Add support for zzzz (and beyond) in format_datetime (#11330)

Summary:
Pull Request resolved: #11330

This diff adds support for JODA's zzzz (or more) patterns (all equivalent) in Presto's
forma_datetime function.

This is used to format long time zone names.

Long time zone names are not available from the IANA time zone database, so we
can't use the tz library to generate these.  Fortunately, unicode provides some
utilities to generate these.

Differential Revision: D64795407

CMakeLists.txt

@@ -472,6 +472,18 @@ add_compile_definitions(FOLLY_HAVE_INT128_T=1)
 set_source(folly)
 resolve_dependency(folly)
 
+if(CMAKE_SYSTEM_NAME MATCHES "Darwin")
+  if(ON_APPLE_M1)
+    list(APPEND CMAKE_PREFIX_PATH "/opt/homebrew/opt/icu4c")
+  else()
+    list(APPEND CMAKE_PREFIX_PATH "/usr/local/opt/icu4c")
+  endif()
+endif()
+
+# ICU is only needed with Boost build from source
+set_source(ICU)
+resolve_dependency(ICU COMPONENTS i18n uc)
+
 if(${VELOX_BUILD_TESTING})
   # Spark qury runner depends on absl, gRPC.
   set_source(absl)

velox/functions/lib/CMakeLists.txt

@@ -24,7 +24,8 @@ velox_link_libraries(velox_functions_util velox_vector velox_common_base)
 velox_add_library(velox_functions_lib_date_time_formatter DateTimeFormatter.cpp
                   DateTimeFormatterBuilder.cpp)
 
-velox_link_libraries(velox_functions_lib_date_time_formatter velox_type_tz)
+velox_link_libraries(velox_functions_lib_date_time_formatter velox_type_tz
+                     ICU::i18n ICU::uc)
 
 velox_add_library(
   velox_functions_lib

velox/functions/lib/DateTimeFormatter.cpp

@@ -1173,10 +1173,10 @@ uint32_t DateTimeFormatter::maxResultSize(const tz::TimeZone* timezone) const {
           // https://en.wikipedia.org/wiki/List_of_tz_database_time_zones
           size += 5;
         } else {
-          VELOX_NYI(
-              "Date format specifier is not yet implemented: {} ({})",
-              getSpecifierName(token.pattern.specifier),
-              token.pattern.minRepresentDigits);
+          // The longest time zone long name is 40, Australian Central Western
+          // Standard Time.
+          // https://www.timeanddate.com/time/zones/
+          size += 50;
         }
 
         break;
@@ -1432,8 +1432,11 @@ int32_t DateTimeFormatter::format(
             std::memcpy(result, abbrev.data(), abbrev.length());
             result += abbrev.length();
           } else {
-            // TODO: implement full name time zone
-            VELOX_NYI("full time zone name is not yet supported");
+            std::string longName = timezone->getLongName(
+                std::chrono::milliseconds(timestamp.toMillis()),
+                tz::TimeZone::TChoose::kEarliest);
+            std::memcpy(result, longName.data(), longName.length());
+            result += longName.length();
           }
         } break;
 

velox/functions/prestosql/tests/DateTimeFunctionsTest.cpp

@@ -3272,6 +3272,12 @@ TEST_F(DateTimeFunctionsTest, formatDateTime) {
   EXPECT_EQ("IST", formatDatetime(parseTimestamp("1970-01-01"), "zzz"));
   EXPECT_EQ("IST", formatDatetime(parseTimestamp("1970-01-01"), "zz"));
   EXPECT_EQ("IST", formatDatetime(parseTimestamp("1970-01-01"), "z"));
+  EXPECT_EQ(
+      "India Standard Time",
+      formatDatetime(parseTimestamp("1970-01-01"), "zzzz"));
+  EXPECT_EQ(
+      "India Standard Time",
+      formatDatetime(parseTimestamp("1970-01-01"), "zzzzzzzzzzzzzzzzzzzzzz"));
 
   // Test daylight savings.
   setQueryTimeZone("America/Los_Angeles");
@@ -3281,16 +3287,47 @@ TEST_F(DateTimeFunctionsTest, formatDateTime) {
   EXPECT_EQ("PDT", formatDatetime(parseTimestamp("2024-03-10 03:00"), "z"));
   EXPECT_EQ("PDT", formatDatetime(parseTimestamp("2024-11-03 01:00"), "z"));
   EXPECT_EQ("PST", formatDatetime(parseTimestamp("2024-11-03 02:00"), "z"));
+  EXPECT_EQ(
+      "Pacific Standard Time",
+      formatDatetime(parseTimestamp("1970-01-01"), "zzzz"));
+  EXPECT_EQ(
+      "Pacific Daylight Time",
+      formatDatetime(parseTimestamp("1970-10-01"), "zzzz"));
+  EXPECT_EQ(
+      "Pacific Standard Time",
+      formatDatetime(parseTimestamp("2024-03-10 01:00"), "zzzz"));
+  EXPECT_EQ(
+      "Pacific Daylight Time",
+      formatDatetime(parseTimestamp("2024-03-10 03:00"), "zzzz"));
+  EXPECT_EQ(
+      "Pacific Daylight Time",
+      formatDatetime(parseTimestamp("2024-11-03 01:00"), "zzzz"));
+  EXPECT_EQ(
+      "Pacific Standard Time",
+      formatDatetime(parseTimestamp("2024-11-03 02:00"), "zzzz"));
+
+  // Test ambiguous time.
+  EXPECT_EQ(
+      "PDT", formatDatetime(parseTimestamp("2024-11-03 01:30:00"), "zzz"));
+  EXPECT_EQ(
+      "Pacific Daylight Time",
+      formatDatetime(parseTimestamp("2024-11-03 01:30:00"), "zzzz"));
 
   // Test a long abbreviation.
   setQueryTimeZone("Asia/Colombo");
   EXPECT_EQ("+0530", formatDatetime(parseTimestamp("1970-10-01"), "z"));
+  EXPECT_EQ(
+      "India Standard Time",
+      formatDatetime(parseTimestamp("1970-10-01"), "zzzz"));
 
-  setQueryTimeZone("Asia/Kolkata");
-  // We don't support more than 3 'z's yet.
-  EXPECT_THROW(
-      formatDatetime(parseTimestamp("1970-01-01"), "zzzz"), VeloxRuntimeError);
+  // Test a long long name.
+  setQueryTimeZone("Australia/Eucla");
+  EXPECT_EQ("+0845", formatDatetime(parseTimestamp("1970-10-01"), "z"));
+  EXPECT_EQ(
+      "Australian Central Western Standard Time",
+      formatDatetime(parseTimestamp("1970-10-01"), "zzzz"));
 
+  setQueryTimeZone("Asia/Kolkata");
   // Literal test cases.
   EXPECT_EQ("hello", formatDatetime(parseTimestamp("1970-01-01"), "'hello'"));
   EXPECT_EQ("'", formatDatetime(parseTimestamp("1970-01-01"), "''"));

velox/type/tz/CMakeLists.txt

@@ -23,4 +23,6 @@ velox_link_libraries(
   velox_external_date
   Boost::regex
   fmt::fmt
-  Folly::folly)
+  Folly::folly
+  ICU::i18n
+  ICU::uc)

velox/type/tz/TimeZoneMap.h

@@ -16,6 +16,15 @@
 
 #pragma once
 
+#include <unicode/locid.h>
+#include <unicode/timezone.h>
+#include <unicode/unistr.h>
+
+// The ICU libraries define TRUE/FALSE macros which frequently conflict with
+// other libraries that use these as enum/variable names.
+#undef TRUE
+#undef FALSE
+
 #include <chrono>
 #include <string>
 #include "velox/common/base/Exceptions.h"
@@ -178,6 +187,37 @@ class TimeZone {
     return getZonedTime(timePoint, choose).get_info().abbrev;
   }
 
+  template <typename TDuration>
+  std::string getLongName(TDuration timestamp, TChoose choose = TChoose::kFail)
+      const {
+    static const icu::Locale locale("en", "US");
+
+    // Get the ICU TimeZone by name
+    std::unique_ptr<icu::TimeZone> tz(icu::TimeZone::createTimeZone(
+        icu::UnicodeString(timeZoneName_.data(), timeZoneName_.length())));
+    VELOX_USER_CHECK_NOT_NULL(tz);
+
+    // According to the documentation this is how to determine if DST applies to
+    // a given timestamp in a given time zone.
+    // https://howardhinnant.github.io/date/tz.html#sys_info
+    date::local_time<TDuration> timePoint{timestamp};
+    bool isDst = getZonedTime(timePoint, choose).get_info().save !=
+        std::chrono::minutes(0);
+
+    // Construct the long name for the time zone.
+    // Note that ICU does not have DST information for many time zones prior to
+    // 1970, so it's important to specify it explicitly.
+    icu::UnicodeString longName;
+    tz->getDisplayName(
+        isDst, icu_53::TimeZone::EDisplayType::LONG, locale, longName);
+
+    // Convert the UnicodeString back to a string and write it out
+    std::string longNameStr;
+    longName.toUTF8String(longNameStr);
+
+    return longNameStr;
+  }
+
  private:
   template <typename TDuration>
   date::zoned_time<TDuration> getZonedTime(