Remove Clang-on-Win fix for definitions of Color's static data members

Remove Clang-on-Win fix for definitions of Color's static data members

Clang does treat the in-class initialization as a definition, just like MSVC.
The problem was that Clang failed to emit the definition unless it was referenced,
which broke the components build. This has been fixed in Clang r213304.

The work-around was added in Blink r176469, and this patch effectively removes that.

BUG=82385
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=178401

[hans, 2014-07-17 20:37:23]

Please take a look.

[Nico, 2014-07-17 20:43:16]

lgtm!

(maybe say "effectively reverts 176469" somewhere)

[hans, 2014-07-17 20:44:43]

On 2014/07/17 20:43:16, Nico (away) wrote:
> lgtm!
> 
> (maybe say "effectively reverts 176469" somewhere)

Done.

[hans, 2014-07-17 20:44:45]

The CQ bit was checked by hans@chromium.org

[commit-bot: I haz the power, 2014-07-17 20:45:29]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/hans@chromium.org/397323002/1

[commit-bot: I haz the power, 2014-07-17 22:42:42]

Change committed as 178401

[Peter Kasting, 2014-07-17 22:58:27]

(1) Instead of having an #if block here, consider using the
STATIC_CONST_MEMBER_DEFINITION macro, which makes MSVC do the right thing.
(2) I thought the C++ standard says compilers are not supposed to emit
definitions automatically, whether a constant is referenced or not?  For
constants whose addresses are not taken, the declaration (i.e. the assignment in
the class declaration) should be sufficient, and no definition should be emitted
or needed.  For constants whose addresses are taken, one is supposed to
explicitly provide a definition, as Color.cpp is doing.

Am I misunderstanding the CL description here?

[Nico, 2014-07-17 23:06:15]

Note that this is a revert; see discussion on
https://codereview.chromium.org/330823006 why it was necessary (clang-cl used to
not be compatible with cl; now it is).

STATIC_CONST_MEMBER_DEFINITION is a base/-only thing, so it can't be used in
blink. (And I think it shouldn't even be in base, but Trung ignored my comments
on the CL that added it.)

[hans, 2014-07-17 23:31:27]

On 2014/07/17 22:58:27, Peter Kasting wrote:
> (2) I thought the C++ standard says compilers are not supposed to emit
> definitions automatically, whether a constant is referenced or not?
> For constants whose addresses are not taken, the declaration (i.e. the
assignment in
> the class declaration) should be sufficient, and no definition should be
emitted
> or needed.  For constants whose addresses are taken, one is supposed to
> explicitly provide a definition, as Color.cpp is doing.

Right, but MSVC doesn't follow the standard in this respect, which is why we
have this #ifdef. At least with my patch, we don't have to work-around the
work-around for Clang on Windows.