[PATCH v5 2/3] Add default implementation of fenv.h and all methods

[Joel Sherrill]

No problem. This needs to be cross-platform and a good foundation to layer
architecture specific implementations.

I changed the name and made the storage static which seems more appropriate.

Version 5007 coming shortly. :)

On Thu, Aug 8, 2019 at 10:42 AM Corinna Vinschen <vinschen@redhat.com>
wrote:

> On Aug  8 17:38, Corinna Vinschen wrote:
> > Hi Joel,
> >
> > I just noticed another problem in terms of using this (partially)
> > with Cygwin:
> >
> > On Aug  8 10:10, joel@rtems.org wrote:
> > > + * NOTE: The extern'ed variable fe_default_env_p is an implementation
> > > + *       detail of this stub.  FE_DFL_ENV must point to an instance of
> > > + *       fenv_t with the default fenv_t. The format of fenv_t and
> where
> > > + *       FE_DFL_ENV is are implementation specific.
> > > + */
> > > +extern const fenv_t *fe_dfl_env_p;
> >
> > Can we please rename this pointer to _fe_dfl_env, as it's already
> > defined in Cygwin?  The reason is that this pointer is exported
> > into user space via ...
> >
> > > +#define FE_DFL_ENV fe_dfl_env_p
> >
> > If you change the name here, Cygwin will not be able to use
>   ^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Make that "If you don't use the _fe_dfl_env name here..."
>
> > > diff --git a/newlib/libm/fenv/fe_dfl_env.c
> b/newlib/libm/fenv/fe_dfl_env.c
> > > [...]
> >
> > this file as storage for the default environment so as not to break
> > backward compatibility with existing executables.
> >
> > > + * This is a non-functional implementation that should be overridden
> > > + * by an architecture specific implementation in
> newlib/libm/machine/ARCH.
> > > + *
> > > + * The implmentation must defined FE_DFL_ENV to point to a default
> > > + * environment of type fenv_t.
> > > + */
> > > +const fenv_t fe_dfl_env = { 0 };
> > > +const fenv_t *fe_dfl_env_p = &fe_dfl_env;
> >
> > The funny thing here is, that this file could be used by all targets,
> > regardless of the definition of fenv_t, without the need to redefine it
> > per target.  Only Cygwin couldn't.  It would have to provide it's own,
> > even if the Cygwin code itself gets moved to i386 and x86_64, unless
> > fe_dfl_env_p is renamed to _fe_dfl_env.
>
>
> Sorry,
> Corinna
>
> --
> Corinna Vinschen
> Cygwin Maintainer
> Red Hat
>