You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 

339 lines
13 KiB

// read_complex().
// General includes.
#include "cl_sysdep.h"
// Specification.
#include "cl_complex_io.h"
// Implementation.
#include <string.h>
#include "cl_input.h"
#include "cl_real_io.h"
#include "cl_float_io.h"
#include "cl_rational_io.h"
#include "cl_integer_io.h"
#include "cl_integer.h"
#include "cl_I.h"
#include "cl_F.h"
#include "cl_C.h"
#include "cl_abort.h"
#undef floor
#include <math.h>
#define floor cln_floor
// Step forward over all digits, to the end of string or to the next non-digit.
static const char * skip_digits (const char * ptr, const char * string_limit, unsigned int base)
{
for ( ; ptr != string_limit; ptr++) {
var char ch = *ptr;
if ((ch >= '0') && (ch <= '9'))
if (ch < '0' + (int)base)
continue;
else
break;
else {
if (base <= 10)
break;
if (((ch >= 'A') && (ch < 'A'-10+(int)base))
|| ((ch >= 'a') && (ch < 'a'-10+(int)base))
)
continue;
else
break;
}
}
return ptr;
}
// Finish reading the "+yi" part of "x+yi" when "x" has already been read.
static const cl_N read_complex_number_rest (const cl_read_flags& flags, const char * string_rest, const char * string, const char * string_limit, const char * * end_of_parse, const cl_R& x);
#define at_end_of_parse(ptr) \
if (end_of_parse) \
{ *end_of_parse = (ptr); } \
else \
{ if ((ptr) != string_limit) { read_number_junk((ptr),string,string_limit); } }
const cl_N read_complex (const cl_read_flags& flags, const char * string, const char * string_limit, const char * * end_of_parse)
{
// If no string_limit is given, it defaults to the end of the string.
if (!string_limit)
string_limit = string + strlen(string);
if (flags.syntax & syntax_rational) {
// Check for rational number syntax.
var unsigned int rational_base = flags.rational_base;
var const char * ptr = string;
if (flags.lsyntax & lsyntax_commonlisp) {
if (ptr == string_limit) goto not_rational_syntax;
if (*ptr == '#') {
// Check for #b, #o, #x, #nR syntax.
ptr++;
if (ptr == string_limit) goto not_rational_syntax;
switch (*ptr) {
case 'b': case 'B':
rational_base = 2; break;
case 'o': case 'O':
rational_base = 8; break;
case 'x': case 'X':
rational_base = 16; break;
default:
var const char * base_end_ptr =
skip_digits(ptr,string_limit,10);
if (base_end_ptr == ptr) goto not_rational_syntax;
if (base_end_ptr == string_limit) goto not_rational_syntax;
if (!((*base_end_ptr == 'r') || (*base_end_ptr == 'R')))
goto not_rational_syntax;
var cl_I base = read_integer(10,0,ptr,0,base_end_ptr-ptr);
if (!((base >= 2) && (base <= 36))) {
fprint(cl_stderr, "Base must be an integer in the range from 2 to 36, not ");
fprint(cl_stderr, base);
fprint(cl_stderr, "\n");
cl_abort();
}
rational_base = FN_to_UL(base); ptr = base_end_ptr;
break;
}
ptr++;
}
}
var const char * ptr_after_prefix = ptr;
var cl_signean sign = 0;
if (ptr == string_limit) goto not_rational_syntax;
switch (*ptr) {
case '-': sign = ~sign;
case '+': ptr++;
default: break;
}
var const char * ptr_after_sign = ptr;
if (flags.syntax & syntax_integer) {
// Check for integer syntax: {'+'|'-'|} {digit}+ {'.'|}
// Allow final dot only in Common Lisp syntax if there was no #<base> prefix.
if ((flags.lsyntax & lsyntax_commonlisp) && (ptr_after_prefix == string)) {
ptr = skip_digits(ptr_after_sign,string_limit,10);
if (ptr != ptr_after_sign)
if (ptr != string_limit)
if (*ptr == '.') {
ptr++;
if ((ptr == string_limit) || !(((*ptr >= '0') && (*ptr <= '9')) || ((*ptr >= 'A') && (*ptr <= 'Z') && (*ptr != 'I')) || ((*ptr >= 'a') && (*ptr <= 'z') && (*ptr != 'i')) || (*ptr == '.') || (*ptr == '_') || (*ptr == '/')))
return read_complex_number_rest(flags,ptr,string,string_limit,end_of_parse,
read_integer(10,sign,ptr_after_sign,0,ptr-ptr_after_sign));
}
}
ptr = skip_digits(ptr_after_sign,string_limit,rational_base);
if ((ptr == string_limit) || !(((*ptr >= '0') && (*ptr <= '9')) || ((*ptr >= 'A') && (*ptr <= 'Z') && (*ptr != 'I')) || ((*ptr >= 'a') && (*ptr <= 'z') && (*ptr != 'i')) || (*ptr == '.') || (*ptr == '_') || (*ptr == '/')))
return read_complex_number_rest(flags,ptr,string,string_limit,end_of_parse,
read_integer(rational_base,sign,ptr_after_sign,0,ptr-ptr_after_sign));
}
if (flags.syntax & syntax_ratio) {
// Check for ratio syntax: {'+'|'-'|} {digit}+ '/' {digit}+
ptr = skip_digits(ptr_after_sign,string_limit,rational_base);
if (ptr != ptr_after_sign)
if (ptr != string_limit)
if (*ptr == '/') {
var const char * ptr_at_slash = ptr;
ptr = skip_digits(ptr_at_slash+1,string_limit,rational_base);
if (ptr != ptr_at_slash+1)
if ((ptr == string_limit) || !(((*ptr >= '0') && (*ptr <= '9')) || ((*ptr >= 'A') && (*ptr <= 'Z') && (*ptr != 'I')) || ((*ptr >= 'a') && (*ptr <= 'z') && (*ptr != 'i')) || (*ptr == '.') || (*ptr == '_') || (*ptr == '/')))
return read_complex_number_rest(flags,ptr,string,string_limit,end_of_parse,
read_rational(rational_base,sign,ptr_after_sign,0,ptr_at_slash-ptr_after_sign,ptr-ptr_after_sign));
}
}
}
not_rational_syntax:
if (flags.syntax & syntax_float) {
// Check for floating-point number syntax:
// {'+'|'-'|} {digit}+ {'.' {digit}* | } expo {'+'|'-'|} {digit}+
// {'+'|'-'|} {digit}* '.' {digit}+ expo {'+'|'-'|} {digit}+
// {'+'|'-'|} {digit}* '.' {digit}+
var const char * ptr = string;
var const unsigned int float_base = 10;
var cl_signean sign = 0;
if (ptr == string_limit) goto not_float_syntax;
switch (*ptr) {
case '-': sign = ~sign;
case '+': ptr++;
default: break;
}
var const char * ptr_after_sign = ptr;
var const char * ptr_after_intpart = skip_digits(ptr_after_sign,string_limit,float_base);
var cl_boolean have_dot = cl_false;
var const char * ptr_before_fracpart = ptr_after_intpart;
var const char * ptr_after_fracpart = ptr_after_intpart;
ptr = ptr_after_intpart;
if (ptr != string_limit)
if (*ptr == '.') {
have_dot = cl_true;
ptr_before_fracpart = ptr+1;
ptr_after_fracpart = skip_digits(ptr_before_fracpart,string_limit,float_base);
}
ptr = ptr_after_fracpart;
var char exponent_marker;
var cl_boolean have_exponent;
var const char * ptr_in_exponent = ptr;
var const char * ptr_after_exponent = ptr;
if ((ptr == string_limit) || !(((*ptr >= '0') && (*ptr <= '9')) || ((*ptr >= 'A') && (*ptr <= 'Z') && (*ptr != 'I')) || ((*ptr >= 'a') && (*ptr <= 'z') && (*ptr != 'i')) || (*ptr == '.') || (*ptr == '/'))) {
// No exponent.
have_exponent = cl_false;
// Must have at least one fractional part digit.
if (ptr_after_fracpart == ptr_before_fracpart) goto not_float_syntax;
exponent_marker = 'E';
} else {
have_exponent = cl_true;
// Must have at least one digit.
if (ptr_after_sign == ptr_after_intpart)
if (ptr_after_fracpart == ptr_before_fracpart)
goto not_float_syntax;
exponent_marker = ((*ptr >= 'a') && (*ptr <= 'z') ? *ptr - 'a' + 'A' : *ptr);
switch (exponent_marker) {
case 'E':
case 'S': case 'F': case 'D': case 'L':
break;
default:
goto not_float_syntax;
}
}
if (have_exponent) {
ptr++;
if (ptr == string_limit) goto not_float_syntax;
switch (*ptr) {
case '-':
case '+': ptr++;
default: break;
}
ptr_in_exponent = ptr;
ptr_after_exponent = skip_digits(ptr_in_exponent,string_limit,10);
if (ptr_after_exponent == ptr_in_exponent) goto not_float_syntax;
}
ptr = ptr_after_exponent;
var const char * ptr_after_prec = ptr;
var cl_float_format_t prec;
if ((ptr != string_limit) && (*ptr == '_')) {
ptr++;
ptr_after_prec = skip_digits(ptr,string_limit,10);
if (ptr_after_prec == ptr) goto not_float_syntax;
var cl_I prec1 = digits_to_I(ptr,ptr_after_prec-ptr,10);
var uintL prec2 = cl_I_to_UL(prec1);
prec = (float_base==10 ? cl_float_format(prec2)
: (cl_float_format_t)((uintL)((1+prec2)*log((double)float_base)*1.442695041)+1)
);
} else {
switch (exponent_marker) {
case 'S': prec = cl_float_format_sfloat; break;
case 'F': prec = cl_float_format_ffloat; break;
case 'D': prec = cl_float_format_dfloat; break;
case 'L': prec = flags.float_flags.default_lfloat_format; break;
case 'E': prec = flags.float_flags.default_float_format; break;
default: NOTREACHED
}
if (flags.float_flags.mantissa_dependent_float_format) {
// Count the number of significant digits.
ptr = ptr_after_sign;
while (ptr < ptr_after_fracpart && (*ptr == '0' || *ptr == '.')) ptr++;
var uintL num_significant_digits =
(ptr_after_fracpart - ptr) - (ptr_before_fracpart > ptr ? 1 : 0);
var uintL prec2 = (num_significant_digits>=2 ? num_significant_digits-2 : 0);
var cl_float_format_t precx =
(float_base==10 ? cl_float_format(prec2)
: (cl_float_format_t)((uintL)((1+prec2)*log((double)float_base)*1.442695041)+1)
);
if ((uintL)precx > (uintL)prec)
prec = precx;
}
}
floatformatcase(prec
, if (!(flags.syntax & syntax_sfloat)) goto not_float_syntax;
, if (!(flags.syntax & syntax_ffloat)) goto not_float_syntax;
, if (!(flags.syntax & syntax_dfloat)) goto not_float_syntax;
, unused len;
if (!(flags.syntax & syntax_lfloat)) goto not_float_syntax;
);
return read_complex_number_rest(flags,ptr_after_prec,string,string_limit,end_of_parse,
read_float(float_base,prec,sign,ptr_after_sign,0,ptr_after_fracpart-ptr_after_sign,ptr_after_exponent-ptr_after_sign,ptr_before_fracpart-ptr_after_sign));
}
not_float_syntax:
if ((flags.syntax & syntax_complex) && (flags.lsyntax & lsyntax_commonlisp)) {
// Check for complex number syntax:
// '#' {'C'|'c'} '(' realpart {' '}+ imagpart ')'
var const char * ptr = string;
if (ptr == string_limit) goto not_complex_syntax;
if (!(*ptr == '#')) goto not_complex_syntax;
ptr++;
if (ptr == string_limit) goto not_complex_syntax;
if (!((*ptr == 'C') || (*ptr == 'c'))) goto not_complex_syntax;
ptr++;
// Modified flags for parsing the realpart and imagpart:
var cl_read_flags flags_for_parts = flags;
flags_for_parts.syntax = (cl_read_syntax_t)((flags_for_parts.syntax & ~syntax_complex) | syntax_maybe_bad);
var const char * end_of_part;
if (ptr == string_limit) goto not_complex_syntax;
if (!(*ptr == '(')) goto not_complex_syntax;
ptr++;
var cl_R realpart = read_real(flags_for_parts,ptr,string_limit,&end_of_part);
if (end_of_part == ptr) goto not_complex_syntax;
ptr = end_of_part;
if (ptr == string_limit) goto not_complex_syntax;
if (!(*ptr == ' ')) goto not_complex_syntax;
ptr++;
while ((ptr != string_limit) && (*ptr == ' ')) { ptr++; }
var cl_R imagpart = read_real(flags_for_parts,ptr,string_limit,&end_of_part);
if (end_of_part == ptr) goto not_complex_syntax;
ptr = end_of_part;
if (ptr == string_limit) goto not_complex_syntax;
if (!(*ptr == ')')) goto not_complex_syntax;
ptr++;
at_end_of_parse(ptr);
return complex(realpart,imagpart);
}
not_complex_syntax:
bad_syntax:
if (flags.syntax & syntax_maybe_bad) {
ASSERT(end_of_parse);
*end_of_parse = string;
return 0; // dummy return
}
read_number_bad_syntax(string,string_limit);
}
static const cl_N read_complex_number_rest (const cl_read_flags& flags, const char * string_rest, const char * string, const char * string_limit, const char * * end_of_parse, const cl_R& x)
{
unused string;
if ((flags.syntax & syntax_complex) && (flags.lsyntax & lsyntax_algebraic)) {
// Finish reading the "+yi" part of "x+yi".
// We allow "y" to begin with a '-'.
// We also allow the '+' to be replaced by '-', but in this case
// "y" may not begin with a '-'.
// We also allow the syntax "xi" (implicit realpart = 0).
var const char * ptr = string_rest;
if (ptr == string_limit) goto not_complex_syntax;
if ((*ptr == 'i') || (*ptr == 'I')) {
ptr++;
at_end_of_parse(ptr);
return complex(0,x);
}
switch (*ptr) {
case '+': ptr++;
case '-': break;
default: goto not_complex_syntax;
}
// Modified flags for parsing the imagpart:
var cl_read_flags flags_for_part = flags;
flags_for_part.syntax = (cl_read_syntax_t)((flags_for_part.syntax & ~syntax_complex) | syntax_maybe_bad);
var const char * end_of_part;
var const cl_R& realpart = x;
var cl_R imagpart = read_real(flags_for_part,ptr,string_limit,&end_of_part);
if (end_of_part == ptr) goto not_complex_syntax;
ptr = end_of_part;
if (ptr == string_limit) goto not_complex_syntax;
if (!((*ptr == 'i') || (*ptr == 'I'))) goto not_complex_syntax;
ptr++;
at_end_of_parse(ptr);
return complex(realpart,imagpart);
}
not_complex_syntax:
at_end_of_parse(string_rest);
return x;
}