Track GL buffer state based on unique resource ID

tests/GpuLayerCacheTest.cpp

Index: tests/GpuLayerCacheTest.cpp
diff --git a/tests/GpuLayerCacheTest.cpp b/tests/GpuLayerCacheTest.cpp
index 2249eba904c7dda92955bd1509a80afd53c2450f..5fc7422e7e780b17edaf10ceeb8ee37e514eb616 100644
--- a/tests/GpuLayerCacheTest.cpp
+++ b/tests/GpuLayerCacheTest.cpp
@@ -107,7 +107,7 @@ static void lock_layer(skiatest::Reporter* reporter,
 // In particular it checks its interaction with the resource cache (w.r.t.
 // locking & unlocking textures).
 // TODO: need to add checks on VRAM usage!
-DEF_GPUTEST_FOR_RENDERING_CONTEXTS(GpuLayerCache, reporter, ctxInfo) {
+DEF_GPUTEST_FOR_NULL_CONTEXT(GpuLayerCache, reporter, ctxInfo) {

[Chris Dalton, 2016/04/06 20:06:21]

This technically isn't a problem right now, but since the GrGLGpu can now create
resources of its own it isn't safe to assert on absolute stats anymore. Null
context should be fine. Is there a reason I'm not seeing that this test needs
real rendering contexts?

[bsalomon, 2016/04/06 20:58:06]

It still passes because we don't trigger those creations in this test? I don't
think making this run on the null GL context will really help at all because it
still has a GrGLGpu which can create (null) resources of its own.

Perhaps this test should track deltas rather then absolute counts?

One thing I've I wanted to do for a long time is beef up the MockContext to
actually be functional (have Mock resource classes). Since it uses its own
MockGpu subclass of GrGpu it would circumvent this problem, but I don't see
anyone doing it anytime soon.

Also, I imagine when Rob finishes the work to have separate DrawTargets for
different DrawContexts this test can be deleted.

     // Add one more layer than can fit in the atlas
     static const int kInitialNumLayers = TestingAccess::NumPlots() + 1;