I’m fairly certain we’ve had lambdas (lisp) for something like 40 years. Type inference (ML/lisp compilers) for 30-40 years. And generics, at least in the sense of parametric polymorphism, (Ada/C++/Java/ML) for about the same amount of time. As far as I can tell, the borrow checker and some features around dependent types are newish, but actually modern features are few and far between.

Lisp had lambdas for 60 years, I think ML has had type inference for about 50 (maybe there are earlier languages that had it?), ML also had generics for that long. So, none of your examples were recent developments by your standards.

Even in C-like languages, generics were part of C++ (or rather templates, in that case) since the beginning, I think (that's 35 years ago). I think you can make a case for type inference reaching the C language family recently, though.

I mean sure "recent" could refer to a span of even 50 years why not. I wouldn't trade my Roslyn compiler for AGOL68.

Comparing a compiler to a language is comparing apples and oranges.

I disagree I think it's more like comparing an apple to an apple tree

Either way, it’s hard to compare a 30 year old language to a modern compiler because the modern compiler has benefited from 30 years of optimizations: you can’t tell how the old language would perform with a similar investment.