[Patch, fortran] PR114535 - [13/14 regression] ICE with elemental finalizer

[Patch, fortran] PR114535 - [13/14 regression] ICE with elemental finalizer

[Paul Richard Thomas]

[-- Attachment #1.1: Type: text/plain, Size: 753 bytes --]

Hi All,

This one is blazingly 'obvious'. I haven't had the heart to investigate why
somebody thought that it is a good idea to check if unreferenced symbols
are finalizable because, I suspect, that 'somebody' was me. Worse, I tried
a couple of other fixes before I hit on the 'obvious' one :-(

The ChangeLog says it all. OK for mainline and then backporting in a couple
of weeks?

Paul

Fortran: Fix ICE in trans-stmt.cc(gfc_trans_call) [PR114535]

2024-04-08  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/114535
* resolve.cc (resolve_symbol): Remove last chunk that checked
for finalization of unreferenced symbols.

gcc/testsuite/
PR fortran/114535
* gfortran.dg/pr114535d.f90: New test.
* gfortran.dg/pr114535iv.f90: Additional source.

[-- Attachment #2: submit.diff --]
[-- Type: text/x-patch, Size: 2526 bytes --]

diff --git a/gcc/fortran/resolve.cc b/gcc/fortran/resolve.cc
index 43315a6a550..4cbf7186119 100644
--- a/gcc/fortran/resolve.cc
+++ b/gcc/fortran/resolve.cc
@@ -17069,15 +17069,6 @@ resolve_symbol (gfc_symbol *sym)
 
   if (sym->param_list)
     resolve_pdt (sym);
-
-  if (!sym->attr.referenced
-      && (sym->ts.type == BT_CLASS || sym->ts.type == BT_DERIVED))
-    {
-      gfc_expr *final_expr = gfc_lval_expr_from_sym (sym);
-      if (gfc_is_finalizable (final_expr->ts.u.derived, NULL))
-	gfc_set_sym_referenced (sym);
-      gfc_free_expr (final_expr);
-    }
 }
 
 
diff --git a/gcc/testsuite/gfortran.dg/pr114535d.f90 b/gcc/testsuite/gfortran.dg/pr114535d.f90
new file mode 100644
index 00000000000..7ce178a1e30
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/pr114535d.f90
@@ -0,0 +1,42 @@
+! { dg-do compile }
+! { dg-compile-aux-modules "pr114535iv.f90" }
+! Contributed by Andrew Benson  <abensonca@gcc.gnu.org>
+!
+module d
+  implicit none
+contains
+  function en() result(dd)
+    use :: iv
+    implicit none
+    type(vs) :: dd
+    dd%i = 1
+  end function en
+end module d
+
+! Delete line 1 and all brands complain that 'vs' is an undefined type.
+! Delete lines 1 and line 2 recreates the original problem.
+module ni
+  implicit none
+contains
+  subroutine iss1()
+!    use :: iv                                        ! line 1
+    use :: d
+    implicit none
+!    type(vs) :: ans; ans = en();                     ! line 2
+  end subroutine iss1
+  subroutine iss2()
+    use :: d
+    implicit none
+  end subroutine iss2
+end module ni ! Used to give an ICE: in gfc_trans_call, at fortran/trans-stmt.cc:400
+
+  use ni
+  use iv
+  type(vs) :: x
+  call iss1()
+  call iss1()
+  if ((ctr .eq. 0) .or. (ctr .ne. 6)) stop 1  ! Depends whether lines 1 & 2 are present
+  call iss2()
+  x = vs(42)
+  if ((ctr .eq. 1) .or. (ctr .ne. 7)) stop 2  ! Make sure destructor available here
+end
diff --git a/gcc/testsuite/gfortran.dg/pr114535iv.f90 b/gcc/testsuite/gfortran.dg/pr114535iv.f90
new file mode 100644
index 00000000000..be629991023
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/pr114535iv.f90
@@ -0,0 +1,18 @@
+! Compiled with pr114535d.f90
+! Contributed by Andrew Benson  <abensonca@gcc.gnu.org>
+!
+module iv
+  type, public :: vs
+     integer :: i
+   contains
+     final :: destructor
+  end type vs
+  integer :: ctr = 0
+contains
+  impure elemental subroutine destructor(s)
+    type(vs), intent(inout) :: s
+    s%i = 0
+    ctr = ctr + 1
+  end subroutine destructor
+end module iv
+

Re: [Patch, fortran] PR114535 - [13/14 regression] ICE with elemental finalizer

[Jerry D]

On 4/8/24 2:45 AM, Paul Richard Thomas wrote:
> Hi All,
> 
> This one is blazingly 'obvious'. I haven't had the heart to investigate why somebody thought that it is a good idea to check if unreferenced symbols are finalizable because, I suspect, that 
> 'somebody' was me. Worse, I tried a couple of other fixes before I hit on the 'obvious' one :-(
> 
> The ChangeLog says it all. OK for mainline and then backporting in a couple of weeks?
> 
> Paul
> 
> Fortran: Fix ICE in trans-stmt.cc(gfc_trans_call) [PR114535]
> 
> 2024-04-08  Paul Thomas  <pault@gcc.gnu.org <mailto:pault@gcc.gnu.org>>
> 
> gcc/fortran
> PR fortran/114535
> * resolve.cc (resolve_symbol): Remove last chunk that checked
> for finalization of unreferenced symbols.
> 
> gcc/testsuite/
> PR fortran/114535
> * gfortran.dg/pr114535d.f90: New test.
> * gfortran.dg/pr114535iv.f90: Additional source.
> 

Yes, OK Paul.  Don't feel bad. It wasn't Tabs. LOL

Jerry