C99：为什么％lli与64位计算机上的int64_t不匹配？

Question

I'm porting 32bit ARM code to 64bit x86 and hit the following error:

format '%lli' expects argument of type 'long long int',
but argument 3 has type 'int64_t'

I found a lot info about using PRIi64 in this case, but nothing about "why I cannot use lli ?".

#include <stdio.h>
#include <inttypes.h>

int main() {
    int64_t hugo = 993;
    long long int fred = 994;
    printf("%s -> %" PRIi64 " lli -> %lli\n", PRIi64, hugo, fred);
    return 0;
}

The code above compiles without error and prints li -> 993 lli -> 994 . So why do I need li for int64_t (8 bytes) and lli for long long int (8 bytes) if both have the same size?

I found one post which says, that it's a portability issue warning . So it is save to turn off this warning? (The only flag I found is -Wno-format which disable quite a lot checks...)

Answer 1

When the Standard was written, many implementations had two or more integer types with the same size and representation; most typically, int would have the same size and representation as either short or long .

Even on platforms where the sizes and representations would match, conversion from int* to a short* or long* requires a cast, but on implementations which used the same representation for two (or more) types, a pointer to either could be used to access data of the other. Support for such usage was not considered an "extension" of C, but more of a "well duh, what else would it do?" that seemed so obvious as to not require documentation(*)

The authors of gcc, however, assume that no pointer to a non-character integer type will be used to access things of any other non-character integer type-- even those whose size and representation are identical . Since printf would use a long long int* to read a parameter for %lld , a long* to read one for %ld , and an int* to read one for %d , the authors of gcc would view any attempt to use a %lld to process an argument of type int64_t as Undefined Behavior on implementations where that type is defined as something other than long long , even if its size and representation would be as expected.

(*)While C89 requires that all "extensions" be documented, the annex which lists common extensions doesn't list it. Since the authors of the Standard cannot plausibly have been unaware that many--if not most--compilers have two or more integer types that share the same size and representation, and allow pointers to those types to be used interchangeably, the only reasonable conclusion would be that they did not regard such behavior as an "extension" that would require documentation.