It's a wonderful idea.<p>If the MIT folks decided, today, to build a Lisp Machine, what would it look like? What would future LMI and Symbolics turn into? How would that affect the way we work?<p>And, while I am at it, what would a Debian GNU/Plan9 or a Ubuntu Plan9 look like?<p>Would it be more successful if it looked more like a traditional Unix desktop and exposed its sophisticated internals only to the power-user?
I think the concept of a lisp OS is a type error. Lisp has interpreters and runtime systems, that run on operating systems. However, Plan 9 suggests ways those interpreters could be better structured.<p>The core features of Plan 9 are that programs interact with their environment by naming things, that names refer to whatever programs or their interpreters mount on them, and that interpreters have a protocol to pass the buck to another progam when they don't know what a name means. To me, those ideas suggest a sequence of titles such as "Mount: the ultimate module system" and "Mount by wire protocol: the ultimate foreign function interface". I don't know what papers would have those titles.<p>The idea would be to decouple lisp interpreters and dialects from lisp libraries. If you could mount any functions that your favorite interpreter lacked, then diversity would cause smaller problems. You could write a single Apache plugin in C that spoke the apply protocol, and suddenly every lisp would support the mechanics of HTTP. Or, the web parts of your program could run in Arc on a PC, and mount procedures written in common lisp and running on a supercomputer for the number crunching. The various lisps might have enough in common to make those things work.
"One lesson of Plan 9 is that the right organizing metaphor is extremely powerful."<p>This sounds like a fundamental idea. Can anyone offer some more real life examples?