Do not share FBOs/VAOs/Transform feedback objects across contexts

gpu/command_buffer/common/gles2_cmd_format.h

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 // This file defines the GLES2 command buffer commands.
 
 #ifndef GPU_COMMAND_BUFFER_COMMON_GLES2_CMD_FORMAT_H_
 #define GPU_COMMAND_BUFFER_COMMON_GLES2_CMD_FORMAT_H_
 
 
@@ skipping 42 matching lines @@
 
 // Command buffer is GPU_COMMAND_BUFFER_ENTRY_ALIGNMENT byte aligned.
 #pragma pack(push, 4)
 static_assert(GPU_COMMAND_BUFFER_ENTRY_ALIGNMENT == 4,
               "pragma pack alignment must be equal to "
               "GPU_COMMAND_BUFFER_ENTRY_ALIGNMENT");
 
 namespace id_namespaces {
 
 // These are used when contexts share resources.
-enum IdNamespaces {
+enum SharedIdNamespaces {
   kBuffers,
-  kFramebuffers,
   kProgramsAndShaders,
   kRenderbuffers,
   kTextures,
-  kQueries,
-  kVertexArrays,
   kSamplers,
   kTransformFeedbacks,

[Zhenyao Mo, 2017/04/19 20:24:28]

Transform feedback objects are also not shared.

[Chandan, 2017/04/20 10:56:07]

Sure. I will add necessary changes for this too.

   kSyncs,
-  kNumIdNamespaces
+  kNumSharedIdNamespaces
 };
 
+enum IdNamespaces { kFramebuffers, kQueries, kVertexArrays, kNumIdNamespaces };

[Zhenyao Mo, 2017/04/19 20:24:28]

Although in general I agree with the spirit of this CL, but I think splitting
into SharedIdNamespaces and IdNamespaces is bad idea.  Now kFramebuffers ==
kBuffers, what mechanism prevents/detects the mis-use of
GetIdHandler(kFramebuffers) and GetIdHandler(kBuffers) get confused?

[Chandan, 2017/04/20 10:56:07]

I had to split this to create IdAllocators only for three(FBO/VAO/Query object)
locally in GLES2Implementation. Is it not necessary to remove kFramebuffers,
kQueries, kVertexArrays from IdNamespaces? Please suggest.

[Zhenyao Mo, 2017/05/09 22:21:54]

Per offline discussion with piman, we can use enum class in c++ 11.  This will
prevent from misuse, say both kFramebuffers and kBuffers are 0.  Of course then
you will need to manually define the mapping from enums to indices, but I think
it's worth it.

[Chandan, 2017/05/10 14:47:37]

By mapping, you mean converting enums to int using static_cast wherever
applicable? Something like:
static_cast<int>(
      id_namespaces::SharedIdNamespaces::kNumSharedIdNamespaces);

+
 enum RangeIdNamespaces { kPaths, kNumRangeIdNamespaces };
 
 // These numbers must not change
 static_assert(kBuffers == 0, "kBuffers should equal 0");
-static_assert(kFramebuffers == 1, "kFramebuffers should equal 1");
-static_assert(kProgramsAndShaders == 2, "kProgramsAndShaders should equal 2");
-static_assert(kRenderbuffers == 3, "kRenderbuffers should equal 3");
-static_assert(kTextures == 4, "kTextures should equal 4");
+static_assert(kProgramsAndShaders == 1, "kProgramsAndShaders should equal 1");
+static_assert(kRenderbuffers == 2, "kRenderbuffers should equal 2");
+static_assert(kTextures == 3, "kTextures should equal 4");

[Zhenyao Mo, 2017/05/09 22:21:54]

These need to be updated.

[Chandan, 2017/05/10 14:47:37]

Need to add static_assert for other enum values as well?

Here too, a static_cast<int> is required?
static_assert(static_cast<int>(SharedIdNamespaces::kBuffers) == 0,
              "kBuffers should equal 0");

 static_assert(kPaths == 0, "kPaths should equal 0");
 
 }  // namespace id_namespaces
 
 // Used for some glGetXXX commands that return a result through a pointer. We
 // need to know if the command succeeded or not and the size of the result. If
 // the command failed its result size will 0.
 template <typename T>
 struct SizedResult {
   typedef T Type;
@@ skipping 196 matching lines @@
 
 #include "../common/gles2_cmd_format_autogen.h"
 
 #pragma pack(pop)
 
 }  // namespace cmd
 }  // namespace gles2
 }  // namespace gpu
 
 #endif  // GPU_COMMAND_BUFFER_COMMON_GLES2_CMD_FORMAT_H_