[ARROW-14210] [C++] CMAKE_AR is not passed to bzip2 thirdparty dependency

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 5.0.0
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/29793
• Assignee: Karl Dunkle Werner
• Reporter: Karl Dunkle Werner
• Created: 03/Oct/21 22:22
• Updated: 11/Jan/23 08:38
• Resolved: 06/Oct/21 01:33

Description

It seems like the AR or CMAKE_AR variables aren't getting passed for the bzip2 build, which causes if to fail if we're doing a BUNDLED build and ar isn't available in the $PATH (e.g. in a conda environment).

To replicate:
1. Download Arrow and start an interactive shell in a container
(docker should be fine if you prefer it to podman)

git clone --depth 1 [e-mail]:apache/arrow.git
podman run -it --rm -v ./arrow:/arrow:Z docker://ursalab/amd64-ubuntu-18.04-conda-python-3.6:worker bash

2. Build Arrow by running this in in the container:

export ARROW_BUILD_TOOLCHAIN=$CONDA_PREFIX
export ARROW_HOME=$CONDA_PREFIX
export PARQUET_HOME=$CONDA_PREFIX
export ARROW_BUILD_TYPE=DEBUG

cd /arrow
mkdir -p cpp/build
pushd cpp/build

cmake \
      -DCMAKE_BUILD_TYPE=$ARROW_BUILD_TYPE \
      -DCMAKE_INSTALL_PREFIX=$ARROW_HOME \
      -DCMAKE_AR=${AR} \
      -DCMAKE_RANLIB=${RANLIB} \
      -DARROW_WITH_BZ2=ON \
      -DARROW_VERBOSE_THIRDPARTY_BUILD=ON \
      -DARROW_JEMALLOC=OFF \
      -DARROW_SIMD_LEVEL=NONE -DARROW_RUNTIME_SIMD_LEVEL=NONE \
      -DARROW_DEPENDENCY_SOURCE=BUNDLED \
      ..
make
# make[3]: ar: No such file or directory
# make[3]: *** [Makefile:48: libbz2.a] Error 127
# make[2]: *** [CMakeFiles/bzip2_ep.dir/build.make:135: bzip2_ep-prefix/src/bzip2_ep-stamp/bzip2_ep-build] Error 2
# make[1]: *** [CMakeFiles/Makefile2:726: CMakeFiles/bzip2_ep.dir/all] Error 2

In the cmake call above, ARROW_JEMALLOC and the SIMD flags are just to skip compiling irrelevant things.

I think this line in ThirdpartyToolchain.cmake needs to be changed to pass CMAKE_AR.
https://github.com/apache/arrow/blob/bad8824d5cda0fd8337c7167729c49af868f93a5/cpp/cmake_modules/ThirdpartyToolchain.cmake#L2211

Other related issues have also needed to pass CMAKE_RANLIB, in addition to CMAKE_AR. I'm not sure if that applies here.

 
Related: ARROW-4471, ARROW-4831

Edit: added ARROW_BUILD_TYPE above.

Issue Links

• links to: GitHub Pull Request #11334