1
0
mirror of https://github.com/osmarks/autobotrobot synced 2025-01-09 15:10:27 +00:00

refactor time unit mappings

This commit is contained in:
osmarks 2020-10-30 14:59:37 +00:00
parent 2da79c6dfe
commit 41c1636398

View File

@ -23,16 +23,40 @@ prefixes = {
} }
number = "(-?[0-9]+(?:\.[0-9]+)?)(" + "|".join(prefixes.keys()) + ")?" number = "(-?[0-9]+(?:\.[0-9]+)?)(" + "|".join(prefixes.keys()) + ")?"
# from here: https://github.com/Rapptz/RoboDanny/blob/18b92ae2f53927aedebc25fb5eca02c8f6d7a874/cogs/utils/time.py time_units = (
short_timedelta_regex = re.compile(f""" ("galacticyears", "cosmicyears", "gy", "[Cc]y"),
(?:(?P<years>{number})(?:years?|y))? # e.g. 2y ("years", "y"),
(?:(?P<months>{number})(?:months?|mo))? # e.g. 2months ("beelifespans", "🐝", "bees?"),
(?:(?P<fortnights>{number})(?:fortnights?|fn|f))? # e.g. 10fn ("months", "mo"),
(?:(?P<weeks>{number})(?:weeks?|w))? # e.g. 10w ("semesters",),
(?:(?P<days>{number})(?:days?|d))? # e.g. 14d ("fortnights", "ft?n?"),
(?:(?P<hours>{number})(?:hours?|h))? # e.g. 12h ("weeks", "w"),
(?:(?P<minutes>{number})(?:minutes?|m))? # e.g. 10m ("days", "d"),
(?:(?P<seconds>{number})(?:seconds?|s))? # e.g. 15s """, re.VERBOSE) ("hours", "h"),
# Wikipedia tells me this is a traditional Chinese timekeeping unit
("ke",),
("minutes", "m"),
("seconds", "s")
)
tu_mappings = {
# dateutil dislikes fractional years, but this is 250My
"galacticyears": (7.8892315e15, "seconds"),
# apparently the average lifespan of a Western honey bee - I'm not very sure whether this is workers/drones/queens or what so TODO
"beelifespans": lambda: (random.randint(122, 152), "days"),
"semesters": (18, "weeks"),
"fortnights": (2, "weeks"),
"ke": (864, "seconds")
}
def rpartfor(u):
if u[0][-1] == "s":
l = [u[0] + "?"]
l.extend(u[1:])
else: l = u
return f"(?:(?P<{u[0]}>{number})(?:{'|'.join(l)}))?"
short_timedelta_regex = re.compile("\n".join(map(rpartfor, time_units)), re.VERBOSE)
def parse_prefixed(s): def parse_prefixed(s):
match = re.match(number, s) match = re.match(number, s)
@ -46,18 +70,21 @@ def parse_short_timedelta(text):
match = short_timedelta_regex.fullmatch(text) match = short_timedelta_regex.fullmatch(text)
if match is None or not match.group(0): raise ValueError("parse failed") if match is None or not match.group(0): raise ValueError("parse failed")
data = { k: parse_prefixed(v) if v else 0 for k, v in match.groupdict().items() } data = { k: parse_prefixed(v) if v else 0 for k, v in match.groupdict().items() }
data["weeks"] += data["fortnights"] * 2 for tu, mapping in tu_mappings.items():
del data["fortnights"] if callable(mapping): mapping = mapping()
qty, resunit = mapping
data[resunit] += qty * data[tu]
del data[tu]
return datetime.datetime.now(tz=datetime.timezone.utc) + relativedelta(**data) return datetime.datetime.now(tz=datetime.timezone.utc) + relativedelta(**data)
cal = parsedatetime.Calendar() cal = parsedatetime.Calendar()
def parse_humantime(text): def parse_humantime(text):
dt_tuple = cal.nlp(text) dt_tuple = cal.nlp(text)
if dt_tuple: return dt_tuple[0][0] if dt_tuple: return dt_tuple[0][0].replace(tzinfo=datetime.timezone.utc)
else: raise ValueError("parse failed") else: raise ValueError("parse failed")
def parse_time(text): def parse_time(text):
try: return datetime.datetime.strptime(text, "%d/%m/%Y") try: return datetime.datetime.strptime(text, "%d/%m/%Y").replace(tzinfo=datetime.timezone.utc)
except: pass except: pass
try: return parse_short_timedelta(text) try: return parse_short_timedelta(text)
except: pass except: pass