Q: I'm porting this program, and it calls a routine drand48, which my library doesn't have. What is it?
A: drand48 is a Unix System V routine which returns floating point random numbers (presumably with 48 bits of precision) in the half-open interval [0, 1)[footnote] . (Its companion seed routine is srand48; neither is in the C Standard.) It's easy to write a low-precision replacement:
#include <stdlib.h> double drand48() { return rand() / (RAND_MAX + 1.); }
To more accurately simulate drand48's semantics, you can try to give it closer to 48 bits worth of precision:
#define PRECISION 2.82e14 /* 2**48, rounded up */ double drand48() { double x = 0; double denom = RAND_MAX + 1.; double need; for(need = PRECISION; need > 1; need /= (RAND_MAX + 1.)) { x += rand() / denom; denom *= RAND_MAX + 1.; } return x; }
Before using code like this, though, beware that it is numerically suspect, particularly if (as is usually the case) the period of rand is on the order of RAND_MAX. (If you have a longer-period random number generator available, such as BSD random, definitely use it when simulating drand48.)
References:
PCS Sec. 11 p. 149