Why errno can set to zero by scanf?(when enter "ctrl+D")

Question

The manual tells us:errno is never set to zero by any system call or library function. But I want to know, Why errno can set to zero by scanf in following codes?(when scanf:enter the"ctrl+D")

#include <stdio.h>
#include <errno.h>
int main()
{
    int i;
    errno = 5;
    printf("errno:%d\n",errno);
    if (scanf("%d", &i) < 1)
        perror("scanf");
    printf("errno:%d\n",errno);
    printf("i:%d\n", i);
    return 0;
}

Answer 1

I can find the following code in the glibc implementation of vfscanf(), (line 589-607 in the linked file, as of the time of writing) which the implementation of scanf() calls:

if (skip_space || (fc != L_('[') && fc != L_('c')
                         && fc != L_('C') && fc != L_('n')))
  {
    /* Eat whitespace. */
    int save_errno = errno;
    __set_errno (0);
    do
      /* We add the additional test for EOF here since otherwise
         inchar will restore the old errno value which might be
         EINTR but does not indicate an interrupt since nothing
         was read at this time. */
      if (__builtin_expect ((c == EOF || inchar () == EOF)
                             && errno == EINTR, 0))
         input_error ();
    while (ISSPACE (c));
    __set_errno (save_errno);
    ungetc (c, s);
    skip_space = 0;
  }

input_error() is #defined as:

#define input_error()         do {     
                          errval = 1; 
                          if (done == 0) done = EOF;
                          goto errout; 
                        } while (0)

where errout is the label for the cleanup code at the end.

So it looks like errno is getting set to 0 prior to the inchar() call, and the old value is later replaced, leaving errno unchanged. But if an error occurs and that if statement executes (notably, if inchar() evaluates to EOF, which is what is happening in this case), it looks like the code to reset errno to its original value may be being skipped. That being said, the condition will only be true if errno == EINTR and therefore not zero, which certainly doesn't appear to be the case, here, so it may have nothing to do with this code, but this is only place I can see errno being set to 0. inchar() itself does mess around with errno, as the comment suggests, and can set errno to inchar_errno, which is initialized to 0 on line 223, so it's also possible there's some other execution path where inchar_errno is not updated but gets assigned to errno anyway.