Merge latest changes from http://www.netlib.org/fp/dtoa.c into dtoa.cc

Merge latest changes from http://www.netlib.org/fp/dtoa.c into dtoa.cc

This should fix the gcc 4.4. strict aliasing issues. More info here:
http://patrakov.blogspot.com/2009/03/dont-use-old-dtoac.html

Also update gcc_warnings.patch to match.

BUG=9104

[Craig Schlenter, 2009-05-02 15:37:25]

Mark, Tony: you were the last people to touch this file so I have picked you as
reviewers ... I hope that is ok.

I've taken the latest dtoa.c from www.netlib.org and merged those changes into
dtoa.cc. I have also updated the warnings patch but I've included the complete
diff to dtoa.cc there as I think it's easier to review that way although
arguably the file name is less accurate now.

The benefits are that most of the gcc warnings and the strict aliasing issues
are fixed.

I don't have any special knowledge of this code and to some degree other than
reviewing the current dtoa diff and the original warnings patch, it's been a
blind cut, paste and compile exercise so it's possible that something will have
broken. I have only compiled on linux in Release mode.

Thank you!

[tony, 2009-05-04 17:28:47]

Hi Craig, thanks for the patch.  For reference, I filed
http://code.google.com/p/chromium/issues/detail?id=11009 for this.

When I tried to update a week or two ago, I had some regressions in
base_unittests.  Do they pass for you?  Also, I think we want to define
NO_HEX_FP.

[Mark Mentovai, 2009-05-04 18:04:53]

Can we push the warnings patch upstream?

[Craig Schlenter, 2009-05-04 18:09:10]

On 2009/05/04 17:28:47, tony wrote:
> Hi Craig, thanks for the patch.  For reference, I filed
> http://code.google.com/p/chromium/issues/detail?id=11009 for this.

Ah, thank you. I didn't know about that. I've been mumbling about the problem
here:
http://code.google.com/p/chromium/issues/detail?id=9104

> When I tried to update a week or two ago, I had some regressions in
> base_unittests.  Do they pass for you?  Also, I think we want to define
> NO_HEX_FP.

Done. base_unittests do pass for me. Does that cover the nan thing you mentioned
in 11009?

Thank you.

[tony, 2009-05-04 18:17:09]

On 2009/05/04 18:09:10, Craig Schlenter wrote:
> On 2009/05/04 17:28:47, tony wrote:
> > Hi Craig, thanks for the patch.  For reference, I filed
> > http://code.google.com/p/chromium/issues/detail?id=11009 for this.
> 
> Ah, thank you. I didn't know about that. I've been mumbling about the problem
> here:
> http://code.google.com/p/chromium/issues/detail?id=9104

Ah, ok.  I've merge the bugs.  Can you add BUG=9104 in the change description?

> > When I tried to update a week or two ago, I had some regressions in
> > base_unittests.  Do they pass for you?  Also, I think we want to define
> > NO_HEX_FP.
> 
> Done. base_unittests do pass for me. Does that cover the nan thing you
mentioned
> in 11009?

Yes, it was in string_util_unittest.cc where I was seeing the inf thing.

> Thank you.

Thanks for the patch!  I will test it on hardy and land it if the tests pass.

Mark: Re warnings patch: It looks like all that's left are our #defines and C++
isms.  I think the warnings are already gone from upstream.

[Craig Schlenter, 2009-05-04 18:35:25]

On 2009/05/04 18:17:09, tony wrote:
> On 2009/05/04 18:09:10, Craig Schlenter wrote:
> > On 2009/05/04 17:28:47, tony wrote:
> > > Hi Craig, thanks for the patch.  For reference, I filed
> > > http://code.google.com/p/chromium/issues/detail?id=11009 for this.
> > 
> > Ah, thank you. I didn't know about that. I've been mumbling about the
problem
> > here:
> > http://code.google.com/p/chromium/issues/detail?id=9104
> 
> Ah, ok.  I've merge the bugs.  Can you add BUG=9104 in the change description?

Done. Will that close the bug or just send an update to it? The bug refers to a
range of gcc 4.4 issues so it probably shouldn't be closed yet.

> > > When I tried to update a week or two ago, I had some regressions in
> > > base_unittests.  Do they pass for you?  Also, I think we want to define
> > > NO_HEX_FP.
> > 
> > Done. base_unittests do pass for me. Does that cover the nan thing you
> mentioned
> > in 11009?
> 
> Yes, it was in string_util_unittest.cc where I was seeing the inf thing.
> 
> > Thank you.
> 
> Thanks for the patch!  I will test it on hardy and land it if the tests pass.
> 
> Mark: Re warnings patch: It looks like all that's left are our #defines and
C++
> isms.  I think the warnings are already gone from upstream.

I'll pull out the useful bits and send it to dmg at acm dot org tomorrow for
inclusion upstream.

Thank you.

[tony, 2009-05-04 20:01:25]

Landed in r15232.  Thanks!

[tony, 2009-05-04 20:06:06]

And reverted in r15233 because the opt build failed to compile.

[tony, 2009-05-04 20:37:53]

Relanded in r15236 with a fix for a gcc warning.