Strange warning when calling sprintf with .* width specifier-CodePudding

For the following code:

#include <stdio.h>

int main() { 
    
    char temp_buffer[8];
    double val = 25.3;

    sprintf(temp_buffer, "%.*g", sizeof(temp_buffer), val);
    printf("%s", temp_buffer);
}

I get the warnings in gcc 11.3 with -Wall flag:

<source>:8:29: warning: field precision specifier '.*' expects argument of type 'int', but argument 3 has type 'long unsigned int' [-Wformat=]
    8 |     sprintf(temp_buffer, "%.*g", sizeof(temp_buffer), val);
      |                           ~~^~   ~~~~~~~~~~~~~~~~~~~
      |                             |    |
      |                             int  long unsigned int
<source>:8:27: warning: '%.*g' directive writing between 1 and 310 bytes into a region of size 8 [-Wformat-overflow=]
    8 |     sprintf(temp_buffer, "%.*g", sizeof(temp_buffer), val);
      |                           ^~~~
<source>:8:26: note: assuming directive output of 12 bytes
    8 |     sprintf(temp_buffer, "%.*g", sizeof(temp_buffer), val);
      |                          ^~~~~~
<source>:8:5: note: 'sprintf' output between 2 and 311 bytes into a destination of size 8
    8 |     sprintf(temp_buffer, "%.*g", sizeof(temp_buffer), val);
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

In fact the size of the destination buffer is too small, but what's with the 'sprintf' output between 2 and 311 bytes into a destination of size 8? Where does that 311 bytes value comes from?

Stranger yet, if I cast the number of the decimal places to int, i.e. (int)sizeof(temp_buffer) the potential overflow numbers drop dramatically:

'sprintf' output between 2 and 16 bytes into a destination of size 8

CodePudding user response：

There are multiple issues in the code:

sprintf expects an int value for the * place holder and you pass a size_t which may have a different size and representation.
the call may have undefined behavior because you request a precision that is the length of the destination array, which may produce output that exceeds said length.

Passing sizeof(temp_buffer) is an error, the compiler seems confused about the actual argument values and makes no particular assumption about the precision value or the number to convert. Yet they seem mistaken when they document the output can be 2 to 311 bytes:

for the value 25.3, the exact representation of the closest IEEE 754 number is 25.300000000000000710542735760100185871124267578125, requiring 52 bytes.
the output for the largest number via printf("%.1000g", -0x1.fffffffffffffp 1023) has 310 characters, thus requiring 311 bytes, which seems to be the reason for the 2 to 311 bytes.
yet a %.*g conversion can actually produce more than 311 bytes: printf("%.1000g", -5e-324) produces 758 characters on both macOS and linux.

When you cast sizeof(temp_buffer) as (int), the compiler determines that the precision is 8 (a non trivial optimisation) and determines that the output can be as small as 2 bytes (a single digit and a null terminator) but no longer than 16 (-, a digit, ., 7 decimals, e, - and as many as 3 exponent digits plus a null terminator. That is still potentially too much for an 8 byte array.

Good job for warning the programmer about this potential undefined behavior!

Use snprintf(), a larger array and pass (int)(sizeof(temp_buffer) - 9) as the precision to get as many decimals as will fit in the worst case. It is difficult to produce as many decimals as will fit in all cases and may require multiple tries or complex postprocessing.

CodePudding user response：

Where does that 311 bytes value comes from?

Compiler is confused.

"%.*g", 8 may output 15^*1 1 (for the null character) characters, but not 311.

I suspect the compiler is making the wrong prediction that printing -DBL_MAX would use 310 1 characters, expect for the fact that %g switches to exponential notation for extreme values - that is the beauty of "%g", limited output. If it did not switch to exponential notation then that extreme output would be like "%.*f", (int) 0.

int main(void) {
  char buffer[1000];
  char temp_buffer[8];
  int len = sprintf(buffer, "%.*g", (int) sizeof(temp_buffer), -DBL_MAX);
  printf("%d <%s>\n", len, buffer);
  len = sprintf(buffer, "%.*f", (int) 0, -DBL_MAX);
  printf("%d <%s>\n", len, buffer);
}

Output

15 <-1.7976931e 308>
310 <-179769313486231570814527423731704356798070565449239578069709514447683386590106403234437238318580897337933920052621128987133479958537240295444402082043505598186819583097828779632178833278408753356733764414284292236482024122876814115690851853178733231033249466834451356736736928870307247739983885979597249860971>

^*1

1 1 1   8-1   1 1  3 --> 15
- d . ddddddd e - eee

CodePudding user response：

Basically what it's trying to tell you is that it's possible to end up with more digits than there is space to store them. What happens if val is a really big number?