Fix R's tzdb

Author: rok

c_glib/test/test-assume-timezone-options.rb

@@ -45,6 +45,7 @@ def test_nonexistent_property
   end
 
   def test_assume_timezone_function
+    omit("std::chrono not available on Windows MinGW") if Gem.win_platform?
     args = [
       Arrow::ArrayDatum.new(build_timestamp_array(:milli, [1504953190000])),
     ]

c_glib/test/test-day-of-week-options.rb

@@ -39,6 +39,7 @@ def test_week_start_property
   end
 
   def test_day_of_week_function_with_count_from_zero_false
+    omit("std::chrono not available on Windows MinGW") if Gem.win_platform?
     args = [
       # 2017-09-09T10:33:10Z (Saturday)
       Arrow::ArrayDatum.new(build_timestamp_array(:milli, [1504953190000])),
@@ -50,6 +51,7 @@ def test_day_of_week_function_with_count_from_zero_false
   end
 
   def test_day_of_week_function_with_week_start
+    omit("std::chrono not available on Windows MinGW") if Gem.win_platform?
     args = [
       # 2017-09-09T10:33:10Z (Saturday)
       Arrow::ArrayDatum.new(build_timestamp_array(:milli, [1504953190000])),

c_glib/test/test-strftime-options.rb

@@ -35,6 +35,7 @@ def test_locale_property
   end
 
   def test_strftime_function
+    omit("std::chrono not available on Windows MinGW") if Gem.win_platform?
     args = [
       Arrow::ArrayDatum.new(build_timestamp_array(:milli, [1504953190854])),
     ]

cpp/src/arrow/config.cc

@@ -78,4 +78,18 @@ RuntimeInfo GetRuntimeInfo() {
   return info;
 }
 
+Status Initialize(const GlobalOptions& options) noexcept {
+  if (options.timezone_db_path.has_value()) {
+#if !USE_OS_TZDB
+    try {
+      arrow_vendored::date::set_install(options.timezone_db_path.value());
+      arrow_vendored::date::reload_tzdb();
+    } catch (const std::runtime_error& e) {
+      return Status::IOError(e.what());
+    }
+#endif
+  }
+  return Status::OK();
+}
+
 }  // namespace arrow

cpp/src/arrow/config.h

@@ -79,4 +79,13 @@ const BuildInfo& GetBuildInfo();
 ARROW_EXPORT
 RuntimeInfo GetRuntimeInfo();
 
+struct GlobalOptions {
+  /// Path to text timezone database. This is only used on Windows MinGW
+  /// builds where std::chrono timezone support is not available.
+  std::optional<std::string> timezone_db_path;
+};
+
+ARROW_EXPORT
+Status Initialize(const GlobalOptions& options) noexcept;
+
 }  // namespace arrow

cpp/src/arrow/util/chrono_internal.h

@@ -33,17 +33,18 @@
 #include <string_view>
 
 // Feature detection for C++20 chrono timezone support
-// We only enable for compilers with FULL support (not partial)
 // https://en.cppreference.com/w/cpp/compiler_support/20.html#cpp_lib_chrono_201907L
 //
+// On Windows with MSVC: std::chrono uses Windows' internal timezone database,
+// eliminating the need for users to install IANA tzdata separately.
+//
+// On Windows with MinGW/GCC: libstdc++ reads tzdata files via TZDIR env var.
+// The tzdata files must be provided (e.g., via the tzdb R package).
+//
 // On non-Windows: GCC libstdc++ has a bug where DST state is incorrectly reset when
 // a timezone transitions between rule sets (e.g., Australia/Broken_Hill around
 // 2000-02-29). Until this is fixed, we use the vendored date.h library.
 // See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116110
-//
-// On Windows: Use std::chrono which accesses Windows' internal timezone database,
-// eliminating the need for users to install IANA tzdata separately. We tolerate
-// the GCC bug here since Windows users are less likely to be using GCC.
 
 #if defined(_WIN32) && defined(__cpp_lib_chrono) && __cpp_lib_chrono >= 201907L
 #  define ARROW_USE_STD_CHRONO 1

r/R/arrow-package.R

@@ -152,6 +152,9 @@ s3_finalizer <- new.env(parent = emptyenv())
     # Disable multithreading on Windows
     # See https://issues.apache.org/jira/browse/ARROW-8379
     options(arrow.use_threads = FALSE)
+
+    # Try to set timezone database for MinGW builds
+    configure_tzdb()
   }
 
   # Set interrupt handlers
@@ -168,6 +171,22 @@ s3_finalizer <- new.env(parent = emptyenv())
   invisible()
 }
 
+configure_tzdb <- function() {
+  # This is needed on Windows MinGW builds where std::chrono timezone support
+  # is not available (older GCC versions). The tzdb R package provides the
+  # IANA timezone database.
+  if (requireNamespace("tzdb", quietly = TRUE)) {
+    tzdb::tzdb_initialize()
+    set_timezone_database(tzdb::tzdb_path("text"))
+  } else {
+    msg <- paste(
+      "The tzdb package is not installed.",
+      "Timezones will not be available to Arrow compute functions."
+    )
+    packageStartupMessage(msg)
+  }
+}
+
 .onAttach <- function(libname, pkgname) {
   # Just to be extra safe, let's wrap this in a try();
   # we don't want a failed startup message to prevent the package from loading

r/R/arrowExports.R

@@ -17,6 +17,8 @@
 
 #include "./arrow_types.h"
 
+#include <optional>
+
 #include <arrow/config.h>
 
 // [[arrow::export]]
@@ -31,3 +33,15 @@ std::vector<std::string> runtime_info() {
   auto info = arrow::GetRuntimeInfo();
   return {info.simd_level, info.detected_simd_level};
 }
+
+// [[arrow::export]]
+void set_timezone_database(cpp11::strings path) {
+  auto paths = cpp11::as_cpp<std::vector<std::string>>(path);
+  if (path.size() != 1) {
+    cpp11::stop("Must provide a single path to the timezone database.");
+  }
+
+  arrow::GlobalOptions options;
+  options.timezone_db_path = std::make_optional(paths[0]);
+  arrow::StopIfNotOk(arrow::Initialize(options));
+}