patch 7.4.2280v7.4.2280

Problem:    printf() doesn't handle infinity float values correctly.
Solution:   Add a table with possible infinity values. (Dominique Pelle)

1 files changed, 48 insertions, 31 deletions

diff --git a/src/message.c b/src/message.c
index 037d45605..12cb2c41a 100644
--- a/src/message.c
+++ b/src/message.c
@@ -3991,6 +3991,28 @@ tv_float(typval_T *tvs, int *idxp)
 #endif
 
 /*
+ * Return the representation of infinity for printf() function:
+ * "-inf", "inf", "+inf", " inf", "-INF", "INF", "+INF" or " INF".
+ */
+    static const char *
+infinity_str(int positive,
+	     char fmt_spec,
+	     int force_sign,
+	     int space_for_positive)
+{
+    static const char *table[] =
+    {
+	"-inf", "inf", "+inf", " inf",
+	"-INF", "INF", "+INF", " INF"
+    };
+    int idx = positive * (1 + force_sign + force_sign * space_for_positive);
+
+    if (ASCII_ISUPPER(fmt_spec))
+	idx += 4;
+    return table[idx];
+}
+
+/*
  * This code was included to provide a portable vsnprintf() and snprintf().
  * Some systems may provide their own, but we always use this one for
  * consistency.
@@ -4008,8 +4030,8 @@ tv_float(typval_T *tvs, int *idxp)
  *
  * Limited support for floating point was added: 'f', 'e', 'E', 'g', 'G'.
  *
- * Length modifiers 'h' (short int) and 'l' (long int) are supported.
- * 'll' (long long int) is not supported.
+ * Length modifiers 'h' (short int) and 'l' (long int) and 'll' (long long int)
+ * are supported.
  *
  * The locale is not used, the string is used as a byte string.  This is only
  * relevant for double-byte encodings where the second byte may be '%'.
@@ -4397,7 +4419,7 @@ vim_vsnprintf(
 		    uvarnumber_T ullong_arg = 0;
 # endif
 
-		    /* only defined for b convertion */
+		    /* only defined for b conversion */
 		    uvarnumber_T bin_arg = 0;
 
 		    /* pointer argument value -only defined for p
@@ -4702,7 +4724,6 @@ vim_vsnprintf(
 		    char	format[40];
 		    int		l;
 		    int		remove_trailing_zeroes = FALSE;
-		    char	*s;
 
 		    f =
 #  if defined(FEAT_EVAL)
@@ -4732,16 +4753,10 @@ vim_vsnprintf(
 			    )
 		    {
 			/* Avoid a buffer overflow */
-			if (f < 0)
-			{
-			    strcpy(tmp, "-inf");
-			    str_arg_l = 4;
-			}
-			else
-			{
-			    strcpy(tmp, "inf");
-			    str_arg_l = 3;
-			}
+			STRCPY(tmp, infinity_str(f > 0.0, fmt_spec,
+					      force_sign, space_for_positive));
+			str_arg_l = STRLEN(tmp);
+			zero_padding = 0;
 		    }
 		    else
 		    {
@@ -4761,23 +4776,25 @@ vim_vsnprintf(
 			}
 			format[l] = fmt_spec;
 			format[l + 1] = NUL;
-			str_arg_l = sprintf(tmp, format, f);
-
-			/* Be consistent: Change "1.#IND" to "nan" and
-			 * "1.#INF" to "inf". */
-			s = *tmp == '-' ? tmp + 1 : tmp;
-			if (STRNCMP(s, "1.#INF", 6) == 0)
-			    STRCPY(s, "inf");
-			else if (STRNCMP(s, "1.#IND", 6) == 0)
-			    STRCPY(s, "nan");
-
-			/* Remove sign before "nan". */
-			if (STRNCMP(tmp, "-nan", 4) == 0)
-			    STRCPY(tmp, "nan");
-
-			/* Add sign before "inf" if needed. */
-			if (isinf(f) == -1 && STRNCMP(tmp, "inf", 3) == 0)
-			    STRCPY(tmp, "-inf");
+
+			if (isnan(f))
+			{
+			    /* Not a number: nan or NAN */
+			    STRCPY(tmp, ASCII_ISUPPER(fmt_spec) ? "NAN"
+								      : "nan");
+			    str_arg_l = 3;
+			    zero_padding = 0;
+			}
+			else if (isinf(f))
+			{
+			    STRCPY(tmp, infinity_str(f > 0.0, fmt_spec,
+					      force_sign, space_for_positive));
+			    str_arg_l = STRLEN(tmp);
+			    zero_padding = 0;
+			}
+			else
+			    /* Regular float number */
+			    str_arg_l = sprintf(tmp, format, f);
 
 			if (remove_trailing_zeroes)
 			{