A basic idea to walk around it in ANGLE is to expand all fields in struct flat recursively like this:

diff --git a/sdk/tests/deqp/data/gles3/shaders/linkage.test b/sdk/tests/deqp/data/gles3/shaders/linkage.test
index 5becf60..28095bd 100644
--- a/sdk/tests/deqp/data/gles3/shaders/linkage.test
+++ b/sdk/tests/deqp/data/gles3/shaders/linkage.test
@@ -381,22 +381,30 @@ group varying "Varying linkage"
                                #version 300 es
                                ${VERTEX_DECLARATIONS}
                                struct S { highp float a; highp vec2 b; };
-                               out S var;
+                               out highp float S_var_a;
+                               out highp vec2 S_var_b;
+                               S var_impl;
                                void main()
                                {
-                                       var.a = 2.0;
-                                       var.b = vec2(1.0, 0.0);
+                                       var_impl.a = 2.0;
+                                       var_impl.b = vec2(1.0, 0.0);
                                        ${VERTEX_OUTPUT}
+                                       S_var_a = var_impl.a;
+                                       S_var_b = var_impl.b;
                                }
                        ""
                        fragment ""
                                #version 300 es
                                ${FRAGMENT_DECLARATIONS}
                                struct S { highp float a; highp vec3 b; };
-                               in S var;
+                               in highp float S_var_a;
+                               in highp vec3 S_var_b;
+                               S var_impl;
                                void main()
                                {
-                                       ${FRAG_COLOR} = vec4(var.a, var.b);
+                                       var_impl.a = S_var_a;
+                                       var_impl.b = S_var_b;
+                                       ${FRAG_COLOR} = vec4(var_impl.a, var_impl.b);
                                }
                        ""
                end


Is it worth a try?

According to Jie's investigation, this is a general MacOSX issue on all GPU vendors. Ken and/or Zhenyao, could you ping Apple guys (through Dean?) to resolve this issue?
 
If that is infeasible or too slow, Jie can take a look and try to workaround this issue in Angle. 

Nah, this test passes on my desktop Mac (AMD Radeon HD - FirePro D500), mac retina laptop (AMD), mac laptop with Intel HD 6000.

On which platforms you saw this failure?

The only failure I saw in linkage.html is differing_interpolation_2, which is due to "centroid" matching rule.

It's a fairly new MacBookPro we have just bought. And it has dual GPUs, Intel and AMD, which can be switched arbitrarily, and either of which can reproduce this failure.

Can you provide the about:gpu info?

Chrome version  Chrome/55.0.2847.0
Operating system        Mac OS X 10.11.6
Software rendering list version 11.10
Driver bug list version 8.90
ANGLE commit id d5da505da77f
2D graphics backend     Skia/55 13a7eee2504e7deb0e27ed3e69a787696d57b037
Command Line Args       --enable-unsafe-es3-apis --flag-switches-begin --flag-switches-end
Driver Information
Initialization time     71
In-process GPU  false
Sandboxed       true
GPU0    VENDOR = 0x1002, DEVICE= 0x6821 *ACTIVE*
GPU1    VENDOR = 0x8086, DEVICE= 0x0d26
Optimus false
AMD switchable  true
Driver vendor
Driver version  1.42.15
Driver date
Pixel shader version    4.10
Vertex shader version   4.10
Max. MSAA samples       8
Machine model name      MacBookPro
Machine model version   11.5
GL_VENDOR       ATI Technologies Inc.
GL_RENDERER     AMD Radeon R9 M370X OpenGL Engine
GL_VERSION      4.1 ATI-1.42.15
GL_EXTENSIONS   GL_ARB_blend_func_extended GL_ARB_draw_buffers_blend GL_ARB_draw_indirect GL_ARB_ES2_compatibility GL_ARB_explicit_attrib_location GL_ARB_gpu_shader_fp64 GL_ARB_gpu_shader5 GL_ARB_instanced_arrays GL_ARB_internalformat_query GL_ARB_occlusion_query2 GL_ARB_sample_shading GL_ARB_sampler_objects GL_ARB_separate_shader_objects GL_ARB_shader_bit_encoding GL_ARB_shader_subroutine GL_ARB_shading_language_include GL_ARB_tessellation_shader GL_ARB_texture_buffer_object_rgb32 GL_ARB_texture_cube_map_array GL_ARB_texture_gather GL_ARB_texture_query_lod GL_ARB_texture_rgb10_a2ui GL_ARB_texture_storage GL_ARB_texture_swizzle GL_ARB_timer_query GL_ARB_transform_feedback2 GL_ARB_transform_feedback3 GL_ARB_vertex_attrib_64bit GL_ARB_vertex_type_2_10_10_10_rev GL_ARB_viewport_array GL_EXT_debug_label GL_EXT_debug_marker GL_EXT_depth_bounds_test GL_EXT_texture_compression_s3tc GL_EXT_texture_filter_anisotropic GL_EXT_texture_mirror_clamp GL_EXT_texture_sRGB_decode GL_APPLE_client_storage GL_APPLE_container_object_shareable GL_APPLE_flush_render GL_APPLE_object_purgeable GL_APPLE_rgb_422 GL_APPLE_row_bytes GL_APPLE_texture_range GL_ATI_texture_mirror_once GL_NV_texture_barrier
Disabled Extensions
Window system binding vendor
Window system binding version
Window system binding extensions
Direct rendering        Yes
Reset notification strategy     0x0000
GPU process crash count 0
Compositor Information
Tile Update Mode        Zero-copy
Partial Raster  Enabled
GpuMemoryBuffers Status
ATC     Software only
ATCIA   Software only
DXT1    Software only
DXT5    Software only
ETC1    Software only
R_8     GPU_READ_CPU_READ_WRITE, GPU_READ_CPU_READ_WRITE_PERSISTENT
BGR_565 Software only
RGBA_4444       Software only
RGBX_8888       Software only
RGBA_8888       GPU_READ, SCANOUT
BGRX_8888       GPU_READ, SCANOUT
BGRA_8888       GPU_READ, SCANOUT, GPU_READ_CPU_READ_WRITE, GPU_READ_CPU_READ_WRITE_PERSISTENT
YVU_420 Software only
YUV_420_BIPLANAR        GPU_READ_CPU_READ_WRITE, GPU_READ_CPU_READ_WRITE_PERSISTENT
UYVY_422        GPU_READ_CPU_READ_WRITE, GPU_READ_CPU_READ_WRITE_PERSISTENT

I have the same laptop. In canary, the failure is as I described, but on ToT, it's what you saw.  It is a regression in Chrome, most likely in ANGLE shader translator.  Can you bisect (or debug) a little?

Okay, what's the version number of your canary?

55.0.2846.4, the latest update

I have failed to bisect it. All builds were good. Actually I had found this issue over 2 weeks when I reported it. So it must be something wrong with my build. The major difference probably was 'debug' and 'release'. So I tried my release build of the same TOT, and it turned to be good. So far most likely it has been a debug build only bug.

The offending assert:

"err: uniqueId(139):     ! Assert failed in uniqueId(139): mUniqueId != 0

Assertion failed: (mUniqueId != 0), function uniqueId, file ../../third_party/angle/src/compiler/translator/Types.h, line 139."

It came from RegenerateStructName.
https://chromium.googlesource.com/angle/angle/+/master/src/compiler/translator/RegenerateStructNames.cpp#25

The TStructure without uniqueId was created while inserting initialization code for struct variables. See https://chromium.googlesource.com/angle/angle/+/master/src/compiler/translator/InitializeVariables.cpp#160
It looks we should have generated unique id for it here. There is another similar case https://chromium.googlesource.com/angle/angle/+/master/src/compiler/translator/Initialize.cpp#484
Should we enforce all such TStructure must have a valid unique id? I think it at least does no harm. 

The following revision refers to this bug:
  https://chromium.googlesource.com/angle/angle/+/6a25746c016b87739606b3d669a565e84d567a80

commit 6a25746c016b87739606b3d669a565e84d567a80
Author: jchen10 <jie.a.chen@intel.com>
Date: Tue Sep 06 00:56:08 2016

Generate uniqueId for all TStructure

If not present, ASSERT error may happen in debug build.

BUG= chromium:643075 

Change-Id: Ia57e3771ab4d2861aefc04287fbbce85232f1f4d
Reviewed-on: https://chromium-review.googlesource.com/381315
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>

[modify] https://crrev.com/6a25746c016b87739606b3d669a565e84d567a80/src/compiler/translator/Types.h
[modify] https://crrev.com/6a25746c016b87739606b3d669a565e84d567a80/src/compiler/translator/ParseContext.cpp
[modify] https://crrev.com/6a25746c016b87739606b3d669a565e84d567a80/src/compiler/translator/Types.cpp

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9bb4e60b43e03ec97398628839dab490c41170d1

commit 9bb4e60b43e03ec97398628839dab490c41170d1
Author: jmadill <jmadill@chromium.org>
Date: Wed Sep 07 02:35:51 2016

Roll ANGLE a4e6f07..5a7e20e

https://chromium.googlesource.com/angle/angle.git/+log/a4e6f07..5a7e20e

BUG= chromium:643075 , chromium:642227 , chromium:638323 , chromium:644057 , 351528 , chromium:593024 

TBR=geofflang@chromium.org

TEST=bots

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.win:win_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2312243003
Cr-Commit-Position: refs/heads/master@{#416828}

[modify] https://crrev.com/9bb4e60b43e03ec97398628839dab490c41170d1/DEPS