The dose makes the poison; we're still a long way from fulling embracing microkernels and capabilities. Security is a holistic property and encompasses finer details too. I want a small TCB. I want capabilities pervasively. And in pursuit of modularity and abstraction, I want to be able to choose the components I want and take those burdens myself. It's a bit silly seeing the nth SIGOPS-SOSP paper on how Linux can be improved by integrating userspace scheduling.
It is the same in safer systems programming languages, we already have the concept since 1961, but apparently making the industry take the right decisions is a tenuous path until something finally makes good ideas stick and gain adoption.
