Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.c > #7104

Re: packing a "variable length struct"

Show all headers | View raw

On 6/27/2011 16:09, Alexander Bartolich wrote:
> Shao Miller wrote:
>> On 6/27/2011 07:19, James Kuyper wrote:
>>> On 06/27/2011 07:01 AM, Alexander Bartolich wrote:
>>> ...
>>>> pointer, i.e. declare
>>>>
>>>>     typedef struct
>>>>     {
>>>>         A_t meta;
>>>>         double centres[1];
>>>>     } B_t;
>>>>
>>>> then allocate sufficient memory and access the array "centres" beyond
>>>> its bounds.
>>>
>>> You should note, however, that while this technique works with most
>>> (?all) C compilers, any program that uses it has undefined behavior.
>>
>> Bounds checking as undefined behaviour?  Say it ain't so!
>
> Imagine a platform where the address range reachable by index/offset
> registers is smaller than the address range covered by size_t.
>

Where does 'size_t' come into things?  Or even 'ptrdiff_t', though you 
didn't bring it up.

Are you suggesting that there are values of 'size_t' for which 'malloc' 
might return a null pointer value?  I can certainly agree to that.

> Exhibit 1: Intel 8086, running in 16-bit mode, memory model "Huge".
> Exhibit 2: The Atmel AVR, a contemporary 8-bit microcontroller.
>
> The compiler can generate much more efficient code if it assumes that
> the array will not be accessed beyond its bounds.
>

Sure.  What bothers me is that Standard text including "bounds"[footnote 
94] and "not an element of an array"[6.5.6p7] and "array 
object"[6.3.2.1p3, 6.5.2.1p2, 6.5.2.1p3, 6.5.2.1p4, 6.5.9p6, etc.] and 
"same array object"[6.5.6p8, 6.5.6p9, 6.5.8p5, 6.5.8p6] appear to lead 
to some challenges.

If we use "effective type"[6.5p6] to determine if something is or is not 
an "array object," that's fine.  But then for:

   #include <stdlib.h>
   #include <assert.h>
   #include <stddef.h>
   #include <stdio.h>

   typedef struct {
       int level;
       long index;
     } A_t;

   typedef struct {
       A_t meta;
       double centres[1];
     } B_t;

   int main(void) {
       void * vp;
       char * finder;
       B_t * foo;
       double (* dbl_arr_ptr)[6];
       double * dbl_ptr;

       /* Certainly room for a 'double[6]' */
       vp = malloc(sizeof *foo + sizeof (double[5]));
       assert(vp);

       /* Find 'centres' */
       finder = vp;
       finder += offsetof(B_t, centres);

       /* Pretend a 'double[6]' is there */
       dbl_arr_ptr = (void *) finder;

       /* Establish the effective type */
       1[*dbl_arr_ptr] = 3.14159;

       /* Pretend we have a 'B_t' at 'vp' */
       foo = vp;

       /* Point one past 'centres' */
       dbl_ptr = foo->centres + 1;

       printf("We have: %f\n", *dbl_ptr);

       free(foo);
       return 0;
     }

Does the 'printf' line invoke undefined behaviour?  What does 'dbl_ptr' 
point to?  Is it one past the single-element 'centres' array or is it 
the second element of an object with effective type 'double[6]'?

Back to comp.lang.c | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

packing a "variable length struct" fishboy <j.j.fishbat@gmail.com> - 2011-06-27 02:46 -0700
  Re: packing a "variable length struct" Ian Collins <ian-news@hotmail.com> - 2011-06-27 22:48 +1200
  Re: packing a "variable length struct" Alexander Bartolich <alexander.bartolich@gmx.at> - 2011-06-27 11:01 +0000
    Re: packing a "variable length struct" James Kuyper <jameskuyper@verizon.net> - 2011-06-27 07:19 -0400
      Re: packing a "variable length struct" Shao Miller <sha0.miller@gmail.com> - 2011-06-27 14:12 -0400
        Re: packing a "variable length struct" Shao Miller <sha0.miller@gmail.com> - 2011-06-27 15:08 -0400
        Re: packing a "variable length struct" Alexander Bartolich <alexander.bartolich@gmx.at> - 2011-06-27 20:09 +0000
          Re: packing a "variable length struct" Shao Miller <sha0.miller@gmail.com> - 2011-06-27 17:45 -0400
  Re: packing a "variable length struct" Eric Sosman <esosman@ieee-dot-org.invalid> - 2011-06-27 07:24 -0400
    Re: packing a "variable length struct" Keith Thompson <kst-u@mib.org> - 2011-06-27 09:06 -0700
      Re: packing a "variable length struct" BGB <cr88192@hotmail.com> - 2011-06-27 15:35 -0700
    Re: packing a "variable length struct" fishboy <j.j.fishbat@gmail.com> - 2011-06-30 23:54 -0700

csiph-web