Thursday 3 January 2013

Functional Programming Whinging

Tim Bray thinks Uncle Bob Martin's post on Functional Programming Basics is the cat's pyjamas.

Meh. "Basics" is the key word in that title - the article is pretty light and fluffy.  Fine if you don't know squat about FP, but it's also accompanied by a whole lot of starry-eyed razzle-dazzle which isn't really justified by the content (and note that I'm not saying it's wrong, just not substantiated).

To be fair, TB does have a few gripes.  Here's a few more:

  • The example used to show how FP wonderfully avoids variables and side-effects is that hoary old one of computing squares of integers.  (I mean really hoary - this was the first program I ever wrote, in WATFIV.  And I at least had cool line printer output!)  How about using something that's a bit more representative of an actual computational problem?  Like say, red-black trees - with deletion!  
  • As TB points out, the people who really need to make algorithms run fast across 64 cores are a small percentage of current coders.  For everyone else, scale-out is a more mundane but pressing problem.  And it's not clear to me whether FP will make that easier.
  • As someone who spends his leisure hours trying to make spatial algorithms more performant, I'm suspicious of anything that promises to automagically make code go faster across multiple cores.  In spatial most interesting problems are not "pleasantly parallel", and many of them are memory-bound as well as being compute-bound.  So advances in performance would seem depend on better algorithms, not a different choice of language.
Back in the day I was pretty keen on FP languages - but I realized after being exposed to Smalltalk and later Java, a lot of their appeal was due to their (necessary) provision of automatic memory management (which was painfully lacking in the "mainstream" languages such as FORTRAN, Pascal, C - oh, and even C++).

But I'm not trying to prove a negative here.  Certainly the FP features of no side-effects and lazy evaluation would seem to offer a lot of benefit for the right class of problems.  And FP or FP-ish languages are more mainstream than ever before.  So perhaps they really will become the mainstream language paradigm.  I just hope I don't have to be coding using layers of inconveniently situated parentheses.

No comments: