Recognize localized AM/PM markers in parse_time()

babel/dates.py

@@ -1377,10 +1377,11 @@ def parse_time(
     # 12 AM and 12 PM are 00:00 and 12:00 respectively.
     hour = int(numbers[indexes['H']])
     if 'a' in format_str:
-        if 'pm' in string.lower():
+        period = _parse_period(string, Locale.parse(locale or LC_TIME))
+        if period == 'pm':
             if hour != 12:
                 hour += 12
-        elif 'am' in string.lower() and hour == 12:
+        elif period == 'am' and hour == 12:
             hour = 0
 
     # Parse up to three numbers from the string.
@@ -1392,6 +1393,35 @@ def parse_time(
     return datetime.time(hour, minute, second)
 
 
+def _parse_period(string: str, locale: Locale) -> Literal['am', 'pm'] | None:
+    """Find which AM/PM day period marker, if any, occurs in `string`.
+
+    The locale's own period names (e.g. ``오후`` for Korean or ``p.m.`` for
+    Spanish) are checked first, longest first so that a short name can't
+    shadow a longer one; the English ``am`` and ``pm`` are accepted as a
+    fallback.  Comparison ignores case and whitespace, since CLDR data uses
+    various non-breaking spaces inside period names.
+    """
+
+    def _normalize(value: str) -> str:
+        return ''.join(value.split()).casefold()
+
+    text = _normalize(string)
+    names = set()
+    for context in locale.day_periods.values():
+        for width in context.values():
+            for period in ('am', 'pm'):
+                if name := _normalize(width.get(period) or ''):
+                    names.add((name, period))
+    for name, period in sorted(names, key=lambda item: (-len(item[0]), item)):
+        if name in text:
+            return period
+    for period in ('pm', 'am'):
+        if period in text:
+            return period
+    return None
+
+
 class DateTimePattern:
     def __init__(self, pattern: str, format: DateTimeFormat):
         self.pattern = pattern

tests/test_dates.py

@@ -540,6 +540,29 @@ def test_parse_time(input, expected):
     assert dates.parse_time(input, locale='en_US') == expected
 
 
+def test_parse_time_localized_am_pm():
+    # The day period marker must be recognized in the locale's own language,
+    # not only as the English "am"/"pm".
+    assert dates.parse_time('오후 3:30', locale='ko_KR', format='short') == time(15, 30)
+    assert dates.parse_time('오전 12:05', locale='ko_KR', format='short') == time(0, 5)
+    assert dates.parse_time('3:30\u202fp.m.', locale='es_US', format='short') == time(15, 30)
+    assert dates.parse_time('12:30\u202fa.m.', locale='es_MX', format='short') == time(0, 30)
+    assert dates.parse_time('3:30\u202fμ.μ.', locale='el_GR', format='short') == time(15, 30)
+    assert dates.parse_time('3:30 م', locale='ar_EG', format='short') == time(15, 30)
+    assert dates.parse_time('3:30 ص', locale='ar_EG', format='short') == time(3, 30)
+    # CLDR names may contain non-breaking spaces; any whitespace is accepted
+    assert dates.parse_time('3:30 p. m.', locale='ca', format='h:mm a') == time(15, 30)
+    # English markers keep working, including with other locales
+    assert dates.parse_time('3:30 PM', locale='de_DE', format='h:mm a') == time(15, 30)
+
+    # Everything format_time() produces on a 12-hour clock parses back
+    for locale in ('en_US', 'ko_KR', 'es_US', 'el_GR', 'ar_EG', 'aa', 'sq', 'fa', 'ca'):
+        for hour in range(24):
+            value = time(hour, 7, 9)
+            formatted = dates.format_time(value, 'h:mm:ss a', locale=locale)
+            assert dates.parse_time(formatted, locale=locale, format='h:mm:ss a') == value, (locale, formatted)
+
+
 def test_parse_time_no_seconds_in_format():
     # parse time using a time format which does not include seconds
     locale = 'cs_CZ'