[PATCH] fortran: Explicitly set name for *LOC default BACK argument [PR108450]

[PATCH] fortran: Explicitly set name for *LOC default BACK argument [PR108450]

[Mikael Morin]

[-- Attachment #1: Type: text/plain, Size: 574 bytes --]

Hello,

this is a fix for a gcc-12 ICE regression.
This ICE rings a bell to me, and I think the change by Tobias which 
triggers it only uncovers a bug that can also happen independently in 
other cases.

The problem is resolution of maxloc expressions is not idempotent, that 
is resolution changes the expression in such a way that it can't be 
successfully resolved again.

I have not tried to prevent multiple resolutions, and fixed instead the 
way the expression is changed.  The patch explains the details.

No regression on x86_64-pc-linux-gnu.
OK for master and 12?

[-- Attachment #2: 0001-fortran-Set-name-for-LOC-default-BACK-argument-PR108.patch --]
[-- Type: text/x-patch, Size: 3027 bytes --]

From e61e2a51e1859f884125670010337f34265997b8 Mon Sep 17 00:00:00 2001
From: Mikael Morin <mikael@gcc.gnu.org>
Date: Sun, 29 Jan 2023 14:38:08 +0100
Subject: [PATCH] fortran: Set name for *LOC default BACK argument [PR108450]

This change fixes an ICE caused by the double resolution of MINLOC,
MAXLOC and FINDLOC expressions which get a default value for the BACK
argument at resolution time.  That argument  is added without name,
and argument reordering code is not prepared to handle unnamed arguments
coming after named ones, so the second resolution causes a NULL pointer
dereference.
The problem is fixed by explicitly setting the argument name.

	PR fortran/108450

gcc/fortran/ChangeLog:

	* check.cc (gfc_check_minloc_maxloc): Explicitly set argument name.
	(gfc_check_findloc): Ditto.

gcc/testsuite/ChangeLog:

	* gfortran.dg/gomp/minmaxloc_1.f90: New test.
---
 gcc/fortran/check.cc                          |  2 ++
 .../gfortran.dg/gomp/minmaxloc_1.f90          | 32 +++++++++++++++++++
 2 files changed, 34 insertions(+)
 create mode 100644 gcc/testsuite/gfortran.dg/gomp/minmaxloc_1.f90

diff --git a/gcc/fortran/check.cc b/gcc/fortran/check.cc
index ebcb8f39852..8c1ae8c2f00 100644
--- a/gcc/fortran/check.cc
+++ b/gcc/fortran/check.cc
@@ -3888,6 +3888,7 @@ gfc_check_minloc_maxloc (gfc_actual_arglist *ap)
     {
       b = gfc_get_logical_expr (gfc_logical_4_kind, NULL, 0);
       ap->next->next->next->next->expr = b;
+      ap->next->next->next->next->name = gfc_get_string ("back");
     }
 
   if (m == NULL && d != NULL && d->ts.type == BT_LOGICAL
@@ -3969,6 +3970,7 @@ gfc_check_findloc (gfc_actual_arglist *ap)
     {
       b = gfc_get_logical_expr (gfc_logical_4_kind, NULL, 0);
       ap->next->next->next->next->next->expr = b;
+      ap->next->next->next->next->next->name = gfc_get_string ("back");
     }
 
   if (m == NULL && d != NULL && d->ts.type == BT_LOGICAL
diff --git a/gcc/testsuite/gfortran.dg/gomp/minmaxloc_1.f90 b/gcc/testsuite/gfortran.dg/gomp/minmaxloc_1.f90
new file mode 100644
index 00000000000..b3691f774de
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/gomp/minmaxloc_1.f90
@@ -0,0 +1,32 @@
+! { dg-do compile }
+!
+! PR fortran/108450
+! This program used to cause an ICE because of the double resolution
+! of the maxloc expression and the addition of a hidden unnamed argument
+! during the first resolution.
+!
+! Original testcase from G. Steinmetz
+
+subroutine s1
+   integer :: a(8) = 0
+   integer :: l
+   integer :: n
+   !$omp atomic
+   n = maxloc(a, mask=l) ! { dg-error ".mask. argument of .maxloc. intrinsic at .1. must be LOGICAL" }
+end
+
+subroutine s2
+   integer :: a(8) = 0
+   integer :: l
+   integer :: n
+   !$omp atomic
+   n = minloc(a, mask=l) ! { dg-error ".mask. argument of .minloc. intrinsic at .1. must be LOGICAL" }
+end
+
+subroutine s3
+   integer :: a(8) = 0
+   integer :: l
+   integer :: n
+   !$omp atomic
+   n = findloc(a, 3, mask=l) ! { dg-error ".mask. argument of .findloc. intrinsic at .1. must be LOGICAL" }
+end
-- 
2.39.0

Re: [PATCH] fortran: Explicitly set name for *LOC default BACK argument [PR108450]

[Harald Anlauf]

Hi Mikael,

Am 29.01.23 um 17:21 schrieb Mikael Morin:
> Hello,
>
> this is a fix for a gcc-12 ICE regression.
> This ICE rings a bell to me, and I think the change by Tobias which
> triggers it only uncovers a bug that can also happen independently in
> other cases.
>
> The problem is resolution of maxloc expressions is not idempotent, that
> is resolution changes the expression in such a way that it can't be
> successfully resolved again.
>
> I have not tried to prevent multiple resolutions, and fixed instead the
> way the expression is changed.  The patch explains the details.
>
> No regression on x86_64-pc-linux-gnu.
> OK for master and 12?

yes, ok for both.

Thanks for the patch!

Harald