patch for -Wno-long-long and early GNAT compilers

[Robert Dewar]

<<Or perhaps the problem might be that GCC changed the way the builtin
setjmp/longjmp was implemented on x86 and the setjmp (in Ada code) is
being compiled with one version while the longjmp (in C code) is being
compiled by a different one.
>>

I agree this is a likely explanation. Note that the GNAT compiler is
full of exception handlers, so the built in setjmp is called all the
time. However, *raising* of exceptions is indeed rare during the
bootstrap process (indeed it used to happen just once, and now, at
least for the moment happens not at all, with the latest patch I
sent along).