[patch, fortran] Fix PR 85896, type confusion with min and max

[patch, fortran] Fix PR 85896, type confusion with min and max

[Thomas Koenig]

[-- Attachment #1: Type: text/plain, Size: 598 bytes --]

Hello world,

the attached patchlet fixes a rejects-valid bug by simply ignoring the
type for max and min during simplification.  This is correct
because setting the type of a generic intrinsic function has
no effect.

It is a rare pleasure to fix a bug by removing code only :-)

Regression-tested. OK for trunk?

Regards

	Thomas

2018-10-30  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR fortran/85896
	* simplify.c (simplify_min_max): Do not convert the type of the
	return expression.

2018-10-30  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR fortran/85896
	* gfortran.dg/min_max_type.f90: New test.

[-- Attachment #2: p1.diff --]
[-- Type: text/x-patch, Size: 981 bytes --]

Index: simplify.c
===================================================================
--- simplify.c	(Revision 265502)
+++ simplify.c	(Arbeitskopie)
@@ -4961,11 +4961,9 @@ static gfc_expr *
 simplify_min_max (gfc_expr *expr, int sign)
 {
   gfc_actual_arglist *arg, *last, *extremum;
-  gfc_intrinsic_sym * specific;
 
   last = NULL;
   extremum = NULL;
-  specific = expr->value.function.isym;
 
   arg = expr->value.function.actual;
 
@@ -4995,15 +4993,6 @@ simplify_min_max (gfc_expr *expr, int sign)
   if (expr->value.function.actual->next != NULL)
     return NULL;
 
-  /* Convert to the correct type and kind.  */
-  if (expr->ts.type != BT_UNKNOWN)
-    return gfc_convert_constant (expr->value.function.actual->expr,
-	expr->ts.type, expr->ts.kind);
-
-  if (specific->ts.type != BT_UNKNOWN)
-    return gfc_convert_constant (expr->value.function.actual->expr,
-	specific->ts.type, specific->ts.kind);
-
   return gfc_copy_expr (expr->value.function.actual->expr);
 }
 

Re: [patch, fortran] Fix PR 85896, type confusion with min and max

[Janne Blomqvist]

On Tue, Oct 30, 2018 at 8:57 PM Thomas Koenig <tkoenig@netcologne.de> wrote:

> Hello world,
>
> the attached patchlet fixes a rejects-valid bug by simply ignoring the
> type for max and min during simplification.  This is correct
> because setting the type of a generic intrinsic function has
> no effect.
>
> It is a rare pleasure to fix a bug by removing code only :-)
>
> Regression-tested. OK for trunk?
>

Ok, thanks.

-- 
Janne Blomqvist