1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/complex.c
marcandre 215b8092a7 * complex.c (nucomp_real_check): raise TypeError instead of ArgumentError
when argument is not a real as expected [ruby-core:28395]

* rational.c (nurat_int_check): ditto (for integers)

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@26805 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2010-03-03 14:18:26 +00:00

2013 lines
42 KiB
C

/*
complex.c: Coded by Tadayoshi Funaba 2008,2009
This implementation is based on Keiju Ishitsuka's Complex library
which is written in ruby.
*/
#include "ruby.h"
#include <math.h>
#define NDEBUG
#include <assert.h>
#define ZERO INT2FIX(0)
#define ONE INT2FIX(1)
#define TWO INT2FIX(2)
VALUE rb_cComplex;
static ID id_abs, id_abs2, id_arg, id_cmp, id_conj, id_convert,
id_denominator, id_divmod, id_eqeq_p, id_expt, id_fdiv, id_floor,
id_idiv, id_imag, id_inspect, id_negate, id_numerator, id_quo,
id_real, id_real_p, id_to_f, id_to_i, id_to_r, id_to_s;
#define f_boolcast(x) ((x) ? Qtrue : Qfalse)
#define binop(n,op) \
inline static VALUE \
f_##n(VALUE x, VALUE y)\
{\
return rb_funcall(x, op, 1, y);\
}
#define fun1(n) \
inline static VALUE \
f_##n(VALUE x)\
{\
return rb_funcall(x, id_##n, 0);\
}
#define fun2(n) \
inline static VALUE \
f_##n(VALUE x, VALUE y)\
{\
return rb_funcall(x, id_##n, 1, y);\
}
#define math1(n) \
inline static VALUE \
m_##n(VALUE x)\
{\
return rb_funcall(rb_mMath, id_##n, 1, x);\
}
#define math2(n) \
inline static VALUE \
m_##n(VALUE x, VALUE y)\
{\
return rb_funcall(rb_mMath, id_##n, 2, x, y);\
}
#define PRESERVE_SIGNEDZERO
inline static VALUE
f_add(VALUE x, VALUE y)
{
#ifndef PRESERVE_SIGNEDZERO
if (FIXNUM_P(y) && FIX2LONG(y) == 0)
return x;
else if (FIXNUM_P(x) && FIX2LONG(x) == 0)
return y;
#endif
return rb_funcall(x, '+', 1, y);
}
inline static VALUE
f_cmp(VALUE x, VALUE y)
{
if (FIXNUM_P(x) && FIXNUM_P(y)) {
long c = FIX2LONG(x) - FIX2LONG(y);
if (c > 0)
c = 1;
else if (c < 0)
c = -1;
return INT2FIX(c);
}
return rb_funcall(x, id_cmp, 1, y);
}
inline static VALUE
f_div(VALUE x, VALUE y)
{
if (FIXNUM_P(y) && FIX2LONG(y) == 1)
return x;
return rb_funcall(x, '/', 1, y);
}
inline static VALUE
f_gt_p(VALUE x, VALUE y)
{
if (FIXNUM_P(x) && FIXNUM_P(y))
return f_boolcast(FIX2LONG(x) > FIX2LONG(y));
return rb_funcall(x, '>', 1, y);
}
inline static VALUE
f_lt_p(VALUE x, VALUE y)
{
if (FIXNUM_P(x) && FIXNUM_P(y))
return f_boolcast(FIX2LONG(x) < FIX2LONG(y));
return rb_funcall(x, '<', 1, y);
}
binop(mod, '%')
inline static VALUE
f_mul(VALUE x, VALUE y)
{
#ifndef PRESERVE_SIGNEDZERO
if (FIXNUM_P(y)) {
long iy = FIX2LONG(y);
if (iy == 0) {
if (FIXNUM_P(x) || TYPE(x) == T_BIGNUM)
return ZERO;
}
else if (iy == 1)
return x;
}
else if (FIXNUM_P(x)) {
long ix = FIX2LONG(x);
if (ix == 0) {
if (FIXNUM_P(y) || TYPE(y) == T_BIGNUM)
return ZERO;
}
else if (ix == 1)
return y;
}
#endif
return rb_funcall(x, '*', 1, y);
}
inline static VALUE
f_sub(VALUE x, VALUE y)
{
#ifndef PRESERVE_SIGNEDZERO
if (FIXNUM_P(y) && FIX2LONG(y) == 0)
return x;
#endif
return rb_funcall(x, '-', 1, y);
}
fun1(abs)
fun1(abs2)
fun1(arg)
fun1(conj)
fun1(denominator)
fun1(floor)
fun1(imag)
fun1(inspect)
fun1(negate)
fun1(numerator)
fun1(real)
fun1(real_p)
fun1(to_f)
fun1(to_i)
fun1(to_r)
fun1(to_s)
fun2(divmod)
inline static VALUE
f_eqeq_p(VALUE x, VALUE y)
{
if (FIXNUM_P(x) && FIXNUM_P(y))
return f_boolcast(FIX2LONG(x) == FIX2LONG(y));
return rb_funcall(x, id_eqeq_p, 1, y);
}
fun2(expt)
fun2(fdiv)
fun2(idiv)
fun2(quo)
inline static VALUE
f_negative_p(VALUE x)
{
if (FIXNUM_P(x))
return f_boolcast(FIX2LONG(x) < 0);
return rb_funcall(x, '<', 1, ZERO);
}
#define f_positive_p(x) (!f_negative_p(x))
inline static VALUE
f_zero_p(VALUE x)
{
switch (TYPE(x)) {
case T_FIXNUM:
return f_boolcast(FIX2LONG(x) == 0);
case T_BIGNUM:
return Qfalse;
case T_RATIONAL:
{
VALUE num = RRATIONAL(x)->num;
return f_boolcast(FIXNUM_P(num) && FIX2LONG(num) == 0);
}
}
return rb_funcall(x, id_eqeq_p, 1, ZERO);
}
#define f_nonzero_p(x) (!f_zero_p(x))
inline static VALUE
f_one_p(VALUE x)
{
switch (TYPE(x)) {
case T_FIXNUM:
return f_boolcast(FIX2LONG(x) == 1);
case T_BIGNUM:
return Qfalse;
case T_RATIONAL:
{
VALUE num = RRATIONAL(x)->num;
VALUE den = RRATIONAL(x)->den;
return f_boolcast(FIXNUM_P(num) && FIX2LONG(num) == 1 &&
FIXNUM_P(den) && FIX2LONG(den) == 1);
}
}
return rb_funcall(x, id_eqeq_p, 1, ONE);
}
inline static VALUE
f_kind_of_p(VALUE x, VALUE c)
{
return rb_obj_is_kind_of(x, c);
}
inline static VALUE
k_numeric_p(VALUE x)
{
return f_kind_of_p(x, rb_cNumeric);
}
inline static VALUE
k_integer_p(VALUE x)
{
return f_kind_of_p(x, rb_cInteger);
}
inline static VALUE
k_fixnum_p(VALUE x)
{
return f_kind_of_p(x, rb_cFixnum);
}
inline static VALUE
k_bignum_p(VALUE x)
{
return f_kind_of_p(x, rb_cBignum);
}
inline static VALUE
k_float_p(VALUE x)
{
return f_kind_of_p(x, rb_cFloat);
}
inline static VALUE
k_rational_p(VALUE x)
{
return f_kind_of_p(x, rb_cRational);
}
inline static VALUE
k_complex_p(VALUE x)
{
return f_kind_of_p(x, rb_cComplex);
}
#define k_exact_p(x) (!k_float_p(x))
#define k_inexact_p(x) k_float_p(x)
#define k_exact_zero_p(x) (k_exact_p(x) && f_zero_p(x))
#define k_exact_one_p(x) (k_exact_p(x) && f_one_p(x))
#define get_dat1(x) \
struct RComplex *dat;\
dat = ((struct RComplex *)(x))
#define get_dat2(x,y) \
struct RComplex *adat, *bdat;\
adat = ((struct RComplex *)(x));\
bdat = ((struct RComplex *)(y))
inline static VALUE
nucomp_s_new_internal(VALUE klass, VALUE real, VALUE imag)
{
NEWOBJ(obj, struct RComplex);
OBJSETUP(obj, klass, T_COMPLEX);
obj->real = real;
obj->imag = imag;
return (VALUE)obj;
}
static VALUE
nucomp_s_alloc(VALUE klass)
{
return nucomp_s_new_internal(klass, ZERO, ZERO);
}
#if 0
static VALUE
nucomp_s_new_bang(int argc, VALUE *argv, VALUE klass)
{
VALUE real, imag;
switch (rb_scan_args(argc, argv, "11", &real, &imag)) {
case 1:
if (!k_numeric_p(real))
real = f_to_i(real);
imag = ZERO;
break;
default:
if (!k_numeric_p(real))
real = f_to_i(real);
if (!k_numeric_p(imag))
imag = f_to_i(imag);
break;
}
return nucomp_s_new_internal(klass, real, imag);
}
#endif
inline static VALUE
f_complex_new_bang1(VALUE klass, VALUE x)
{
assert(!k_complex_p(x));
return nucomp_s_new_internal(klass, x, ZERO);
}
inline static VALUE
f_complex_new_bang2(VALUE klass, VALUE x, VALUE y)
{
assert(!k_complex_p(x));
assert(!k_complex_p(y));
return nucomp_s_new_internal(klass, x, y);
}
#ifdef CANONICALIZATION_FOR_MATHN
#define CANON
#endif
#ifdef CANON
static int canonicalization = 0;
void
nucomp_canonicalization(int f)
{
canonicalization = f;
}
#endif
inline static void
nucomp_real_check(VALUE num)
{
switch (TYPE(num)) {
case T_FIXNUM:
case T_BIGNUM:
case T_FLOAT:
case T_RATIONAL:
break;
default:
if (!k_numeric_p(num) || !f_real_p(num))
rb_raise(rb_eTypeError, "not a real");
}
}
inline static VALUE
nucomp_s_canonicalize_internal(VALUE klass, VALUE real, VALUE imag)
{
#ifdef CANON
#define CL_CANON
#ifdef CL_CANON
if (k_exact_zero_p(imag) && canonicalization)
return real;
#else
if (f_zero_p(imag) && canonicalization)
return real;
#endif
#endif
if (f_real_p(real) && f_real_p(imag))
return nucomp_s_new_internal(klass, real, imag);
else if (f_real_p(real)) {
get_dat1(imag);
return nucomp_s_new_internal(klass,
f_sub(real, dat->imag),
f_add(ZERO, dat->real));
}
else if (f_real_p(imag)) {
get_dat1(real);
return nucomp_s_new_internal(klass,
dat->real,
f_add(dat->imag, imag));
}
else {
get_dat2(real, imag);
return nucomp_s_new_internal(klass,
f_sub(adat->real, bdat->imag),
f_add(adat->imag, bdat->real));
}
}
/*
* call-seq:
* Complex.rect(real[, imag]) -> complex
* Complex.rectangular(real[, imag]) -> complex
*
* Returns a complex object which denotes the given rectangular form.
*/
static VALUE
nucomp_s_new(int argc, VALUE *argv, VALUE klass)
{
VALUE real, imag;
switch (rb_scan_args(argc, argv, "11", &real, &imag)) {
case 1:
nucomp_real_check(real);
imag = ZERO;
break;
default:
nucomp_real_check(real);
nucomp_real_check(imag);
break;
}
return nucomp_s_canonicalize_internal(klass, real, imag);
}
inline static VALUE
f_complex_new1(VALUE klass, VALUE x)
{
assert(!k_complex_p(x));
return nucomp_s_canonicalize_internal(klass, x, ZERO);
}
inline static VALUE
f_complex_new2(VALUE klass, VALUE x, VALUE y)
{
assert(!k_complex_p(x));
return nucomp_s_canonicalize_internal(klass, x, y);
}
/*
* call-seq:
* Complex(x[, y]) -> numeric
*
* Returns x+i*y;
*/
static VALUE
nucomp_f_complex(int argc, VALUE *argv, VALUE klass)
{
return rb_funcall2(rb_cComplex, id_convert, argc, argv);
}
#define imp1(n) \
extern VALUE rb_math_##n(VALUE x);\
inline static VALUE \
m_##n##_bang(VALUE x)\
{\
return rb_math_##n(x);\
}
#define imp2(n) \
extern VALUE rb_math_##n(VALUE x, VALUE y);\
inline static VALUE \
m_##n##_bang(VALUE x, VALUE y)\
{\
return rb_math_##n(x, y);\
}
imp2(atan2)
imp1(cos)
imp1(cosh)
imp1(exp)
imp2(hypot)
#define m_hypot(x,y) m_hypot_bang(x,y)
extern VALUE rb_math_log(int argc, VALUE *argv);
static VALUE
m_log_bang(VALUE x)
{
return rb_math_log(1, &x);
}
imp1(sin)
imp1(sinh)
imp1(sqrt)
static VALUE
m_cos(VALUE x)
{
if (f_real_p(x))
return m_cos_bang(x);
{
get_dat1(x);
return f_complex_new2(rb_cComplex,
f_mul(m_cos_bang(dat->real),
m_cosh_bang(dat->imag)),
f_mul(f_negate(m_sin_bang(dat->real)),
m_sinh_bang(dat->imag)));
}
}
static VALUE
m_sin(VALUE x)
{
if (f_real_p(x))
return m_sin_bang(x);
{
get_dat1(x);
return f_complex_new2(rb_cComplex,
f_mul(m_sin_bang(dat->real),
m_cosh_bang(dat->imag)),
f_mul(m_cos_bang(dat->real),
m_sinh_bang(dat->imag)));
}
}
#if 0
static VALUE
m_sqrt(VALUE x)
{
if (f_real_p(x)) {
if (f_positive_p(x))
return m_sqrt_bang(x);
return f_complex_new2(rb_cComplex, ZERO, m_sqrt_bang(f_negate(x)));
}
else {
get_dat1(x);
if (f_negative_p(dat->imag))
return f_conj(m_sqrt(f_conj(x)));
else {
VALUE a = f_abs(x);
return f_complex_new2(rb_cComplex,
m_sqrt_bang(f_div(f_add(a, dat->real), TWO)),
m_sqrt_bang(f_div(f_sub(a, dat->real), TWO)));
}
}
}
#endif
inline static VALUE
f_complex_polar(VALUE klass, VALUE x, VALUE y)
{
assert(!k_complex_p(x));
assert(!k_complex_p(y));
return nucomp_s_canonicalize_internal(klass,
f_mul(x, m_cos(y)),
f_mul(x, m_sin(y)));
}
/*
* call-seq:
* Complex.polar(abs[, arg]) -> complex
*
* Returns a complex object which denotes the given polar form.
*/
static VALUE
nucomp_s_polar(int argc, VALUE *argv, VALUE klass)
{
VALUE abs, arg;
switch (rb_scan_args(argc, argv, "11", &abs, &arg)) {
case 1:
nucomp_real_check(abs);
arg = ZERO;
break;
default:
nucomp_real_check(abs);
nucomp_real_check(arg);
break;
}
return f_complex_polar(klass, abs, arg);
}
/*
* call-seq:
* cmp.real -> real
*
* Returns the real part.
*/
static VALUE
nucomp_real(VALUE self)
{
get_dat1(self);
return dat->real;
}
/*
* call-seq:
* cmp.imag -> real
* cmp.imaginary -> real
*
* Returns the imaginary part.
*/
static VALUE
nucomp_imag(VALUE self)
{
get_dat1(self);
return dat->imag;
}
/*
* call-seq:
* -cmp -> complex
*
* Returns negation of the value.
*/
static VALUE
nucomp_negate(VALUE self)
{
get_dat1(self);
return f_complex_new2(CLASS_OF(self),
f_negate(dat->real), f_negate(dat->imag));
}
inline static VALUE
f_addsub(VALUE self, VALUE other,
VALUE (*func)(VALUE, VALUE), ID id)
{
if (k_complex_p(other)) {
VALUE real, imag;
get_dat2(self, other);
real = (*func)(adat->real, bdat->real);
imag = (*func)(adat->imag, bdat->imag);
return f_complex_new2(CLASS_OF(self), real, imag);
}
if (k_numeric_p(other) && f_real_p(other)) {
get_dat1(self);
return f_complex_new2(CLASS_OF(self),
(*func)(dat->real, other), dat->imag);
}
return rb_num_coerce_bin(self, other, id);
}
/*
* call-seq:
* cmp + numeric -> complex
*
* Performs addition.
*/
static VALUE
nucomp_add(VALUE self, VALUE other)
{
return f_addsub(self, other, f_add, '+');
}
/*
* call-seq:
* cmp - numeric -> complex
*
* Performs subtraction.
*/
static VALUE
nucomp_sub(VALUE self, VALUE other)
{
return f_addsub(self, other, f_sub, '-');
}
/*
* call-seq:
* cmp * numeric -> complex
*
* Performs multiplication.
*/
static VALUE
nucomp_mul(VALUE self, VALUE other)
{
if (k_complex_p(other)) {
VALUE real, imag;
get_dat2(self, other);
real = f_sub(f_mul(adat->real, bdat->real),
f_mul(adat->imag, bdat->imag));
imag = f_add(f_mul(adat->real, bdat->imag),
f_mul(adat->imag, bdat->real));
return f_complex_new2(CLASS_OF(self), real, imag);
}
if (k_numeric_p(other) && f_real_p(other)) {
get_dat1(self);
return f_complex_new2(CLASS_OF(self),
f_mul(dat->real, other),
f_mul(dat->imag, other));
}
return rb_num_coerce_bin(self, other, '*');
}
inline static VALUE
f_divide(VALUE self, VALUE other,
VALUE (*func)(VALUE, VALUE), ID id)
{
if (k_complex_p(other)) {
int flo;
get_dat2(self, other);
flo = (k_float_p(adat->real) || k_float_p(adat->imag) ||
k_float_p(bdat->real) || k_float_p(bdat->imag));
if (f_gt_p(f_abs(bdat->real), f_abs(bdat->imag))) {
VALUE r, n;
r = (*func)(bdat->imag, bdat->real);
n = f_mul(bdat->real, f_add(ONE, f_mul(r, r)));
if (flo)
return f_complex_new2(CLASS_OF(self),
(*func)(self, n),
(*func)(f_negate(f_mul(self, r)), n));
return f_complex_new2(CLASS_OF(self),
(*func)(f_add(adat->real,
f_mul(adat->imag, r)), n),
(*func)(f_sub(adat->imag,
f_mul(adat->real, r)), n));
}
else {
VALUE r, n;
r = (*func)(bdat->real, bdat->imag);
n = f_mul(bdat->imag, f_add(ONE, f_mul(r, r)));
if (flo)
return f_complex_new2(CLASS_OF(self),
(*func)(f_mul(self, r), n),
(*func)(f_negate(self), n));
return f_complex_new2(CLASS_OF(self),
(*func)(f_add(f_mul(adat->real, r),
adat->imag), n),
(*func)(f_sub(f_mul(adat->imag, r),
adat->real), n));
}
}
if (k_numeric_p(other) && f_real_p(other)) {
get_dat1(self);
return f_complex_new2(CLASS_OF(self),
(*func)(dat->real, other),
(*func)(dat->imag, other));
}
return rb_num_coerce_bin(self, other, id);
}
#define rb_raise_zerodiv() rb_raise(rb_eZeroDivError, "divided by 0")
/*
* call-seq:
* cmp / numeric -> complex
* cmp.quo(numeric) -> complex
*
* Performs division.
*
* For example:
*
* Complex(10.0) / 3 #=> (3.3333333333333335+(0/1)*i)
* Complex(10) / 3 #=> ((10/3)+(0/1)*i) # not (3+0i)
*/
static VALUE
nucomp_div(VALUE self, VALUE other)
{
return f_divide(self, other, f_quo, id_quo);
}
#define nucomp_quo nucomp_div
/*
* call-seq:
* cmp.fdiv(numeric) -> complex
*
* Performs division as each part is a float, never returns a float.
*
* For example:
*
* Complex(11,22).fdiv(3) #=> (3.6666666666666665+7.333333333333333i)
*/
static VALUE
nucomp_fdiv(VALUE self, VALUE other)
{
return f_divide(self, other, f_fdiv, id_fdiv);
}
static VALUE
m_log(VALUE x)
{
if (f_real_p(x) && f_positive_p(x))
return m_log_bang(x);
return rb_complex_new2(m_log_bang(f_abs(x)), f_arg(x));
}
static VALUE
m_exp(VALUE x)
{
VALUE ere, im;
if (f_real_p(x))
return m_exp_bang(x);
ere = m_exp_bang(f_real(x));
im = f_imag(x);
return rb_complex_new2(f_mul(ere, m_cos_bang(im)),
f_mul(ere, m_sin_bang(im)));
}
inline static VALUE
f_reciprocal(VALUE x)
{
return f_quo(ONE, x);
}
/*
* call-seq:
* cmp ** numeric -> complex
*
* Performs exponentiation.
*
* For example:
*
* Complex('i') ** 2 #=> (-1+0i)
* Complex(-8) ** Rational(1,3) #=> (1.0000000000000002+1.7320508075688772i)
*/
static VALUE
nucomp_expt(VALUE self, VALUE other)
{
if (k_exact_zero_p(other))
return f_complex_new_bang1(CLASS_OF(self), ONE);
if (k_rational_p(other) && f_one_p(f_denominator(other)))
other = f_numerator(other); /* c14n */
if (k_complex_p(other)) {
get_dat1(other);
if (k_exact_zero_p(dat->imag))
other = dat->real; /* c14n */
}
if (k_complex_p(other)) {
VALUE r, theta, nr, ntheta;
get_dat1(other);
r = f_abs(self);
theta = f_arg(self);
nr = m_exp_bang(f_sub(f_mul(dat->real, m_log_bang(r)),
f_mul(dat->imag, theta)));
ntheta = f_add(f_mul(theta, dat->real),
f_mul(dat->imag, m_log_bang(r)));
return f_complex_polar(CLASS_OF(self), nr, ntheta);
}
if (k_fixnum_p(other)) {
if (f_gt_p(other, ZERO)) {
VALUE x, z;
long n;
x = self;
z = x;
n = FIX2LONG(other) - 1;
while (n) {
long q, r;
while (1) {
get_dat1(x);
q = n / 2;
r = n % 2;
if (r)
break;
x = f_complex_new2(CLASS_OF(self),
f_sub(f_mul(dat->real, dat->real),
f_mul(dat->imag, dat->imag)),
f_mul(f_mul(TWO, dat->real), dat->imag));
n = q;
}
z = f_mul(z, x);
n--;
}
return z;
}
return f_expt(f_reciprocal(self), f_negate(other));
}
if (k_numeric_p(other) && f_real_p(other)) {
VALUE r, theta;
if (k_bignum_p(other))
rb_warn("in a**b, b may be too big");
r = f_abs(self);
theta = f_arg(self);
return f_complex_polar(CLASS_OF(self), f_expt(r, other),
f_mul(theta, other));
}
return rb_num_coerce_bin(self, other, id_expt);
}
/*
* call-seq:
* cmp == object -> true or false
*
* Returns true if cmp equals object numerically.
*/
static VALUE
nucomp_eqeq_p(VALUE self, VALUE other)
{
if (k_complex_p(other)) {
get_dat2(self, other);
return f_boolcast(f_eqeq_p(adat->real, bdat->real) &&
f_eqeq_p(adat->imag, bdat->imag));
}
if (k_numeric_p(other) && f_real_p(other)) {
get_dat1(self);
return f_boolcast(f_eqeq_p(dat->real, other) && f_zero_p(dat->imag));
}
return f_eqeq_p(other, self);
}
/* :nodoc: */
static VALUE
nucomp_coerce(VALUE self, VALUE other)
{
if (k_numeric_p(other) && f_real_p(other))
return rb_assoc_new(f_complex_new_bang1(CLASS_OF(self), other), self);
if (TYPE(other) == T_COMPLEX)
return rb_assoc_new(other, self);
rb_raise(rb_eTypeError, "%s can't be coerced into %s",
rb_obj_classname(other), rb_obj_classname(self));
return Qnil;
}
/*
* call-seq:
* cmp.abs -> real
* cmp.magnitude -> real
*
* Returns the absolute part of its polar form.
*/
static VALUE
nucomp_abs(VALUE self)
{
get_dat1(self);
if (f_zero_p(dat->real)) {
VALUE a = f_abs(dat->imag);
if (k_float_p(dat->real) && !k_float_p(dat->imag))
a = f_to_f(a);
return a;
}
if (f_zero_p(dat->imag)) {
VALUE a = f_abs(dat->real);
if (!k_float_p(dat->real) && k_float_p(dat->imag))
a = f_to_f(a);
return a;
}
return m_hypot(dat->real, dat->imag);
}
/*
* call-seq:
* cmp.abs2 -> real
*
* Returns square of the absolute value.
*/
static VALUE
nucomp_abs2(VALUE self)
{
get_dat1(self);
return f_add(f_mul(dat->real, dat->real),
f_mul(dat->imag, dat->imag));
}
/*
* call-seq:
* cmp.arg -> float
* cmp.angle -> float
* cmp.phase -> float
*
* Returns the angle part of its polar form.
*/
static VALUE
nucomp_arg(VALUE self)
{
get_dat1(self);
return m_atan2_bang(dat->imag, dat->real);
}
/*
* call-seq:
* cmp.rect -> array
* cmp.rectangular -> array
*
* Returns an array; [cmp.real, cmp.imag].
*/
static VALUE
nucomp_rect(VALUE self)
{
get_dat1(self);
return rb_assoc_new(dat->real, dat->imag);
}
/*
* call-seq:
* cmp.polar -> array
*
* Returns an array; [cmp.abs, cmp.arg].
*/
static VALUE
nucomp_polar(VALUE self)
{
return rb_assoc_new(f_abs(self), f_arg(self));
}
/*
* call-seq:
* cmp.conj -> complex
* cmp.conjugate -> complex
*
* Returns the complex conjugate.
*/
static VALUE
nucomp_conj(VALUE self)
{
get_dat1(self);
return f_complex_new2(CLASS_OF(self), dat->real, f_negate(dat->imag));
}
#if 0
/* :nodoc: */
static VALUE
nucomp_true(VALUE self)
{
return Qtrue;
}
#endif
/*
* call-seq:
* cmp.real? -> false
*
* Returns false.
*/
static VALUE
nucomp_false(VALUE self)
{
return Qfalse;
}
#if 0
/* :nodoc: */
static VALUE
nucomp_exact_p(VALUE self)
{
get_dat1(self);
return f_boolcast(k_exact_p(dat->real) && k_exact_p(dat->imag));
}
/* :nodoc: */
static VALUE
nucomp_inexact_p(VALUE self)
{
return f_boolcast(!nucomp_exact_p(self));
}
#endif
extern VALUE rb_lcm(VALUE x, VALUE y);
/*
* call-seq:
* cmp.denominator -> integer
*
* Returns the denominator (lcm of both denominator, real and imag).
*
* See numerator.
*/
static VALUE
nucomp_denominator(VALUE self)
{
get_dat1(self);
return rb_lcm(f_denominator(dat->real), f_denominator(dat->imag));
}
/*
* call-seq:
* cmp.numerator -> numeric
*
* Returns the numerator.
*
* For example:
*
* 1 2 3+4i <- numerator
* - + -i -> ----
* 2 3 6 <- denominator
*
* c = Complex('1/2+2/3i') #=> ((1/2)+(2/3)*i)
* n = c.numerator #=> (3+4i)
* d = c.denominator #=> 6
* n / d #=> ((1/2)+(2/3)*i)
* Complex(Rational(n.real, d), Rational(n.imag, d))
* #=> ((1/2)+(2/3)*i)
* See denominator.
*/
static VALUE
nucomp_numerator(VALUE self)
{
VALUE cd;
get_dat1(self);
cd = f_denominator(self);
return f_complex_new2(CLASS_OF(self),
f_mul(f_numerator(dat->real),
f_div(cd, f_denominator(dat->real))),
f_mul(f_numerator(dat->imag),
f_div(cd, f_denominator(dat->imag))));
}
/* :nodoc: */
static VALUE
nucomp_hash(VALUE self)
{
st_index_t v, h[2];
VALUE n;
get_dat1(self);
n = rb_hash(dat->real);
h[0] = NUM2LONG(n);
n = rb_hash(dat->imag);
h[1] = NUM2LONG(n);
v = rb_memhash(h, sizeof(h));
return LONG2FIX(v);
}
/* :nodoc: */
static VALUE
nucomp_eql_p(VALUE self, VALUE other)
{
if (k_complex_p(other)) {
get_dat2(self, other);
return f_boolcast((CLASS_OF(adat->real) == CLASS_OF(bdat->real)) &&
(CLASS_OF(adat->imag) == CLASS_OF(bdat->imag)) &&
f_eqeq_p(self, other));
}
return Qfalse;
}
#ifndef HAVE_SIGNBIT
#ifdef signbit
#define HAVE_SIGNBIT 1
#endif
#endif
inline static VALUE
f_signbit(VALUE x)
{
switch (TYPE(x)) {
case T_FLOAT: {
#ifdef HAVE_SIGNBIT
double f = RFLOAT_VALUE(x);
return f_boolcast(!isnan(f) && signbit(f));
#else
char s[2];
double f = RFLOAT_VALUE(x);
if (isnan(f)) return Qfalse;
(void)snprintf(s, sizeof s, "%.0f", f);
return f_boolcast(s[0] == '-');
#endif
}
}
return f_negative_p(x);
}
inline static VALUE
f_tpositive_p(VALUE x)
{
return f_boolcast(!f_signbit(x));
}
static VALUE
f_format(VALUE self, VALUE (*func)(VALUE))
{
VALUE s, impos;
get_dat1(self);
impos = f_tpositive_p(dat->imag);
s = (*func)(dat->real);
rb_str_cat2(s, !impos ? "-" : "+");
rb_str_concat(s, (*func)(f_abs(dat->imag)));
if (!rb_isdigit(RSTRING_PTR(s)[RSTRING_LEN(s) - 1]))
rb_str_cat2(s, "*");
rb_str_cat2(s, "i");
return s;
}
/*
* call-seq:
* cmp.to_s -> string
*
* Returns the value as a string.
*/
static VALUE
nucomp_to_s(VALUE self)
{
return f_format(self, f_to_s);
}
/*
* call-seq:
* cmp.inspect -> string
*
* Returns the value as a string for inspection.
*/
static VALUE
nucomp_inspect(VALUE self)
{
VALUE s;
s = rb_usascii_str_new2("(");
rb_str_concat(s, f_format(self, f_inspect));
rb_str_cat2(s, ")");
return s;
}
/* :nodoc: */
static VALUE
nucomp_marshal_dump(VALUE self)
{
VALUE a;
get_dat1(self);
a = rb_assoc_new(dat->real, dat->imag);
rb_copy_generic_ivar(a, self);
return a;
}
/* :nodoc: */
static VALUE
nucomp_marshal_load(VALUE self, VALUE a)
{
get_dat1(self);
dat->real = RARRAY_PTR(a)[0];
dat->imag = RARRAY_PTR(a)[1];
rb_copy_generic_ivar(self, a);
return self;
}
/* --- */
VALUE
rb_complex_raw(VALUE x, VALUE y)
{
return nucomp_s_new_internal(rb_cComplex, x, y);
}
VALUE
rb_complex_new(VALUE x, VALUE y)
{
return nucomp_s_canonicalize_internal(rb_cComplex, x, y);
}
VALUE
rb_complex_polar(VALUE x, VALUE y)
{
return f_complex_polar(rb_cComplex, x, y);
}
static VALUE nucomp_s_convert(int argc, VALUE *argv, VALUE klass);
VALUE
rb_Complex(VALUE x, VALUE y)
{
VALUE a[2];
a[0] = x;
a[1] = y;
return nucomp_s_convert(2, a, rb_cComplex);
}
/*
* call-seq:
* cmp.to_i -> integer
*
* Returns the value as an integer if possible.
*/
static VALUE
nucomp_to_i(VALUE self)
{
get_dat1(self);
if (k_inexact_p(dat->imag) || f_nonzero_p(dat->imag)) {
VALUE s = f_to_s(self);
rb_raise(rb_eRangeError, "can't convert %s into Integer",
StringValuePtr(s));
}
return f_to_i(dat->real);
}
/*
* call-seq:
* cmp.to_f -> float
*
* Returns the value as a float if possible.
*/
static VALUE
nucomp_to_f(VALUE self)
{
get_dat1(self);
if (k_inexact_p(dat->imag) || f_nonzero_p(dat->imag)) {
VALUE s = f_to_s(self);
rb_raise(rb_eRangeError, "can't convert %s into Float",
StringValuePtr(s));
}
return f_to_f(dat->real);
}
/*
* call-seq:
* cmp.to_r -> rational
*
* Returns the value as a rational if possible.
*/
static VALUE
nucomp_to_r(VALUE self)
{
get_dat1(self);
if (k_inexact_p(dat->imag) || f_nonzero_p(dat->imag)) {
VALUE s = f_to_s(self);
rb_raise(rb_eRangeError, "can't convert %s into Rational",
StringValuePtr(s));
}
return f_to_r(dat->real);
}
/*
* call-seq:
* cmp.rationalize([eps]) -> rational
*
* Returns the value as a rational if possible. An optional argument
* eps is always ignored.
*/
static VALUE
nucomp_rationalize(int argc, VALUE *argv, VALUE self)
{
rb_scan_args(argc, argv, "01", NULL);
return nucomp_to_r(self);
}
/*
* call-seq:
* nil.to_c -> (0+0i)
*
* Returns zero as a complex.
*/
static VALUE
nilclass_to_c(VALUE self)
{
return rb_complex_new1(INT2FIX(0));
}
/*
* call-seq:
* num.to_c -> complex
*
* Returns the value as a complex.
*/
static VALUE
numeric_to_c(VALUE self)
{
return rb_complex_new1(self);
}
static VALUE comp_pat0, comp_pat1, comp_pat2, a_slash, a_dot_and_an_e,
null_string, underscores_pat, an_underscore;
#define WS "\\s*"
#define DIGITS "(?:[0-9](?:_[0-9]|[0-9])*)"
#define NUMERATOR "(?:" DIGITS "?\\.)?" DIGITS "(?:[eE][-+]?" DIGITS ")?"
#define DENOMINATOR DIGITS
#define NUMBER "[-+]?" NUMERATOR "(?:\\/" DENOMINATOR ")?"
#define NUMBERNOS NUMERATOR "(?:\\/" DENOMINATOR ")?"
#define PATTERN0 "\\A" WS "(" NUMBER ")@(" NUMBER ")" WS
#define PATTERN1 "\\A" WS "([-+])?(" NUMBER ")?[iIjJ]" WS
#define PATTERN2 "\\A" WS "(" NUMBER ")(([-+])(" NUMBERNOS ")?[iIjJ])?" WS
static void
make_patterns(void)
{
static const char comp_pat0_source[] = PATTERN0;
static const char comp_pat1_source[] = PATTERN1;
static const char comp_pat2_source[] = PATTERN2;
static const char underscores_pat_source[] = "_+";
if (comp_pat0) return;
comp_pat0 = rb_reg_new(comp_pat0_source, sizeof comp_pat0_source - 1, 0);
rb_gc_register_mark_object(comp_pat0);
comp_pat1 = rb_reg_new(comp_pat1_source, sizeof comp_pat1_source - 1, 0);
rb_gc_register_mark_object(comp_pat1);
comp_pat2 = rb_reg_new(comp_pat2_source, sizeof comp_pat2_source - 1, 0);
rb_gc_register_mark_object(comp_pat2);
a_slash = rb_usascii_str_new2("/");
rb_gc_register_mark_object(a_slash);
a_dot_and_an_e = rb_usascii_str_new2(".eE");
rb_gc_register_mark_object(a_dot_and_an_e);
null_string = rb_usascii_str_new2("");
rb_gc_register_mark_object(null_string);
underscores_pat = rb_reg_new(underscores_pat_source,
sizeof underscores_pat_source - 1, 0);
rb_gc_register_mark_object(underscores_pat);
an_underscore = rb_usascii_str_new2("_");
rb_gc_register_mark_object(an_underscore);
}
#define id_match rb_intern("match")
#define f_match(x,y) rb_funcall(x, id_match, 1, y)
#define id_aref rb_intern("[]")
#define f_aref(x,y) rb_funcall(x, id_aref, 1, y)
#define id_post_match rb_intern("post_match")
#define f_post_match(x) rb_funcall(x, id_post_match, 0)
#define id_split rb_intern("split")
#define f_split(x,y) rb_funcall(x, id_split, 1, y)
#define id_include_p rb_intern("include?")
#define f_include_p(x,y) rb_funcall(x, id_include_p, 1, y)
#define id_count rb_intern("count")
#define f_count(x,y) rb_funcall(x, id_count, 1, y)
#define id_gsub_bang rb_intern("gsub!")
#define f_gsub_bang(x,y,z) rb_funcall(x, id_gsub_bang, 2, y, z)
static VALUE
string_to_c_internal(VALUE self)
{
VALUE s;
s = self;
if (RSTRING_LEN(s) == 0)
return rb_assoc_new(Qnil, self);
{
VALUE m, sr, si, re, r, i;
int po;
m = f_match(comp_pat0, s);
if (!NIL_P(m)) {
sr = f_aref(m, INT2FIX(1));
si = f_aref(m, INT2FIX(2));
re = f_post_match(m);
po = 1;
}
if (NIL_P(m)) {
m = f_match(comp_pat1, s);
if (!NIL_P(m)) {
sr = Qnil;
si = f_aref(m, INT2FIX(1));
if (NIL_P(si))
si = rb_usascii_str_new2("");
{
VALUE t;
t = f_aref(m, INT2FIX(2));
if (NIL_P(t))
t = rb_usascii_str_new2("1");
rb_str_concat(si, t);
}
re = f_post_match(m);
po = 0;
}
}
if (NIL_P(m)) {
m = f_match(comp_pat2, s);
if (NIL_P(m))
return rb_assoc_new(Qnil, self);
sr = f_aref(m, INT2FIX(1));
if (NIL_P(f_aref(m, INT2FIX(2))))
si = Qnil;
else {
VALUE t;
si = f_aref(m, INT2FIX(3));
t = f_aref(m, INT2FIX(4));
if (NIL_P(t))
t = rb_usascii_str_new2("1");
rb_str_concat(si, t);
}
re = f_post_match(m);
po = 0;
}
r = INT2FIX(0);
i = INT2FIX(0);
if (!NIL_P(sr)) {
if (f_include_p(sr, a_slash))
r = f_to_r(sr);
else if (f_gt_p(f_count(sr, a_dot_and_an_e), INT2FIX(0)))
r = f_to_f(sr);
else
r = f_to_i(sr);
}
if (!NIL_P(si)) {
if (f_include_p(si, a_slash))
i = f_to_r(si);
else if (f_gt_p(f_count(si, a_dot_and_an_e), INT2FIX(0)))
i = f_to_f(si);
else
i = f_to_i(si);
}
if (po)
return rb_assoc_new(rb_complex_polar(r, i), re);
else
return rb_assoc_new(rb_complex_new2(r, i), re);
}
}
static VALUE
string_to_c_strict(VALUE self)
{
VALUE a = string_to_c_internal(self);
if (NIL_P(RARRAY_PTR(a)[0]) || RSTRING_LEN(RARRAY_PTR(a)[1]) > 0) {
VALUE s = f_inspect(self);
rb_raise(rb_eArgError, "invalid value for convert(): %s",
StringValuePtr(s));
}
return RARRAY_PTR(a)[0];
}
#define id_gsub rb_intern("gsub")
#define f_gsub(x,y,z) rb_funcall(x, id_gsub, 2, y, z)
/*
* call-seq:
* str.to_c -> complex
*
* Returns a complex which denotes the string form. The parser
* ignores leading whitespaces and trailing garbage. Any digit
* sequences can be separated by an underscore. Returns zero for null
* or garbage string.
*
* For example:
*
* '9'.to_c #=> (9+0i)
* '2.5'.to_c #=> (2.5+0i)
* '2.5/1'.to_c #=> ((5/2)+0i)
* '-3/2'.to_c #=> ((-3/2)+0i)
* '-i'.to_c #=> (0-1i)
* '45i'.to_c #=> (0+45i)
* '3-4i'.to_c #=> (3-4i)
* '-4e2-4e-2i'.to_c #=> (-400.0-0.04i)
* '-0.0-0.0i'.to_c #=> (-0.0-0.0i)
* '1/2+3/4i'.to_c #=> ((1/2)+(3/4)*i)
* 'ruby'.to_c #=> (0+0i)
*/
static VALUE
string_to_c(VALUE self)
{
VALUE s, a, backref;
backref = rb_backref_get();
rb_match_busy(backref);
s = f_gsub(self, underscores_pat, an_underscore);
a = string_to_c_internal(s);
rb_backref_set(backref);
if (!NIL_P(RARRAY_PTR(a)[0]))
return RARRAY_PTR(a)[0];
return rb_complex_new1(INT2FIX(0));
}
static VALUE
nucomp_s_convert(int argc, VALUE *argv, VALUE klass)
{
VALUE a1, a2, backref;
rb_scan_args(argc, argv, "11", &a1, &a2);
if (NIL_P(a1) || (argc == 2 && NIL_P(a2)))
rb_raise(rb_eTypeError, "can't convert nil into Complex");
backref = rb_backref_get();
rb_match_busy(backref);
switch (TYPE(a1)) {
case T_FIXNUM:
case T_BIGNUM:
case T_FLOAT:
break;
case T_STRING:
a1 = string_to_c_strict(a1);
break;
}
switch (TYPE(a2)) {
case T_FIXNUM:
case T_BIGNUM:
case T_FLOAT:
break;
case T_STRING:
a2 = string_to_c_strict(a2);
break;
}
rb_backref_set(backref);
switch (TYPE(a1)) {
case T_COMPLEX:
{
get_dat1(a1);
if (k_exact_zero_p(dat->imag))
a1 = dat->real;
}
}
switch (TYPE(a2)) {
case T_COMPLEX:
{
get_dat1(a2);
if (k_exact_zero_p(dat->imag))
a2 = dat->real;
}
}
switch (TYPE(a1)) {
case T_COMPLEX:
if (argc == 1 || (k_exact_zero_p(a2)))
return a1;
}
if (argc == 1) {
if (k_numeric_p(a1) && !f_real_p(a1))
return a1;
/* expect raise exception for consistency */
if (!k_numeric_p(a1))
return rb_convert_type(a1, T_COMPLEX, "Complex", "to_c");
}
else {
if ((k_numeric_p(a1) && k_numeric_p(a2)) &&
(!f_real_p(a1) || !f_real_p(a2)))
return f_add(a1,
f_mul(a2,
f_complex_new_bang2(rb_cComplex, ZERO, ONE)));
}
{
VALUE argv2[2];
argv2[0] = a1;
argv2[1] = a2;
return nucomp_s_new(argc, argv2, klass);
}
}
/* --- */
/*
* call-seq:
* num.real -> self
*
* Returns self.
*/
static VALUE
numeric_real(VALUE self)
{
return self;
}
/*
* call-seq:
* num.imag -> 0
* num.imaginary -> 0
*
* Returns zero.
*/
static VALUE
numeric_imag(VALUE self)
{
return INT2FIX(0);
}
/*
* call-seq:
* num.abs2 -> real
*
* Returns square of self.
*/
static VALUE
numeric_abs2(VALUE self)
{
return f_mul(self, self);
}
#define id_PI rb_intern("PI")
/*
* call-seq:
* num.arg -> 0 or float
* num.angle -> 0 or float
* num.phase -> 0 or float
*
* Returns 0 if the value is positive, pi otherwise.
*/
static VALUE
numeric_arg(VALUE self)
{
if (f_positive_p(self))
return INT2FIX(0);
return rb_const_get(rb_mMath, id_PI);
}
/*
* call-seq:
* num.rect -> array
*
* Returns an array; [num, 0].
*/
static VALUE
numeric_rect(VALUE self)
{
return rb_assoc_new(self, INT2FIX(0));
}
/*
* call-seq:
* num.polar -> array
*
* Returns an array; [num.abs, num.arg].
*/
static VALUE
numeric_polar(VALUE self)
{
return rb_assoc_new(f_abs(self), f_arg(self));
}
/*
* call-seq:
* num.conj -> self
* num.conjugate -> self
*
* Returns self.
*/
static VALUE
numeric_conj(VALUE self)
{
return self;
}
/*
* call-seq:
* flo.arg -> 0 or float
* flo.angle -> 0 or float
* flo.phase -> 0 or float
*
* Returns 0 if the value is positive, pi otherwise.
*/
static VALUE
float_arg(VALUE self)
{
if (isnan(RFLOAT_VALUE(self)))
return self;
if (f_tpositive_p(self))
return INT2FIX(0);
return rb_const_get(rb_mMath, id_PI);
}
/*
* A complex number can be represented as a paired real number with
* imaginary unit; a+bi. Where a is real part, b is imaginary part
* and i is imaginary unit. Real a equals complex a+0i
* mathematically.
*
* In ruby, you can create complex object with Complex, Complex::rect,
* Complex::polar or to_c method.
*
* Complex(1) #=> (1+0i)
* Complex(2, 3) #=> (2+3i)
* Complex.polar(2, 3) #=> (-1.9799849932008908+0.2822400161197344i)
* 3.to_c #=> (3+0i)
*
* You can also create complex object from floating-point numbers or
* strings.
*
* Complex(0.3) #=> (0.3+0i)
* Complex('0.3-0.5i') #=> (0.3-0.5i)
* Complex('2/3+3/4i') #=> ((2/3)+(3/4)*i)
* Complex('1@2') #=> (-0.4161468365471424+0.9092974268256817i)
*
* 0.3.to_c #=> (0.3+0i)
* '0.3-0.5i'.to_c #=> (0.3-0.5i)
* '2/3+3/4i'.to_c #=> ((2/3)+(3/4)*i)
* '1@2'.to_c #=> (-0.4161468365471424+0.9092974268256817i)
*
* A complex object is either an exact or an inexact number.
*
* Complex(1, 1) / 2 #=> ((1/2)+(1/2)*i)
* Complex(1, 1) / 2.0 #=> (0.5+0.5i)
*/
void
Init_Complex(void)
{
#undef rb_intern
#define rb_intern(str) rb_intern_const(str)
assert(fprintf(stderr, "assert() is now active\n"));
id_abs = rb_intern("abs");
id_abs2 = rb_intern("abs2");
id_arg = rb_intern("arg");
id_cmp = rb_intern("<=>");
id_conj = rb_intern("conj");
id_convert = rb_intern("convert");
id_denominator = rb_intern("denominator");
id_divmod = rb_intern("divmod");
id_eqeq_p = rb_intern("==");
id_expt = rb_intern("**");
id_fdiv = rb_intern("fdiv");
id_floor = rb_intern("floor");
id_idiv = rb_intern("div");
id_imag = rb_intern("imag");
id_inspect = rb_intern("inspect");
id_negate = rb_intern("-@");
id_numerator = rb_intern("numerator");
id_quo = rb_intern("quo");
id_real = rb_intern("real");
id_real_p = rb_intern("real?");
id_to_f = rb_intern("to_f");
id_to_i = rb_intern("to_i");
id_to_r = rb_intern("to_r");
id_to_s = rb_intern("to_s");
rb_cComplex = rb_define_class("Complex", rb_cNumeric);
rb_define_alloc_func(rb_cComplex, nucomp_s_alloc);
rb_undef_method(CLASS_OF(rb_cComplex), "allocate");
#if 0
rb_define_private_method(CLASS_OF(rb_cComplex), "new!", nucomp_s_new_bang, -1);
rb_define_private_method(CLASS_OF(rb_cComplex), "new", nucomp_s_new, -1);
#else
rb_undef_method(CLASS_OF(rb_cComplex), "new");
#endif
rb_define_singleton_method(rb_cComplex, "rectangular", nucomp_s_new, -1);
rb_define_singleton_method(rb_cComplex, "rect", nucomp_s_new, -1);
rb_define_singleton_method(rb_cComplex, "polar", nucomp_s_polar, -1);
rb_define_global_function("Complex", nucomp_f_complex, -1);
rb_undef_method(rb_cComplex, "%");
rb_undef_method(rb_cComplex, "<");
rb_undef_method(rb_cComplex, "<=");
rb_undef_method(rb_cComplex, "<=>");
rb_undef_method(rb_cComplex, ">");
rb_undef_method(rb_cComplex, ">=");
rb_undef_method(rb_cComplex, "between?");
rb_undef_method(rb_cComplex, "div");
rb_undef_method(rb_cComplex, "divmod");
rb_undef_method(rb_cComplex, "floor");
rb_undef_method(rb_cComplex, "ceil");
rb_undef_method(rb_cComplex, "modulo");
rb_undef_method(rb_cComplex, "remainder");
rb_undef_method(rb_cComplex, "round");
rb_undef_method(rb_cComplex, "step");
rb_undef_method(rb_cComplex, "truncate");
rb_undef_method(rb_cComplex, "i");
#if 0 /* NUBY */
rb_undef_method(rb_cComplex, "//");
#endif
rb_define_method(rb_cComplex, "real", nucomp_real, 0);
rb_define_method(rb_cComplex, "imaginary", nucomp_imag, 0);
rb_define_method(rb_cComplex, "imag", nucomp_imag, 0);
rb_define_method(rb_cComplex, "-@", nucomp_negate, 0);
rb_define_method(rb_cComplex, "+", nucomp_add, 1);
rb_define_method(rb_cComplex, "-", nucomp_sub, 1);
rb_define_method(rb_cComplex, "*", nucomp_mul, 1);
rb_define_method(rb_cComplex, "/", nucomp_div, 1);
rb_define_method(rb_cComplex, "quo", nucomp_quo, 1);
rb_define_method(rb_cComplex, "fdiv", nucomp_fdiv, 1);
rb_define_method(rb_cComplex, "**", nucomp_expt, 1);
rb_define_method(rb_cComplex, "==", nucomp_eqeq_p, 1);
rb_define_method(rb_cComplex, "coerce", nucomp_coerce, 1);
rb_define_method(rb_cComplex, "abs", nucomp_abs, 0);
rb_define_method(rb_cComplex, "magnitude", nucomp_abs, 0);
rb_define_method(rb_cComplex, "abs2", nucomp_abs2, 0);
rb_define_method(rb_cComplex, "arg", nucomp_arg, 0);
rb_define_method(rb_cComplex, "angle", nucomp_arg, 0);
rb_define_method(rb_cComplex, "phase", nucomp_arg, 0);
rb_define_method(rb_cComplex, "rectangular", nucomp_rect, 0);
rb_define_method(rb_cComplex, "rect", nucomp_rect, 0);
rb_define_method(rb_cComplex, "polar", nucomp_polar, 0);
rb_define_method(rb_cComplex, "conjugate", nucomp_conj, 0);
rb_define_method(rb_cComplex, "conj", nucomp_conj, 0);
#if 0
rb_define_method(rb_cComplex, "~", nucomp_conj, 0); /* gcc */
#endif
rb_define_method(rb_cComplex, "real?", nucomp_false, 0);
#if 0
rb_define_method(rb_cComplex, "complex?", nucomp_true, 0);
rb_define_method(rb_cComplex, "exact?", nucomp_exact_p, 0);
rb_define_method(rb_cComplex, "inexact?", nucomp_inexact_p, 0);
#endif
rb_define_method(rb_cComplex, "numerator", nucomp_numerator, 0);
rb_define_method(rb_cComplex, "denominator", nucomp_denominator, 0);
rb_define_method(rb_cComplex, "hash", nucomp_hash, 0);
rb_define_method(rb_cComplex, "eql?", nucomp_eql_p, 1);
rb_define_method(rb_cComplex, "to_s", nucomp_to_s, 0);
rb_define_method(rb_cComplex, "inspect", nucomp_inspect, 0);
rb_define_method(rb_cComplex, "marshal_dump", nucomp_marshal_dump, 0);
rb_define_method(rb_cComplex, "marshal_load", nucomp_marshal_load, 1);
/* --- */
rb_define_method(rb_cComplex, "to_i", nucomp_to_i, 0);
rb_define_method(rb_cComplex, "to_f", nucomp_to_f, 0);
rb_define_method(rb_cComplex, "to_r", nucomp_to_r, 0);
rb_define_method(rb_cComplex, "rationalize", nucomp_rationalize, -1);
rb_define_method(rb_cNilClass, "to_c", nilclass_to_c, 0);
rb_define_method(rb_cNumeric, "to_c", numeric_to_c, 0);
make_patterns();
rb_define_method(rb_cString, "to_c", string_to_c, 0);
rb_define_private_method(CLASS_OF(rb_cComplex), "convert", nucomp_s_convert, -1);
/* --- */
rb_define_method(rb_cNumeric, "real", numeric_real, 0);
rb_define_method(rb_cNumeric, "imaginary", numeric_imag, 0);
rb_define_method(rb_cNumeric, "imag", numeric_imag, 0);
rb_define_method(rb_cNumeric, "abs2", numeric_abs2, 0);
rb_define_method(rb_cNumeric, "arg", numeric_arg, 0);
rb_define_method(rb_cNumeric, "angle", numeric_arg, 0);
rb_define_method(rb_cNumeric, "phase", numeric_arg, 0);
rb_define_method(rb_cNumeric, "rectangular", numeric_rect, 0);
rb_define_method(rb_cNumeric, "rect", numeric_rect, 0);
rb_define_method(rb_cNumeric, "polar", numeric_polar, 0);
rb_define_method(rb_cNumeric, "conjugate", numeric_conj, 0);
rb_define_method(rb_cNumeric, "conj", numeric_conj, 0);
rb_define_method(rb_cFloat, "arg", float_arg, 0);
rb_define_method(rb_cFloat, "angle", float_arg, 0);
rb_define_method(rb_cFloat, "phase", float_arg, 0);
rb_define_const(rb_cComplex, "I",
f_complex_new_bang2(rb_cComplex, ZERO, ONE));
}
/*
Local variables:
c-file-style: "ruby"
End:
*/