member function swap vs std::move

[m0shbear]

Suppose I have a container type C<T>, which implements
MoveConstructable, MoveAssignable, and member function swap(C<T>&).

In which case is it more elegant to use
C<T> c1;
// fill c1 with data
C<T> c2 = std::move(c1);

over

C<T> c1;
// fill c1 with data
C<T> c2;
c2.swap(c1);

?

Granted, the .swap technique was a hack for moving containers in pre-C+
+11 classes, but now that C++11 is out, does it have any merit for new
classes?

 

[88888 Dihedral]

If the object in any class was stored as one or more continuos blocks of memory
in the compiled or interpreted executables, then the action of swap of two objects in the same class can be faster than the C++ way.

The multi-class inheritance allowed in C++ adds more overheads in any
copy contructor or assignment operations.

In C the assignment operation allowed in any parensis does not add too much
overhead, since C types are supposed to be register holdables in HW.

But the same teasing trick in C++ is just slowing down some basic operations
in objects of muti-level multi-class-inherant classes.

Thus, it requires more skills to write and maintain C++ codes.

 

[Fabrizio J Bonsignore]

What about the memory-runtime tradeoff? Can you be more specific as to
where you spend more runtime to save more memory and viceversa? Swap
should always be more memory intensive; move can be done purely
runtime with a single intermediate hold variable, independently of how
many castings/polymorphisms/new-copy constructors you might have to
call. (Particularly with swap you would be invoking copy constructors
a gallore, so they may be factorized out of the comparison function,
but it would depend on your pointer-auto object regime). What
framework did you say you are using this in?

Danilo J Bonsignore

 

[Fabrizio J Bonsignore]

(Oh, if you just move:: and there IS a guaranteed zero object in the
array, you can always use the zero object already there as swap memory
and regenerate it lastly with a copy assignment-new object. Keeping an
index/pointer would be more efficient than holding a whole auto
object. You can also reserve index zero for Zero object and use it as
swap memory/working memory already there, cf, the reserved variables
in WM$ structures).

Danilo J Bonsignore

 

[88888 Dihedral]

在 2012å¹´2月6日星期一UTC+8上åˆ10æ—¶17分02秒，Fabrizio J Bonsignore写é“：

What about the memory-runtime tradeoff? Can you be more specific as to
where you spend more runtime to save more memory and viceversa? Swap
should always be more memory intensive; move can be done purely
runtime with a single intermediate hold variable, independently of how
many castings/polymorphisms/new-copy constructors you might have to
call. (Particularly with swap you would be invoking copy constructors
a gallore, so they may be factorized out of the comparison function,
but it would depend on your pointer-auto object regime). What
framework did you say you are using this in?

Danilo J Bonsignore

Condsider a class inherited in 6 levels from the root and at each level of
2 to 6 mutilevel inheritances of classes, then the growth of an expoential rate of memory used by an object for its properties and methods from all super classes in the class level is unavoiadable. Thus, for an object in a class that has to support all its upper classes functionality eat the heap like a hog
very often.

This helps the DRAM industry a lot, but it does not help in training professional programmers to produce better SW. Unless the programmers will
gain profits from the DRAM sellers, I can't agree with those teasing tricks
as a SW programmer.

 

[Juha Nieminen]

If the object in any class was stored as one or more continuos blocks of memory
in the compiled or interpreted executables, then the action of swap of two objects in the same class can be faster than the C++ way.

Your text is quite incoherent.

I believe that what you are *trying* to say is that if all objects were
handled by reference (as they are eg. in Java or Objective-C) rather than
by value (as in C++), then swapping two objects would be way faster.

Except that in this case you wouldn't be swapping the contents of the
objects, but just the two references. Any third reference would not see
the swap.

 

[88888 Dihedral]

在 2012å¹´2月7日星期二UTC+8下åˆ2æ—¶19分55秒，Juha Nieminen写é“：

Your text is quite incoherent.

I believe that what you are *trying* to say is that if all objects were
handled by reference (as they are eg. in Java or Objective-C) rather than
by value (as in C++), then swapping two objects would be way faster.

Except that in this case you wouldn't be swapping the contents of the
objects, but just the two references. Any third reference would not see
the swap.

I agree with you. The refefrence scheme in swapping is faster.