Generate symbols in release builds

[epoger, 2011-09-27 21:45:56]

http://codereview.appspot.com/5142050/diff/1/common_conditions.gypi
File common_conditions.gypi (left):

http://codereview.appspot.com/5142050/diff/1/common_conditions.gypi#oldcode31
common_conditions.gypi:31: 'DebugInformationFormat': '1', # debugOldStyleInfo
(/Z7)
The reason I set this to use /Z7 in the first place was
http://code.google.com/p/skia/issues/detail?id=320 .  Do you have a good
alternate resolution to that issue?

[bsalomon, 2011-09-27 22:36:30]

On 2011/09/27 21:45:56, epoger wrote:
> http://codereview.appspot.com/5142050/diff/1/common_conditions.gypi
> File common_conditions.gypi (left):
> 
> http://codereview.appspot.com/5142050/diff/1/common_conditions.gypi#oldcode31
> common_conditions.gypi:31: 'DebugInformationFormat': '1', # debugOldStyleInfo
> (/Z7)
> The reason I set this to use /Z7 in the first place was
> http://code.google.com/p/skia/issues/detail?id=320 .  Do you have a good
> alternate resolution to that issue?

Checkout the new version. It puts the pdb files in the output directory. I also
figured out a bunch of other flags that we can set, including raising the
warning level. This finds a number of things that should be fixed as well as
some annoying ones that can be worked around.

[bungeman, 2011-09-28 13:06:11]

Should add LinkTimeCodeGeneration to VCLinkerTool for release.

http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi
File common_conditions.gypi (right):

http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi#newcode61
common_conditions.gypi:61: 'VCLinkerTool': {
Should set (I think 1 is correct for enabling, will need to check)

'LinkTimeCodeGeneration': '1' #/LTCG

otherwise I don't think WholeProgramOptimization above will really do anything.

[bsalomon, 2011-09-28 13:13:51]

http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi
File common_conditions.gypi (right):

http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi#newcode61
common_conditions.gypi:61: 'VCLinkerTool': {
On 2011/09/28 13:06:11, bungeman wrote:
> Should set (I think 1 is correct for enabling, will need to check)
> 
> 'LinkTimeCodeGeneration': '1' #/LTCG
> 
> otherwise I don't think WholeProgramOptimization above will really do
anything.

Done.

[epoger, 2011-09-28 13:39:48]

On 2011/09/28 13:13:51, bsalomon wrote:
> http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi
> File common_conditions.gypi (right):
> 
>
http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi#newcode61
> common_conditions.gypi:61: 'VCLinkerTool': {
> On 2011/09/28 13:06:11, bungeman wrote:
> > Should set (I think 1 is correct for enabling, will need to check)
> > 
> > 'LinkTimeCodeGeneration': '1' #/LTCG
> > 
> > otherwise I don't think WholeProgramOptimization above will really do
> anything.
> 
> Done.

LGTM

[bungeman, 2011-09-28 13:48:34]

On 2011/09/28 13:13:51, bsalomon wrote:
> http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi
> File common_conditions.gypi (right):
> 
>
http://codereview.appspot.com/5142050/diff/8001/common_conditions.gypi#newcode61
> common_conditions.gypi:61: 'VCLinkerTool': {
> On 2011/09/28 13:06:11, bungeman wrote:
> > Should set (I think 1 is correct for enabling, will need to check)
> > 
> > 'LinkTimeCodeGeneration': '1' #/LTCG
> > 
> > otherwise I don't think WholeProgramOptimization above will really do
> anything.
> 
> Done.

LGTM

[bsalomon, 2011-09-28 14:34:02]

Checked in at r2346. I removed -g from the gcc release build flags because it
embeds the symbols in the executable. We can make separate sym files but it
isn't as simple as setting a single file.

I also commented out the change to the floating point model because it would
require new gm baseline images.

[bsalomon, 2011-09-28 14:34:48]

On 2011/09/28 14:34:02, bsalomon wrote:
> Checked in at r2346. I removed -g from the gcc release build flags because it
> embeds the symbols in the executable. We can make separate sym files but it
> isn't as simple as setting a single file.

err.. I mean "single flag".

> 
> I also commented out the change to the floating point model because it would
> require new gm baseline images.