Dear experts,
is the following where-construct allowed ?

real, dimension(045) :: A
real, dimension(100) :: B

! fill A with numbers - invalid numbers are indicated by -9999.

! now fill B
where( A /= -9999. )
B = A
else where
B = -9999.
end where

For this construct it is always Size(B) > Size(A) with shown
dimensions.

The Absoft Compiler does not complain (v9.0)

Thanks
-Ralf

On 12 sep, 10:06, "ralf.schaa" gmail.com> wrote:

On Sep 12, 7:06 pm, "ralf.schaa" gmail.com> wrote:
[snip]

and an update from Gfortran 4.3.1 and CVF 6.6C: both complain about
the different array sizes

On Sep 12, 7:40 pm, Arjen Markus wrote:

[snip]

yeah, I tested it just a minute ago ...

thanks Arjen - I give up the where-construct. It's a pity since it is
fancy and short ...

-Ralf

program test
implicit none
INTEGER n
real, dimension(045) :: A
real, dimension(100) :: B

ralf.schaa gmail.com> wrote:

[where with non-conformable arrays]

While the answer was implied by other replies, nobody explicitly
answered this from a standard perspective other than for Arjen's "I do
not think so."

I'll verify Arjen's answer. No, it is not allowed. (And it wouldn't be
consistent to allow it any more than lots of other whole array things
with nonconformable arrays; one could imagine definitions that used just
part of the arrays in some cases, but that's not how the language does
it anywhere.)

It isn't the kind of error that compilers are required to catch, but of
course, doing so where practical is a desirable feature in a compiler.

You can still use the WHERE statement to do what you want cleanly and
simply (2 short lines).

! Fill all elements of B() with invalid numbers
! Then reset values in B() that correspond to valid elements in A()

B = -9999.
WHERE( A /= -9999. ) B(1:size(A)) = A

Dan

On Sep 13, 2:48 am, Dan aol.com> wrote:

And ineffiently: many elements of B will be set twice. That's what the
elsewhere clause is for.

Regards,

Mike Metcalf

"Dan" aol.com> wrote in message
news:615961ed-04e1-4d20-927d-1d31891d620f@a3g2000prm.googlegroups.com...

Way use where at all?

B(1:min(size(A),size(B))) = A(1:min(size(A),size(B)))
B(size(A)+1:)= -9999.

will do the work even if size(A) > size(B).

For the OP's dimensions its enough with:

michaelmetcalf@compuserve.com wrote:

It is hard to make generalizations on efficiency, but
it is likely that testing a value is slower than
assigning a value. One possible exception would be
avoiding the write to the swap file if you are
thrashing virtual storage. There are better fixes
in that case, though.

-- glen