mirror of
https://github.com/ynput/ayon-core.git
synced 2026-01-01 16:34:53 +01:00
Python modules for Ftrack "api" and "action handler"
This commit is contained in:
parent
f6537932e3
commit
68c78deeda
515 changed files with 135019 additions and 21 deletions
344
pype/vendor/arrow/parser.py
vendored
Normal file
344
pype/vendor/arrow/parser.py
vendored
Normal file
|
|
@ -0,0 +1,344 @@
|
|||
# -*- coding: utf-8 -*-
|
||||
from __future__ import absolute_import
|
||||
from __future__ import unicode_literals
|
||||
|
||||
from datetime import datetime
|
||||
from dateutil import tz
|
||||
import re
|
||||
|
||||
try:
|
||||
from functools import lru_cache
|
||||
except ImportError: # pragma: no cover
|
||||
from backports.functools_lru_cache import lru_cache # pragma: no cover
|
||||
|
||||
from arrow import locales
|
||||
|
||||
|
||||
class ParserError(RuntimeError):
|
||||
pass
|
||||
|
||||
|
||||
class DateTimeParser(object):
|
||||
|
||||
_FORMAT_RE = re.compile('(YYY?Y?|MM?M?M?|Do|DD?D?D?|d?d?d?d|HH?|hh?|mm?|ss?|S+|ZZ?Z?|a|A|X)')
|
||||
_ESCAPE_RE = re.compile('\[[^\[\]]*\]')
|
||||
|
||||
_ONE_OR_MORE_DIGIT_RE = re.compile('\d+')
|
||||
_ONE_OR_TWO_DIGIT_RE = re.compile('\d{1,2}')
|
||||
_FOUR_DIGIT_RE = re.compile('\d{4}')
|
||||
_TWO_DIGIT_RE = re.compile('\d{2}')
|
||||
_TZ_RE = re.compile('[+\-]?\d{2}:?(\d{2})?')
|
||||
_TZ_NAME_RE = re.compile('\w[\w+\-/]+')
|
||||
|
||||
|
||||
_BASE_INPUT_RE_MAP = {
|
||||
'YYYY': _FOUR_DIGIT_RE,
|
||||
'YY': _TWO_DIGIT_RE,
|
||||
'MM': _TWO_DIGIT_RE,
|
||||
'M': _ONE_OR_TWO_DIGIT_RE,
|
||||
'DD': _TWO_DIGIT_RE,
|
||||
'D': _ONE_OR_TWO_DIGIT_RE,
|
||||
'HH': _TWO_DIGIT_RE,
|
||||
'H': _ONE_OR_TWO_DIGIT_RE,
|
||||
'hh': _TWO_DIGIT_RE,
|
||||
'h': _ONE_OR_TWO_DIGIT_RE,
|
||||
'mm': _TWO_DIGIT_RE,
|
||||
'm': _ONE_OR_TWO_DIGIT_RE,
|
||||
'ss': _TWO_DIGIT_RE,
|
||||
's': _ONE_OR_TWO_DIGIT_RE,
|
||||
'X': re.compile('\d+'),
|
||||
'ZZZ': _TZ_NAME_RE,
|
||||
'ZZ': _TZ_RE,
|
||||
'Z': _TZ_RE,
|
||||
'S': _ONE_OR_MORE_DIGIT_RE,
|
||||
}
|
||||
|
||||
MARKERS = ['YYYY', 'MM', 'DD']
|
||||
SEPARATORS = ['-', '/', '.']
|
||||
|
||||
def __init__(self, locale='en_us', cache_size=0):
|
||||
|
||||
self.locale = locales.get_locale(locale)
|
||||
self._input_re_map = self._BASE_INPUT_RE_MAP.copy()
|
||||
self._input_re_map.update({
|
||||
'MMMM': self._choice_re(self.locale.month_names[1:], re.IGNORECASE),
|
||||
'MMM': self._choice_re(self.locale.month_abbreviations[1:],
|
||||
re.IGNORECASE),
|
||||
'Do': re.compile(self.locale.ordinal_day_re),
|
||||
'dddd': self._choice_re(self.locale.day_names[1:], re.IGNORECASE),
|
||||
'ddd': self._choice_re(self.locale.day_abbreviations[1:],
|
||||
re.IGNORECASE),
|
||||
'd': re.compile(r"[1-7]"),
|
||||
'a': self._choice_re(
|
||||
(self.locale.meridians['am'], self.locale.meridians['pm'])
|
||||
),
|
||||
# note: 'A' token accepts both 'am/pm' and 'AM/PM' formats to
|
||||
# ensure backwards compatibility of this token
|
||||
'A': self._choice_re(self.locale.meridians.values())
|
||||
})
|
||||
if cache_size > 0:
|
||||
self._generate_pattern_re =\
|
||||
lru_cache(maxsize=cache_size)(self._generate_pattern_re)
|
||||
|
||||
def parse_iso(self, string):
|
||||
|
||||
has_time = 'T' in string or ' ' in string.strip()
|
||||
space_divider = ' ' in string.strip()
|
||||
|
||||
if has_time:
|
||||
if space_divider:
|
||||
date_string, time_string = string.split(' ', 1)
|
||||
else:
|
||||
date_string, time_string = string.split('T', 1)
|
||||
time_parts = re.split('[+-]', time_string, 1)
|
||||
has_tz = len(time_parts) > 1
|
||||
has_seconds = time_parts[0].count(':') > 1
|
||||
has_subseconds = re.search('[.,]', time_parts[0])
|
||||
|
||||
if has_subseconds:
|
||||
formats = ['YYYY-MM-DDTHH:mm:ss%sS' % has_subseconds.group()]
|
||||
elif has_seconds:
|
||||
formats = ['YYYY-MM-DDTHH:mm:ss']
|
||||
else:
|
||||
formats = ['YYYY-MM-DDTHH:mm']
|
||||
else:
|
||||
has_tz = False
|
||||
# generate required formats: YYYY-MM-DD, YYYY-MM-DD, YYYY
|
||||
# using various separators: -, /, .
|
||||
l = len(self.MARKERS)
|
||||
formats = [separator.join(self.MARKERS[:l-i])
|
||||
for i in range(l)
|
||||
for separator in self.SEPARATORS]
|
||||
|
||||
if has_time and has_tz:
|
||||
formats = [f + 'Z' for f in formats]
|
||||
|
||||
if space_divider:
|
||||
formats = [item.replace('T', ' ', 1) for item in formats]
|
||||
|
||||
return self._parse_multiformat(string, formats)
|
||||
|
||||
def _generate_pattern_re(self, fmt):
|
||||
|
||||
# fmt is a string of tokens like 'YYYY-MM-DD'
|
||||
# we construct a new string by replacing each
|
||||
# token by its pattern:
|
||||
# 'YYYY-MM-DD' -> '(?P<YYYY>\d{4})-(?P<MM>\d{2})-(?P<DD>\d{2})'
|
||||
tokens = []
|
||||
offset = 0
|
||||
|
||||
# Extract the bracketed expressions to be reinserted later.
|
||||
escaped_fmt = re.sub(self._ESCAPE_RE, "#", fmt)
|
||||
# Any number of S is the same as one.
|
||||
escaped_fmt = re.sub('S+', 'S', escaped_fmt)
|
||||
escaped_data = re.findall(self._ESCAPE_RE, fmt)
|
||||
|
||||
fmt_pattern = escaped_fmt
|
||||
|
||||
for m in self._FORMAT_RE.finditer(escaped_fmt):
|
||||
token = m.group(0)
|
||||
try:
|
||||
input_re = self._input_re_map[token]
|
||||
except KeyError:
|
||||
raise ParserError('Unrecognized token \'{0}\''.format(token))
|
||||
input_pattern = '(?P<{0}>{1})'.format(token, input_re.pattern)
|
||||
tokens.append(token)
|
||||
# a pattern doesn't have the same length as the token
|
||||
# it replaces! We keep the difference in the offset variable.
|
||||
# This works because the string is scanned left-to-right and matches
|
||||
# are returned in the order found by finditer.
|
||||
fmt_pattern = fmt_pattern[:m.start() + offset] + input_pattern + fmt_pattern[m.end() + offset:]
|
||||
offset += len(input_pattern) - (m.end() - m.start())
|
||||
|
||||
final_fmt_pattern = ""
|
||||
a = fmt_pattern.split("#")
|
||||
b = escaped_data
|
||||
|
||||
# Due to the way Python splits, 'a' will always be longer
|
||||
for i in range(len(a)):
|
||||
final_fmt_pattern += a[i]
|
||||
if i < len(b):
|
||||
final_fmt_pattern += b[i][1:-1]
|
||||
|
||||
return tokens, re.compile(final_fmt_pattern, flags=re.IGNORECASE)
|
||||
|
||||
def parse(self, string, fmt):
|
||||
|
||||
if isinstance(fmt, list):
|
||||
return self._parse_multiformat(string, fmt)
|
||||
|
||||
fmt_tokens, fmt_pattern_re = self._generate_pattern_re(fmt)
|
||||
|
||||
match = fmt_pattern_re.search(string)
|
||||
if match is None:
|
||||
raise ParserError('Failed to match \'{0}\' when parsing \'{1}\''
|
||||
.format(fmt_pattern_re.pattern, string))
|
||||
parts = {}
|
||||
for token in fmt_tokens:
|
||||
if token == 'Do':
|
||||
value = match.group('value')
|
||||
else:
|
||||
value = match.group(token)
|
||||
self._parse_token(token, value, parts)
|
||||
return self._build_datetime(parts)
|
||||
|
||||
def _parse_token(self, token, value, parts):
|
||||
|
||||
if token == 'YYYY':
|
||||
parts['year'] = int(value)
|
||||
elif token == 'YY':
|
||||
value = int(value)
|
||||
parts['year'] = 1900 + value if value > 68 else 2000 + value
|
||||
|
||||
elif token in ['MMMM', 'MMM']:
|
||||
parts['month'] = self.locale.month_number(value.lower())
|
||||
|
||||
elif token in ['MM', 'M']:
|
||||
parts['month'] = int(value)
|
||||
|
||||
elif token in ['DD', 'D']:
|
||||
parts['day'] = int(value)
|
||||
|
||||
elif token in ['Do']:
|
||||
parts['day'] = int(value)
|
||||
|
||||
elif token.upper() in ['HH', 'H']:
|
||||
parts['hour'] = int(value)
|
||||
|
||||
elif token in ['mm', 'm']:
|
||||
parts['minute'] = int(value)
|
||||
|
||||
elif token in ['ss', 's']:
|
||||
parts['second'] = int(value)
|
||||
|
||||
elif token == 'S':
|
||||
# We have the *most significant* digits of an arbitrary-precision integer.
|
||||
# We want the six most significant digits as an integer, rounded.
|
||||
# FIXME: add nanosecond support somehow?
|
||||
value = value.ljust(7, str('0'))
|
||||
|
||||
# floating-point (IEEE-754) defaults to half-to-even rounding
|
||||
seventh_digit = int(value[6])
|
||||
if seventh_digit == 5:
|
||||
rounding = int(value[5]) % 2
|
||||
elif seventh_digit > 5:
|
||||
rounding = 1
|
||||
else:
|
||||
rounding = 0
|
||||
|
||||
parts['microsecond'] = int(value[:6]) + rounding
|
||||
|
||||
elif token == 'X':
|
||||
parts['timestamp'] = int(value)
|
||||
|
||||
elif token in ['ZZZ', 'ZZ', 'Z']:
|
||||
parts['tzinfo'] = TzinfoParser.parse(value)
|
||||
|
||||
elif token in ['a', 'A']:
|
||||
if value in (
|
||||
self.locale.meridians['am'],
|
||||
self.locale.meridians['AM']
|
||||
):
|
||||
parts['am_pm'] = 'am'
|
||||
elif value in (
|
||||
self.locale.meridians['pm'],
|
||||
self.locale.meridians['PM']
|
||||
):
|
||||
parts['am_pm'] = 'pm'
|
||||
|
||||
@staticmethod
|
||||
def _build_datetime(parts):
|
||||
|
||||
timestamp = parts.get('timestamp')
|
||||
|
||||
if timestamp:
|
||||
tz_utc = tz.tzutc()
|
||||
return datetime.fromtimestamp(timestamp, tz=tz_utc)
|
||||
|
||||
am_pm = parts.get('am_pm')
|
||||
hour = parts.get('hour', 0)
|
||||
|
||||
if am_pm == 'pm' and hour < 12:
|
||||
hour += 12
|
||||
elif am_pm == 'am' and hour == 12:
|
||||
hour = 0
|
||||
|
||||
return datetime(year=parts.get('year', 1), month=parts.get('month', 1),
|
||||
day=parts.get('day', 1), hour=hour, minute=parts.get('minute', 0),
|
||||
second=parts.get('second', 0), microsecond=parts.get('microsecond', 0),
|
||||
tzinfo=parts.get('tzinfo'))
|
||||
|
||||
def _parse_multiformat(self, string, formats):
|
||||
|
||||
_datetime = None
|
||||
|
||||
for fmt in formats:
|
||||
try:
|
||||
_datetime = self.parse(string, fmt)
|
||||
break
|
||||
except ParserError:
|
||||
pass
|
||||
|
||||
if _datetime is None:
|
||||
raise ParserError('Could not match input to any of {0} on \'{1}\''.format(formats, string))
|
||||
|
||||
return _datetime
|
||||
|
||||
@staticmethod
|
||||
def _map_lookup(input_map, key):
|
||||
|
||||
try:
|
||||
return input_map[key]
|
||||
except KeyError:
|
||||
raise ParserError('Could not match "{0}" to {1}'.format(key, input_map))
|
||||
|
||||
@staticmethod
|
||||
def _try_timestamp(string):
|
||||
|
||||
try:
|
||||
return float(string)
|
||||
except:
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
def _choice_re(choices, flags=0):
|
||||
return re.compile('({0})'.format('|'.join(choices)), flags=flags)
|
||||
|
||||
|
||||
class TzinfoParser(object):
|
||||
|
||||
_TZINFO_RE = re.compile('([+\-])?(\d\d):?(\d\d)?')
|
||||
|
||||
@classmethod
|
||||
def parse(cls, string):
|
||||
|
||||
tzinfo = None
|
||||
|
||||
if string == 'local':
|
||||
tzinfo = tz.tzlocal()
|
||||
|
||||
elif string in ['utc', 'UTC']:
|
||||
tzinfo = tz.tzutc()
|
||||
|
||||
else:
|
||||
|
||||
iso_match = cls._TZINFO_RE.match(string)
|
||||
|
||||
if iso_match:
|
||||
sign, hours, minutes = iso_match.groups()
|
||||
if minutes is None:
|
||||
minutes = 0
|
||||
seconds = int(hours) * 3600 + int(minutes) * 60
|
||||
|
||||
if sign == '-':
|
||||
seconds *= -1
|
||||
|
||||
tzinfo = tz.tzoffset(None, seconds)
|
||||
|
||||
else:
|
||||
tzinfo = tz.gettz(string)
|
||||
|
||||
if tzinfo is None:
|
||||
raise ParserError('Could not parse timezone expression "{0}"'.format(string))
|
||||
|
||||
return tzinfo
|
||||
Loading…
Add table
Add a link
Reference in a new issue