fix: during first leap seconds of the day UTC date can be wrong

pynmea2/types/proprietary/tnl.py

@@ -189,11 +189,12 @@ class TNLEVT(TNL, DatetimeFix):
         0 	Talker ID $PTNL
         1 	Message ID EVT
         2 	Event time. UTC time of event in format hhmmss.ssssss
-        3 	Port number. Port event markers receiver: "1" or "2" (optional), if two ports are available.
+        3 	Port number. Port event markers receiver: "1" or "2" (optional),
+                if two ports are available.
         4 	NNNNNN. Incremental number of events on each independent port.
         5 	WWWW. Week number of event (since 06 January 1980).
         6 	Day of week. Days denoted 0 = Sunday…6 = Saturday.
-        7 	Leap second. UTC Leap Second offset from GPS time, Currently 18 seconds as of 07 July 2017.
+        7 	Leap second. UTC Leap Second offset from GPS time,
         8 	The checksum data, always begins with *
 
         Example message:
@@ -212,8 +213,15 @@ class TNLEVT(TNL, DatetimeFix):
 
     # We can derive the date from GPS Week Number and Day of the Week
     # Presumingly 1024 overflow was taken into accound by the GPS unit
+    # Taking into account that the GPS day could have overflown during
+    # leap seconds, if 0 <= UTC timestamp < leap seconds
     @property
     def datestamp(self):
         gps_epoch = datetime.date(year=1980, month=1, day=6)
-        return gps_epoch + datetime.timedelta(
-                weeks=self.gps_week_num, days=self.gps_day_num)
+        gps_week_day_delta = datetime.timedelta(weeks=self.gps_week_num, days=self.gps_day_num)
+        utc_time_delta = datetime.timedelta(
+                hours=self.timestamp.hour, minutes=self.timestamp.minute,
+                seconds=self.timestamp.second)
+        utc_leap_delta = datetime.timedelta(seconds=self.leap_secs)
+
+        return gps_epoch + gps_week_day_delta + utc_time_delta - utc_leap_delta