ARROW-14393: [C++] GTest linking errors during the source release verification

[kszucs]

Pin budled gtest to 1.10.0 as required by the thirdparty toolchain: https://github.com/apache/arrow/blob/master/cpp/cmake_modules/ThirdpartyToolchain.cmake#L1814

vcpkg crossbow build which was failing with the same reason as the source release verification script: https://github.com/ursacomputing/crossbow/runs/3951554360

[github-actions]

https://issues.apache.org/jira/browse/ARROW-14393

[github-actions]

⚠️ Ticket has not been started in JIRA, please click 'Start Progress'.

[kszucs]

It fixes the earlier failing vcpkg build and should fix the windows verification build as well because we install gtest=1.10 using conda.

[ursabot]

Benchmark runs are scheduled for baseline = 80ecf33 and contender = 4ac62d5. 4ac62d5 is a master commit associated with this PR. Results will be available as each benchmark for each run completes.
Conbench compare runs links:
[Finished ⬇️0.0% ⬆️0.0%] ec2-t3-xlarge-us-east-2
[Finished ⬇️0.0% ⬆️0.0%] ursa-i9-9960x
[Finished ⬇️0.27% ⬆️0.04%] ursa-thinkcentre-m75q
Supported benchmarks:
ursa-i9-9960x: langs = Python, R, JavaScript
ursa-thinkcentre-m75q: langs = C++, Java
ec2-t3-xlarge-us-east-2: cloud = True

[pitrou]

Why was this done? If you're using the conda package, you shouldn't use the bundled version, and vice-versa.

[lidavidm]

IIRC the issue was: we required GTest 1.11, but still installed 1.10 via Conda, and in the end we would use the headers from 1.10 but link to 1.11, because the -I directive for the Conda prefix's include dir came first. There's a longer discussion on Zulip.

[pitrou]

ARROW_GTEST_BUILD_VERSION is not the required version, it's only the version used when building a bundled version. It shouldn't conflict with the conda-installed version.

[pitrou]

However, it's an error to both install it with Conda and ask CMake to build a bundled version. I have no clue where this happens, because the release script is a mess.

[lidavidm]

Ah, sorry, I misunderstood then - but regardless, the failing builds were building GTest from source, so we still had the version mismatch.

The Zulip discussion starts here sorry, here is the full thread

The better fix would probably have been to prioritize the bundled include directory over the Conda one.