ninja win: Add msvs emulation for FloatingPointModel

ninja win: Add msvs emulation for FloatingPointModel

BUG=
R=scottmg@chromium.org

Committed: https://code.google.com/p/gyp/source/detail?r=2003

[Shezan Baig (Bloomberg), 2014-11-13 19:50:51]

sbaig1@bloomberg.net changed reviewers:
+ scottmg@chromium.org

[scottmg, 2014-11-13 21:35:55]

https://codereview.chromium.org/720153005/diff/1/test/win/gyptest-cl-floating...
File test/win/gyptest-cl-floating-point-model.py (right):

https://codereview.chromium.org/720153005/diff/1/test/win/gyptest-cl-floating...
test/win/gyptest-cl-floating-point-model.py:21: # It's hard to map flags to
output contents in a non-fragile way, so just
I believe cl defines _M_FP_FAST, _M_FP_PRECISE, and _M_FP_STRICT depending on
/fp mode. It's not a ton better, but at least it could run on msvs too that way.

[Shezan Baig (Bloomberg), 2014-11-13 21:42:56]

On 2014/11/13 21:35:55, scottmg wrote:
>
https://codereview.chromium.org/720153005/diff/1/test/win/gyptest-cl-floating...
> File test/win/gyptest-cl-floating-point-model.py (right):
> 
>
https://codereview.chromium.org/720153005/diff/1/test/win/gyptest-cl-floating...
> test/win/gyptest-cl-floating-point-model.py:21: # It's hard to map flags to
> output contents in a non-fragile way, so just
> I believe cl defines _M_FP_FAST, _M_FP_PRECISE, and _M_FP_STRICT depending on
> /fp mode. It's not a ton better, but at least it could run on msvs too that
way.


Interesting!  I was looking for some kind of predefined macro to detect this,
but couldn't find anything.  Where did you find this out from? :)

[scottmg, 2014-11-13 21:45:39]

float.h mentions _M_FP_FAST, but yeah, I guess it's undocumented. I think it's
probably ok to rely on though to make this test a little better.

[Shezan Baig (Bloomberg), 2014-11-13 22:28:51]

On 2014/11/13 21:45:39, scottmg wrote:
> float.h mentions _M_FP_FAST, but yeah, I guess it's undocumented. I think it's
> probably ok to rely on though to make this test a little better.

I've uploaded a new test.  The test uncovered a bug in my original
implementation, where I was missing the 'default' value for FloatingPointModel. 
msvs specifies /fp:precise by default on the cl commmand-line if
FloatingPointModel is not set, so my new patch emulates that.

[scottmg, 2014-11-13 23:12:57]

Thanks, LGTM

[Shezan Baig (Bloomberg), 2014-11-13 23:17:48]

Committed patchset #2 (id:20001) manually as 2003 (presubmit successful).