Tricky C problem

This is infuriating. When I run the below (compiled with Norcroft RISC OS ARM C vsn 5.83 [01 Jun 2020]) I get

8d2c 7
8d2c 14

where I expect the two addresses to be different. When I look at the disassembly, it seems to be calculating the address of s.t->y before the loop, so nothing you do inside the loop (in particular making the assignment s = f(s)) can change its value.

What’s going on here??

Edit: If I compile with -g the addresses come out different. Am I doing something undefined in C, or is this a compiler bug?

#include <stdio.h>
#include <stdbool.h>

typedef struct
{
  int x1;
  int x2;
  int x3;
  int x4;
  int x5;
  int x6;
  char y[1];
} t_t;

typedef struct
{
  char *a;
  t_t *t;
} s_t;

static t_t t1 = {1,2,3,4,5,6,{7}};
static t_t t2 = {8,9,10,11,12,13,{14}};

s_t f(s_t s)
{
  s.a = NULL;
  s.t = &t2;
  return s;
}

s_t g(void)
{
  s_t s = {(char*)-1,&t1};
  return s;
}

bool done(s_t s)
{
  return s.a == NULL;
}

int main(int argc, char **argv)
{
  s_t s = g();
  while (!done(s))
  {
    printf("%x %d\n", s.t->y, s.t->y[0]);
    if (s.t->y[0] == 14)
     break;
    else
    {
      s = f(s);
      printf("%x %d\n", s.t->y, s.t->y[0]);
    }
  }
  return 0;
}

I tried our luck with DDE30 but …

*cc -v
Norcroft RISC OS ARM C vsn 5.85 [22 Oct 2020]

cc -c -depend !Depend -throwback   -ff -fah -o o.Julie c.Julie
Norcroft RISC OS ARM C vsn 5.85 [22 Oct 2020]
"c.Julie", line 25: Warning: extern 'f' not declared in header
"c.Julie", line 32: Warning: extern 'g' not declared in header
"c.Julie", line 38: Warning: extern 'done' not declared in header
"c.Julie", line 47: Warning: actual type <pointer> mismatches format '%x'
"c.Julie", line 53: Warning: actual type <pointer> mismatches format '%x'
"c.Julie", line 57: Warning: variable 'argv' declared but not used
"c.Julie", line 57: Warning: variable 'argc' declared but not used
c.Julie: 7 warnings, 0 errors, 0 serious errors
link  -o Julie o.Julie  CLIB:o.stubs
squeeze  Julie
Julie: application built
*Julie
8d20 7
8d20 14

… and then with gcc which looks more hopeful …

*gcc julie.c
*a/out
8a7c 7
8a98 14
*

This is a problem in C, your call to the function f (s) does not modify s.
It is necessary to pass s by reference.
Add a function f2() which does the same thing, but on the variable passed as a reference and not its copy.

  void f2(s_t *s)
  {
    s->a= NULL;
    s->t = &t2;
  }

Sorry for my English

No, it is a compiler bug. All members of s must be treated as having been modified by function f as that structure is its return value.

Jean-Michel: NULL != -1

I agree with Jean-Michel,

the function f should be defined as:

void f(s_t *s)
{
  s->a = NULL;
  s->t = &t2;
}

and in main obviously replace s = f(s) with:

f(&s);

This is how you would do what you want in C.

“I agree with Jean-Michel,”

Thanks.

NULL != -1
exact but,in the Julie program:
bool done(s_t s) { return (s.a == NULL); }
It’s more homogeneous. NULL is used to initialize pointers to 0.

just a note, GCC would compile Julie’s code and it would work as she expects, but I am not 100% sure if this is actually a correct behaviour or not, hence I do not want to state there is a bug in DDE !CC…

The correct C syntax for what she wants to do is what Jean-Michael described.

The NULL vs Bool is also correct, Bool should be either -1 or 0. If you want to be pedantic and stick to C99 standard then you should use stdbool.h provided values, which are 0 for false and 1 for true.

Thanks guys, I’ve filed a ticket. It seems to come down to returning a struct from a function, and not having the pointer members of the structs at the start of the structs.