What is an encoding error for sprintf that should return -1?

user_14751543 0
Viewed 467

I understand snprintf will return a negative value when "an encoding error occurs"

But what is a simple example of such an "encoding error" that will produce that result?

I'm working with gcc 10.2.0 C compiler, and I've tried malformed format specifiers, unreasonably large numbers for field length, and even null format strings.

  • Malformed format specifiers just get printed literally
  • Unreasonably large numbers as length specifiers produce fatal errors
  • Null format strings also produce fatal errors

This relates to repeatedly doing something like:

length += snprintf(...

to build up a formatted string.

That might be safe if it is certain not to return a negative value.

Advancing the buffer pointer by a negative length could cause it to go out of bounds. But I'm looking for a case where that would actually happen. If there is such a case then the added complexity of this may be warranted:

length += result = snprintf(...

So far I couldn't find a scenario where it would be worth adding complexity for a check of a value that the compiler may never produce. Maybe you can give a simple example of one.

language-lawyer error-handling gcc c
2 Answers

What is an encoding error for sprintf that should return -1?

On my machine, "%ls" did not like the 0xFFFF - certainly an encoding error.

  char buf[42];
  wchar_t s[] = { 0xFFFF,49,50,51,0 };
  int i = snprintf(buf, sizeof buf, "<%ls>", s);
  printf("%d\n", i);

Output

-1

Below code returned -1, but not so much due to encoding error as for pathological format.

#include <stdio.h>

int main() {
  size_t n = 0xFFFFFFFFLLu + 1;
  char *fmt = malloc(n);
  if (fmt == NULL) {
    puts("OOM");
    return -42;
  }
  memset(fmt, 'x', n);
  fmt[n - 1] = '\0';
  char buf[42];
  int i = snprintf(buf, sizeof buf, fmt);
  printf("%d %x\n", i, (unsigned) i);
  free(fmt);
  return 7;
}

Output

-1 ffffffff

I did get a surprising -1 when passing a too big a size, even though the snprintf() only needed 6 bytes.

  char buf[42];
  int i = snprintf(buf, 4299195472, "Hello");
  printf("%d\n", i);

Output

-1

I was able to come up with a short example returning -1 on a *fprintf() to stdout due to orientation conflict.

#include <wchar.h>
#include <stdio.h>

int main() {
  int w = wprintf(L"Hello wide world\n");
  wprintf(L"%d\n", w);
  int s = printf("Hello world\n");
  wprintf(L"%d\n", s);
}

Output

Hello wide world
17
-1
user_24103590

Normally you only expect an error from printf and family when an output error occurs. From the Linux man page:

If an output error is encountered, a negative value is returned.

So if you are outputting to a FILE and an output error of some kind (EPIPE, EIO) occurs, you'll get a negative return value. For s[n]printf, since there's no output, there would never be a negative return value.

The standard talks about the possibility of an "encoding error", but only defines what that means with respect to wide character streams, with a note that byte streams might need to convert to wide streams in some cases.

An encoding error occurs if the character sequence presented to the underlying mbrtowc function does not form a valid (generalized) multibyte character, or if the code value passed to the underlying wcrtomb does not correspond to a valid (generalized) multibyte character. The wide character input/output functions and the byte input/output functions store the value of the macro EILSEQ in errno if and only if an encoding error occurs.

That would seem to imply that you can get an encoding error if you use the %ls or %lc formats to convert a wide string or characters to bytes. Not sure if there are any other cases where it could occur.

user_164060
Related