Update the generator to work around a MSVC limitation.

Update the C++ generator to work around a MSVC integral constant bug

MSVC doesn't use the C++11 type rules for integral constants. A literal
-2147483648 is parsed in two steps:
1) MSVC evalutes 2147483648. Since it is greater than 2^31-1, it decides
   the type of the integral constant should be unsigned int.
2) It then applies unary minus to an unsigned int, which doesn't do
  anything.
To work around this, the C++ binding generator detects this edge case
and splits the constant in two.

BUG=445618
R=hansmuller@google.com

Committed: https://chromium.googlesource.com/external/mojo/+/e3719475d5971283d1d2250533d53066b2ff9797

[dcheng, 2014-12-31 00:32:31]

dcheng@chromium.org changed reviewers:
+ hansmuller@chromium.org

[dcheng, 2014-12-31 00:32:32]

This seems silly for mojom files to have to hack around. It seems like it should
belong in the generator.

I didn't handle the corresponding case for -2^63: if that's desired, it would be
easy to handle that here too.

[hansmuller1, 2014-12-31 00:45:55]

hansmuller@google.com changed reviewers:
+ hansmuller@google.com

[hansmuller1, 2014-12-31 00:45:56]

Beautiful, LGTM. Nit: shouldn't the constant expression have spaces, like:
return '(-%d - 1)' % (2**31 - 1). It might also be useful to emit a comment that
points at the bug, in case anyone ever reads the generated code.

[dcheng, 2014-12-31 00:50:11]

Good point: updated the spaces (and fixed a minor grammar typo).

The comment seems like it'd be hard, unless you're ok with just sticking it in a
/* */ block. Does that sound OK?

[hansmuller1, 2014-12-31 00:55:52]

On 2014/12/31 00:50:11, dcheng wrote:
> Good point: updated the spaces (and fixed a minor grammar typo).
> 
> The comment seems like it'd be hard, unless you're ok with just sticking it in
a
> /* */ block. Does that sound OK?

I think that would be fine.

[dcheng, 2014-12-31 01:03:01]

On 2014/12/31 at 00:55:52, hansmuller wrote:
> On 2014/12/31 00:50:11, dcheng wrote:
> > Good point: updated the spaces (and fixed a minor grammar typo).
> > 
> > The comment seems like it'd be hard, unless you're ok with just sticking it
in a
> > /* */ block. Does that sound OK?
> 
> I think that would be fine.

Done. Thanks for the quick review!

[dcheng, 2015-01-05 21:53:38]

Committed patchset #3 (id:40001) manually as
e3719475d5971283d1d2250533d53066b2ff9797 (presubmit successful).