Using smaller buffer size for number serialization.

Issue #88 and #61

parson.c

@@ -41,7 +41,9 @@
 
 #define STARTING_CAPACITY 16
 #define MAX_NESTING       2048
-#define FLOAT_FORMAT      "%1.17g"
+
+#define FLOAT_FORMAT "%1.17g" /* do not increase precision without incresing NUM_BUF_SIZE */
+#define NUM_BUF_SIZE 64 /* double printed with "%1.17g" shouldn't be longer than 25 bytes so let's be paranoid and use 64 */
 
 #define SIZEOF_TOKEN(a)       (sizeof(a) - 1)
 #define SKIP_CHAR(str)        ((*str)++)
@@ -1416,7 +1418,7 @@ JSON_Value * json_value_deep_copy(const JSON_Value *value) {
 }
 
 size_t json_serialization_size(const JSON_Value *value) {
-    char num_buf[1100]; /* recursively allocating buffer on stack is a bad idea, so let's do it only once */
+    char num_buf[NUM_BUF_SIZE]; /* recursively allocating buffer on stack is a bad idea, so let's do it only once */
     int res = json_serialize_to_buffer_r(value, NULL, 0, 0, num_buf);
     return res < 0 ? 0 : (size_t)(res + 1);
 }
@@ -1476,7 +1478,7 @@ char * json_serialize_to_string(const JSON_Value *value) {
 }
 
 size_t json_serialization_size_pretty(const JSON_Value *value) {
-    char num_buf[1100]; /* recursively allocating buffer on stack is a bad idea, so let's do it only once */
+    char num_buf[NUM_BUF_SIZE]; /* recursively allocating buffer on stack is a bad idea, so let's do it only once */
     int res = json_serialize_to_buffer_r(value, NULL, 0, 1, num_buf);
     return res < 0 ? 0 : (size_t)(res + 1);
 }