Fix velox dylib loading paths for conda build on macos (#355)

Summary:
In this PR I'm attempting to introduce a temporary workaround to fix the velox dylib issue, via a post-build hack. This is to make `conda install torcharrow` work out of the box.

Currently, you will encounter an error of dylibs not found when velox is dlopen-ing them. This is because the location of those dylibs were determined at compile time. For example: `/usr/local/opt/gflags/lib/libgflags.2.2.dylib` was installed to a default location by `brew install`.

We can specify the dylib dependencies in the conda recipe, but they will be installed to the conda environment locations, rather than the system ones (`/user/local/...`). As a result, the velox dylib (`_torcharrow.*.so`) won't be able to find those libraries.

On macos there is a tool called `install_name_tool` that can change the dylib paths, my plan is to modify those dylib paths from absolute ones to `loader_path/...` so that the velox .so can find the dylibs installed by conda.

Pull Request resolved: #355

Reviewed By: wenleix

Differential Revision: D36956804

Pulled By: bearzx

fbshipit-source-id: 2551910eaec0c52b59b769de67c52aae633bbbc6

Author: bearzx

.github/workflows/nightly-conda.yml

@@ -136,6 +136,29 @@ jobs:
           scripts/build_mac_dep.sh ranges_v3 fmt double_conversion folly re2
           ./packaging/build_conda.sh
 
+      - name: Fix Velox Dylib Paths
+        shell: bash -l {0}
+        env:
+          PYTHON_VERSION: ${{ matrix.python-version }}
+        run: |
+          cd conda-bld/osx-64
+          pkg_name=`ls ./torcharrow-* | sed -n -e 's/.*\(torcharrow.*\).tar.bz2/\1/p'`
+          mkdir ./${pkg_name}
+          tar -xf ./${pkg_name}.tar.bz2 -C ./${pkg_name}
+          rm ./${pkg_name}.tar.bz2
+          cd ./${pkg_name}
+
+          source ../../../packaging/fix_conda_dylib_paths.sh
+          conda_lib_folder=lib/python${PYTHON_VERSION}/site-packages/torcharrow
+          so_name=`ls ${conda_lib_folder}/_torcharrow.* | sed -n -e 's/.*\(_torcharrow.*\.so\)/\1/p'`
+          fix_velox_dylib_paths ${conda_lib_folder}/${so_name}
+          otool -L ${conda_lib_folder}/${so_name}
+
+          tar -cjf ../${pkg_name}.tar.bz2 *
+          cd ..
+          rm -rf ./${pkg_name}
+          cd ../..
+
       - name: Conda Upload
         shell: bash -l {0}
         env:

packaging/build_conda.sh

@@ -19,4 +19,4 @@ setup_env 0.1
 setup_conda_pytorch_constraint
 
 mkdir -p conda-bld
-conda build $CONDA_CHANNEL_FLAGS --no-anaconda-upload --output-folder conda-bld --python "$PYTHON_VERSION" packaging/torcharrow
+conda build "$CONDA_CHANNEL_FLAGS" -c conda-forge --no-anaconda-upload --output-folder conda-bld --python "$PYTHON_VERSION" packaging/torcharrow

packaging/fix_conda_dylib_paths.sh

@@ -0,0 +1,39 @@
+#!/usr/bin/env bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+function fix_velox_dylib_paths() {
+  velox_so=$1
+
+  # other libs
+  libs="libgflags libglog libz libssl libcrypto libbz2 liblz4 libzstd libsodium"
+
+  for libx in ${libs}
+  do
+    liby=$(otool -L "${velox_so}" | sed -n -e "s/\\(.*${libx}\..*dylib\\).*/\\1/p" | tr -d '[:blank:]')
+
+    install_name_tool -change "${liby}" "@loader_path/../../../${libx}.dylib" "${velox_so}"
+  done
+
+  libx=libevent
+  # libevent
+  liby=$(otool -L "${velox_so}" | sed -n -e "s/\\(.*${libx}-.*dylib\\).*/\\1/p" | tr -d '[:blank:]')
+  install_name_tool -change "${liby}" "@loader_path/../../../${libx}.dylib" "${velox_so}"
+
+  # boost libs
+  boost_libs="libboost_context libboost_filesystem libboost_atomic libboost_regex libboost_system libboost_thread"
+
+  for libx in ${boost_libs}
+  do
+    liby=$(otool -L "${velox_so}" | sed -n -e "s/\\(.*${libx}.*dylib\\).*/\\1/p" | tr -d '[:blank:]')
+
+    install_name_tool -change "${liby}" "@loader_path/../../../${libx}.dylib" "${velox_so}"
+  done
+
+  libx=libboost_program
+  liby=$(otool -L "${velox_so}" | sed -n -e "s/\\(.*${libx}.*dylib\\).*/\\1/p" | tr -d '[:blank:]')
+  install_name_tool -change "${liby}" "@loader_path/../../../${libx}_options.dylib" "${velox_so}"
+}

packaging/torcharrow/meta.yaml

@@ -20,20 +20,22 @@ requirements:
     - typing
     - tabulate
     - pyarrow
-    - arrow
+    - arrow-cpp==8.0.0
     - cffi
-    - glog==0.4.0
+    - glog==0.6.0
     - libsodium
-    {{ environ.get('CONDA_PYTORCH_CONSTRAINT') }}
+    - libboost
+    - libevent
+    - bzip2
+    - lz4-c
 
 build:
   string: py{{py}}
   script_env:
     - BUILD_VERSION
     - CPU_TARGET
+    - PYTHON_VERSION
 
-#GitHub Actions usually provide Mac without AVX2 support, for now just skip the test until Velox can be run without AVX2
-#pytest --no-header -v torcharrow/test
 test:
   imports:
     - torcharrow