Have GMs read the GrContext via a setter/getter rather than a global.

Have GMs read the GrContext via a setter/getter rather than a global.

GetGr's current global state makes things rather tricky to run GPU GMs in
parallel (DM).  This API change will let me feed the right GrContext to the
right GM in DM.

I'm not planning on changing the status quo in GM-the-tool: the new getters and
setters still just return the same global.

BUG=skia:1590
R=bsalomon@google.com, robertphillips@google.com

Committed: https://code.google.com/p/skia/source/detail?r=11306

[robertphillips, 2013-09-16 18:13:37]

lgtm modulo question

https://codereview.chromium.org/23567032/diff/4001/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.chromium.org/23567032/diff/4001/gm/gmmain.cpp#newcode1588
gm/gmmain.cpp:1588: autogr.set(gr);
Why not just "gr"?

[mtklein, 2013-09-16 18:17:42]

https://codereview.chromium.org/23567032/diff/4001/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.chromium.org/23567032/diff/4001/gm/gmmain.cpp#newcode1588
gm/gmmain.cpp:1588: autogr.set(gr);
On 2013/09/16 18:13:38, robertphillips wrote:
> Why not just "gr"?

No good reason.  Got rid of GetGr() entirely.

[bsalomon, 2013-09-16 18:26:58]

With a little bit of work the GrContext is accessible from the canvas:

        GrContext* context = NULL;
        SkBaseDevice* device = canvas->getTopDevice();
        GrRenderTarget* rt = device->accessRenderTarget();
        if (NULL != rt) {
            context = rt->getContext();
        }

[robertphillips, 2013-09-16 18:39:29]

On 2013/09/16 18:26:58, bsalomon wrote:
> With a little bit of work the GrContext is accessible from the canvas:
> 
>         GrContext* context = NULL;
>         SkBaseDevice* device = canvas->getTopDevice();
>         GrRenderTarget* rt = device->accessRenderTarget();
>         if (NULL != rt) {
>             context = rt->getContext();
>         }

Do you think SkDevice should have an accessRenderTarget method?

[bsalomon, 2013-09-16 19:02:55]

On 2013/09/16 18:39:29, robertphillips wrote:
> On 2013/09/16 18:26:58, bsalomon wrote:
> > With a little bit of work the GrContext is accessible from the canvas:
> > 
> >         GrContext* context = NULL;
> >         SkBaseDevice* device = canvas->getTopDevice();
> >         GrRenderTarget* rt = device->accessRenderTarget();
> >         if (NULL != rt) {
> >             context = rt->getContext();
> >         }
> 
> Do you think SkDevice should have an accessRenderTarget method?

I think it's useful. Do you think it should be removed? I wonder if SkCanvas
should have a GrContext getter to make this cleaner and to make it possible to
get GrContext should we ever get to the point of making SkDevice a non-public
class.

[mtklein, 2013-09-16 19:54:55]

On 2013/09/16 19:02:55, bsalomon wrote:
> On 2013/09/16 18:39:29, robertphillips wrote:
> > On 2013/09/16 18:26:58, bsalomon wrote:
> > > With a little bit of work the GrContext is accessible from the canvas:
> > > 
> > >         GrContext* context = NULL;
> > >         SkBaseDevice* device = canvas->getTopDevice();
> > >         GrRenderTarget* rt = device->accessRenderTarget();
> > >         if (NULL != rt) {
> > >             context = rt->getContext();
> > >         }
> > 
> > Do you think SkDevice should have an accessRenderTarget method?
> 
> I think it's useful. Do you think it should be removed? I wonder if SkCanvas
> should have a GrContext getter to make this cleaner and to make it possible to
> get GrContext should we ever get to the point of making SkDevice a non-public
> class.

I uploaded a new patchset based on Brian's suggestion.  What do you guys think? 
I'm perfectly happy with both #3 and #4, but #4 is nicer to the tool-writer.

[bsalomon, 2013-09-16 20:38:04]

On 2013/09/16 19:54:55, mtklein wrote:
> On 2013/09/16 19:02:55, bsalomon wrote:
> > On 2013/09/16 18:39:29, robertphillips wrote:
> > > On 2013/09/16 18:26:58, bsalomon wrote:
> > > > With a little bit of work the GrContext is accessible from the canvas:
> > > > 
> > > >         GrContext* context = NULL;
> > > >         SkBaseDevice* device = canvas->getTopDevice();
> > > >         GrRenderTarget* rt = device->accessRenderTarget();
> > > >         if (NULL != rt) {
> > > >             context = rt->getContext();
> > > >         }
> > > 
> > > Do you think SkDevice should have an accessRenderTarget method?
> > 
> > I think it's useful. Do you think it should be removed? I wonder if SkCanvas
> > should have a GrContext getter to make this cleaner and to make it possible
to
> > get GrContext should we ever get to the point of making SkDevice a
non-public
> > class.
> 
> I uploaded a new patchset based on Brian's suggestion.  What do you guys
think? 
> I'm perfectly happy with both #3 and #4, but #4 is nicer to the tool-writer.

lgtm @ 4. I think we will eventually want to make this a method on SkCanvas
rather than a GM-specific helper.

[mtklein, 2013-09-16 20:40:15]

On 2013/09/16 20:38:04, bsalomon wrote:
> On 2013/09/16 19:54:55, mtklein wrote:
> > On 2013/09/16 19:02:55, bsalomon wrote:
> > > On 2013/09/16 18:39:29, robertphillips wrote:
> > > > On 2013/09/16 18:26:58, bsalomon wrote:
> > > > > With a little bit of work the GrContext is accessible from the canvas:
> > > > > 
> > > > >         GrContext* context = NULL;
> > > > >         SkBaseDevice* device = canvas->getTopDevice();
> > > > >         GrRenderTarget* rt = device->accessRenderTarget();
> > > > >         if (NULL != rt) {
> > > > >             context = rt->getContext();
> > > > >         }
> > > > 
> > > > Do you think SkDevice should have an accessRenderTarget method?
> > > 
> > > I think it's useful. Do you think it should be removed? I wonder if
SkCanvas
> > > should have a GrContext getter to make this cleaner and to make it
possible
> to
> > > get GrContext should we ever get to the point of making SkDevice a
> non-public
> > > class.
> > 
> > I uploaded a new patchset based on Brian's suggestion.  What do you guys
> think? 
> > I'm perfectly happy with both #3 and #4, but #4 is nicer to the tool-writer.
> 
> lgtm @ 4. I think we will eventually want to make this a method on SkCanvas
> rather than a GM-specific helper.

SGTM

[mtklein, 2013-09-16 20:42:21]

Committed patchset #4 manually as r11306 (presubmit successful).