Time is the New Memory:
The time problem is not easy to see in today’s mainstream languages because there are no constructs that make time explicit. It is implicit in the system. We don’t even know that’s what we’re doing when we use locks to try to make this work.
I’ve been thinking about how we represent time in programs for a while. The problem is that concurrent programs are all about time, but mostly, we only use two mechanisms to represent it in our programs.
The semi-explicit way is through locks. When we insert locks around some bit of code, we are giving hints to the system that things should only proceed in a certain order. This ordering gives us a notion of time, but it’s not horribly comforting.
The completely implicit way we represent time in our programs is by ordering calls to functions and the lines of code within those functions. Line 10 always executes before line 11, etc.
The problem that Rich Hickey, who has some fantastic ideas about this time stuff, has in the article I quoted is that time is managed manually and implicitly. When you start writing large concurrent programs, this falls apart. We need better constructs to deal with it.
Think of it like the shift from unstructured programming to structured programming to object-oriented programming. At first we just had a long code listing; no functions, just line after line of code. This became mentally untenable, so we shifted to structured, procedural programming. But some of our data was global and it was often hard to tell what functions belong to what data. So we moved to object-oriented programming and encapsulation.
Hopefully Rich Hickey, Simon Peyton-Jones and other functional programming folks can lead us to is a nice way to structure our programs around time. I’m eager to have my brain melted by what they conjure up.
