OwnPtr, OwnArrayPtr: Use copy/move-and-swap for assignment operators

OwnPtr, OwnArrayPtr: Use copy/move-and-swap for assignment operators

Rationals:
- decrease of repeated code
- consistency with RefPtr

[Mikhail, 2013-09-11 11:08:27]

Could you please have a look?

[cevans, 2013-09-11 17:50:05]

On 2013/09/11 11:08:27, mikhail.pozdnyakov wrote:
> Could you please have a look?

Does this do anything to binary size? I find that gcc is most sensitive to
binary size, but clang binary sizes are also useful as if they change, you might
have crossed some inlining threshold.

[Nico, 2013-09-11 18:00:36]

https://codereview.chromium.org/23440024/diff/5001/Source/wtf/OwnPtr.h
File Source/wtf/OwnPtr.h (right):

https://codereview.chromium.org/23440024/diff/5001/Source/wtf/OwnPtr.h#newcod...
Source/wtf/OwnPtr.h:168: OwnPtr ptr = std::move(o);
As mentioned several times, we can't use C++11 library features.

This doesn't build on Mac which doesn't have a C++11 standard library:

../../third_party/WebKit/Source/wtf/OwnPtr.h:168:27: error: no member named
'move' in namespace 'std'
        OwnPtr ptr = std::move(o);
                     ~~~~~^

[Mikhail, 2013-09-12 08:15:28]

On 2013/09/11 18:00:36, Nico wrote:
> https://codereview.chromium.org/23440024/diff/5001/Source/wtf/OwnPtr.h
> File Source/wtf/OwnPtr.h (right):
> 
>
https://codereview.chromium.org/23440024/diff/5001/Source/wtf/OwnPtr.h#newcod...
> Source/wtf/OwnPtr.h:168: OwnPtr ptr = std::move(o);
> As mentioned several times, we can't use C++11 library features.
> 
Yeah, I just should not have touched code inside "#if
COMPILER_SUPPORTS(CXX_RVALUE_REFERENCES)". I'll update the cl accordingly.

[Mikhail, 2013-09-12 15:07:27]

On 2013/09/11 17:50:05, cevans wrote:
> On 2013/09/11 11:08:27, mikhail.pozdnyakov wrote:
> > Could you please have a look?
> 
> Does this do anything to binary size? I find that gcc is most sensitive to
> binary size, but clang binary sizes are also useful as if they change, you
might
> have crossed some inlining threshold.

Good point, thank you for mentioning this! 
I had believed that compiler would optimize it out but the change indeed
slightly increases libwebkit.so size: from to 41804392 to 41804520 bytes (used
gcc 4.6).

I would still vote for consistence: either we use copy/move-and-swap for
assignment operators in all the WTF smart pointer classes (1), or we do not use
it at all and revert also r156871(https://codereview.chromium.org/23666004/)
(2).

To my mind variant #1 is preferable as it makes code more compact and readable
even though it affects binary size a bit.

What do you think?