Re: [PATCH] PR fortran/84734 -- Fix ICE on invalid code

Re: [PATCH] PR fortran/84734 -- Fix ICE on invalid code

[Steve Kargl]

[-- Attachment #1: Type: text/plain, Size: 1351 bytes --]

On Fri, Mar 09, 2018 at 09:13:10PM -0800, Steve Kargl wrote:
> In fixing PR fortran/83633, it seems the patch I committed
> introduced an ICE for nonsensical invalid Fortran.  The 
> attached patch cures the ICE and now (re)issues an error
> message.
> 
> The basic problem seems to boil down to the recursive 
> calling of gfc_simplify_expr reduces "huge(1_8)+1_8" to
> "constant + constant".  When the chain of gfc_simplify_expr
> tries to reduces this expression an overflow occurs.  An
> error message is queud but never emitted, and the result
> is set to NULL and both constants are freed.  The NULL is
> passed back up through the chain of gfc_simplify_expr.
> At some point that NULL pointer is referenced.  The patch
> works around the problem by passing the result with the
> overflow value up the chain.  
> 
> Regression tested on x86_64-*-freebsd.  I intend to commit
> this patch tomorrow, which on my clock is only 2.75 hours
> away.
> 
> 2018-03-09  Steven G. Kargl  <kargl@gcc.gnu.org>
> 
>         PR fortran/84734
>         * arith.c (check_result, eval_intrinsic):  If result overflows, pass
>         the expression up the chain instead of a NULL pointer.
> 
> 2018-03-09  Steven G. Kargl  <kargl@gcc.gnu.org>
> 
>         PR fortran/84734
>         * gfortran.dg/pr84734.f90: New test.
> 

Now with an attached patch.

-- 
Steve

[-- Attachment #2: pr84734.diff --]
[-- Type: text/x-diff, Size: 1096 bytes --]

Index: gcc/fortran/arith.c
===================================================================
--- gcc/fortran/arith.c	(revision 258367)
+++ gcc/fortran/arith.c	(working copy)
@@ -555,10 +555,10 @@ check_result (arith rc, gfc_expr *x, gfc_expr *r, gfc_
       val = ARITH_OK;
     }
 
-  if (val != ARITH_OK)
-    gfc_free_expr (r);
-  else
+  if (val == ARITH_OK || val == ARITH_OVERFLOW)
     *rp = r;
+  else
+    gfc_free_expr (r);
 
   return val;
 }
@@ -1603,8 +1603,12 @@ eval_intrinsic (gfc_intrinsic_op op,
   if (rc != ARITH_OK)
     {
       gfc_error (gfc_arith_error (rc), &op1->where);
+      if (rc == ARITH_OVERFLOW)
+	goto done;
       return NULL;
     }
+
+done:
 
   gfc_free_expr (op1);
   gfc_free_expr (op2);
Index: gcc/testsuite/gfortran.dg/pr84734.f90
===================================================================
--- gcc/testsuite/gfortran.dg/pr84734.f90	(nonexistent)
+++ gcc/testsuite/gfortran.dg/pr84734.f90	(working copy)
@@ -0,0 +1,4 @@
+! { dg-do compile }
+! PR fortran/84734
+   integer :: b(huge(1_8)+1_8) = 0 ! { dg-error "Arithmetic overflow" }
+   end

[PATCH] PR fortran/84734 -- Fix ICE on invalid code

[Steve Kargl]

In fixing PR fortran/83633, it seems the patch I committed
introduced an ICE for nonsensical invalid Fortran.  The 
attached patch cures the ICE and now (re)issues an error
message.

The basic problem seems to boil down to the recursive 
calling of gfc_simplify_expr reduces "huge(1_8)+1_8" to
"constant + constant".  When the chain of gfc_simplify_expr
tries to reduces this expression an overflow occurs.  An
error message is queud but never emitted, and the result
is set to NULL and both constants are freed.  The NULL is
passed back up through the chain of gfc_simplify_expr.
At some point that NULL pointer is referenced.  The patch
works around the problem by passing the result with the
overflow value up the chain.  

Regression tested on x86_64-*-freebsd.  I intend to commit
this patch tomorrow, which on my clock is only 2.75 hours
away.

2018-03-09  Steven G. Kargl  <kargl@gcc.gnu.org>

        PR fortran/84734
        * arith.c (check_result, eval_intrinsic):  If result overflows, pass
        the expression up the chain instead of a NULL pointer.

2018-03-09  Steven G. Kargl  <kargl@gcc.gnu.org>

        PR fortran/84734
        * gfortran.dg/pr84734.f90: New test.

-- 
Steve