(read1): Don't assume that atof ("-0.0") yields -0.0.
authorPaul Eggert <eggert@twinsun.com>
Mon, 31 Aug 1998 21:18:28 +0000 (21:18 +0000)
committerPaul Eggert <eggert@twinsun.com>
Mon, 31 Aug 1998 21:18:28 +0000 (21:18 +0000)
Handle leading '-' uniformly for zeros, infinities, and NaNs.

src/lread.c

index 8ff3bd4..04f34e8 100644 (file)
@@ -2106,22 +2106,35 @@ read1 (readcharfun, pch, first_in_list)
 #ifdef LISP_FLOAT_TYPE
            if (isfloat_string (read_buffer))
              {
+               /* Compute NaN and infinities using 0.0 in a variable,
+                  to cope with compilers that think they are smarter
+                  than us.  */
                double zero = 0.0;
-               double value = atof (read_buffer);
-               if (read_buffer[0] == '-' && value == 0.0)
-                 value *= -1.0;
-               /* The only way this can be true, after isfloat_string
+
+               double value;
+
+               /* Negate the value ourselves.  This treats 0, NaNs,
+                  and infinity properly on IEEE floating point hosts,
+                  and works around a common bug where atof ("-0.0")
+                  drops the sign.  */
+               int negative = read_buffer[0] == '-';
+
+               /* The only way p[-1] can be 'F' or 'N', after isfloat_string
                   returns 1, is if the input ends in e+INF or e+NaN.  */
-               if (p[-1] == 'F' || p[-1] == 'N')
+               switch (p[-1])
                  {
-                   if (p[-1] == 'N')
-                     value = zero / zero;
-                   else if (read_buffer[0] == '-')
-                     value = - 1.0 / zero;
-                   else
-                     value = 1.0 / zero;
+                 case 'F':
+                   value = 1.0 / zero;
+                   break;
+                 case 'N':
+                   value = zero / zero;
+                   break;
+                 default:
+                   value = atof (read_buffer + negative);
+                   break;
                  }
-               return make_float (value);
+
+               return make_float (negative ? - value : value);
              }
 #endif
          }