[jandrese]

I just skimmed that paper but I don't see a section on how to avoid the bottleneck when your functional program is trapped on a Von Neumann machine. It seems to me that functional constructs have to be mapped down to structures that will suffer the Von Newmann bottleneck if they are being executed on a Von Newmann machine. But there isn't any apparent discussion of an alternative machine architecture, just the computer language.

Indeed one of the complaints you sometimes see about functional programming is the amount of memory churn it produces on a Von Newmann architecture.

[carapace]

I'm actually working on that. I've come up with a way to dynamically create dataflow graphs on banks of hardware interconnected by latching sort-nets and programmed by a simple, pure, functional, "concatinative" programming language called Joy.

[jandrese]

Didn't the MIT LISP machines achieve this back in the 80s?