Implement concept of unstable ABI into crossgen2

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.

[davidwrighton]

• Use for the Vector types based on current defined stability rules
• Will prevent compilation of methods with not yet stable abis, or calls to methods with said unstable apis

Fixes#13522

[MichalStrehovsky]

We're already doing something similar for Vector<T>: we basically just say that the layout in Indeterminate and there's code checking for that that aborts crossgening the method whenever a piece of layout info is needed and it ends up Indeterminate.

Do we need to introduce a new general purpose mechanism here? Could we just have a FieldLayoutAlgorithm that returns Indeterminate layout for these same as for Vector<T> but only do it if the version bubble doesn't include CoreLib?

Cc @dotnet/crossgen-contrib

[janvorli]

It looks ok, but I wonder about the same thing that @MichalStrehovsky said. It seems it would make the change more local.

[davidwrighton]

Unfortunately, making the types have indeterminate size also prevents usage of the Vector intrinsics. This approach allows the intrinsics to continue to work and use the Vector128/256 types while simultaneously not permitting these apis to cross an ABI boundary.

[janvorli]

Makes sense.