Re: [PATCH] Fortran: fix passing a procedure pointer to c_funloc [PR124652]

Le 01/04/2026 à 22:38, Harald Anlauf a écrit :
> Dear all,
> 
> the attached almost obvious patch fixes actually two issues:
> - when C_FUNLOC is applied to a procedure pointer instead of
>    a procedure, and the procedure pointer is a dummy, we need
>    to dereference it;
> - a procedure pointer dummy with intent(out) was clobbered,
>    but the clobber was using the dereferenced pointer, which
>    did fail in gimple at -O1 and higher (and is actually
>    detecting wrong code).
>    The solution is to   not clobber (and we have a couple of
>    similar exceptions).
> 
> Regtested on x86_64-pc-linux-gnu.  OK for mainline?
> 
> As the above issues can be naughty, is it OK to backport to 15-branch?
> 
> Thanks,
> Harald
> 
> pr124652.diff
> 
> From b76f5b9108d8ac1dfaa54817d42fae292451ff02 Mon Sep 17 00:00:00 2001
> From: Harald Anlauf <[email protected]>
> Date: Wed, 1 Apr 2026 22:28:02 +0200
> Subject: [PATCH] Fortran: fix passing a procedure pointer to c_funloc
>  [PR124652]
> 
> 	PR fortran/124652
> 
> gcc/fortran/ChangeLog:
> 
> 	* trans-expr.cc (gfc_conv_procedure_call): Do not clobber a
> 	procedure pointer intent(out) argument.
> 	* trans-intrinsic.cc (conv_isocbinding_function): When passing to
> 	C_FUNLOC a procedure pointer that is a dummy, dereference it.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* gfortran.dg/c_funloc_tests_10.f90: New test.
> ---
>  gcc/fortran/trans-expr.cc                     |  1 +
>  gcc/fortran/trans-intrinsic.cc                |  3 +
>  .../gfortran.dg/c_funloc_tests_10.f90         | 86 +++++++++++++++++++
>  3 files changed, 90 insertions(+)
>  create mode 100644 gcc/testsuite/gfortran.dg/c_funloc_tests_10.f90
> 
> diff --git a/gcc/fortran/trans-expr.cc b/gcc/fortran/trans-expr.cc
> index 52918961584..3945d9eaa67 100644
> --- a/gcc/fortran/trans-expr.cc
> +++ b/gcc/fortran/trans-expr.cc
> @@ -7549,6 +7549,7 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
>  				  && !e->ts.u.derived->attr.alloc_comp
>  				  && !e->ts.u.derived->attr.pdt_type
>  				  && !gfc_is_finalizable (e->ts.u.derived, NULL)))
> +			  && e->ts.type != BT_PROCEDURE
>  			  && !sym->attr.elemental)
>  			{
>  			  tree var;

I agree, if the argument is a dummy procedure or dummy procedure pointer 
we don't want the clobber, but if the argument is a local procedure 
pointer, we do want the clobber, don't we?

I think we should check that the var clobbered is a VAR_DECL, or maybe 
a PARM_DECL or RESULT_DECL (or a conversion of such a DECL?).

If you don't have time to refine it further, please go ahead with your 
patch as is.  At least it prevents the ICE.

Mikael