Re: [boost] [property_tree] Copy constructor and assignment operator in xml_writer_settings class

Subject: Re: [boost] [property_tree] Copy constructor and assignment operator in xml_writer_settings class
From: Mateusz Loskot (mateusz_at_[hidden])
Date: 2010-10-18 11:13:21

On 18/10/10 15:27, David Abrahams wrote:
> At Mon, 18 Oct 2010 15:02:39 +0100,
> Mateusz Loskot wrote:
>>
>>>> Looking at the current definition [1], I'm not convinced.
>>>> According to C++(n3092)/12.8/25, assignment operator for this class
>>>> is defined as deleted due to "non-static data members of const
>>>> non-class type".
>>>
>>> I have no idea what your point is.
>>
>> My point is that the class definition suggests its copying might
>> not be kosher. It is valid to copy-construct it but not assign.
>
> That's because it has members that are, um, const.  Assignment is a
> mutating operation.  Construction is not.

Yes, it's clear to me.

>>> Do you want me to prevent copy construction too, just because copy
>>> assignment is not possible? Or do you want me to make the members
>>> non-const so that copy assignment is possible?
>>
>> In my opinion, copy constructor and assignment operator
>> walk together to make a type fully copyable.
>
> The proper term here is "regular," I think (see
> http://www.stepanovpapers.com/DeSt98.pdf).

Interesting.

> But then you need equality
> and a bunch of other operations to make it meaningful.

I wasn't looking that far.

>> Thus, IMO, either the
>> definition should change or both copy operations disabled.
>
> I disagree.  We have movable but noncopyable types.  There's nothing
> wrong in principle with copyable but not-assignable types.

Yes, I forgot and confused CopyConstructible != Assignable

So, xml_writer_settings should be considered as CopyConstructible
having implicitly declared copy constructor, but not Assignable
having implicitly disabled assignment operator.
No explicit motions are required here.

>>> If the former, why? If the latter, I suppose I could do that. I'm
>>> not the one who made the members const in the first place.
>>
>> I would suggest to either disable both copy operations (does it make
>> sense to have copy ctor without assignment operator?) or change the
>> definition.
>>
>> I'd also like to point that I have asked generally:
>> What is recommended or required policy regarding copy
>> operations, it is copy ctor and assignment operator.
>> So, I have asked for comments and discussions too, as my
>> understanding of this matter may be incorrect.
>
> There's no required policy.  As for recommendations, I say strive for
> regularity, but sometimes it's just not appropriate.

Understood. I'm still digesting Stepanov's elements of programming
and finding their applications in practice.

Thanks David!

Sebastian, my apologies for wasting your time.

Best regards,
-- 
Mateusz Loskot, http://mateusz.loskot.net
Charter Member of OSGeo, http://osgeo.org
Member of ACCU, http://accu.org