(j3.2006) How to assign to the declared-type part of a polymorphic allocatable array?

[Van Snyder]

Here's an ugly solution, that gets uglier and uglier as more extensions
of a type are created:

On Wed, 2016-09-21 at 19:57 -0700, Van Snyder wrote:
> Is there a way to assign to the declared-type part of a polymorphic
> allocatable array, other than one component at a time?
> 
> I have a type H_t (for horizontal geolocation) that contains latitude
> and longitude.  It's agnostic whether the latitude is geocentric or
> geodetic.  It has extensions H_Geoc (for geocentric latitude) and H_Geod
> (for geodetic latitude), H_V_t (including a height component V, agnostic
> whether geocentric or geodetic), H_V_Geoc (geocentric latitude and
> height), H_V_Geod (geodetic latitude and height), and others.
> 
> I have an object
> 
>   class(h_t), allocatable :: Geo(:)
> 
> which I allocate
> 
>   if ( geodetic ) then
>     allocate ( h_v_geod :: geo(n) )
>   else
>     allocate ( h_v_geoc :: geo(n) )
>   end if

select type ( geo )
class is ( h_geoc ) ! h_v_geoc is an extension of h_geoc
  geo%h_t = z%ZOT_to_geo()
class is ( h_geod ) ! h_v_geod is an extension of h_geod
  geo%h_t = z%ZOT_to_geo()
end select

I can't use a class default block because therein the declared type of
geo would be h_t, and it wouldn't have a nonpolymorphic h_t component to
which I can assign.

> I then want to fill the horizontal geolocations by applying an elemental
> type-bound function to an array of a type ZOT_t, from which latitudes
> and longitudes are computed.  The type ZOT_t represents a uniform
> triangular grid on a (-1:1 X -1:1) square, which is then projected by a
> zenitial orthogonal transformation onto a sphere.  The grid is developed
> to a specified refinement, without regard to whether latitudes computed
> from it will be ultimately interpreted as geocentric or geodetic.
> Therefore, the result of the type-bound function ZOT_to_Geo is
> type(H_t), not class(H_t).  So after allocating Geo, I'd like to assign
> to its horizontal geolocation components using
> 
>   geo%h_t = z%ZOT_to_geo()
> 
> (size(z) == n) but we've only provided for a parent component, not a
> "self" component, so geo%h_t doesn't exist.  I can't use
> 
>   geo = z%ZOT_to_geo()
> 
> for two reasons.  First, geo is polymorphic.  Second, it has a "v"
> component, that I'm not yet ready to fill.
> 
> Is there something simpler than this:
> 
>   block
>     type(h_t) :: T
>     integer :: I
>     do i = 1, n
>       t = z(i)%ZOT_to_geo()
>       geo(i)%lat = t%lat
>       geo(i)%lon = g%lon
>     end do
>   end block
> 
> Ignoring for a while the "v" component of h_v_geod, I tried
> 
>   allocate ( geo, source=h_geod(h_t=z%zot_to_geo()) )
> 
> but that doesn't work because structure constructors aren't elemental
> (how could they be?).  Maybe I could overload h_geod with an elemental
> function; I haven't tried that yet.
> 
> I tried
> 
>   allocate ( h_geod :: geo, source = z%zot_to_geo() )
> 
> but that doesn't work because it has both a type spec and a source.
> 
> I did finally make this work, but it seems like a needlessly wordy
> kludge:
> 
>   allocate ( geo, source = [ ( h_v_geod(h_t=z(i)%zot_to_geo(),v=0), i = 1, n ) ] )
> 
> Isn't there something simpler?
> 
> [BTW, it doesn't work with ifort 15.0.2.164.  It complains about a
> missing field [LAT] not being initialized.  Maybe it's been repaired
> already.  I have not yet been able to convince our SA to install the
> latest Intel release.]
> 
> 
> _______________________________________________
> J3 mailing list
> J3 at mailman.j3-fortran.org
> http://mailman.j3-fortran.org/mailman/listinfo/j3