Fix DateTimeFormatter to set weekDateFormat only if it has week and day of week along with year. (facebookincubator#12456)

Summary:

This is a targeted fix that ensures that date_parse formats which contain %a but without additional specifiers like %v does not set weekDateFormat. An example of such a format is  '%d-%b-%Y (%a)' which when parsing a string like '11-Mar-2024 (Mon)' will incorrectly ignore the month and day and instead return a date like '2024-01-01' . 

There are likely other cases where additional or inconsistent specifiers are handled incorrectly. A full fix of DateTimeFormatter is beyond scope of this PR, but I will send out additional PR's in the future to address these issues. 

Mitigates S493660.

Reviewed By: pedroerp

Differential Revision: D70253624

Author: Krishna Pai

velox/functions/lib/DateTimeFormatter.cpp

@@ -60,6 +60,7 @@ struct Date {
   bool isYearOfEra = false; // Year of era cannot be zero or negative.
   bool hasYear = false; // Whether year was explicitly specified.
   bool hasDayOfWeek = false; // Whether dayOfWeek was explicitly specified.
+  bool hasWeek = false; // Whether week was explicitly specified.
 
   int32_t hour = 0;
   int32_t minute = 0;
@@ -870,9 +871,7 @@ int32_t parseFromPattern(
       return -1;
     }
     cur += size;
-    if (!date.weekOfMonthDateFormat) {
-      date.weekDateFormat = true;
-    }
+    date.hasDayOfWeek = true;
     date.dayOfYearFormat = false;
     if (!date.hasYear) {
       date.hasYear = true;
@@ -1095,6 +1094,7 @@ int32_t parseFromPattern(
           return -1;
         }
         date.year = number;
+        date.hasWeek = true;
         date.weekDateFormat = true;
         date.dayOfYearFormat = false;
         date.centuryFormat = false;
@@ -1107,6 +1107,7 @@ int32_t parseFromPattern(
           return -1;
         }
         date.week = number;
+        date.hasWeek = true;
         date.weekDateFormat = true;
         date.dayOfYearFormat = false;
         date.weekOfMonthDateFormat = false;
@@ -1637,6 +1638,10 @@ Expected<DateTimeResult> DateTimeFormatter::parse(
 
   // Convert the parsed date/time into a timestamp.
   Expected<int64_t> daysSinceEpoch;
+
+  // Ensure you use week date format only when you have year and at least week.
+  date.weekDateFormat = date.hasYear && date.hasWeek;
+
   if (date.weekDateFormat) {
     daysSinceEpoch =
         util::daysSinceEpochFromWeekDate(date.year, date.week, date.dayOfWeek);

velox/functions/prestosql/tests/DateTimeFunctionsTest.cpp

@@ -4530,6 +4530,12 @@ TEST_F(DateTimeFunctionsTest, dateParse) {
   EXPECT_EQ(
       Timestamp(-66600, 0), dateParse("1969-12-31+11:00", "%Y-%m-%d+%H:%i"));
 
+  setQueryTimeZone("America/Los_Angeles");
+  // Tests if it uses weekdateformat if %v not present but %a is present.
+  EXPECT_EQ(
+      Timestamp(1730707200, 0),
+      dateParse("04-Nov-2024 (Mon)", "%d-%b-%Y (%a)"));
+
   VELOX_ASSERT_THROW(dateParse("", "%y+"), "Invalid date format: ''");
   VELOX_ASSERT_THROW(dateParse("1", "%y+"), "Invalid date format: '1'");
   VELOX_ASSERT_THROW(dateParse("116", "%y+"), "Invalid date format: '116'");