LDM Notes for September 21st, 2022

[333fred]

https://github.com/dotnet/csharplang/blob/main/meetings/2022/LDM-2022-09-21.md

Agenda

• Unchampioned issue triage
  • Track subtype exhaustiveness for classes with only private constructors
  • ReadOnlySpan initialization from static data
  • Embedded Language Indicators for raw string literals
  • Implicit Parameters
• Unsafer Unsafeness
  • Conclusion

[HaloFour]

There are some potentially interesting cases in this with ref assemblies as well, since private constructors are normally dropped.

If a type only had private constructors wouldn't dropping it from a reference assembly break the contract since the type would be left with only a public parameterless constructor?

[HaloFour]

Ah, this would also be a problem:

If there are no InternalsVisibleTo attributes, internal function members are also removed.

Can the rules regarding how these assemblies are emitted be changed? Could a private constructor, if only just a stub, be considered observable metadata that should be retained in an assembly where a type has nested subtypes?

[HaloFour]

I would think that the decisions made here would impact the ability to design "reference DUs", defined by a closed type hierarchy.

[333fred]

These are excellent questions that need to be answered 😉.

[sab39]

I always assumed that the default no-argument public constructor was synthesized by the compiler, and that at the IL/runtime level a class might have no constructors at all, in which case there wouldn't be an automatic default one either. There's no reason to emit any constructor at all for a static class, for example. There's nothing saying that a reference assembly has to be expressible in C# source code, is there?

[HaloFour]

@sab39

Right, the problem is that if a reference assembly elides private protected and internal constructors that it becomes impossible to tell based on that metadata whether those nested types represent a closed hierarchy for the purposes of exhaustiveness checking.