Hi Uwe, I apologize for taking your time and thank you for confirming that boost-cpp is no longer a runtime dependency. I see now that the issue was that our .condarc file placed the conda/anaconda channel about conda/conda-forge channel causing it to download from that channel which for some reason must still list boost-cpp as a dependency. Explicitly using conda-forge as the channel via "conda install -c conda-forge pyarrow=2.0.0" now no longer pulls in boost-cpp.
Thanks again, Alex On Mon, Feb 22, 2021 at 11:53 AM Uwe L. Korn <[email protected]> wrote: > Hello Alex, > > continuing this here instead of StackOverflow. Can you share your source > of conda packages: Do you use conda-forge? Can you share the output of > `conda list` of your current environment? > > What stucks me is that we have never built pyarrow on conda-forge against > boost-cpp 1.73, only 1.72 and 1.74. We though have removed boost-cpp as a > runtime dependency so I suspect the boost-cpp version change comes from > packages that aren't from conda-forge but from conda-forge. > > Best, > Uwe > > On Mon, Feb 22, 2021, at 1:48 AM, Alex Tulenko wrote: > > Hi, > > I am in the process of upgrading a project from pyarrow 0.8.0 to 2.0.0. As > part of this upgrade I see boost-cpp 1.63 is installed into the > environment. I have a number of libraries that were linked against an older > version of boost-cpp and so I would like to avoid pulling in this new > version of boost-cpp. > > I am generally looking for guidance on how to best handle this situation. > Running ldd on the libarrow*.so's installed into our conda env I can't > actually see any references to the libboost*.so's. > > This leads to my first question: is boost-cpp is a runtime requirement for > pyarrow 2.0.0? (maybe only for certain components?) If it is a runtime > requirement is there a recipe for building pyarrow/arrow-cpp with static > linkings or any version of pyarrow that can be built without boost-cpp? > > I am open to any suggestions on how to get around the boost-cpp dependency. > > Thanks! > Alex > > >
