apacheGH-41450: [R][CI] rhub/container follow ons (apache#41451)

More CI changes: * GitHub Issue: apache#41450 (specifically use the rhub containers approach for clang sanitizer, remove some of our work arounds) * Remove CentOS 7 CI support for R Authored-by: Jonathan Keane <jkeane@gmail.com> Signed-off-by: Jonathan Keane <jkeane@gmail.com>
assignUser · May 12, 2024 · 6d03215 · 6d03215
1 parent 37bd413
commit 6d03215
Show file tree

Hide file tree

Showing 14 changed files with 15 additions and 97 deletions.
diff --git a/.env b/.env
@@ -86,9 +86,6 @@ ARROW_R_DEV=TRUE
 R_PRUNE_DEPS=FALSE
 TZ=UTC
 
-# Any non-empty string will install devtoolset-${DEVTOOLSET_VERSION}
-DEVTOOLSET_VERSION=
-
 # Used through docker-compose.yml and serves as the default version for the
 # ci/scripts/install_vcpkg.sh script. Prefer to use short SHAs to keep the
 # docker tags more readable.

diff --git a/.github/workflows/r.yml b/.github/workflows/r.yml
@@ -192,12 +192,11 @@ jobs:
       fail-fast: false
       matrix:
         config:
-          - { org: "rhub", image: "ubuntu-gcc12", tag: "latest", devtoolset: "" }
+          - { org: "rhub", image: "ubuntu-gcc12", tag: "latest" }
     env:
       R_ORG: ${{ matrix.config.org }}
       R_IMAGE: ${{ matrix.config.image }}
       R_TAG: ${{ matrix.config.tag }}
-      DEVTOOLSET_VERSION: ${{ matrix.config.devtoolset }}
     steps:
       - name: Checkout Arrow
         uses: actions/checkout@3df4ab11eba7bda6032a0b82a6bb43b11571feac # v4.0.0

diff --git a/ci/docker/linux-r.dockerfile b/ci/docker/linux-r.dockerfile
@@ -27,9 +27,6 @@ ENV R_BIN=${r_bin}
 ARG r_dev=FALSE
 ENV ARROW_R_DEV=${r_dev}
 
-ARG devtoolset_version=
-ENV DEVTOOLSET_VERSION=${devtoolset_version}
-
 ARG r_prune_deps=FALSE
 ENV R_PRUNE_DEPS=${r_prune_deps}
 

diff --git a/ci/scripts/java_jni_manylinux_build.sh b/ci/scripts/java_jni_manylinux_build.sh
@@ -35,9 +35,6 @@ echo "=== Clear output directories and leftovers ==="
 rm -rf ${build_dir}
 
 echo "=== Building Arrow C++ libraries ==="
-devtoolset_version=$(rpm -qa "devtoolset-*-gcc" --queryformat %{VERSION} | \
-                       grep -o "^[0-9]*")
-devtoolset_include_cpp="/opt/rh/devtoolset-${devtoolset_version}/root/usr/include/c++/${devtoolset_version}"
 : ${ARROW_ACERO:=ON}
 export ARROW_ACERO
 : ${ARROW_BUILD_TESTS:=ON}

diff --git a/ci/scripts/r_docker_configure.sh b/ci/scripts/r_docker_configure.sh
@@ -67,26 +67,6 @@ sloppiness = include_file_ctime
 hash_dir = false" >> ~/.ccache/ccache.conf
 fi
 
-# Special hacking to try to reproduce quirks on centos using non-default build
-# tooling.
-if [[ -n "$DEVTOOLSET_VERSION" ]]; then
-  $PACKAGE_MANAGER install -y centos-release-scl
-  $PACKAGE_MANAGER install -y "devtoolset-$DEVTOOLSET_VERSION"
-
-  # Enable devtoolset here so that `which gcc` finds the right compiler below
-  source /opt/rh/devtoolset-${DEVTOOLSET_VERSION}/enable
-
-  # Build images which require the devtoolset don't have CXX17 variables
-  # set as the system compiler doesn't support C++17
-  if [ ! "`{R_BIN} CMD config CXX17`" ]; then
-    mkdir -p ~/.R
-    echo "CC = $(which gcc) -fPIC" >> ~/.R/Makevars
-    echo "CXX17 = $(which g++) -fPIC" >> ~/.R/Makevars
-    echo "CXX17STD = -std=c++17" >> ~/.R/Makevars
-    echo "CXX17FLAGS = ${CXX11FLAGS}" >> ~/.R/Makevars
-  fi
-fi
-
 if [ -f "${ARROW_SOURCE_HOME}/ci/scripts/r_install_system_dependencies.sh" ]; then
   "${ARROW_SOURCE_HOME}/ci/scripts/r_install_system_dependencies.sh"
 fi

diff --git a/ci/scripts/r_sanitize.sh b/ci/scripts/r_sanitize.sh
@@ -46,6 +46,8 @@ unset ARROW_R_DEV
 export ARROW_R_VERBOSE_TEST=TRUE
 
 export UBSAN_OPTIONS="print_stacktrace=1,suppressions=/arrow/r/tools/ubsan.supp"
+# From the old rhub image https://github.com/r-hub/rhub-linux-builders/blob/master/fedora-clang-devel-san/Dockerfile
+export ASAN_OPTIONS="alloc_dealloc_mismatch=0:detect_leaks=0:detect_odr_violation=0"
 
 # run tests
 pushd tests

diff --git a/ci/scripts/r_test.sh b/ci/scripts/r_test.sh
@@ -48,9 +48,6 @@ if [ "$ARROW_USE_PKG_CONFIG" != "false" ]; then
 fi
 
 export _R_CHECK_COMPILATION_FLAGS_KNOWN_="${_R_CHECK_COMPILATION_FLAGS_KNOWN_} ${ARROW_R_CXXFLAGS}"
-# These should generally be picked up, but are slightly wrong in rhub's containers it appears
-# https://github.com/r-hub/containers/pull/63
-export _R_CHECK_COMPILATION_FLAGS_KNOWN_="${_R_CHECK_COMPILATION_FLAGS_KNOWN_} -Wno-parentheses -Werror=format-security -Wp,-D_FORTIFY_SOURCE=3"
 
 if [ "$ARROW_R_DEV" = "TRUE" ]; then
   # These are sometimes used in the Arrow C++ build and are not a problem

diff --git a/dev/tasks/r/azure.linux.yml b/dev/tasks/r/azure.linux.yml
@@ -38,7 +38,6 @@ jobs:
       export R_ORG={{ r_org }}
       export R_IMAGE={{ r_image }}
       export R_TAG={{ r_tag }}
-      export DEVTOOLSET_VERSION={{ devtoolset_version|default("") }}
       export R_CUSTOM_CCACHE={{ r_custom_ccache|default("false") }}
       docker-compose pull --ignore-pull-failures r
       docker-compose build r

diff --git a/dev/tasks/r/github.packages.yml b/dev/tasks/r/github.packages.yml
@@ -299,14 +299,14 @@ jobs:
           # choosing a binary on this OS. If libarrow_binary is TRUE, we're on
           # an OS that is not in the allowlist, so we have to opt-in to use the
           # binary. Other env vars used in r_docker_configure.sh can be added
-          # here (like devtoolset) and wired up in the later steps.
+          # here and wired up in the later steps.
           - {image: "rhub/ubuntu-clang", libarrow_binary: "TRUE"}
           # fedora-clang-devel cannot use binaries bc of libc++ (uncomment to see the error)
           # - {image: "rhub/fedora-clang-devel", libarrow_binary: "TRUE"}
           - {image: "rhub/ubuntu-release"} # currently ubuntu-22.04
           - {image: "rocker/r-ver:4.0.0"} # ubuntu-20.04
-          - {image: "rstudio/r-base:4.1-focal"} # ubuntu-20.04
-          - {image: "rstudio/r-base:4.2-centos7", devtoolset: "8"}
+          - {image: "rstudio/r-base:4.1-focal"}
+          - {image: "rstudio/r-base:4.2-jammy"}
           - {image: "rstudio/r-base:4.3-noble"}
     steps:
       # Get the arrow checkout just for the docker config scripts
@@ -317,7 +317,6 @@ jobs:
       - name: Install system requirements
         env:
           ARROW_R_DEV: "TRUE" # To install curl/openssl in r_docker_configure.sh
-          DEVTOOLSET_VERSION: {{ '${{ matrix.config.devtoolset }}' }}
         shell: bash
         run: |
           # Make sure R is on the path for the R-hub devel versions (where RPREFIX is set in its dockerfile)

diff --git a/dev/tasks/tasks.yml b/dev/tasks/tasks.yml
@@ -1410,15 +1410,6 @@ tasks:
         GCC_VERSION: 12
       image: ubuntu-r-only-r
 
-  test-r-rstudio-r-base-4.2-centos7-devtoolset-8:
-    ci: azure
-    template: r/azure.linux.yml
-    params:
-      r_org: rstudio
-      r_image: r-base
-      r_tag: 4.2-centos7
-      devtoolset_version: 8
-
   test-r-minimal-build:
     ci: azure
     template: r/azure.linux.yml
@@ -1436,13 +1427,13 @@ tasks:
         R_PRUNE_DEPS: TRUE
       image: ubuntu-r-sanitizer
 
-  test-fedora-r-clang-sanitizer:
+  test-r-clang-sanitizer:
     ci: github
     template: docker-tests/github.linux.yml
     params:
       env:
         R_PRUNE_DEPS: TRUE
-      image: fedora-r-clang-sanitizer
+      image: r-clang-sanitizer
 
   {% for go_version, staticcheck in [("1.21", "v0.4.7"), ("1.22", "latest")] %}
   test-debian-12-go-{{ go_version }}:

diff --git a/docker-compose.yml b/docker-compose.yml
@@ -162,7 +162,7 @@ x-hierarchy:
   - ubuntu-r-valgrind
   - ubuntu-swift
   - ubuntu-verify-rc
-  - fedora-r-clang-sanitizer
+  - r-clang-sanitizer
   - r
   - r-revdepcheck
   # helper services
@@ -1472,7 +1472,6 @@ services:
       args:
         base: ${R_ORG}/${R_IMAGE}:${R_TAG}
         r_dev: ${ARROW_R_DEV}
-        devtoolset_version: ${DEVTOOLSET_VERSION}
         tz: ${TZ}
         r_prune_deps: ${R_PRUNE_DEPS}
         r_custom_ccache: ${R_CUSTOM_CCACHE}
@@ -1484,7 +1483,6 @@ services:
       ARROW_R_DEV: ${ARROW_R_DEV}
       # To test for CRAN release, delete ^^ these two env vars so we download the Apache release
       ARROW_USE_PKG_CONFIG: "false"
-      devtoolset_version: ${DEVTOOLSET_VERSION}
     volumes:
       - .:/arrow:delegated
     command: >
@@ -1517,19 +1515,17 @@ services:
       /bin/bash -c "
         /arrow/ci/scripts/r_sanitize.sh /arrow"
 
-  fedora-r-clang-sanitizer:
-    image: ${REPO}:r-rhub-fedora-clang-devel-latest
+  r-clang-sanitizer:
+    image: ${REPO}:r-rhub-clang-devel-latest
     build:
       context: .
       dockerfile: ci/docker/linux-r.dockerfile
       cache_from:
-        - ${REPO}:r-rhub-fedora-clang-devel-latest
+        - ${REPO}:r-rhub-clang-devel-latest
       args:
-        # TODO: change this to rhub/clang-asan
-        base: rhub/fedora-clang-devel-san
+        base: rhub/clang-asan
         r_dev: ${ARROW_R_DEV}
-        devtoolset_version: ${DEVTOOLSET_VERSION}
-        r_bin: RDsan
+        r_bin: R
         tz: ${TZ}
         r_prune_deps: ${R_PRUNE_DEPS}
     shm_size: *shm-size

diff --git a/r/tools/test-nixlibs.R b/r/tools/test-nixlibs.R
@@ -158,10 +158,6 @@ test_that("check_allowlist", {
 })
 
 test_that("find_latest_nightly()", {
-  skip_if(
-    getRversion() > "4.4.0",
-    "long last version components (>8) fail to max on r-devel"
-  )
   tf <- tempfile()
   tf_uri <- paste0("file://", tf)
   on.exit(unlink(tf))

diff --git a/r/tools/ubsan.supp b/r/tools/ubsan.supp
@@ -16,3 +16,4 @@
 # under the License.
 
 vptr:include/c++/8/bits/shared_ptr_base.h
+function:cleancall.c
diff --git a/r/vignettes/install.Rmd b/r/vignettes/install.Rmd
@@ -28,35 +28,6 @@ For `gcc`, this generally means version 7 or newer. Most contemporary Linux
 distributions have a new enough compiler; however, CentOS 7 is a notable
 exception, as it ships with gcc 4.8.
 
-If you are on CentOS 7, to build arrow you will need to install a newer `devtoolset`, and you'll need to update R's Makevars to define the `CXX17` variables. This script installs `devtoolset-8` and configures R to be able to use C++17:
-
-```
-#!/usr/bin/env bash
-
-yum install -y centos-release-scl
-yum install -y devtoolset-8
-# Optional: also install cloud storage dependencies, as described below
-yum install -y libcurl-devel openssl-devel
-
-source /opt/rh/devtoolset-8/enable
-
-if [ ! `R CMD config CXX17` ]; then
-  mkdir -p ~/.R
-  echo "CC = $(which gcc) -fPIC" >> ~/.R/Makevars
-  echo "CXX17 = $(which g++) -fPIC" >> ~/.R/Makevars
-  echo "CXX17STD = -std=c++17" >> ~/.R/Makevars
-  echo "CXX17FLAGS = ${CXX11FLAGS}" >> ~/.R/Makevars
-fi
-```
-
-Note that the C++17 compiler is only required at *build* time. You don't need
-to enable the devtoolset every time you load the package. What's more, if you
-install a binary package from RStudio Package Manager (see method 1a below), you
-do not need to set up any of this. Likewise, if you `R CMD INSTALL --build`
-arrow on a CentOS machine with the newer compilers, you can take the binary
-package it produces and install it on any other CentOS machine without those
-compilers.
-
 ### Libraries
 
 Optional support for reading from cloud storage--AWS S3 and
@@ -517,10 +488,6 @@ The install script should work everywhere, so if libarrow fails to compile,
 please [report an issue](https://issues.apache.org/jira/projects/ARROW/issues)
 so that we can improve the script.
 
-### Known installation issues
-
-* On CentOS, building the package requires a more modern `devtoolset` than the default system compilers. See "System dependencies" above.
-
 ## Contributing
 
 We are constantly working to make the installation process as painless as
Original file line number	Diff line number	Diff line change
Expand Up		@@ -16,3 +16,4 @@
		# under the License.

		vptr:include/c++/8/bits/shared_ptr_base.h
		function:cleancall.c