Spec 2005 and gfortran

[THOMAS Paul Richard 169137]

These bugs are amongst the most-wanted of all:

In pr15502 there are:

1] 12840 - Unable to find scalarization loop specifier

The likes of m = SUM((/ (m+1,m=0,order) /)) fail with a gfc_todo.

2] 17917 - ICE on module equivalence

I submitted a partial patch for this.  The equivalences only reside in the
scope in which they were associated; they need to be treated like module
common blocks.

3] 18108 - generic interfaces in modules where implicit none has been
declared.

As far as I can tell, I have fixed this with my fix for PR16940. 
http://gcc.gnu.org/ml/gcc-cvs/2005-07/msg00704.html and
 http://gcc.gnu.org/ml/gcc-cvs/2005-07/msg00824.html
I missed this PR completely.

4] 21063 - ICE in gfc_conv_ss_descriptor, at fortran/trans-array.c:1224
after using maxloc function; 

ie. this is broken -
real, dimension(100) :: foo
.......
write(*,*) foo(maxloc(dummy))

This is related to the failure of the likes of 
write (*,*) foo((/2,3,7/)) because the gcc_assert only allows variables
through gfc_conv_ss_descriptor.

This bug breaks the Polyhedron testsuite in a number of places.  Also, the
setting of index vectors by initialisers or functions, seems like exactly
the sort of thing that one ones to do with f95!

If I undertake to finish the fix to PR17917 as my next job, how about
somebody, with familiarity of the scalarizer, doing likewise for PR12840 and
PR21063 - Paul, Steven, .....?

Paul T