I've talked to members of the Roslyn team and asked about some features I would have found interesting to get added to the C# team (immutable, non-nullable data-types, type-inference on return-types, etc).
They seem very much on top of their game, and their response has been that lots of these features would probably introduce problems with existing base-class libraries and classes. If they were to introduce those features, they would have to do it at release 1, not 10 years later. They agreed it would have made the platform better, but now it's too late.
As for the "deep architectural" changes required, Roslyn is indeed that change. But the team was very clear on wanting to deliver in stages. 1: Deliver the new compiler and then 2: later when it all has proven to work well, then start using this new architecture to deliver new features.
This is generally accurate, but the last paragraph is a little suspect -- we're iterating on C# 6 right now and using the Roslyn compiler to build the features necessary for it.
I don't think compiler architecture is seriously stopping features in their tracks -- there's only a cost/benefit analysis. 'MichaelGG's suggestion on inferred members, for example, has implications on the contracts of types and usability/user interaction with the language, so it's not just a question of if we can do it, but if we want to do it.
They seem very much on top of their game, and their response has been that lots of these features would probably introduce problems with existing base-class libraries and classes. If they were to introduce those features, they would have to do it at release 1, not 10 years later. They agreed it would have made the platform better, but now it's too late.
As for the "deep architectural" changes required, Roslyn is indeed that change. But the team was very clear on wanting to deliver in stages. 1: Deliver the new compiler and then 2: later when it all has proven to work well, then start using this new architecture to deliver new features.