Tuesday, June 29, 2010

Weekly notes

Fascinosum & Tremendum

OMG, first IDEs took programming, now they're trying take programming languages. Really, ditch your IDE. Stomp on it. Burn it.

I'm afraid of what will happen when PHBs start to demand DSLs from their Maven-toting programming morons. And it's happening right now! However, since this can only further the Lisp cause in the longer run, I'm quite happy.

GADTs

GADTs are teh hotness. As you know, all Haskellers' favorite pastime is to implement an evaluator for the addition of numbers. There are literally a thousand papers on this topic. With GADTs, they can now exploit the host language type system to make sure that their petty additions are correctly typed. Of course, geeks of other languages are trembling in the face of a number addition gap, so they want to typecheck GADTs in their languages, too. And maybe even throw in a Visitor, or two.

One way of bringing GADT-like power to lesser languages is via type constructor polymorphism, as discussed on The Axis of Eval previously. However, the hordes from the Evil Northwest are proposing a different solution. Instead of higher-kinding type bounds, they are proposing a rather humble where clause. Be assured that your correspondent will report back with further information.

The mother lode of pointer tagging schemes

Nuff said.

Relatedly, from Stress-testing Control Structures for Dynamic Dispatch in Java ('02):
[O]ur results show that strength reduction of control structures is likely to be beneficial regardless of the hardware and JVM, when the number of possible receiver types can be determined to be small. For numbers of possible types up to 4, if sequences are most efficient. Between 4 and 10, binary tree dispatch is generally preferable. For more types, the best implementation is a classical table-based implementation such as currently provided by most JVMs for virtual calls. These are safe, conservative bets, that generally provide a significant improvement and, when not optimal, result only in a small decrease in performance.
And of course, Agner's The microarchitecture of Intel, AMD and VIA CPUs: An optimization guide for assembly programmers and compiler makers.

Hack we must!

No comments: