Add printf formatting macros for nk data types

[PROP65]

Adds macros from formatting the nk data types

Macro	Type
NK_FMT_CHAR	nk_char
NK_FMT_UCHAR	nk_uchar
NK_FMT_BYTE	nk_byte
NK_FMT_SHORT	nk_short
NK_FMT_USHORT	nk_ushort
NK_FMT_INT	nk_int
NK_FMT_UINT	nk_uint
NK_FMT_SIZE	nk_size
NK_FMT_PTR	nk_ptr

This won't completely prevent issues like #727 but it should help.

Needs to be tested on Windows.

[PROP65]

Here's a test program to check for warnings.

https://gist.github.com/PROP65/88cf2ec23a5052350f2e79857dd9fd8e

[RobLoach]

Seems like a good solution. Those warnings have been pretty annoying in the build sometimes.

[RobLoach]

Seems good to me. Going to leave this open for a bit for others to review too.

[klei1984]

I am using msys2/Mingw64 toolchain on Windows which has some problems with the NK_INCLUDE_FIXED_TYPES feature. See in-line review comments.

[klei1984]

On Windows 64 bit, msys2/mingw64 this line gives the following warning message if demo is built in -std=c89 mode.

Compiler settings:

CFLAGS += -std=c89 -Wall -Wextra -pedantic -ggdb -O0 -DSDL_DISABLE_IMMINTRIN_H
CFLAGS += `sdl2-config --cflags`

Warning message:

In file included from main.c:63:
../../demo/common/overview.c: In function 'overview':
../../demo/common/overview.c:447:33: warning: ISO C does not support the 'I64' ms_printf length modi
fier [-Wformat=]
  447 |                 sprintf(buffer, "%" NK_FMT_SIZE, sum);
      |                                 ^~~                  

NK_INCLUDE_FIXED_TYPES is active in the above context.

Disabling the macro in the 64 bit build environment fails via assertions like:

NK_STATIC_ASSERT(sizeof(nk_size) >= sizeof(void*));
NK_STATIC_ASSERT(sizeof(nk_ptr) >= sizeof(void*));

The potential reason is this:

    #elif defined(__GNUC__) || defined(__clang__)
      #if defined(__x86_64__) || defined(__ppc64__) || defined(__PPC64__) || defined(__aarch64__)
        #define NK_SIZE_TYPE unsigned long
      #else

On Windows 64 bit, msys2/mingw64 unsigned long is 4 bytes.

Architecture of built software:

architecture: i386:x86-64, flags 0x0000013b:
HAS_RELOC, EXEC_P, HAS_DEBUG, HAS_SYMS, HAS_LOCALS, D_PAGED
start address 0x00000001400013f0

The 32 bit build environment, using msys2/mingw32, builds without errors and warnings.

[PROP65]

I added checks for MinGW; the static asserts should be fixed. Since there isn't any type in c89 that is guaranteed to be at least 64 bits, it may not be possible on 64 bit windows targets to format nk_size in a way that is c89 compliant.

[klei1984]

I checked both NK_INCLUDE_FIXED_TYPES active and inactive in 32/64 bit and c89/c99/c11/c17/c23 combinations.

• I can confirm that the demo I checked earlier is now not tripping on the assertion tests in any checked compilation modes.

The rest are warnings and as noted by PRP65, most are C ISO standards related.

• c89 raises long long warnings in 64 bit mode which is perfectly fine.
• in case fixed types macro is disabled, 32 bit mode in c89 mode warns about no support for I32 printf flag.
• in case fixed types macro is disabled, 32/64 bit modes in c99/C11/C17/C23 modes warn about no support for I printf flag.
• in case fixed types macro is disabled, 64 bit mode in c99/C11/C17/C23 modes warn about NK_FMT_SIZE being incompatible with size_t. Warning example below.

../../demo/common/overview.c:447:33: warning: format '%u' expects argument of type 'unsigned int', but argument 3 has type 'size_t' {aka 'long long unsigned int'} [-Wformat=]
  447 |                 sprintf(buffer, "%" NK_FMT_SIZE, sum);
      |                                 ^~~              ~~~
      |                                                  |
      |                                                  size_t {aka long long unsigned int}

[klei1984]

Warning message:

../../demo/common/overview.c:459:33: warning: ISO C does not support the 'I64' ms_printf length modi
fier [-Wformat=]
  459 |                 sprintf(buffer, "%" NK_FMT_SIZE, sum);
      |                           

[RobLoach]

IS this good to go, now?

[sleeptightAnsiC]

Is this really necessary?

Nuklear does not depend on exact-int-size printf formatting anywhere in its sources, only demos and examples really need it. There is roughly one place in a library where a simple formatting (%s/%d/%f/etc) is being used, and this PR is not even applying to it. Demos/examples are relaxed from C89 rule and there would be nothing wrong if you simply include <inttypes.h> somewhere inside of demo/common/overview.c. In fact most of those already require C99, and this is even mentioned here. Adding all those macros into library headers brings so much complexity and possible issues. I don't believe it's worth it.

@RobLoach Don't get me wrong, but I don't think this should be ever merged. The linked issue (#727) is only about demo/common/overview.c. Fixing it by adding all those symbols to the library is an over-kill.