we need extra precision in calculating center_of_mass

we need extra precision here

This is not an issue in theory, I've seen such bug in practice, please refer to
the following bug for a simplified test case.

BUG=http://code.google.com/p/skia/issues/detail?id=497
TEST=

[guanqun, 2012-02-16 15:05:30]

Please review. Thanks!

[guanqun, 2012-02-16 15:17:44]

Some explaination here:

The simplified test case would generate 8 control points (forming a circle) .
The values of these points ranges from 999 to 1001. According to the original
implementation, fanPt calculates to 1007.xx due to the limitation of being
float. Changing it to double makes it correct, the fanPt calculated would be
around (1000, 1000).

Thanks!

[guanqun, 2012-02-21 02:33:27]

Sorry to bother, any comments on this bug?

I've seen in revision 3216, cheapComputeDirection fallbacks to double
computation as well. So I'm afraid this should be issue, right?

I've applied this patch on latest chromium, and it fixes the bug.

Thanks!

[bsalomon, 2012-02-21 18:20:07]

On 2012/02/21 02:33:27, guanqun wrote:
> Sorry to bother, any comments on this bug?
> 
> I've seen in revision 3216, cheapComputeDirection fallbacks to double
> computation as well. So I'm afraid this should be issue, right?
> 
> I've applied this patch on latest chromium, and it fixes the bug.
> 
> Thanks!

Hi I think your change is correct. I'm out of the office today. I'll be back
tomorrow and can land the change. I'm also wondering if the computation can be
reordered to avoid the underflows.