Support ISO time intervals

Previously, the only way an interval could be specified was through an
ambiguous date, which has limited flexibility. Conveniently, the
aniso8601 library provides a function to parse the various ISO interval
formats¹.

Note that this supports time information where all other date format
support and logic is based on date without time information. This is
fine because it is trivial to ignore the time information and have
everything work as expected. It seems far-fetched for now, but maybe
someday we will end up using time information, since date is inherently
ambiguous without at least a time zone.

¹ <https://en.wikipedia.org/wiki/ISO_8601#Time_intervals>

Author: victorlin

Diff for: augur/dates/__init__.py

@@ -1,3 +1,4 @@
+import aniso8601
 import argparse
 import datetime
 from textwrap import dedent
@@ -121,6 +122,25 @@ def get_numerical_date_from_value(value, fmt=None, min_max_year=None):
         except InvalidDate as error:
             raise AugurError(str(error)) from error
         return [treetime.utils.numeric_date(d) for d in ambig_date]
+    if '/' in value:
+        # ISO interval
+        try:
+            start, end = aniso8601.parse_interval(value)
+        except ValueError as e:
+            raise AugurError(f"Invalid date {value!r}: {e}") from e
+
+        # ignore time parts - we don't use this info, and dropping it allows for comparisons
+        if isinstance(start, datetime.datetime):
+            start = start.date()
+        if isinstance(end, datetime.datetime):
+            end = start.date()
+
+        # sort because aniso8601.parse_interval can return dates out of order
+        # https://bitbucket.org/nielsenb/aniso8601/src/a4f767de4429b246356b7a3bc98b24c68ce49477/aniso8601/interval.py#lines-236:240
+        if start > end:
+            start, end = end, start
+
+        return list(map(treetime.utils.numeric_date, (start, end)))
     try:
         return treetime.utils.numeric_date(datetime.datetime.strptime(value, fmt))
     except:

Diff for: setup.py

@@ -51,6 +51,7 @@
     package_data = {'augur': ['data/*']},
     python_requires = '>={}'.format('.'.join(str(n) for n in py_min_version)),
     install_requires = [
+        "aniso8601 >=10.0.0, ==10.*",
         "bcbio-gff >=0.7.1, ==0.7.*",
         # TODO: Remove biopython >= 1.80 pin if it is added to bcbio-gff: https://github.com/chapmanb/bcbb/issues/142
         "biopython >=1.80, ==1.*",

Diff for: tests/dates/test_dates.py

@@ -60,6 +60,55 @@ def test_get_numerical_date_from_value_current_day_limit(self):
             == pytest.approx(2000.138, abs=1e-3)
         )
 
+    def test_get_numerical_date_from_value_interval(self):
+        # Valid ISO dates form an interval.
+        assert dates.get_numerical_date_from_value("2019-01-02/2019-03-04") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=1, day=2)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=3, day=4)), abs=1e-3),
+        ]
+
+        # Time parts are valid but ignored.
+        assert dates.get_numerical_date_from_value("2019-01-02T13:00:00Z/2019-03-04") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=1, day=2)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=3, day=4)), abs=1e-3),
+        ]
+
+        # Shorthands are valid.
+        assert dates.get_numerical_date_from_value("2019-01-02/03-04") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=1, day=2)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=3, day=4)), abs=1e-3),
+        ]
+
+        # Reduced precision dates are not supported.
+        with pytest.raises(AugurError):
+            dates.get_numerical_date_from_value("2019-01/2019-03")
+
+        # This reduced precision date works.
+        assert dates.get_numerical_date_from_value("2019-01/2019-03-04") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=1, day=1)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=3, day=4)), abs=1e-3),
+        ]
+
+        # This one does not.
+        with pytest.raises(AugurError):
+            dates.get_numerical_date_from_value("2019-01-02/2019-03")
+
+        # Start and duration is valid.
+        assert dates.get_numerical_date_from_value("2019-01-02/P1M") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=1, day=2)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=2, day=1)), abs=1e-3),
+        ]
+
+        # Duration and end is valid.
+        assert dates.get_numerical_date_from_value("P1M/2019-03-04") == [
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=2, day=2)), abs=1e-3),
+            pytest.approx(dates.numeric_date(datetime.date(year=2019, month=3, day=4)), abs=1e-3),
+        ]
+
+        # Numerical dates are not valid.
+        with pytest.raises(AugurError):
+            dates.get_numerical_date_from_value("2019.0/2019-06-01")
+
     def test_is_date_ambiguous(self):
         """is_date_ambiguous should return true for ambiguous dates and false for valid dates."""
         # Test complete date strings with ambiguous values.