Phillip Helbig writes:
> To me, it seems that pointing to an object and having the size
> determined at run time are two such completely different concepts that
> I'm wondering why the former is used for the latter at all....
>
> When the present kludges are no longer necessary, what about making them
> obsolete and then deleting them? Otherwise, they are bound to cause
> much more confusion than the FORTRAN IV stuff still kicking around.
The present kludge of using pointers where you'd really prefer
allocatables is because of omissions in the language (namely the
omission of the ability to use allocatables in some important
contexts). I can't imagine what it is that you would be talking about
obsoleting. Well, I suppose one could say that we are obsoleting the
omission. There are roles for pointers and there are roles for
allocatables. I would vehemently oppose deleting either.
Perhaps you are talking about disallowing pointers to arrays. I
suppose that's my best guess about what you must mean. (Because I
really can't think of anything else you could mean). If so, I would
vehemently oppose that also. Doing so would involve major work to
provide suitable replacement functionality such as the current ability
to provide pointers to array slices.
About now, Giles will probably jump in and note that he never did
like allowing pointers to array slices and that the functionaility
ought to have been provided by something more like the IDENTIFY
proposal in some of the f8x drafts. I don't entirely agree, but
I'll not argue the point.
What I will argue is that, even if you decided to do something
like that, it would be a major addition to the language. Such
an addition is not part of any currently approved proposal. One
would have to do something about adding in whatever was going to
be the replacement functionality before one is likely to get very
far in a discussion of obsoleting pointers to array slices.
Replacement functionality is not in the f2k plans. So you are
either talking about the next revision after f2k, or you are
talking about trashing the f2k plans enough that its time scale
would be majorly set back. So if replacement functionality is
added in f2k+1 (whatever that might be), you'd need to be
talking at least f2k+2 before it was reasonable to consider
obsolescence.
That schedule assume that you can get agreement on such a direction.
I don't think you'd find anything particularly close to consensus
about taking such a direction. (For instance, I don't like it).
--
Richard Maine
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|