LDM Notes for November 30th, 2022

[333fred]

https://github.com/dotnet/csharplang/blob/main/meetings/2022/LDM-2022-11-30.md

Agenda

• Revise MemberNotNull for required
• params Span<T>

[RikkiGibson]

Should users be able to turn off the feature at the compilation-level such that backing arrays are always heap allocated?

I would expect that params ReadOnlySpan<T> methods will not be invokable in expanded form before the C# version the feature releases in. Is that right?

Maybe in some cases people will be able to turn off the feature for triage purposes by downgrading language version (not TargetFramework), and making their code go back to binding to the array overload.

[333fred]

I don't think that downgrading langversion is a viable testing switch here.

[sab39]

What would be the value of turning it off at the compilation level? I could see wanting to turn it off on a case-by-case basis, potentially, but turning it off globally would affect calls to BCL methods, calls to methods in the code that's being compiled, and calls to methods in third-party libraries that use the feature. Even in the scenario described, where the stack allocation is causing problems, not every use of the feature will cause problems, and switching all of them to heap allocation would surely change the performance characteristics of the program in other ways, obscuring the very thing we're trying to make measurable.

Switching it for calls to a particular method might be more helpful in that testing scenario, but that sounds too specialized and niche to be worth the engineering effort.