Pre-DR: variadic standard library functions
===========================================

To what extent should the variadic functions in <stdio.h> and
<wchar.h> be required to act as if they use the <stdarg.h> macros to
access arguments?  (The problem is well-known from comp.std.c, but
there seems to be a lack of consensus on both the meaning and the
intent of the standard.  Furthermore, this issue is involved in one
implementor's claims of nonconformance of other library
implementations <http://www.dinkumware.com/conform_c.html> [1].)

There are at least three cases involved:

1. Arguments of correct promoted type, and in range for unpromoted
   type, but not of the unpromoted type.

#include <stdio.h>

void
f(void)
{
  // Suppose that unsigned char promotes to int.
  printf("%hhu", 0);
}

Here, %hhu expects an unsigned char, which has been promoted to int,
but has been passed an int not derived from an unsigned char.  Is this
defined, or should it be?

2. Arguments of correct promoted type, but out of range for unpromoted
   type.

#include <stdio.h>

void
f(void)
{
  // Suppose that UCHAR_MAX == 255.
  printf("%hhu", 256);
}

Here, the argument is of type int, the correct promoted type, but even
printf implemented using the <stdarg.h> macros only can tell that it
did not arise from an unsigned char.  Is this defined, or should it
be?  If not, what purpose does the wording about printf converting the
argument back to unsigned char before printing serve?  (The disputes
over conformance relate to failure to make such a conversion back to
unsigned char.)

3. Arguments of incorrect type, allowed for va_arg by 7.15.1.1#2.

#include <stdio.h>

void
f(void)
{
  printf("%x", 0);
}

Here, an int argument is passed where an unsigned int is expected, but
the value is representable in both types.  Is this defined, or should
it be?


[1] Some claims of nonconformance in this document are based on
misunderstandings of the standard.  In particular, all the claims of
problems with implementing <stdint.h> INTN_C macros by concatenation
are wrong, as the parmaeter to such macros is specified as a "decimal,
octal, or hexadecimal constant", not a general integer constant
expression (and is unsuffixed by virtue of the syntax in 6.4.4.1,
something missed in N1046).  However, the question on what is defined
for printf %hh is a genuine uncertainty about the standard.