The most generic way to initialise an automatic variable to zero (as it would be if in static scope) is to use:

type_t variable = {0,};

This operation is defined by C90, and works whether the variable is a scalar or aggregate type as demonstrated below. This can replace memset(&var,0,sizeof(var)) which assumes that initialised variables are always represented by 0 in static scope (Not a terrible assumption TBH).

Note the code below along with the assembly generated by gcc-4.1.2 shows that if any element values are not specified, they're initialised as is any unused padding space between elements of aggregate types. However if all elements are explicitly initialised the padding is left untouched.

void f(void)
{
    typedef int opaque_t;
    opaque_t o = { 0, }; /* can init scalar types like this */

    typedef struct { char c; int a; int b; } s_t;
    s_t s1 = { 0,      };
 /* movl        $0, -16(%ebp)
    movl        $0, -12(%ebp)
    movl        $0, -8(%ebp) */

    s_t s2 = { 0, 0, 0 };
 /* movb        $0, -28(%ebp)
    movl        $0, -24(%ebp)
    movl        $0, -20(%ebp) */

    s_t s3 = { 1, 0,   };
 /* movl        $0, -52(%ebp)
    movl        $0, -48(%ebp)
    movl        $0, -44(%ebp)
    movb        $1, -52(%ebp) */
}

Unfortunately the gcc -Wmissing-field-initializers option (which is included in -Wextra) issues a warning for this useful construct before gcc 4.7. I had requested that this warning should be suppressed, which was done in April 2011. With earlier versions of GCC you can enable this warning by writing your own declaration macro to work around the warning, but that's very awkward.

Note the discussion above pertains to C, not C++. C++ does not guarantee that the gaps between members is initialized, and in fact GCC 4.9 changed so that it no longer is initialized.