[Bug fortran/122977] [15/16 Regression] Wrongly rejects associate to a contiguous pointer

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122977

--- Comment #6 from anlauf at gcc dot gnu.org ---
(In reply to Steve Kargl from comment #5)
> On Wed, Dec 03, 2025 at 08:44:39PM +0000, anlauf at gcc dot gnu.org wrote:
> > 
> > --- Comment #4 from anlauf at gcc dot gnu.org ---
> > (In reply to Steve Kargl from comment #3)
> > > 
> > > I haven't looked at the issue with gfc_is_simply_contiguous() in
> > > comment #2, but if it passes regression testing, and you have a
> > > testcase (or two) feel free to commit.  The comment in the patch
> > > explains the rationale.
> > 
> > The change in comment#2 regresses on gfortran.dg/associate_11.f90
> > because it will pack the associate variable before passing to a subroutine.
> > I don't have a simple answer yet.
> 
> I see.  In looking at the symbol_attribute struct in gfortran.h,
> does the packing lead to attr.temporary or attr.artificial being
> set?  These are the only ones that look like they may convey
> some needed info; otherwise, we could add a new packed_arg attribute.

Unfortunately I do not fully understand what you mean.

Determining the properties of the associate-variable at compile time
will need more work.  Associating with a contiguous pointer will make
it contiguous.  As of yet, associating it with a contiguous array could be
determined for some cases, but we have no means to pass that information.

(We could mis-use the contiguous attribute and set it in resolve_assoc_var
if the target is non-pointer but known (or guessed) to be contiguous.
Or add a new attribute contiguous_assoc_target - or some longer name... :)

I am open for opinions.