Some examples can break due to dependency on headers coming from different files

[sleeptightAnsiC]

Many demo examples make an assumption that they can use symbols coming from a different file, due to indirect header inclusion.

For example, Nuklear/demo/common/overview.c uses libm math symbols, but never includes <math.h> or any similar header:

Nuklear/demo/common/overview.c

Lines 1 to 3 in fcd64f8

	#include <limits.h> /* INT_MAX */
	#include <time.h> /* struct tm, localtime */

Nuklear/demo/common/overview.c

Lines 754 to 759 in fcd64f8

	for (id = 0, i = 0; i < 32; ++i) {
	nk_chart_push_slot(ctx, (float)fabs(sin(id)), 0);
	nk_chart_push_slot(ctx, (float)cos(id), 1);
	nk_chart_push_slot(ctx, (float)sin(id), 2);
	id += step;
	}

This breaks, if I try to provide my own NK_SIN/NK_COS/STBTT_sqrt/etc in one of the demo files, and then remove the math header there.

This has happened to me when working on updated sdl3_renderer. In newer version of SDL3, <SDL3/SDL.h> does not pull <math.h> any more, and provides its own version of math symbols.

Note that Nuklear/demo/common/overview.c and <math.h> is just an example here. More demos are affected by this issue due to all kinds of headers.

[sleeptightAnsiC]

Note: header inclusion can be traced by compiler: GCC and Clang can do this with -H flag

Another note: https://include-what-you-use.org/ can help with issues like this one, but I don't believe it works when including source files.

[RobLoach]

Definitely something we should fix up. I think optimally we'd be defining our own NK_INT_MAX, and maybe even move away from time.h entirely.