GH-45166: [CI][C++] Upgrade Alpine Linux to 3.18 from 3.16 (#45168)

### Rationale for this change

Alpine Linux 3.16 reached EOL.

### What changes are included in this PR?

Use Alpine Linux 3.18 that is the oldest maintained version.

### Are these changes tested?

Yes.

### Are there any user-facing changes?

No.
* GitHub Issue: #45166

Authored-by: Sutou Kouhei <kou@clear-code.com>
Signed-off-by: Jacob Wujciak-Jens <jacob@wujciak.de>

Author: kou

.env

@@ -47,7 +47,7 @@ ULIMIT_CORE=-1
 
 # Default versions for platforms
 ALMALINUX=8
-ALPINE_LINUX=3.16
+ALPINE_LINUX=3.18
 DEBIAN=12
 FEDORA=39
 UBUNTU=22.04

ci/docker/alpine-linux-3.16-cpp.dockerfile ci/docker/alpine-linux-3.18-cpp.dockerfile

@@ -16,9 +16,10 @@
 # under the License.
 
 ARG arch=amd64
-FROM ${arch}/alpine:3.16
+FROM ${arch}/alpine:3.18
 
 RUN apk add \
+        apache-orc-dev \
         bash \
         benchmark-dev \
         boost-dev \
@@ -39,8 +40,8 @@ RUN apk add \
         grpc-dev \
         gtest-dev \
         libxml2-dev \
-        llvm13-dev \
-        llvm13-static \
+        llvm16-dev \
+        llvm16-static \
         lz4-dev \
         make \
         musl-locales \
@@ -61,6 +62,7 @@ RUN apk add \
         thrift-dev \
         tzdata \
         utf8proc-dev \
+        xsimd-dev \
         zlib-dev \
         zstd-dev && \
     rm -rf /var/cache/apk/* && \
@@ -98,6 +100,5 @@ ENV ARROW_ACERO=ON \
     ARROW_WITH_ZSTD=ON \
     AWSSDK_SOURCE=BUNDLED \
     google_cloud_cpp_storage_SOURCE=BUNDLED \
-    ORC_SOURCE=BUNDLED \
-    PATH=/usr/lib/ccache/:$PATH \
-    xsimd_SOURCE=BUNDLED
+    MUSL_LOCPATH=/usr/share/i18n/locales/musl \
+    PATH=/usr/lib/ccache/bin:$PATH

cpp/cmake_modules/ThirdpartyToolchain.cmake

@@ -1311,10 +1311,12 @@ if(ARROW_USE_BOOST)
     endif()
   endforeach()
 
+  set(BOOST_PROCESS_HAVE_V2 FALSE)
   if(TARGET Boost::process)
     # Boost >= 1.86
     target_compile_definitions(Boost::process INTERFACE "BOOST_PROCESS_HAVE_V1")
     target_compile_definitions(Boost::process INTERFACE "BOOST_PROCESS_HAVE_V2")
+    set(BOOST_PROCESS_HAVE_V2 TRUE)
   else()
     # Boost < 1.86
     add_library(Boost::process INTERFACE IMPORTED)
@@ -1329,6 +1331,7 @@ if(ARROW_USE_BOOST)
     endif()
     if(Boost_VERSION VERSION_GREATER_EQUAL 1.80)
       target_compile_definitions(Boost::process INTERFACE "BOOST_PROCESS_HAVE_V2")
+      set(BOOST_PROCESS_HAVE_V2 TRUE)
       # Boost < 1.86 has a bug that
       # boost::process::v2::process_environment::on_setup() isn't
       # defined. We need to build Boost Process source to define it.
@@ -1341,6 +1344,20 @@ if(ARROW_USE_BOOST)
       endif()
     endif()
   endif()
+  if(BOOST_PROCESS_HAVE_V2
+     AND # We can't use v2 API on Windows because v2 API doesn't support
+         # process group[1] and GCS testbench uses multiple processes[2].
+         #
+         # [1] https://github.com/boostorg/process/issues/259
+         # [2] https://github.com/googleapis/storage-testbench/issues/669
+         (NOT WIN32)
+     AND # We can't use v2 API with musl libc with Boost Process < 1.86
+         # because Boost Process < 1.86 doesn't support musl libc[3].
+         #
+         # [3] https://github.com/boostorg/process/commit/aea22dbf6be1695ceb42367590b6ca34d9433500
+         (NOT (ARROW_WITH_MUSL AND (Boost_VERSION VERSION_LESS 1.86))))
+    target_compile_definitions(Boost::process INTERFACE "BOOST_PROCESS_USE_V2")
+  endif()
 
   message(STATUS "Boost include dir: ${Boost_INCLUDE_DIRS}")
 endif()

cpp/src/arrow/compute/kernels/scalar_temporal_test.cc

@@ -2154,13 +2154,23 @@ TEST_F(ScalarTemporalTest, StrftimeOtherLocale) {
   auto options = StrftimeOptions("%d %B %Y %H:%M:%S", "fr_FR.UTF-8");
   const char* milliseconds = R"(
       ["1970-01-01T00:00:59.123", "2021-08-18T15:11:50.456", null])";
+#ifdef ARROW_WITH_MUSL
+  // musl-locales uses Capital case for month name.
+  // musl doesn't use "," for milliseconds separator.
+  const char* expected = R"(
+      ["01 Janvier 1970 00:00:59.123", "18 Août 2021 15:11:50.456", null])";
+#else
   const char* expected = R"(
       ["01 janvier 1970 00:00:59,123", "18 août 2021 15:11:50,456", null])";
+#endif
   CheckScalarUnary("strftime", timestamp(TimeUnit::MILLI, "UTC"), milliseconds, utf8(),
                    expected, &options);
 }
 
 TEST_F(ScalarTemporalTest, StrftimeInvalidLocale) {
+#ifdef ARROW_WITH_MUSL
+  GTEST_SKIP() << "musl doesn't report an error for invalid locale";
+#endif
 #ifdef __EMSCRIPTEN__
   GTEST_SKIP() << "Emscripten doesn't build with multiple locales as default";
 #endif

cpp/src/arrow/testing/process.cc

@@ -32,17 +32,6 @@
 // work if windows.h is already included.
 #  include <boost/asio/io_context.hpp>
 
-#  ifdef BOOST_PROCESS_HAVE_V2
-// We can't use v2 API on Windows because v2 API doesn't support
-// process group [1] and GCS testbench uses multiple processes [2].
-//
-// [1] https://github.com/boostorg/process/issues/259
-// [2] https://github.com/googleapis/storage-testbench/issues/669
-#    ifndef _WIN32
-#      define BOOST_PROCESS_USE_V2
-#    endif
-#  endif
-
 #  ifdef BOOST_PROCESS_USE_V2
 #    ifdef BOOST_PROCESS_NEED_SOURCE
 // Workaround for https://github.com/boostorg/process/issues/312

docker-compose.yml

@@ -217,7 +217,7 @@ services:
     #   docker compose build alpine-linux-cpp
     #   docker compose run --rm alpine-linux-cpp
     # Parameters:
-    #   ALPINE_LINUX: 3.16
+    #   ALPINE_LINUX: 3.18
     #   ARCH: amd64, arm64v8, ...
     image: ${REPO}:${ARCH}-alpine-linux-${ALPINE_LINUX}-cpp
     build: