While looking at Andrew's patch for fixing the long double issue
on Darwin PPC, I noticed something interesting in the symbols in the
object files created for gcc/libgfortran/generated. Some of these
object files have symbols with the proper $LDBL128 appended on Darwin PPC.
For example...
nm rrspacing_r16.o | grep expl
U _frexpl$LDBL128
U _ldexpl$LDBL128
This contrasts with other object files created from gcc/libgfortran/generated...
nm _exp_r16.o | grep expl
U _expl
The difference seems to be that rrspacing_r16.c explicitly calls frexpl()
whereas _exp_r16.c calls exp() rather than expl().
This leads me to wonder if this problem with large_real_kind_2 failing
on Darwin PPC isn't due to various *_r10.c and *_r16.c files making calls
to foo() rather than fool(). Is it possible that this issue (and the
necessary fix) could really be that simple?
Jack
