Re: MIT discovered issue with gcc

[Octavio Alvarez <alvarezp@alvarezp.ods.org>]

I want to correct a mistake I made in my last message, where my logic of 
using assert() and NDEBUG was upside-down.

This doesn't affect the logic of my statement, as the use of assert() 
was just an idea to let the compiler know regarding assumptions it could 
make.

On 28/11/13 14:41, Octavio Alvarez wrote:
> As an idea, and just thinking out loud, it could be a bad idea, how
> about this:
>
> #undef NDEBUG
> int f(int a) {
>      assert(a == 4);
>      int b = a + 7;
>      int c = a - b + 7;
>      if (3/c > 4)
>          return 5;
>      return 10;
> }
>
> Note that NDEBUG is undefined, effectively not resulting in any actual
> code, but the compiler could take it as an optimization hint.

I meant the other way around:

#define NDEBUG
int f(int a) {
     assert(a == 4);
     int b = a + 7;
     int c = a - b + 7;
     if (3/c > 4)
         return 5;
     return 10;
}

Note that NDEBUG is defined, effectively not resulting in any actual
code, but the compiler could take it as an optimization hint.

> void foo(int s, int e) {
>           assert(s < e);
>           int i;
>           for (i = s ; i != e ; i++) {
>                   something();
>           }
> }
>
> Even if NDEBUG is not defined. Or something similar, like a conditional
> comment:

This, again, should have been:

#define NDEBUG
void foo(int s, int e) {
          assert(s < e);
          int i;
          for (i = s ; i != e ; i++) {
                  something();
          }
}

Even if NDEBUG is defined. Or something similar, like a conditional comment:

> void foo(int s, int e) {
>           /* GCC_ASSERT(s < e) */
>           int i;
>           for (i = s ; i != e ; i++) {
>                   something();
>           }
> }
>
> ... or whatever hint is included.

Please excuse this mistake.

Best regards.