Customized sprintf,is it worth the effort? - c

Question

The code below is extracted from nginx,which basically rewrites sprintf,in fact nginx also rewrites some other string functions,is it worth the effort?
u_char *
ngx_vslprintf(u_char *buf, u_char *last, const char *fmt, va_list args)
{
u_char *p, zero;
int d;
double f, scale;
size_t len, slen;
int64_t i64;
uint64_t ui64;
ngx_msec_t ms;
ngx_uint_t width, sign, hex, max_width, frac_width, n;
ngx_str_t *v;
ngx_variable_value_t *vv;
while (*fmt && buf < last) {
/*
* "buf < last" means that we could copy at least one character:
* the plain character, "%%", "%c", and minus without the checking
*/
if (*fmt == '%') {
i64 = 0;
ui64 = 0;
zero = (u_char) ((*++fmt == '0') ? '0' : ' ');
width = 0;
sign = 1;
hex = 0;
max_width = 0;
frac_width = 0;
slen = (size_t) -1;
while (*fmt >= '0' && *fmt <= '9') {
width = width * 10 + *fmt++ - '0';
}
for ( ;; ) {
switch (*fmt) {
case 'u':
sign = 0;
fmt++;
continue;
case 'm':
max_width = 1;
fmt++;
continue;
case 'X':
hex = 2;
sign = 0;
fmt++;
continue;
case 'x':
hex = 1;
sign = 0;
fmt++;
continue;
case '.':
fmt++;
while (*fmt >= '0' && *fmt <= '9') {
frac_width = frac_width * 10 + *fmt++ - '0';
}
break;
case '*':
slen = va_arg(args, size_t);
fmt++;
continue;
default:
break;
}
break;
}
switch (*fmt) {
case 'V':
v = va_arg(args, ngx_str_t *);
len = ngx_min(((size_t) (last - buf)), v->len);
buf = ngx_cpymem(buf, v->data, len);
fmt++;
continue;
case 'v':
vv = va_arg(args, ngx_variable_value_t *);
len = ngx_min(((size_t) (last - buf)), vv->len);
buf = ngx_cpymem(buf, vv->data, len);
fmt++;
continue;
case 's':
p = va_arg(args, u_char *);
if (slen == (size_t) -1) {
while (*p && buf < last) {
*buf++ = *p++;
}
} else {
len = ngx_min(((size_t) (last - buf)), slen);
buf = ngx_cpymem(buf, p, len);
}
fmt++;
continue;
case 'O':
i64 = (int64_t) va_arg(args, off_t);
sign = 1;
break;
case 'P':
i64 = (int64_t) va_arg(args, ngx_pid_t);
sign = 1;
break;
case 'T':
i64 = (int64_t) va_arg(args, time_t);
sign = 1;
break;
case 'M':
ms = (ngx_msec_t) va_arg(args, ngx_msec_t);
if ((ngx_msec_int_t) ms == -1) {
sign = 1;
i64 = -1;
} else {
sign = 0;
ui64 = (uint64_t) ms;
}
break;
case 'z':
if (sign) {
i64 = (int64_t) va_arg(args, ssize_t);
} else {
ui64 = (uint64_t) va_arg(args, size_t);
}
break;
case 'i':
if (sign) {
i64 = (int64_t) va_arg(args, ngx_int_t);
} else {
ui64 = (uint64_t) va_arg(args, ngx_uint_t);
}
if (max_width) {
width = NGX_INT_T_LEN;
}
break;
case 'd':
if (sign) {
i64 = (int64_t) va_arg(args, int);
} else {
ui64 = (uint64_t) va_arg(args, u_int);
}
break;
case 'l':
if (sign) {
i64 = (int64_t) va_arg(args, long);
} else {
ui64 = (uint64_t) va_arg(args, u_long);
}
break;
case 'D':
if (sign) {
i64 = (int64_t) va_arg(args, int32_t);
} else {
ui64 = (uint64_t) va_arg(args, uint32_t);
}
break;
case 'L':
if (sign) {
i64 = va_arg(args, int64_t);
} else {
ui64 = va_arg(args, uint64_t);
}
break;
case 'A':
if (sign) {
i64 = (int64_t) va_arg(args, ngx_atomic_int_t);
} else {
ui64 = (uint64_t) va_arg(args, ngx_atomic_uint_t);
}
if (max_width) {
width = NGX_ATOMIC_T_LEN;
}
break;
case 'f':
f = va_arg(args, double);
if (f < 0) {
*buf++ = '-';
f = -f;
}
ui64 = (int64_t) f;
buf = ngx_sprintf_num(buf, last, ui64, zero, 0, width);
if (frac_width) {
if (buf < last) {
*buf++ = '.';
}
scale = 1.0;
for (n = frac_width; n; n--) {
scale *= 10.0;
}
/*
* (int64_t) cast is required for msvc6:
* it can not convert uint64_t to double
*/
ui64 = (uint64_t) ((f - (int64_t) ui64) * scale + 0.5);
buf = ngx_sprintf_num(buf, last, ui64, '0', 0, frac_width);
}
fmt++;
continue;
#if !(NGX_WIN32)
case 'r':
i64 = (int64_t) va_arg(args, rlim_t);
sign = 1;
break;
#endif
case 'p':
ui64 = (uintptr_t) va_arg(args, void *);
hex = 2;
sign = 0;
zero = '0';
width = NGX_PTR_SIZE * 2;
break;
case 'c':
d = va_arg(args, int);
*buf++ = (u_char) (d & 0xff);
fmt++;
continue;
case 'Z':
*buf++ = '\0';
fmt++;
continue;
case 'N':
#if (NGX_WIN32)
*buf++ = CR;
#endif
*buf++ = LF;
fmt++;
continue;
case '%':
*buf++ = '%';
fmt++;
continue;
default:
*buf++ = *fmt++;
continue;
}
if (sign) {
if (i64 < 0) {
*buf++ = '-';
ui64 = (uint64_t) -i64;
} else {
ui64 = (uint64_t) i64;
}
}
buf = ngx_sprintf_num(buf, last, ui64, zero, hex, width);
fmt++;
} else {
*buf++ = *fmt++;
}
}
return buf;
}
IMO writing similar stuff only causes waste of memory,your idea?

Answer 1

Whether it's "worth it" is subjective - there are lots of things to consider:
What advantages does it give over sprintf? Do you need them?
Are you willing to live with whatever shortcomings may be there?
Do you understand the code well enough to be able to fix it if problems are found?
Is it as fast as sprintf? If not, do you care?
You may also want to consider writing a cover function that adds whatever extra functionality you need but then just calls sprintf, rather than re-implement the entire function.

Answer 2

I think this kind of thing is not only wasteful but actively harmful. The name includes printf, which would lead a reasonable person first seeing code that's using it to assume its format strings are printf-compatible. But in fact they're only a very poor approximation of printf semantics. This could lead to extremely serious, security-critical bugs, which might go undetected if the code only appears in a non-common-usage case. The C standard has a perfectly safe and usable snprintf function which should always be used when this functionality is needed. And shame on whoever designed this junk in the name of "security"...