handling setjmp/longjmp in new front-end

[Richard Kenner]

    Because we don't construct the CFG properly.  The non-local
    goto handling (around which __builtin_longjmp is built)
    assumes that a non-local goto leaves the current function.

OK, that shouldn't be an issue.

    What are the odds of getting Ada to use GCC's language
    independant exception handling code?

Unknown.  The first step is understanding precisely what that code does.

The original design of that code had Ada in mind, but it's unclear how much
of the design document was actually implemented.