Hi, On Sat, Sep 12, 2020 at 05:00:54PM +0200, Paul Boddie wrote:
> I asked previously about whether anyone noticed that Fiasco had gained support > for capabilities and got nothing substantial back in response, other than > perhaps an implicit acknowledgement that people probably hadn't noticed. > > Certainly, the OC variant of Fiasco has been around for at least ten years, so > perhaps you might be able to shed some more light on whether its availability > alleviated any of the various "technical constraints". We are well aware that several newer L4 variants got some kind of capability support. Indeed this started happening around the same time it became apparent that the original l4hurd wasn't viable -- and the orginal l4hurd developers did invest significant effort looking into some upcoming capability-enabled L4 variants. (Along with the Coyotos kernel.) However, as challenges became apparent with these newer kernels as well, the bigger takeaway was that a design based on someone else's kernel will always be constrained in some ways -- much like the orignal Mach-based Hurd... The conclusion being that while any new attempt to get the Hurd on a more modern kernel should certainly use learnings from other projects, it would likely be better off implementing its own kernel, specifically tailored to the envisioned design. (I wrote about this multiple times in the past -- and I'm pretty sure at least one of my previous comments ended up on the website somewhere...) -antrik-
