This may be lower-level, with all the explicit type conversions, but it doesn't look much more complicated to me. Maybe I'm fooling myself?
variable a variable b variable c fvariable discriminant
: b^2 b @ dup * ; : 4ac 4 a @ c @ * * ; : s>f s>d d>f ;
: -b b @ negate s>f ; : /2a a @ 2* s>f f/ ; : f. f>d d. ;
: quadeq c ! b ! a ! b^2 4ac - s>f fsqrt discriminant f!
-b discriminant f@ f+ /2a
-b discriminant f@ f- /2a ;
Usage:
: kragen@inexorable:~/devel/inexorable-misc ; gforth quadeq.fs
Gforth 0.6.2, Copyright (C) 1995-2003 Free Software Foundation, Inc.
Gforth comes with ABSOLUTELY NO WARRANTY; for details type `license'
Type `bye' to exit
2 4 -30 quadeq f. f. -5. 3. ok
I'd dispense with the floating-point nonsense altogether, but ANS Forth, like C, only defines a square-root routine in floating-point. It's a little more code to define in Forth than in C:
variable n : improve n @ over / + 2/ ; : far? - abs 1 > ;
: sqrt dup n ! dup if begin dup improve swap over far? while repeat then
dup dup * n @ > if 1- then ;
But that hardly seems like a strong recommendation.
Edit: someone else's cute Newton's-method square-root is at http://jasonwoof.org/sqrt.fs --- it looks quite a bit simpler!
On further inspection, the jasonwoof code is quite a bit simpler not only because it has a better approach (which it does!) but also because it leaves out the checks for 0 and off-by-one. So it crashes on 0 and gives the wrong answer for 6, saying 3 when the answer correct to two places is 2.45. You can correct the wrong answers with the same dup dup * n @ > if 1- then business that I was using, assuming you want floor(sqrt(n)) instead of, say, round(sqrt(n)) or ceil(sqrt(n)).
Edit: someone else's cute Newton's-method square-root is at http://jasonwoof.org/sqrt.fs --- it looks quite a bit simpler!