[cap-talk] cap-talk Digest, Vol 66, Issue 45

[Leo Meyerovich]

Memory safety is a property; managed code is when we use a particular 
mechanism for achieving it.

For an alternative mechanism, memory safety can be statically verified 
with region types. No runtime checks would be necessary.

There are a lot of was to achieve it. Singularity's ownership types and 
protocols, region types, dependent types, process isolation, binary 
translation, etc. There's a variety of reasons for picking different 
ones: the program domain, verification needs, performance vs. 
expressiveness constraints, and granularity of safety, etc. all impact 
where you should look. Finally, process isolation with a VM-per-process 
has historically been the way to go (and often another VM underneath it 
all). The research community, security nuts, and performance nuts hate 
generally this architecture, but there have been proven QoS, hardening, 
robustness, separation of concerns, etc. benefits.

- Leo


[[sorry for not responding to the individual message; I get the list as 
a digest and haven't figured out individual responses yet]]