Enable warnings-as-errors on Windows.

Enable warnings-as-errors on Windows.
Committed: https://code.google.com/p/skia/source/detail?r=7094

[bsalomon, 2013-01-08 19:52:10]

This builds clean for me in both Release and Debug.

[robertphillips, 2013-01-08 19:54:49]

LGTM

[reed1, 2013-01-08 20:04:47]

me no likie the neg infinity change... unless we really need it.

[Humper, 2013-01-08 20:08:00]

On 2013/01/08 20:04:47, reed1 wrote:
> me no likie the neg infinity change... unless we really need it.

negInf does seem a little hamfisted, but it's also probably too broad to turn
off the warning that it creates in whatever header file that #defines it, since
that's probably included almost everywhere.

I prefer zero warnings to -inf so this is fine with me but we're starting to (or
maybe way past the point of) philosophical debate about code cleanliness :)

[bsalomon, 2013-01-08 20:12:12]

On 2013/01/08 20:08:00, Humper wrote:
> On 2013/01/08 20:04:47, reed1 wrote:
> > me no likie the neg infinity change... unless we really need it.
> 
> negInf does seem a little hamfisted, but it's also probably too broad to turn
> off the warning that it creates in whatever header file that #defines it,
since
> that's probably included almost everywhere.
> 
> I prefer zero warnings to -inf so this is fine with me but we're starting to
(or
> maybe way past the point of) philosophical debate about code cleanliness :)

The error comes when LTCG actually resolves gIEEEInfinity to a compile time
constant. The linker's code generator negates infinity and sees that value is
(shockingly) negative infinity and says:

d:\src\skia6\tests\pathtest.cpp(256): warning C4756: overflow in constant
arithmetic

An alternative is to #pragma push/disable/pop around the usage in PathTest.cpp
(assuming that the LTCG respects the disable). I'm okay with that.

[Humper, 2013-01-08 20:15:05]

On 2013/01/08 20:12:12, bsalomon wrote:
> On 2013/01/08 20:08:00, Humper wrote:
> > On 2013/01/08 20:04:47, reed1 wrote:
> > > me no likie the neg infinity change... unless we really need it.
> > 
> > negInf does seem a little hamfisted, but it's also probably too broad to
turn
> > off the warning that it creates in whatever header file that #defines it,
> since
> > that's probably included almost everywhere.
> > 
> > I prefer zero warnings to -inf so this is fine with me but we're starting to
> (or
> > maybe way past the point of) philosophical debate about code cleanliness :)
> 
> The error comes when LTCG actually resolves gIEEEInfinity to a compile time
> constant. The linker's code generator negates infinity and sees that value is
> (shockingly) negative infinity and says:
> 
> d:\src\skia6\tests\pathtest.cpp(256): warning C4756: overflow in constant
> arithmetic
> 
> An alternative is to #pragma push/disable/pop around the usage in PathTest.cpp
> (assuming that the LTCG respects the disable). I'm okay with that.

also assuming we don't actually CARE about the overflow ;)

[bsalomon, 2013-01-08 20:29:35]

On 2013/01/08 20:15:05, Humper wrote:
> On 2013/01/08 20:12:12, bsalomon wrote:
> > On 2013/01/08 20:08:00, Humper wrote:
> > > On 2013/01/08 20:04:47, reed1 wrote:
> > > > me no likie the neg infinity change... unless we really need it.
> > > 
> > > negInf does seem a little hamfisted, but it's also probably too broad to
> turn
> > > off the warning that it creates in whatever header file that #defines it,
> > since
> > > that's probably included almost everywhere.
> > > 
> > > I prefer zero warnings to -inf so this is fine with me but we're starting
to
> > (or
> > > maybe way past the point of) philosophical debate about code cleanliness
:)
> > 
> > The error comes when LTCG actually resolves gIEEEInfinity to a compile time
> > constant. The linker's code generator negates infinity and sees that value
is
> > (shockingly) negative infinity and says:
> > 
> > d:\src\skia6\tests\pathtest.cpp(256): warning C4756: overflow in constant
> > arithmetic
> > 
> > An alternative is to #pragma push/disable/pop around the usage in
PathTest.cpp
> > (assuming that the LTCG respects the disable). I'm okay with that.
> 
> also assuming we don't actually CARE about the overflow ;)

Mike said "uncle", landing as is.