What is the expected behavior of referring to previously set fields, when declaring initial values for a struct using compound literal: Specifically is it ok to: struct foo v = { .v1 = ..., .v2 = .v1+1 };

More complete example:

struct s {
    int i1 ;
    int i2 ;
} ;

void func(void)
{
     struct s v1 = { 3 } ;
     v1 = (struct s) ( .i1 = 4, .i2 = v1.i1+10 }
     // always v1.i2=13, NOT 14

     // Is this legal ?
     struct s v2 = ( .i1 = 1, .i2 = v2.i1+10 }
     // Under GCC, v2.i2 = 11

}

The first case – the standard is clear that “temporary” compound literal is created, based on the value of v1 before the assignment (v1.i1=3), resulting in v2.i2 = 13. The explanation is that this is basically:

     struct s v1 = { 3 } ;
     struct s temp = ( .i1 = 4, .i2 = v1.i1+10 }
     v1 = temp ;

For the second case, at least with GCC, it looks as if it’s OK to reference previously set values in the SAME statement – in this case i2 is set to 11, implying that this was NOT implemented using temporary variable which would have results in v2.i2 = 10;

     struct s temp = ( .i1 = 4, .i2 = v1.i1+10 }
     struct s v1 = temp ;

My question: what does the C99/C23 standard says about the second case ? is it required behavior, implementation specific behavior or undefined behavior which compilers should flag as warning/error ?

I tried the above with GCC9 (-Wall, -Wextra), also with Coverity – no issue was raises.