Determine whether generic math interfaces need to express variance

[stephentoub]

See discussion in #74025 (comment). The question is whether the new generic math interfaces being introduced in .NET 7 should express co/contravariance (in and out on the various generic type parameters), e.g. 197d96f. It's not clear if there are meaningful scenarios for it or if adding them now could conflict with future language work, but it could also be a breaking change to add them later.

Seems likely we should just leave them as-is, but we should make an explicit decision.

Tagging subscribers to this area: @dotnet/area-system-numerics
See info in area-owners.md if you want to be subscribed.

Issue Details

---

See discussion in #74025 (comment). The question is whether the new generic math interfaces being introduced in .NET 7 should express co/contravariance (in and out on the various generic type parameters), e.g. 197d96f. It's not clear if there are meaningful scenarios for it or if adding them now could conflict with future language work, but it could also be a breaking change to add them later.

Seems likely we should just leave them as-is, but we should make an explicit decision.

Author:	stephentoub
Assignees:	tannergooding
Labels:	area-System.Numerics
Milestone:	7.0.0

[hez2010]

I don't see any use case against co/contravariance on generic math. Generally we are using generic math against structs. And even if we support it, currently we have no way to call a static interface member on interface type directly without a type parameter.

[jeffhandley]

@tannergooding (and @dotnet/area-system-numerics), have you reached a conclusion on this? If we decide to add the expressions, we would need to do so before the RC2 snap.

[dakersnar]

@tannergooding and I discussed this on Friday and have decided we are not going to include this feature. He plans on sending a longer explanation here but for now we can remove the blocking-release tag.

[jeffhandley]

I want to keep this in the 7.0.0 milestone, but labeled as documentation. We need to have notes about variance in the docs and/or here so that folks who ask about this again can understand why we chose not to express it.