jkotas]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkotas]

/azp run runtime-nativeaot-outerloop

[jkotas]

@MichalStrehovsky PTLA

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkotas]

This bug is not a recent regression for AOT compilers. The test introduced by this change fails with PublishAot and PublishReadyToRun in .NET 8. Any thoughts about whether the backport meets the bar?

[EgorBo]

This bug is not a recent regression for AOT compilers. The test introduced by this change fails with PublishAot and PublishReadyToRun in .NET 8. Any thoughts about whether the backport meets the bar?

hm.. doesn't repro for me with R2R on net8.0, but does repro with NativeAOT

[jkotas]

hm.. doesn't repro for me with R2R on net8.0, but does repro with NativeAOT

Good catch. The pattern matching of typeof(X) == t.GetType() checks seems to be broken in R2R (you can take a look why is that).

The pattern matching of typeof(X) == typeof(Y) works well and hits the bug in .NET 8:

C:\repro>type Program.cs
using System.Runtime.CompilerServices;
public class Test96876
{
public static void Main()
{
Console.WriteLine(foo<string>());
}
[MethodImpl(MethodImplOptions.NoInlining)]
static bool foo<T>() => typeof(T[]) == typeof(string[]);
}
C:\repro>dotnet run
True
C:\repro>dotnet publish /p:PublishReadyToRun=true
MSBuild version 17.8.3+195e7f5a3 for .NET
Determining projects to restore...
Restored C:\repro\repro.csproj (in 174 ms).
repro -> C:\repro\bin\Release\net8.0\win-x64\repro.dll
repro -> C:\repro\bin\Release\net8.0\win-x64\publish\
C:\repro>C:\repro\bin\Release\net8.0\win-x64\publish\repro.exe
False
C:\repro>rmdir /s /q bin obj
C:\repro>dotnet publish /p:PublishAot=true
MSBuild version 17.8.3+195e7f5a3 for .NET
Determining projects to restore...
Restored C:\repro\repro.csproj (in 288 ms).
repro -> C:\repro\bin\Release\net8.0\win-x64\repro.dll
Generating native code
repro -> C:\repro\bin\Release\net8.0\win-x64\publish\
C:\repro>C:\repro\bin\Release\net8.0\win-x64\publish\repro.exe
False

[jkotas]

The pattern matching of typeof(X) == t.GetType() checks seems to be broken in R2R

Opened #97134

[jkotas]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[MichalStrehovsky]

This bug is not a recent regression for AOT compilers. The test introduced by this change fails with PublishAot and PublishReadyToRun in .NET 8. Any thoughts about whether the backport meets the bar?

We also shipped with it in .NET 7. Not a regression from previous release, no customer reports. Doesn't feel like we have a strong case for backport.