> Richard Maine wrote:
>

>>
>> Yes. Heck, it wasnt too awfully long ago that Fortran had *NO* dynamic
>> allocation of any kind. Dynamic allocation wasn't standard in Fortran
>> until f90. Prior to that, some codes used nonstandard nonportable hacks
>> of various kinds.

>
>
> I work regularly with a legacy F77 program that has author-written
> DMA. (I'm not the author.)
> The program structure permits a stack sort of allocation.
> At the beginning you allocate one humongous array X, and
> put the array and an integer pointer to the "next available element"
> in common. When you need some more memory, you start loading
> stuff in the next available element, and update the pointer to
> the next place above where you filled. To deallocate, you just set
> the pointer back where it was. You can only free stuff
> from the top of the stack down; no cutting holes in the middle.
> But that's okay for the way this calculation works. To pass a workspace
> array of length 100 to a subroutine, you do
> temp = nextavail
> nextavail = nextavail+100 ! requests allocation
> call checkstorage(nextavail) !checks to make you've not overflowed
> call workroutine(x(temp)) ! passes array address
> nextavail = temp !deallocates
>
> The reason I'm writing this: this simple scheme IS portable.
> I've used it on a dozen architectures with almost as many
> Fortran compilers and it has never caused trouble. So for a limited
> class of needs, there was a portable way to do allocations.
>

> -George <- likes F90 allocations better :-)