[llvm-commits] preserve pointer roots in globals

[Dan Gohman]

On Jun 29, 2012, at 2:01 AM, Chandler Carruth <chandlerc at google.com> wrote:

> On Fri, Jun 29, 2012 at 12:34 AM, Duncan Sands <baldrick at free.fr> wrote:
> 
> > This patch seems completely backwards to how it should be to me.
> 
> This is actually I think everyone's first reaction... and it's not unreasonable. We've been debating this pretty heatedly ourselves, so let me try to give some background.

Since this code is so prone to leading everyone to such reactions, it'd be
useful to capture your explanation in a comment in the source itself.

Thanks,

Dan