Support for timestamps

Signed-off-by: Julien Jerphanion <[email protected]>

.github/workflows/osx.yml

@@ -18,7 +18,7 @@ jobs:
       fail-fast: false
       matrix:
         os:
-        - 13
+        - 14
         config:
         - { name: Debug }
         - { name: Release }

CMakeLists.txt

@@ -43,6 +43,22 @@ message(STATUS "Building sparrow v${${PROJECT_NAME}_VERSION}")
 
 OPTION(BUILD_TESTS "sparrow test suite" OFF)
 
+include(CheckCXXSymbolExists)
+
+if(cxx_std_20 IN_LIST CMAKE_CXX_COMPILE_FEATURES)
+  set(header version)
+else()
+  set(header ciso646)
+endif()
+
+check_cxx_symbol_exists(_LIBCPP_VERSION ${header} LIBCPP)
+if(LIBCPP)
+  message(STATUS "Using libc++")
+  # Allow the use of not visible yet availabile features, such
+  # as some formatter for new types.
+  add_compile_definitions(_LIBCPP_DISABLE_AVAILABILITY)
+endif()
+
 # Linter options
 # =============
 
@@ -55,6 +71,16 @@ endif()
 # Dependencies
 # ============
 
+set(SPARROW_INTERFACE_DEPENDENCIES "" CACHE STRING "List of dependencies to be linked to the sparrow target")
+
+# libc++ does not provide all the elements of the C++20 standard library,
+# so we need to use `HowardHinnant/date`.
+# See: https://libcxx.llvm.org/Status/Cxx20.html#note-p0355
+if (LIBCPP)
+    find_package(date CONFIG REQUIRED)
+    list(APPEND SPARROW_INTERFACE_DEPENDENCIES date::date)
+endif()
+
 # Build
 # =====
 
@@ -65,7 +91,7 @@ set(SPARROW_HEADERS
     ${SPARROW_INCLUDE_DIR}/sparrow/buffer.hpp
     ${SPARROW_INCLUDE_DIR}/sparrow/buffer_view.hpp
     ${SPARROW_INCLUDE_DIR}/sparrow/config.hpp
-	${SPARROW_INCLUDE_DIR}/sparrow/contracts.hpp
+    ${SPARROW_INCLUDE_DIR}/sparrow/contracts.hpp
     ${SPARROW_INCLUDE_DIR}/sparrow/data_traits.hpp
     ${SPARROW_INCLUDE_DIR}/sparrow/data_type.hpp
     ${SPARROW_INCLUDE_DIR}/sparrow/dynamic_bitset.hpp
@@ -86,6 +112,8 @@ target_include_directories(sparrow INTERFACE
     $<BUILD_INTERFACE:${SPARROW_INCLUDE_DIR}>
     $<INSTALL_INTERFACE:include>)
 
+target_link_libraries(sparrow INTERFACE ${SPARROW_INTERFACE_DEPENDENCIES})
+
 # We do not use non-standard C++
 set_target_properties(sparrow PROPERTIES CMAKE_CXX_EXTENSIONS OFF)
 target_compile_features(sparrow INTERFACE cxx_std_20)

environment-dev.yml

@@ -5,5 +5,7 @@ dependencies:
   # Build
   - cmake
   - ninja
+  # Alternative of `std::chrono` introduced in C++20.
+  - howardhinnant_date
   # Tests
   - doctest

include/sparrow/data_traits.hpp

@@ -125,6 +125,12 @@ namespace sparrow
         using default_layout = variable_size_binary_layout<value_type, std::span<byte_t>, const std::span<byte_t>>;  // FIXME: this is incorrect, change when we have the right types
     };
 
+    template <>
+    struct arrow_traits<timestamp> : common_native_types_traits<timestamp>
+    {
+        static constexpr data_type type_id = data_type::TIMESTAMP;
+    };
+
     namespace predicate
     {
 
@@ -147,4 +153,4 @@ namespace sparrow
         } constexpr has_arrow_traits;
     }
 
-}
+}

include/sparrow/data_type.hpp

@@ -14,6 +14,15 @@
 
 #pragma once
 
+#if __cpp_lib_chrono <= 201907L
+    // P0355R7 (Extending chrono to Calendars and Time Zones) has not been entirely implemented in libc++ yet.
+    // See: https://libcxx.llvm.org/Status/Cxx20.html#note-p0355
+    // For now, we use HowardHinnant/date as a replacement if we are compiling with libc++.
+    // TODO: remove this once libc++ has full support for P0355R7.
+#   include <chrono>
+#else
+#   include <date/tz.h>
+#endif
 #include <climits>
 #include <cstdint>
 #include <optional>
@@ -45,6 +54,12 @@ namespace sparrow
     using float64_t = std::float64_t;
 #endif
 
+#if __cpp_lib_chrono <= 201907L
+    using timestamp = std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds>;
+#else
+    using timestamp = date::zoned_time<std::chrono::nanoseconds>;
+#endif
+
     // We need to be sure the current target platform is setup to support correctly these types.
     static_assert(sizeof(float16_t) == 2);
     static_assert(sizeof(float32_t) == 4);
@@ -82,14 +97,15 @@ namespace sparrow
         BINARY,
         // Fixed-size binary. Each value occupies the same number of bytes
         FIXED_SIZE_BINARY,
+        // Number of nanoseconds since the UNIX epoch with an optional timezone.
+        // See: https://arrow.apache.org/docs/python/timestamps.html#timestamps
+        TIMESTAMP,
     };
 
     /// C++ types value representation types matching Arrow types.
     // NOTE: this needs to be in sync-order with `data_type`
     using all_base_types_t = mpl::typelist<
-        std::nullopt_t  // REVIEW: not sure about if we need to have this one? for representing NA? is this
-                        // the right type?
-        ,
+        std::nullopt_t,
         bool,
         std::uint8_t,
         std::int8_t,
@@ -103,7 +119,8 @@ namespace sparrow
         float32_t,
         float64_t,
         std::string,
-        std::vector<byte_t>
+        std::vector<byte_t>,
+        sparrow::timestamp
         // TODO: add missing fundamental types here
         >;