[C++] Boost minimum version needs increasing for clang16 builds

[thisisnic]

Although the bundled boost version was increased in #33890, the clang16 build is still failing on CRAN because the machine it's being tested on has an older version already installed, and so the bundled version is not being used, and the build is failing.

We need to set the minimum version of Boost to a certain value if Arrow C++ is being compiled with clang16 in arrow/cpp/cmake_modules/ThirdpartyToolchain.cmake.

Currently the minimum value is determined by this block:

arrow/cpp/cmake_modules/ThirdpartyToolchain.cmake

Lines 1060 to 1064 in 4a5fadb

	if(ARROW_BUILD_TESTS)
	set(ARROW_BOOST_REQUIRED_VERSION "1.64")
	else()
	set(ARROW_BOOST_REQUIRED_VERSION "1.58")
	endif()

[thisisnic]

I've been able to replicate the CRAN failure locally by using this Dockerfile, in which I install Boost (version 1.74 is the default), so it should be possible to test a fix by rerunning this:

FROM rhub/clang16:latest

RUN apt-get update && apt-get install -y cmake git libssl-dev libboost-dev

# Checkout the Arrow + switch to a branch
RUN git clone https://github.com/thisisnic/arrow.git /arrow && \
  cd /arrow && \
  git switch r-11.0.0.1

RUN git config --global user.email "you@example.com" && \
  git config --global user.name "Your Name"

RUN mkdir /arrow-build && \
  cd /arrow-build && \
  cmake /arrow/cpp \
  -DCMAKE_C_COMPILER="clang-16"  -DCMAKE_CXX_COMPILER="clang++-16" \
  -DCMAKE_CXX_FLAGS="-stdlib=libc++" \
  -DARROW_CSV=ON -DARROW_DATASET=ON -DARROW_FILESYSTEM=ON \
  -DARROW_COMPUTE=ON -DARROW_PARQUET=ON -DBoost_SOURCE=AUTO

# Pass even if the build fails so that we can copy error logs from /arrow-build
ENV VERBOSE 1
RUN cd /arrow-build && cmake --build . > /build_log.txt || true
RUN cp /arrow-build/thrift_ep-prefix/src/thrift_ep-stamp/thrift_ep-build-*.log / || true

[paleolimbot]

Can we also fix this by adding -DBOOST_SOURCE=BUNDLED to the cmake command that we issue when doing a bundled build in the R package?

[thisisnic]

@paleolimbot I don't know where that is in the R package, but sounds like that would work. Mind commenting on the PR with that, now it's up?

[assignUser]

@paleolimbot that should work but is a workaround and not a fix imo.

[paleolimbot]

I like your approach in the PR!

It looks like we tried to ensure that system boost wasn't used but it doesn't work for some reason:

arrow/r/inst/build_arrow_static.sh

Line 54 in 4a5fadb

${CMAKE} -DARROW_BOOST_USE_SHARED=OFF \

[assignUser]

That's probably because the var is called Boost_NO_SYSTEM_PATHS (might have changed at some point?)

[assignUser]

Wait I skipped over the ARROW bit in the beginning xD I need to take a look. The above var turns off system boost in FindBoost.cmake used by find_package.

[nealrichardson]

IIUC -DARROW_BOOST_USE_SHARED=OFF is about using shared vs. static libraries (which happens to be irrelevant for us now since we only need boost headers for the R build). I don't think it's about using system packages.

If it were me, I'd add -DBOOST_SOURCE=BUNDLED and resubmit. @assignUser 's PR would also work (and should be merged for 12.0.0), though it's also a workaround, just at a lower level. The best long-term fix is to make a couple of contributions to Thrift so that it doesn't need boost (or at least the parts we need of it don't use boost).

[js8544]

Side note: I've dealt with the std::unary_function error before when upgrading our CI to clang-15 in #33029 and https://github.com/apache/arrow/blob/master/cpp/src/arrow/flight/test_util.cc#L35. Back in clang-15 it was a warning but -Werror caused build failures. Adding a #define BOOST_NO_CXX98_FUNCTION_BASE before including boost headers would be a simple workaround.