Win64: Only use __builtin_setjmp/__builtin_longjmp when compiling with GCC

[qmfrederik]

Using __builtin_setjmp when compiling with Clang on MSYS2 will result in compiler warnings and runtime crashes. It seems to work on GCC, though.

Compiler warning on Clang:

warning: incompatible pointer types passing 'jmp_buf' (aka 'struct _SETJMP_FLOAT128[16]') to parameter of type 'void **' [-Wincompatible-pointer-types]
                NS_DURING
                ^~~~~~~~~
note: expanded from macro 'NS_DURING'
                    if (!setjmp(NSLocalHandler.jumpState)) {
                                ^~~~~~~~~~~~~~~~~~~~~~~~
note: expanded from macro 'setjmp'
#define setjmp(X)       __builtin_setjmp(X)
                                         ^
1 warning generated.

[qmfrederik]

Actually, I think the best approach may be to remove this alltogether? setjmp and __builtin_setjmp have slightly different semantics. Using __builtin_setjmp as a replacement for setjmp apparently breaks interoperability between GNUstep compiled using gcc and other software compiled with clang.

[rfm]

Yes, I see this is a hack specifically for the state of mingw64 as it was in 2016 ... so I agree that if current mingw64 is working without the hack we should simply remove it.

[qmfrederik]

@rfm I compiled libs-base on the UCRT64 environment of MSYS2, using gcc as the compiler, and using the 'traditional'/setjmp-based exceptions (e.g. without -fexceptions and fojbc-exceptions). All tests (make check) passed. I ran the test suite of one of our applications, which passed as well.

That seems to be an indication that it's OK to remove this workaround, 7 years after it was introduced.