Planet JFX

Numeric Types[]

In attendance: Joe Darcy, Lubo, Brian, Per, and Robert.

Joe's comments --

  • long to/from double is a major source of conversion headaches
  • float is too low of precision to be useful
  • single numeric type (double) is a nightmare: array indexed by 2.5; isPrime, etc
  • smaller numeric types (short, byte, etc) are not "true" JVM types -- and have little use in the space
  • the combination of double and int is a good choice in the space -- provides good interop with libraries and best performance
  • alternative is double and infinite precision integer
  • infinite precision integer could be achieved with an extra Object field and two xor tests for overflow

Group consensus --

  • double and int used now
  • decision on infinite precision integer can be deferred without architectural impact
  • idea that infinite precision integer could be optionally supported (for small devices) was rejected on WORA grounds


Back to OpenJFX Compiler

Comments[]

I'd note that if you're accessing an array by index in jfx you're probably avoiding more set-oriented, declarative equivalents to something. It seems to me that it should be substantially easier for a runtime to optimize set notation and functions, versus procedural code.

var numbers = [ 1,4,7,4,3,7,9];
var odds = select n from numbers where n % 2 == 1;

Rossjudson 21:37, 1 March 2008 (UTC)