Re: [gfortran] Add malloc and free intrinsics

[Brooks Moses <bmoses at stanford dot edu>]

(I haven't been around long enough to make anything like an official 
interpretation on this matter, but I think my unofficial opinions might 
be useful, so....)

Janne Blomqvist wrote:
> On Tue, Sep 27, 2005 at 09:34:17AM +0200, François-Xavier Coudert wrote:
> > > I'd like this feature to be marked with GFC_STD_LEGACY rather than
> > > GFC_STD_GNU.
> >
> > Well, in my understanding GFC_STD_LEGACY was to be used for features
> > we only support because they were implemented in g77.

My understanding is that this is almost correct, but should read "in g77 
or other in legacy compilers".  (Otherwise, it would make more sense to 
just call it GFC_STD_G77.)

> Isn't that what we have GFC_STD_GNU for? Or in general, non-standard
> intrinsics that we provide since they provide some useful
> functionality that the standard by itself doesn't provide. E.g. the
> unix intrinsics, extra math functions etc.

I agree with your second sentence -- GFC_STD_GNU is for non-standard 
things that we provide because we think they're useful.

This is, in many cases, different from things that were included in g77 
because the implementers of g77 thought them useful.  So, I disagree 
with your first sentence -- GFC_STD_GNU is _not_ for things we support 
just because g77 did.  It should only include things from g77 that are 
still useful in light of Fortran-95's changes to the language.

> My understanding is that GFC_STD_LEGACY essentially is "stuff we
> support since it's useful for legacy code but which the current
> standard we support (F95) provides in another incompatible form".
>
> E.g. cray pointers/malloc/free vs. allocatables/pointers/(de)allocate,
> and Hollerith vs. character.
>
> Well, who knows the official definition of STD_LEGACY and STD_GNU?

Is there one?

If not, I propose the definitions I gave at the top of this post.  :)

The core distinction is that GFC_STD_GNU is stuff that we hope people 
will use when writing new code, and GFC_STD_LEGACY is stuff that we hope 
they won't.


(In any case, I'll agree with Francois that malloc/free and Cray 
pointers should be treated the same way, whatever the end result is.)

- Brooks