--std vs __intN

[DJ Delorie]

Currently, __int128 is disabled if __STRICT_ANSI__ via --std.

I'm working on making __int<N> generic within gcc, to support targets
that have non-standard type sizes (msp430's pointer is 20 bits).  This
means, for example, that size_t and sizeof(void*) may be one of these
__int<N> types.

This means if __int<N> is disabled via --std, most of the standard
library breaks because the pointer type isn't listed in any of the
headers' templates.

If I *don't* disable __int<N> when --std says to, some of the
testsuite tests - which test for that case - fail.

What to do?