Guard visibility variables on AIX

[Jake-Egan]

Summary

The addition of these visibility variables caused a test-suite build failure on AIX because visibility options aren't supported. This patch guards the variables on AIX.

[lebedev.ri]

Please can you submit this upstream first?

[daltenty]

In D122316#3402566, @lebedev.ri wrote:

Please can you submit this upstream first?

The added visibility options are breaking all the AIX bots:
https://lab.llvm.org/buildbot/#/builders/214

Either this needs to be fixed locally or we'll need to revert the benchmark update in D121343

[lebedev.ri]

Right. But that's not the question i asked, i believe?

[daltenty]

Right. But that's not the question i asked, i believe?

Sorry if I've misunderstood. We'll make sure this get posted to Google Benchmark, but do you want to wait for that to land upstream first? If so, shall we revert D121343 to unblock the bots?

[daltenty]

LGTM as an interm fix. We'll revert and replace this once we have a fix in Google Benchmark.

[mtrofin]

In D122316#3403326, @daltenty wrote:

LGTM as an interm fix. We'll revert and replace this once we have a fix in Google Benchmark.

Apologies if I missed it, has the google/benchmarks PR been submitted? Thanks!

[Jake-Egan]

In D122316#3457227, @mtrofin wrote:

In D122316#3403326, @daltenty wrote:

LGTM as an interm fix. We'll revert and replace this once we have a fix in Google Benchmark.

Apologies if I missed it, has the google/benchmarks PR been submitted? Thanks!

Hi, we are looking at fixing the visibility issue instead: https://discourse.llvm.org/t/rfc-adding-exported-visibility-style-to-the-ir-to-model-xcoff-exported-visibility/61853/4. Then we can revert this patch after.

[mtrofin]

In D122316#3464814, @Jake-Egan wrote:

In D122316#3457227, @mtrofin wrote:

In D122316#3403326, @daltenty wrote:

LGTM as an interm fix. We'll revert and replace this once we have a fix in Google Benchmark.

Apologies if I missed it, has the google/benchmarks PR been submitted? Thanks!

Hi, we are looking at fixing the visibility issue instead: https://discourse.llvm.org/t/rfc-adding-exported-visibility-style-to-the-ir-to-model-xcoff-exported-visibility/61853/4. Then we can revert this patch after.

Ah - got it, thank you!

MicroBenchmarks/libs/benchmark/CMakeLists.txt

 # This option can be used to disable building and running unit tests which depend on gtest
 # in cases where it is not possible to build or find a valid version of gtest.
 option(BENCHMARK_ENABLE_GTEST_TESTS "Enable building the unit tests which depend on gtest" ON)
 option(BENCHMARK_USE_BUNDLED_GTEST "Use bundled GoogleTest. If disabled, the find_package(GTest) will be used." ON)
 
 option(BENCHMARK_ENABLE_LIBPFM "Enable performance counters provided by libpfm" OFF)
 
 # Export only public symbols
+if(NOT ${CMAKE_SYSTEM_NAME} MATCHES "AIX")
   set(CMAKE_CXX_VISIBILITY_PRESET hidden)
   set(CMAKE_VISIBILITY_INLINES_HIDDEN ON)
+endif()
 
 if(MSVC)
     # As of CMake 3.18, CMAKE_SYSTEM_PROCESSOR is not set properly for MSVC and
     # cross-compilation (e.g. Host=x86_64, target=aarch64) requires using the
     # undocumented, but working variable.
     # See https://gitlab.kitware.com/cmake/cmake/-/issues/15170
     set(CMAKE_SYSTEM_PROCESSOR ${MSVC_CXX_ARCHITECTURE_ID})
     if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "ARM")