Comment from fsamuel on the patch after revert: "Dynamically sized arrays on the stack is not a standard C++ feature IIRC." Compiler difference, then?

+andybons for crossover team

For the sake of other folks who look into this, links to the various builds:

Passing builds on the CQ:
debug compile: https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_compile_dbg_ng/builds/296920
regular compile: https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/332273
x64 compile: https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_x64_rel_ng/builds/317654
clang compile: https://build.chromium.org/p/tryserver.chromium.win/builders/win_clang/builds/122120

Failing build on the CI:
regular compile: https://build.chromium.org/p/chromium/builders/Win/builds/49156

Actual compile failure message:
[21777/54633] CXX obj/cc/cc_perftests/surface_aggregator_perftest.obj
FAILED: obj/cc/cc_perftests/surface_aggregator_perftest.obj 
ninja -t msvc -e environment.x86 -- C:\b\c\cipd\goma/gomacc.exe "C:\b\depot_tools\win_toolchain\vs_files\d5dc33b15d1b2c086f2f6632e2fd15882f80dbd3\VC\bin\amd64_x86/cl.exe" /nologo /showIncludes /FC @obj/cc/cc_perftests/surface_aggregator_perftest.obj.rsp /c ../../cc/surfaces/surface_aggregator_perftest.cc /Foobj/cc/cc_perftests/surface_aggregator_perftest.obj /Fd"obj/cc/cc_perftests_cc.pdb"
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(50): error C2131: expression did not evaluate to a constant
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(50): note: failure was caused by non-constant arguments or reference to a non-constant symbol
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(50): note: see usage of 'num_surfaces'
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(52): error C2228: left of '.reset' must have class/struct/union
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(52): note: type is 'std::unique_ptr<cc::SurfaceFactory,std::default_delete<_Ty>> [num_surfaces]'
        with
        [
            _Ty=cc::SurfaceFactory
        ]
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(99): error C2039: 'SubmitCompositorFrame': is not a member of 'std::unique_ptr<cc::SurfaceFactory,std::default_delete<_Ty>>'
        with
        [
            _Ty=cc::SurfaceFactory
        ]
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(50): note: see declaration of 'std::unique_ptr<cc::SurfaceFactory,std::default_delete<_Ty>>'
        with
        [
            _Ty=cc::SurfaceFactory
        ]
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(138): error C2039: 'EvictSurface': is not a member of 'std::unique_ptr<cc::SurfaceFactory,std::default_delete<_Ty>>'
        with
        [
            _Ty=cc::SurfaceFactory
        ]
c:\b\c\b\win_archive\src\cc\surfaces\surface_aggregator_perftest.cc(50): note: see declaration of 'std::unique_ptr<cc::SurfaceFactory,std::default_delete<_Ty>>'
        with
        [
            _Ty=cc::SurfaceFactory
        ]

surface_aggregator_perftest.cc is built as part of //cc:cc_perftests. 

//cc:cc_perftests is not built by default in the commit queue. Due to the slowness of the Windows compile, we do not build everything on every job, only the targets that are explicitly configured to be built and that are affected by the patch (building everything would make the CQ jobs take too long and so we've intentionally chosen not to do this).

The "Win" bot on the p/chromium waterfall, on the other hand, builds "all". 

So, this all played out the way I would've expected it to. This happens from time to time and is a known tradeoff of the way we've done things.