[PATCH] Fix PR70054

[PATCH] Fix PR70054

[Richard Biener]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 3157 bytes --]


The following patch adjusts strict_aliasing_warning to use
proper alias_set_subset_of instead of relying on alias_sets_conflict_p
as after the PR66110 fix aggregates with a char[] member do not
automatically behave like having alias-set zero.  As a side-effect
the test will be somewhat stricter as well accessing a 'long' with
a struct { int i; long l; } * will now warn while it previously
didn't:

struct S { int i; long l; };
long x;
struct S foo () { return *(struct S *)&x; }

now warns:

t.c: In function Β‘fooΒ’:
t.c:3:35: warning: dereferencing type-punned pointer will break 
strict-aliasing rules [-Wstrict-aliasing]
 struct S foo () { return *(struct S *)&x; }
                                   ^
Bootstrap and regtest running on x86_64-unknown-linux-gnu, ok for trunk?

Thanks,
Richard.

2016-03-03  Richard Biener  <rguenther@suse.de>

	PR c++/70054
	* c-common.c (strict_aliasing_warning): Use alias_set_subset_of
	instead of alias_sets_conflict_p.

	* g++.dg/warn/Wstrict-aliasing-bogus-union-2.C: New testcase.
	* gcc.dg/Wstrict-aliasing-struct-member.c: New testcase.

Index: gcc/c-family/c-common.c
===================================================================
*** gcc/c-family/c-common.c	(revision 233902)
--- gcc/c-family/c-common.c	(working copy)
*************** strict_aliasing_warning (tree otype, tre
*** 1568,1574 ****
            alias_set_type set2 = get_alias_set (TREE_TYPE (type));
  
            if (set1 != set2 && set2 != 0
! 	      && (set1 == 0 || !alias_sets_conflict_p (set1, set2)))
  	    {
  	      warning (OPT_Wstrict_aliasing, "dereferencing type-punned "
  		       "pointer will break strict-aliasing rules");
--- 1568,1574 ----
            alias_set_type set2 = get_alias_set (TREE_TYPE (type));
  
            if (set1 != set2 && set2 != 0
! 	      && (set1 == 0 || !alias_set_subset_of (set2, set1)))
  	    {
  	      warning (OPT_Wstrict_aliasing, "dereferencing type-punned "
  		       "pointer will break strict-aliasing rules");
Index: gcc/testsuite/g++.dg/warn/Wstrict-aliasing-bogus-union-2.C
===================================================================
*** gcc/testsuite/g++.dg/warn/Wstrict-aliasing-bogus-union-2.C	(revision 0)
--- gcc/testsuite/g++.dg/warn/Wstrict-aliasing-bogus-union-2.C	(working copy)
***************
*** 0 ****
--- 1,14 ----
+ /* { dg-do compile { target c++11 } } */
+ /* { dg-options "-Wstrict-aliasing=2 -O2 -Wall" } */
+ 
+ #include <type_traits>
+ 
+ struct foo
+ {
+   std::aligned_storage<sizeof(long), alignof(long)>::type raw;
+ 
+   long& cooked()
+     {
+       return *static_cast<long*>(static_cast<void*>(&raw)); /* { dg-bogus "strict-aliasing" } */
+     }
+ };
Index: gcc/testsuite/gcc.dg/Wstrict-aliasing-struct-member.c
===================================================================
--- gcc/testsuite/gcc.dg/Wstrict-aliasing-struct-member.c	(revision 0)
+++ gcc/testsuite/gcc.dg/Wstrict-aliasing-struct-member.c	(working copy)
@@ -0,0 +1,6 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -Wall" } */
+
+struct S { int i; long l; };
+long x;
+struct S foo () { return *(struct S *)&x; } /* { dg-warning "will break strict-aliasing" } */

Re: [PATCH] Fix PR70054

[Jeff Law]

On 03/03/2016 05:35 AM, Richard Biener wrote:
>
> The following patch adjusts strict_aliasing_warning to use
> proper alias_set_subset_of instead of relying on alias_sets_conflict_p
> as after the PR66110 fix aggregates with a char[] member do not
> automatically behave like having alias-set zero.  As a side-effect
> the test will be somewhat stricter as well accessing a 'long' with
> a struct { int i; long l; } * will now warn while it previously
> didn't:
>
> struct S { int i; long l; };
> long x;
> struct S foo () { return *(struct S *)&x; }
>
> now warns:
>
> t.c: In function Β‘fooΒ’:
> t.c:3:35: warning: dereferencing type-punned pointer will break
> strict-aliasing rules [-Wstrict-aliasing]
>   struct S foo () { return *(struct S *)&x; }
>                                     ^
> Bootstrap and regtest running on x86_64-unknown-linux-gnu, ok for trunk?
>
> Thanks,
> Richard.
>
> 2016-03-03  Richard Biener  <rguenther@suse.de>
>
> 	PR c++/70054
> 	* c-common.c (strict_aliasing_warning): Use alias_set_subset_of
> 	instead of alias_sets_conflict_p.
>
> 	* g++.dg/warn/Wstrict-aliasing-bogus-union-2.C: New testcase.
> 	* gcc.dg/Wstrict-aliasing-struct-member.c: New testcase.
The PR doesn't have a regression marker, but from reading the PR it's 
clearly a regression.

OK for the trunk,

Thanks,
Jeff

Re: [PATCH] Fix PR70054

[Richard Biener]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1505 bytes --]

On Fri, 4 Mar 2016, Jeff Law wrote:

> On 03/03/2016 05:35 AM, Richard Biener wrote:
> > 
> > The following patch adjusts strict_aliasing_warning to use
> > proper alias_set_subset_of instead of relying on alias_sets_conflict_p
> > as after the PR66110 fix aggregates with a char[] member do not
> > automatically behave like having alias-set zero.  As a side-effect
> > the test will be somewhat stricter as well accessing a 'long' with
> > a struct { int i; long l; } * will now warn while it previously
> > didn't:
> > 
> > struct S { int i; long l; };
> > long x;
> > struct S foo () { return *(struct S *)&x; }
> > 
> > now warns:
> > 
> > t.c: In function Β‘fooΒ’:
> > t.c:3:35: warning: dereferencing type-punned pointer will break
> > strict-aliasing rules [-Wstrict-aliasing]
> >   struct S foo () { return *(struct S *)&x; }
> >                                     ^
> > Bootstrap and regtest running on x86_64-unknown-linux-gnu, ok for trunk?
> > 
> > Thanks,
> > Richard.
> > 
> > 2016-03-03  Richard Biener  <rguenther@suse.de>
> > 
> > 	PR c++/70054
> > 	* c-common.c (strict_aliasing_warning): Use alias_set_subset_of
> > 	instead of alias_sets_conflict_p.
> > 
> > 	* g++.dg/warn/Wstrict-aliasing-bogus-union-2.C: New testcase.
> > 	* gcc.dg/Wstrict-aliasing-struct-member.c: New testcase.
> The PR doesn't have a regression marker, but from reading the PR it's clearly
> a regression.

Yes, it is - sorry for not updating the bugzilla appropriately.

> OK for the trunk,

Thanks,
Richard.