[patch] fix c++/18016 - warn about self-initialization in constructor init-list

[patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 593 bytes --]

Here's my attempt to fix PR c++/18016 so that we get a warning from

struct S {
  int i;
  S() : i(i) { }
};

-Winit-self is broken for C++ (PR c++/34772) so I made this warn with
-Wuninitialized.  If you want to suppress the warning just don't put a
mem-initializer in the constructor (which will work at least until
someone fixes PR c++/2972)

cp/ChangeLog:

        PR c++/18016
        * init.c (perform_member_init): Check for self-initialization.


testsuite/ChangeLog:

        PR c++/18016
        * g++.dg/warn/pr18016.C: New.


tested x86_64-linux with no regressions, ok for trunk?

[-- Attachment #2: 18016.txt --]
[-- Type: text/plain, Size: 1272 bytes --]

Index: cp/init.c
===================================================================
--- cp/init.c	(revision 168023)
+++ cp/init.c	(working copy)
@@ -449,6 +449,17 @@ perform_member_init (tree member, tree i
   if (decl == error_mark_node)
     return;
 
+  if (warn_uninitialized && init && init != NULL_TREE
+      && TREE_CODE (init) == TREE_LIST && TREE_CHAIN (init) == NULL_TREE)
+    {
+      tree val = TREE_VALUE (init);
+      if (TREE_CODE (val) == COMPONENT_REF && TREE_OPERAND (val, 1) == member
+	  && TREE_OPERAND (val, 0) == current_class_ref)
+	warning_at (DECL_SOURCE_LOCATION (current_function_decl),
+	            OPT_Wuninitialized, "%qD is initialized with itself",
+		    member);
+    }
+
   if (init == void_type_node)
     {
       /* mem() means value-initialization.  */
Index: testsuite/g++.dg/warn/pr18016.C
===================================================================
--- testsuite/g++.dg/warn/pr18016.C	(revision 0)
+++ testsuite/g++.dg/warn/pr18016.C	(revision 0)
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-Wuninitialized" } */
+
+class X {
+  int i;
+  X() : i(i) { }   // { dg-warning "initialized with itself" }
+  X(int i) : i(i) { }
+  X(const X& x) : i(x.i) { }
+};
+
+// { dg-prune-output "In constructor" }

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Paolo Carlini]

Hi,

+  if (warn_uninitialized && init && init != NULL_TREE

I think the last check in this line is redundant.

Paolo.

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jonathan Wakely]

On 21 December 2010 17:04, Paolo Carlini wrote:
> Hi,
>
> +  if (warn_uninitialized && init && init != NULL_TREE
>
> I think the last check in this line is redundant.

Thanks, Paolo. I should have checked the definition of NULL_TREE.

I'm re-testing with that change, but it shouldn't affect the results.

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 417 bytes --]

On 21 December 2010 17:11, Jonathan Wakely wrote:
> On 21 December 2010 17:04, Paolo Carlini wrote:
>> Hi,
>>
>> +  if (warn_uninitialized && init && init != NULL_TREE
>>
>> I think the last check in this line is redundant.
>
> Thanks, Paolo. I should have checked the definition of NULL_TREE.
>
> I'm re-testing with that change, but it shouldn't affect the results.

Updated patch, tested x86_64-linux

[-- Attachment #2: 18016-2.txt --]
[-- Type: text/plain, Size: 1251 bytes --]

Index: cp/init.c
===================================================================
--- cp/init.c	(revision 168023)
+++ cp/init.c	(working copy)
@@ -449,6 +449,17 @@ perform_member_init (tree member, tree i
   if (decl == error_mark_node)
     return;
 
+  if (warn_uninitialized && init && TREE_CODE (init) == TREE_LIST
+      && TREE_CHAIN (init) == NULL_TREE)
+    {
+      tree val = TREE_VALUE (init);
+      if (TREE_CODE (val) == COMPONENT_REF && TREE_OPERAND (val, 1) == member
+	  && TREE_OPERAND (val, 0) == current_class_ref)
+	warning_at (DECL_SOURCE_LOCATION (current_function_decl),
+	            OPT_Wuninitialized, "%qD is initialized with itself",
+		    member);
+    }
+
   if (init == void_type_node)
     {
       /* mem() means value-initialization.  */
Index: testsuite/g++.dg/warn/pr18016.C
===================================================================
--- testsuite/g++.dg/warn/pr18016.C	(revision 0)
+++ testsuite/g++.dg/warn/pr18016.C	(revision 0)
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-Wuninitialized" } */
+
+class X {
+  int i;
+  X() : i(i) { }   // { dg-warning "initialized with itself" }
+  X(int i) : i(i) { }
+  X(const X& x) : i(x.i) { }
+};
+
+// { dg-prune-output "In constructor" }

[patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 937 bytes --]

I didn't get this reviewed previously so I'll try again for 4.7 ...


---------- Forwarded message ----------


Here's my attempt to fix PR c++/18016 so that we get a warning from

struct S {
 int i;
 S() : i(i) { }
};

As I said in the audit trail, -Winit-self is broken for C++ (PR
c++/34772) so I made this warn with
-Wuninitialized and not affected by -Winit-self.

Even if -Winit-self wasn't broken I don't think it applies here,
there's no valid reason to initialize a member with itself, if you
really want to leave a member uninitialized then don't use a
mem-initializer in the constructor (which will work at least until
someone fixes PR c++/2972)

cp/ChangeLog:

       PR c++/18016
       * init.c (perform_member_init): Check for self-initialization.


testsuite/ChangeLog:

       PR c++/18016
       * g++.dg/warn/pr18016.C: New.


tested x86_64-linux with no regressions, ok for trunk?

[-- Attachment #2: 18016-3.txt --]
[-- Type: text/plain, Size: 1203 bytes --]

Index: cp/init.c
===================================================================
--- cp/init.c	(revision 174044)
+++ cp/init.c	(working copy)
@@ -501,6 +501,17 @@
   if (decl == error_mark_node)
     return;
 
+  if (warn_uninitialized && init && TREE_CODE (init) == TREE_LIST
+      && TREE_CHAIN (init) == NULL_TREE)
+    {
+      tree val = TREE_VALUE (init);
+      if (TREE_CODE (val) == COMPONENT_REF && TREE_OPERAND (val, 1) == member
+	  && TREE_OPERAND (val, 0) == current_class_ref)
+	warning_at (DECL_SOURCE_LOCATION (current_function_decl),
+		    OPT_Wuninitialized, "%qD is initialized with itself",
+		    member);
+    }
+
   if (init == void_type_node)
     {
       /* mem() means value-initialization.  */
Index: testsuite/g++.dg/warn/pr18016.C
===================================================================
--- testsuite/g++.dg/warn/pr18016.C	(revision 0)
+++ testsuite/g++.dg/warn/pr18016.C	(revision 0)
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-Wuninitialized" } */
+
+class X {
+  int i;
+  X() : i(i) { }   // { dg-warning "initialized with itself" }
+  X(int i) : i(i) { }
+  X(const X& x) : i(x.i) { }
+};
+
+// { dg-prune-output "In constructor" }

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jason Merrill]

On 05/22/2011 05:11 PM, Jonathan Wakely wrote:
> As I said in the audit trail, -Winit-self is broken for C++ (PR
> c++/34772)

Not anymore.  :)

> Even if -Winit-self wasn't broken I don't think it applies here,
> there's no valid reason to initialize a member with itself, if you
> really want to leave a member uninitialized then don't use a
> mem-initializer in the constructor (which will work at least until
> someone fixes PR c++/2972)

...at which point we would want it to be controlled by -Winit-self, so 
let's use -Winit-self now.

OK with that change.

Jason

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 1492 bytes --]

On 23 May 2011 03:07, Jason Merrill wrote:
> On 05/22/2011 05:11 PM, Jonathan Wakely wrote:
>>
>> As I said in the audit trail, -Winit-self is broken for C++ (PR
>> c++/34772)
>
> Not anymore.  :)

Aha!

>> Even if -Winit-self wasn't broken I don't think it applies here,
>> there's no valid reason to initialize a member with itself, if you
>> really want to leave a member uninitialized then don't use a
>> mem-initializer in the constructor (which will work at least until
>> someone fixes PR c++/2972)
>
> ...at which point we would want it to be controlled by -Winit-self, so let's
> use -Winit-self now.

I still think this warning shouldn't be tied to -Winit-self. That
isn't included in -Wall or -Wextra, and it implies we're saying that a
self-referencing mem-initializer is a GNU extension to silence a
warning, when I think it's always a mistake and never done
intentionally.  But I can live with that.

> OK with that change.

Thanks, I've attached what I committed.  I only checked
warn_init_self, not both warn_uninitialized && warn_init_self, I think
that's right.

I've also just noticed that my change doesn't catch references which
are initialized with themselves.  I won't try to fix that now, but
would that be done by checking for INDIRECT_REF?

I also had some thoughts about making init-self work for class types e.g.
  std::string s = s;
Would checking if the copy constructor's parameter is *this be a
simple way to catch that?

[-- Attachment #2: 18016-4.txt --]
[-- Type: text/plain, Size: 1211 bytes --]

Index: cp/init.c
===================================================================
--- cp/init.c	(revision 174044)
+++ cp/init.c	(working copy)
@@ -501,6 +501,17 @@
   if (decl == error_mark_node)
     return;
 
+  if (warn_init_self && init && TREE_CODE (init) == TREE_LIST
+      && TREE_CHAIN (init) == NULL_TREE)
+    {
+      tree val = TREE_VALUE (init);
+      if (TREE_CODE (val) == COMPONENT_REF && TREE_OPERAND (val, 1) == member
+	  && TREE_OPERAND (val, 0) == current_class_ref)
+	warning_at (DECL_SOURCE_LOCATION (current_function_decl),
+		    OPT_Wuninitialized, "%qD is initialized with itself",
+		    member);
+    }
+
   if (init == void_type_node)
     {
       /* mem() means value-initialization.  */
Index: testsuite/g++.dg/warn/pr18016.C
===================================================================
--- testsuite/g++.dg/warn/pr18016.C	(revision 0)
+++ testsuite/g++.dg/warn/pr18016.C	(revision 0)
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-Wuninitialized -Winit-self" } */
+
+class X {
+  int i;
+  X() : i(i) { }   // { dg-warning "initialized with itself" }
+  X(int i) : i(i) { }
+  X(const X& x) : i(x.i) { }
+};
+
+// { dg-prune-output "In constructor" }

Re: [patch] fix c++/18016 - warn about self-initialization in constructor init-list

[Jason Merrill]

On 05/23/2011 04:20 AM, Jonathan Wakely wrote:
> I still think this warning shouldn't be tied to -Winit-self. That
> isn't included in -Wall or -Wextra, and it implies we're saying that a
> self-referencing mem-initializer is a GNU extension to silence a
> warning, when I think it's always a mistake and never done
> intentionally.

I agree, but I think the answer to that is to enable -Winit-self with 
-Wextra in C++ mode.

> Thanks, I've attached what I committed.  I only checked
> warn_init_self, not both warn_uninitialized&&  warn_init_self, I think
> that's right.

I think so.

> I've also just noticed that my change doesn't catch references which
> are initialized with themselves.  I won't try to fix that now, but
> would that be done by checking for INDIRECT_REF?

Yes, or you can use REFERENCE_REF_P.

> I also had some thoughts about making init-self work for class types e.g.
>    std::string s = s;
> Would checking if the copy constructor's parameter is *this be a
> simple way to catch that?

Well, in this snippet there's no *this involved, the parameter would be 
the decl being initialized.

Jason