On Thu, 2017-04-06 at 09:49 +1200, John Harper wrote:
> REAL64 is the same as KIND(1D0) in many machines, including
> all three that I have access to, but it is apparently the same as
> KIND(1E0) in some.
This is related to the reason that TYPEALIAS, i.e., a new name for an
existing type, but not a new type, was a pointless idea.
If you define two kind type parameters, say SELECTED_REAL_KIND(6) and
SELECTED_REAL_KIND(12), and use them in two specific procedures to
create "single precision" and "double precision" specific procedures for
a generic, on machines for which REAL64 is the same as REAL(KIND(1.0e0))
you can't compile the generic interface.
With new types instead of new names for two identical declarations,
generic resolution would work.
|