Isn't the failure in gfortran.dg/edit_real_1.f90 likely due to
the fact that gfortran (unlike the gcc and g++ compiler) doesn't
generate the $LDBL128 versions of the symbols for function calls
related to long doubles?
I don't think so (but I may be wrong). As far as I can tell
edit_real_1.f90 deals only with doubles. Using -fdump-tree-original
I get:
static real8 C.967 = 1.79769313486231570814527423731704356798070567526e+308;
for both avatars of gfortran. So the problem is either with the building
of C.967 (does not it use gmp/mpfr?) or in the I/O, but I don't see where
the code should need $LDBL128 versions.
Dominique