Passing temprary by reference requires copy constructor in 3.4?

Passing temprary by reference requires copy constructor in 3.4?

[Peter Cech]

Hi,

trying to compile

class EventObject
{
public:
  EventObject() {}

private:
  EventObject(const EventObject&);
};

class AnyEvent : public EventObject
{
  public:
    AnyEvent() {}

  private:
    AnyEvent(const AnyEvent&);
};

void invoke(const EventObject &) {}

void caller1()
{
  AnyEvent event;
  invoke(event);
}

void caller2()
{
  invoke(AnyEvent());
}

leads to following error with both gcc (GCC) 3.4.3  (Gentoo Linux 3.4.3,
ssp-3.4.3-0, pie-8.7.6.6) and gcc (GCC) 4.0.0 20041121 (experimental)

a.cpp: In function `void caller2()':
a.cpp:16: error: `AnyEvent::AnyEvent(const AnyEvent&)' is private
a.cpp:29: error: within this context

Why gcc requires copy constructor in such situation?

Regards,
Peter Cech

Re: Passing temprary by reference requires copy constructor in 3.4?

[Andrew Pinski]

On Nov 26, 2004, at 2:10 PM, Peter Cech wrote:

> Hi,
> a.cpp: In function `void caller2()':
> a.cpp:16: error: `AnyEvent::AnyEvent(const AnyEvent&)' is private
> a.cpp:29: error: within this context
>
> Why gcc requires copy constructor in such situation?

It does not require the copy constructor, just requires it be 
accessible.
I would read the changes page for 3.4 which documents why this changed.
Mainly this is required by the C++ standard.

Thanks,
Andrew Pinski

Re: Passing temprary by reference requires copy constructor in 3.4?

[Mikael Kilpeläinen]

On Friday 26 November 2004 21:16, Andrew Pinski wrote:
>
> It does not require the copy constructor, just requires it be
> accessible.
> I would read the changes page for 3.4 which documents why this changed.
> Mainly this is required by the C++ standard.

Indeed, however seems that std commitee is removing the restriction. Hence I 
would like to see a flag to disable the error (if there is not one already 
and I am just not aware of it). I am refering to the DR 391.


Mikael

Re: Passing temprary by reference requires copy constructor in 3.4?

[Andrew Pinski]

On Nov 26, 2004, at 2:55 PM, Mikael Kilpeläinen wrote:

> On Friday 26 November 2004 21:16, Andrew Pinski wrote:
>>
>> It does not require the copy constructor, just requires it be
>> accessible.
>> I would read the changes page for 3.4 which documents why this 
>> changed.
>> Mainly this is required by the C++ standard.
>
> Indeed, however seems that std commitee is removing the restriction. 
> Hence I
> would like to see a flag to disable the error (if there is not one 
> already
> and I am just not aware of it). I am refering to the DR 391.

Why, the DR is still active so GCC is correct as of right now.
If and when the DR goes to defect, we will implement the needed
change.  But currently GCC is following the standard.

Thanks,
Andrew Pinski

Re: Passing temprary by reference requires copy constructor in 3.4?

[Mikael Kilpeläinen]

On Friday 26 November 2004 22:03, Andrew Pinski wrote:
> Why, the DR is still active so GCC is correct as of right now.
> If and when the DR goes to defect, we will implement the needed
> change.  But currently GCC is following the standard.

Yes, I was not saying that gcc is wrong. That was not my intention, I just 
wanted to point out that it is likely to happen and the flag could be helpful 
for certain people. Of course I might be wrong. I just saw fine opportunity 
to talk about this :)


Mikael

Re: Passing temprary by reference requires copy constructor in 3.4?

[Paolo Bonzini]

>> Indeed, however seems that std commitee is removing the restriction. 
>> Hence I would like to see a flag to disable the error (if there is not one 
>> already and I am just not aware of it). I am refering to the DR 391.
> 
> Why, the DR is still active so GCC is correct as of right now.
> If and when the DR goes to defect, we will implement the needed
> change.  But currently GCC is following the standard.

It could be material for -fpermissive.  But surely not now that mainline 
is stage3.

Paolo