(j3.2006) (SC22WG5.3816) [ukfortran] Ballot on the technical content of the TR
Aleksandar Donev
donev1 at llnl.gov
Wed Dec 10 16:48:47 EST 2008
Hi Jim,
> This may sound like arguing for some other vendors, but Sun just said
> they don't have the rank information in their descriptors.
A few vendors also does not have type information, which is also
required to be carried around by BIND(C) routines in the TR. Think
of "TYPE(*), DIMENSION(:)" and how that would be passed.
There is the option that type and rank info be removed from the
descriptor structures in the TR. It has been discussed before. IMO, it
seriously limits the usefulness of descriptors, since the user has to
construct his own descriptors to pass the type and rank around (like
the hidden length for characters). But, it will eliminate the "this or
that vendor is disadvantaged" argument. Ultimately, it comes to the
good old users vs. vendor argument...
I personally can go along with either, since at least there is a
reasonable technical argument for either. But of course, I prefer if
compilers put type/rank info into the descriptor, based on the
actual---much safer and easier to use.
> Is there
> any evidence this assumed-rank is needed beyond IGNORE_TKR?
I am not sure what "evidence" means in this context? You want statistics
polling users? You want an example of where it would be useful? I
hardly think anything I say will make a difference---people usually
have a pre-determined mind about this, just like those oldies that
say "don't use assumed-shape arrays" have made their mind and will
never change it...
Best,
Aleks
More information about the J3
mailing list