[patch, fortran] PR39239 Warning about EQUIVALENCE and VOLATILE

[patch, fortran] PR39239 Warning about EQUIVALENCE and VOLATILE

[Nicolas Koenig]

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

Hello everyone,

a simple patch to throw a warning if not all and not none of the 
equivalence objects are volatile. (And the according modification of 
gfortran.dg/volatile11.f90)

Nicolas

Regression tested for:

GNU Fortran (GCC) 7.0.1 20170311 (experimental)

Changelog:

2017-03-13  Nicolas Koenig  <koenigni@student.ethz.ch>

                 PR fortran/39239
                 * resolve.c (resolve_equivalence): Warn if not either 
none or all equivalence objects are volatile
                 * gfortran.dg/volatile11.f90: Changed test to test for 
the new warning



[-- Attachment #2: p7.diff --]
[-- Type: text/x-patch, Size: 2395 bytes --]

Index: fortran/resolve.c
===================================================================
--- fortran/resolve.c	(revision 246070)
+++ fortran/resolve.c	(working copy)
@@ -15560,7 +15560,7 @@ resolve_equivalence (gfc_equiv *eq)
   locus *last_where = NULL;
   seq_type eq_type, last_eq_type;
   gfc_typespec *last_ts;
-  int object, cnt_protected;
+  int object, cnt_protected, cnt_volatile;
   const char *msg;
 
   last_ts = &eq->expr->symtree->n.sym->ts;
@@ -15569,6 +15569,8 @@ resolve_equivalence (gfc_equiv *eq)
 
   cnt_protected = 0;
 
+  cnt_volatile = 0;
+
   for (object = 1; eq; eq = eq->eq, object++)
     {
       e = eq->expr;
@@ -15641,6 +15643,17 @@ resolve_equivalence (gfc_equiv *eq)
 
       sym = e->symtree->n.sym;
 
+      if(sym->attr.volatile_)
+        cnt_volatile++;
+      if(cnt_volatile > 0 && cnt_volatile != object)
+	{
+	  gfc_warning (0, "Either all or none of the objects in "
+	  	       "the EQUIVALENCE set at %L shall have the "
+		       "VOLATILE attribute",
+		       &e->where);
+	  break;
+	}
+
       if (sym->attr.is_protected)
 	cnt_protected++;
       if (cnt_protected > 0 && cnt_protected != object)
Index: testsuite/gfortran.dg/volatile11.f90
===================================================================
--- testsuite/gfortran.dg/volatile11.f90	(revision 246070)
+++ testsuite/gfortran.dg/volatile11.f90	(working copy)
@@ -1,8 +1,9 @@
 ! { dg-do compile }
-! { dg-options "-O2 -fdump-tree-optimized" }
+! { dg-options "-Wall -O2 -fdump-tree-optimized" }
 ! Tests that volatile can be applied to members of common blocks or
 ! equivalence groups (PR fortran/35037)
 !
+
 subroutine wait1
   logical event
   volatile event
@@ -14,26 +15,10 @@ subroutine wait1
 end subroutine
 
 subroutine wait2
-  logical event, foo
-  volatile event
-  equivalence (event, foo)
-  event = .false.
-  do
-    if (event) print *, 'NotOptimizedAway2'
-  end do
-end subroutine
-
-subroutine wait3
   logical event
   integer foo
   volatile foo
-  equivalence (event, foo)
-  event = .false.
-  do
-    if (event) print *, 'IsOptimizedAway'
-  end do
+  equivalence (event, foo) ! { dg-warning "in the EQUIVALENCE set" } 
 end subroutine
 
 ! { dg-final { scan-tree-dump "NotOptimizedAway1" "optimized" } } */
-! { dg-final { scan-tree-dump "NotOptimizedAway2" "optimized" } } */
-! { dg-final { scan-tree-dump-not "IsOptimizedAway" "optimized" } } */

Re: [patch, fortran] PR39239 Warning about EQUIVALENCE and VOLATILE

[Jerry DeLisle]

On 03/14/2017 01:17 PM, Nicolas Koenig wrote:
> Hello everyone,
> 
> a simple patch to throw a warning if not all and not none of the equivalence
> objects are volatile. (And the according modification of
> gfortran.dg/volatile11.f90)
> 
> Nicolas
> 
> Regression tested for:
> 
> GNU Fortran (GCC) 7.0.1 20170311 (experimental)
> 
> Changelog:
> 
> 2017-03-13  Nicolas Koenig  <koenigni@student.ethz.ch>
> 
>                 PR fortran/39239
>                 * resolve.c (resolve_equivalence): Warn if not either none or
> all equivalence objects are volatile
>                 * gfortran.dg/volatile11.f90: Changed test to test for the new
> warning
> 
> 

Hi Nicolas,

Thanks for starting in on this.

Since this results in a warning, maybe change the wording from 'shall' to should.

I did not dig into the Fortran Standards so I assume it need not be an error.

Also when you submit a patch, please also let us know what platform you
regression tested on, such as x86-64-linux, or Windows, or similar. (You can get
the whole string from subdirectory names in build directory. On mine its
x86_64-pc-linux-gnu) Sometimes we accidentally break things on different
platforms. so this way we can see it tested ok over here and seems to fail over
there.

Your patch has changed some of the scan dumps and I am wondering if you have
deleted something we use to check for?

Jerry

Re: [patch, fortran] PR39239 Warning about EQUIVALENCE and VOLATILE

[Nicolas Koenig]

[-- Attachment #1: Type: text/plain, Size: 1823 bytes --]



On 03/14/2017 10:42 PM, Jerry DeLisle wrote:
> On 03/14/2017 01:17 PM, Nicolas Koenig wrote:
>> Hello everyone,
>>
>> a simple patch to throw a warning if not all and not none of the equivalence
>> objects are volatile. (And the according modification of
>> gfortran.dg/volatile11.f90)
>>
>> Nicolas
>>
>> Regression tested for:
>>
>> GNU Fortran (GCC) 7.0.1 20170311 (experimental)
>>
>> Changelog:
>>
>> 2017-03-13  Nicolas Koenig  <koenigni@student.ethz.ch>
>>
>>                  PR fortran/39239
>>                  * resolve.c (resolve_equivalence): Warn if not either none or
>> all equivalence objects are volatile
>>                  * gfortran.dg/volatile11.f90: Changed test to test for the new
>> warning
>>
>>
> Hi Nicolas,
>
> Thanks for starting in on this.
>
> Since this results in a warning, maybe change the wording from 'shall' to should.
>
> I did not dig into the Fortran Standards so I assume it need not be an error.
>
> Also when you submit a patch, please also let us know what platform you
> regression tested on, such as x86-64-linux, or Windows, or similar. (You can get
> the whole string from subdirectory names in build directory. On mine its
> x86_64-pc-linux-gnu) Sometimes we accidentally break things on different
> platforms. so this way we can see it tested ok over here and seems to fail over
> there.
>
> Your patch has changed some of the scan dumps and I am wondering if you have
> deleted something we use to check for?
>
> Jerry
>
Hello Jerry,
I have to thank for the kind feedback.
Attached is a reworked version of the patch with the changes applied. It 
also should  have the same scan dump now, one of the test cases was 
edited stupidly.
The regression tests for both the old as well as the new test have been 
performed on an x86-64-linux (x86_64-pc-linux-gnu).
Nicolas


[-- Attachment #2: p8.diff --]
[-- Type: text/x-patch, Size: 2073 bytes --]

Index: fortran/resolve.c
===================================================================
--- fortran/resolve.c	(revision 246143)
+++ fortran/resolve.c	(working copy)
@@ -15560,7 +15560,7 @@ resolve_equivalence (gfc_equiv *eq)
   locus *last_where = NULL;
   seq_type eq_type, last_eq_type;
   gfc_typespec *last_ts;
-  int object, cnt_protected;
+  int object, cnt_protected, cnt_volatile;
   const char *msg;
 
   last_ts = &eq->expr->symtree->n.sym->ts;
@@ -15569,6 +15569,8 @@ resolve_equivalence (gfc_equiv *eq)
 
   cnt_protected = 0;
 
+  cnt_volatile = 0;
+
   for (object = 1; eq; eq = eq->eq, object++)
     {
       e = eq->expr;
@@ -15641,6 +15643,17 @@ resolve_equivalence (gfc_equiv *eq)
 
       sym = e->symtree->n.sym;
 
+      if(sym->attr.volatile_)
+        cnt_volatile++;
+      if(cnt_volatile > 0 && cnt_volatile != object)
+	{
+	  gfc_warning (0, "Either all or none of the objects in "
+	  	       "the EQUIVALENCE set at %L should have the "
+		       "VOLATILE attribute",
+		       &e->where);
+	  break;
+	}
+
       if (sym->attr.is_protected)
 	cnt_protected++;
       if (cnt_protected > 0 && cnt_protected != object)
Index: testsuite/gfortran.dg/volatile11.f90
===================================================================
--- testsuite/gfortran.dg/volatile11.f90	(revision 246140)
+++ testsuite/gfortran.dg/volatile11.f90	(working copy)
@@ -3,6 +3,7 @@
 ! Tests that volatile can be applied to members of common blocks or
 ! equivalence groups (PR fortran/35037)
 !
+
 subroutine wait1
   logical event
   volatile event
@@ -16,7 +17,7 @@ end subroutine
 subroutine wait2
   logical event, foo
   volatile event
-  equivalence (event, foo)
+  equivalence (event, foo) ! { dg-warning "in the EQUIVALENCE set" }
   event = .false.
   do
     if (event) print *, 'NotOptimizedAway2'
@@ -27,7 +28,7 @@ subroutine wait3
   logical event
   integer foo
   volatile foo
-  equivalence (event, foo)
+  equivalence (event, foo) ! { dg-warning "in the EQUIVALENCE set" }
   event = .false.
   do
     if (event) print *, 'IsOptimizedAway'