// tcc -m32 minimal.c

#include <stdio.h>

void printf_stack_pointer(int ln) {
    unsigned int i;
    asm("\t movl %%esp,%0" : "=r"(i));
    printf("At line %i, SP is %u\n", ln, i);
}

#define PRINTF_STACK_POINTER printf_stack_pointer(__LINE__);


__stdcall int some_stdcall_func(int foo, int bar, int baz) {
	printf("Hello from stdcall: %i\n", foo);
	return 12;
}


int main() {

	PRINTF_STACK_POINTER
	// Baseline stack pointer for comparison.

	some_stdcall_func(1, 0, 0);
	PRINTF_STACK_POINTER
	// No problem. Stack pointer is unchanged, as it should be.


	((int __stdcall (*)(int, int, int))some_stdcall_func) (2, 0, 0);
	PRINTF_STACK_POINTER
	// Still no problem.


	((int(*__stdcall)(int, int, int))some_stdcall_func) (3, 0, 0);
	PRINTF_STACK_POINTER
	// Still no problem.


	((int(__stdcall*)(int, int, int))some_stdcall_func) (4, 0, 0);
	PRINTF_STACK_POINTER
	// Problem! Notice that SP is now 12 ( e.g. sizeof(int)*3 ) bytes higher.

	// My guess: For whatever reason TCC thinks this is a cdecl function
	// (caller-cleanup) and has generated code to recover the stack.

	// Of course, since the function is actually callee-cleanup, it has already
	// did that. The stack has been "cleaned" twice and is now sizeof(args) too high.

	// Seems to happen when the calling convention attribute is inside the brackets
	// AND before the asterisk.
	
}