lib/c-strtod.c

   1 /* Convert string to floating-point number, using the C locale.
   2
   3    Copyright (C) 2003-2004, 2006, 2009-2024 Free Software Foundation, Inc.
   4
   5    This program is free software: you can redistribute it and/or modify
   6    it under the terms of the GNU General Public License as published by
   7    the Free Software Foundation, either version 3 of the License, or
   8    (at your option) any later version.
   9
  10    This program is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13    GNU General Public License for more details.
  14
  15    You should have received a copy of the GNU General Public License
  16    along with this program.  If not, see <https://www.gnu.org/licenses/>.  */
  17
  18 /* Written by Paul Eggert and Bruno Haible.  */
  19
  20 #include <config.h>
  21
  22 #include "c-strtod.h"
  23
  24 #include <errno.h>
  25 #include <locale.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28
  29 #if FLOAT
  30 # define C_STRTOD c_strtof
  31 # define DOUBLE float
  32 # define STRTOD_L strtof_l
  33 # define HAVE_GOOD_STRTOD_L (HAVE_STRTOF_L && !GNULIB_defined_strtof_function)
  34 # define STRTOD strtof
  35 #elif LONG
  36 # define C_STRTOD c_strtold
  37 # define DOUBLE long double
  38 # define STRTOD_L strtold_l
  39 # define HAVE_GOOD_STRTOD_L (HAVE_STRTOLD_L && !GNULIB_defined_strtold_function)
  40 # define STRTOD strtold
  41 #else
  42 # define C_STRTOD c_strtod
  43 # define DOUBLE double
  44 # define STRTOD_L strtod_l
  45 # define HAVE_GOOD_STRTOD_L (HAVE_STRTOD_L && !GNULIB_defined_strtod_function)
  46 # define STRTOD strtod
  47 #endif
  48
  49 #if defined LC_ALL_MASK && (HAVE_GOOD_STRTOD_L || HAVE_WORKING_USELOCALE)
  50
  51 /* Cache for the C locale object.
  52    Marked volatile so that different threads see the same value
  53    (avoids locking).  */
  54 static volatile locale_t c_locale_cache;
  55
  56 /* Return the C locale object, or (locale_t) 0 with errno set
  57    if it cannot be created.  */
  58 static locale_t
  59 c_locale (void)
  60 {
  61   if (!c_locale_cache)
  62     c_locale_cache = newlocale (LC_ALL_MASK, "C", (locale_t) 0);
  63   return c_locale_cache;
  64 }
  65
  66 #else
  67
  68 # if HAVE_NL_LANGINFO
  69 #  include <langinfo.h>
  70 # endif
  71 # include "c-ctype.h"
  72
  73 /* Determine the decimal-point character according to the current locale.  */
  74 static char
  75 decimal_point_char (void)
  76 {
  77   const char *point;
  78   /* Determine it in a multithread-safe way.  We know nl_langinfo is
  79      multithread-safe on glibc systems and Mac OS X systems, but is not required
  80      to be multithread-safe by POSIX.  sprintf(), however, is multithread-safe.
  81      localeconv() is rarely multithread-safe.  */
  82 # if HAVE_NL_LANGINFO && (__GLIBC__ || defined __UCLIBC__ || (defined __APPLE__ && defined __MACH__))
  83   point = nl_langinfo (RADIXCHAR);
  84 # elif 1
  85   char pointbuf[5];
  86   sprintf (pointbuf, "%#.0f", 1.0);
  87   point = &pointbuf[1];
  88 # else
  89   point = localeconv () -> decimal_point;
  90 # endif
  91   /* The decimal point is always a single byte: either '.' or ','.  */
  92   return (point[0] != '\0' ? point[0] : '.');
  93 }
  94
  95 #endif
  96
  97 DOUBLE
  98 C_STRTOD (char const *nptr, char **endptr)
  99 {
 100   DOUBLE r;
 101
 102 #if defined LC_ALL_MASK && (HAVE_GOOD_STRTOD_L || HAVE_WORKING_USELOCALE)
 103
 104   locale_t locale = c_locale ();
 105   if (!locale)
 106     {
 107       if (endptr)
 108         *endptr = (char *) nptr;
 109       return 0; /* errno is set here */
 110     }
 111
 112 # if HAVE_GOOD_STRTOD_L
 113
 114   r = STRTOD_L (nptr, endptr, locale);
 115
 116 # else /* HAVE_WORKING_USELOCALE */
 117
 118   locale_t old_locale = uselocale (locale);
 119   if (old_locale == (locale_t)0)
 120     {
 121       if (endptr)
 122         *endptr = (char *) nptr;
 123       return 0; /* errno is set here */
 124     }
 125
 126   r = STRTOD (nptr, endptr);
 127
 128   int saved_errno = errno;
 129   if (uselocale (old_locale) == (locale_t)0)
 130     /* We can't switch back to the old locale.  The thread is hosed.  */
 131     abort ();
 132   errno = saved_errno;
 133
 134 # endif
 135
 136 #else
 137
 138   char decimal_point = decimal_point_char ();
 139
 140   if (decimal_point == '.')
 141     /* In this locale, C_STRTOD and STRTOD behave the same way.  */
 142     r = STRTOD (nptr, endptr);
 143   else
 144     {
 145       /* Start and end of the floating-point number.  */
 146       char const *start;
 147       char const *end;
 148       /* Position of the decimal point '.' in the floating-point number.
 149          Either decimal_point_p == NULL or start <= decimal_point_p < end.  */
 150       char const *decimal_point_p = NULL;
 151       /* Set to true if we encountered decimal_point while parsing.  */
 152       int seen_decimal_point = 0;
 153
 154       /* Parse
 155          1. a sequence of white-space characters,
 156          2. a subject sequence possibly containing a floating-point number,
 157          as described in POSIX
 158          <https://pubs.opengroup.org/onlinepubs/9699919799/functions/strtod.html>.
 159        */
 160       {
 161         char const *p;
 162
 163         p = nptr;
 164
 165         /* Parse a sequence of white-space characters.  */
 166         while (*p != '\0' && c_isspace ((unsigned char) *p))
 167           p++;
 168         start = p;
 169         end = p;
 170
 171         /* Start to parse a subject sequence.  */
 172         if (*p == '+' || *p == '-')
 173           p++;
 174         if (*p == '0')
 175           {
 176             end = p + 1;
 177             if (p[1] == 'x' || p[1] == 'X')
 178               {
 179                 size_t num_hex_digits = 0;
 180                 p += 2;
 181                 /* Parse a non-empty sequence of hexadecimal digits optionally
 182                    containing the decimal point character '.'.  */
 183                 while (*p != '\0')
 184                   {
 185                     if (c_isxdigit ((unsigned char) *p))
 186                       {
 187                         num_hex_digits++;
 188                         p++;
 189                       }
 190                     else if (*p == '.')
 191                       {
 192                         if (decimal_point_p == NULL)
 193                           {
 194                             decimal_point_p = p;
 195                             p++;
 196                           }
 197                         else
 198                           break;
 199                       }
 200                     else
 201                       break;
 202                   }
 203                 seen_decimal_point = (*p == decimal_point);
 204                 if (num_hex_digits > 0)
 205                   {
 206                     end = p;
 207                     /* Parse the character 'p' or the character 'P', optionally
 208                        followed by a '+' or '-' character, and then followed by
 209                        one or more decimal digits.  */
 210                     if (*p == 'p' || *p == 'P')
 211                       {
 212                         p++;
 213                         if (*p == '+' || *p == '-')
 214                           p++;
 215                         if (*p != '\0' && c_isdigit ((unsigned char) *p))
 216                           {
 217                             p++;
 218                             while (*p != '\0' && c_isdigit ((unsigned char) *p))
 219                               p++;
 220                             end = p;
 221                           }
 222                       }
 223                   }
 224                 else
 225                   decimal_point_p = NULL;
 226                 goto done_parsing;
 227               }
 228           }
 229         {
 230           size_t num_digits = 0;
 231           /* Parse a non-empty sequence of decimal digits optionally containing
 232              the decimal point character '.'.  */
 233           while (*p != '\0')
 234             {
 235               if (c_isdigit ((unsigned char) *p))
 236                 {
 237                   num_digits++;
 238                   p++;
 239                 }
 240               else if (*p == '.')
 241                 {
 242                   if (decimal_point_p == NULL)
 243                     {
 244                       decimal_point_p = p;
 245                       p++;
 246                     }
 247                   else
 248                     break;
 249                 }
 250               else
 251                 break;
 252             }
 253           seen_decimal_point = (*p == decimal_point);
 254           if (num_digits > 0)
 255             {
 256               end = p;
 257               /* Parse the character 'e' or the character 'E', optionally
 258                  followed by a '+' or '-' character, and then followed by one or
 259                  more decimal digits.  */
 260               if (*p == 'e' || *p == 'E')
 261                 {
 262                   p++;
 263                   if (*p == '+' || *p == '-')
 264                     p++;
 265                   if (*p != '\0' && c_isdigit ((unsigned char) *p))
 266                     {
 267                       p++;
 268                       while (*p != '\0' && c_isdigit ((unsigned char) *p))
 269                         p++;
 270                       end = p;
 271                     }
 272                 }
 273             }
 274           else
 275             decimal_point_p = NULL;
 276         }
 277       }
 278      done_parsing:
 279       if (end == start || (decimal_point_p == NULL && !seen_decimal_point))
 280         /* If end == start, we have not parsed anything.  The given string
 281            might be "INF", "INFINITY", "NAN", "NAN(chars)", or invalid.
 282            We can pass it to STRTOD.
 283            If end > start and decimal_point_p == NULL, we have parsed a
 284            floating-point number, and it does not have a '.' (and not a ','
 285            either, of course).  If !seen_decimal_point, we did not
 286            encounter decimal_point while parsing.  In this case, the
 287            locale-dependent STRTOD function can handle it.  */
 288         r = STRTOD (nptr, endptr);
 289       else
 290         {
 291           /* Create a modified floating-point number, in which the character '.'
 292              is replaced with the locale-dependent decimal_point.  */
 293           size_t len = end - start;
 294           char *buf;
 295           char *buf_malloced = NULL;
 296           char stackbuf[1000];
 297           char *end_in_buf;
 298
 299           if (len < sizeof (stackbuf))
 300             buf = stackbuf;
 301           else
 302             {
 303               buf = malloc (len + 1);
 304               if (buf == NULL)
 305                 {
 306                   /* Out of memory.
 307                      Callers may not be prepared to see errno == ENOMEM.  But
 308                      they should be prepared to see errno == EINVAL.  */
 309                   errno = EINVAL;
 310                   if (endptr != NULL)
 311                     *endptr = (char *) nptr;
 312                   return 0;
 313                 }
 314               buf_malloced = buf;
 315             }
 316
 317           memcpy (buf, start, len);
 318           if (decimal_point_p != NULL)
 319             buf[decimal_point_p - start] = decimal_point;
 320           buf[len] = '\0';
 321
 322           r = STRTOD (buf, &end_in_buf);
 323           if (endptr != NULL)
 324             *endptr = (char *) (start + (end_in_buf - buf));
 325
 326           if (buf_malloced != NULL)
 327             {
 328               int saved_errno = errno;
 329               free (buf_malloced);
 330               errno = saved_errno;
 331             }
 332         }
 333     }
 334
 335 #endif
 336
 337   return r;
 338 }