[Bug fortran/123321] Namelist input of variable accessed by use association with renaming fails

[van.snyder at sbcglobal dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123321

--- Comment #13 from van.snyder at sbcglobal dot net <van.snyder at sbcglobal dot net> ---
On Sat, 2025-12-27 at 23:18 +0000, kargl at gcc dot gnu.org wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123321
> 
> --- Comment #11 from Steve Kargl <kargl at gcc dot gnu.org> ---
> (In reply to van.snyder@sbcglobal.net from comment #9)
> > On Sat, 2025-12-27 at 22:06 +0000, kargl at gcc dot gnu.org wrote:
> > > >      C8107 (R871) The namelist-group-name shall not be a name
> > > >      accessed by use association.
> > > > 
> > > > and it appears that it is not handled correctly.
> > > > 
> > > > gfortran.dg/namelist_use_only.f90 is using a namelist-group-
> > > > name
> > > > from the
> > > > module.
> > > 
> > > Need to think about this a bit more.  I may be misinterpreting
> > > C8107.
> > 
> > 
> > In the bug I posted, the namelist-group-name is local to the main
> > program, not accessed by use association. It's a namelist-group-
> > object,
> > which 8.9 "Namelist Statement" p5 (F18, and F25 draft) explicitly
> > says
> > can be accessed by use association, that apparently can't be
> > renamed
> > and then the local name used in a namelist statement.
> 
> Thanks for the follow up.  Every time I look at a namelist bug,
> I have to go relearn what a namelist is and how to use it. :).

I have taken to using namelist to process command-line arguments. I put
the "&nml " in the beginning of a character variable, then read the
command-line argument after it. If it contains "=" I append "/" and
read with namelist, and if not I do other stuff with it. The standard
says character data in namelist has to be quoted, and shells "help"
with quoting, so I need to add more quotes. I saw a but report that
gfortran allows character data in namelist without quotes. That's
harmless and helpful; a feature not a bug.

Here's an example:

./Invert_Design -o 7/test 'prefix="7/D-" n_design=7 rotate=f show=t
max_poisson=5 coolant_temperature=700 distance_method="M" h_value=0.7'
….

> Now, the interesting case is gfortran.dg/namelist_use_only.f90, which
> I think is invalid but gfortran currently accepts!  That code reduces
> to

Yes, this one should be prohibited by C8103 (in the F18 standard).

I don't see a reason for C8103, so long as every namelist-group-object
is accessible where the namelist-stmt appears.

>   module foo
>     implicit none
>     integer i
>     namelist /nml1/i
>   end module foo
> 
>   program bar
>     use foo, only : k => i, nml1
>     implicit none
>     character(len=10) :: str ="&nml1 i=1/"
>     read(str,nml=nml1)
>     if (k /= 1) stop 1
>   end program bar
> 
> Here, 'i' is renamed to 'k' and the namelist 'nml1' is use
> associated.
> gfortran compiles the above and the binary executes.  Note, 'i' is
> inaccessible, so the read() statement I believe is invalid due to
> 
>   F2023, 8.9 NAMELIST statement
>   ...
>   5 A namelist group object shall either be accessed by use or host
> association
>     or shall have its declared type, kind type parameters of the
> declared type,
>     and rank specified by previous statements in the same scoping
> unit or
>     by the implicit typing rules in effect for the scoping unit.
> 
> With my patch, the above code yields
> % gfcx -o z w2.f90 
> % ./z
> At line 11 of file w2.f90
> Fortran runtime error: Cannot match namelist object name i

Other than getting the namelist-group-name by use association, this is
what I would expect.