You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
101 lines
3.4 KiB
101 lines
3.4 KiB
import re
|
|
|
|
from .parser import _next_significant, _to_token_iterator
|
|
|
|
|
|
def parse_nth(input):
|
|
"""Parse `<An+B> <https://drafts.csswg.org/css-syntax-3/#anb>`_,
|
|
as found in `:nth-child()
|
|
<https://drafts.csswg.org/selectors/#nth-child-pseudo>`_
|
|
and related Selector pseudo-classes.
|
|
|
|
Although tinycss2 does not include a full Selector parser,
|
|
this bit of syntax is included as it is particularly tricky to define
|
|
on top of a CSS tokenizer.
|
|
|
|
:type input: :obj:`str` or :term:`iterable`
|
|
:param input: A string or an iterable of :term:`component values`.
|
|
:returns:
|
|
A ``(a, b)`` tuple of integers, or :obj:`None` if the input is invalid.
|
|
|
|
"""
|
|
tokens = _to_token_iterator(input, skip_comments=True)
|
|
token = _next_significant(tokens)
|
|
if token is None:
|
|
return
|
|
token_type = token.type
|
|
if token_type == 'number' and token.is_integer:
|
|
return parse_end(tokens, 0, token.int_value)
|
|
elif token_type == 'dimension' and token.is_integer:
|
|
unit = token.lower_unit
|
|
if unit == 'n':
|
|
return parse_b(tokens, token.int_value)
|
|
elif unit == 'n-':
|
|
return parse_signless_b(tokens, token.int_value, -1)
|
|
else:
|
|
match = N_DASH_DIGITS_RE.match(unit)
|
|
if match:
|
|
return parse_end(tokens, token.int_value, int(match.group(1)))
|
|
elif token_type == 'ident':
|
|
ident = token.lower_value
|
|
if ident == 'even':
|
|
return parse_end(tokens, 2, 0)
|
|
elif ident == 'odd':
|
|
return parse_end(tokens, 2, 1)
|
|
elif ident == 'n':
|
|
return parse_b(tokens, 1)
|
|
elif ident == '-n':
|
|
return parse_b(tokens, -1)
|
|
elif ident == 'n-':
|
|
return parse_signless_b(tokens, 1, -1)
|
|
elif ident == '-n-':
|
|
return parse_signless_b(tokens, -1, -1)
|
|
elif ident[0] == '-':
|
|
match = N_DASH_DIGITS_RE.match(ident[1:])
|
|
if match:
|
|
return parse_end(tokens, -1, int(match.group(1)))
|
|
else:
|
|
match = N_DASH_DIGITS_RE.match(ident)
|
|
if match:
|
|
return parse_end(tokens, 1, int(match.group(1)))
|
|
elif token == '+':
|
|
token = next(tokens) # Whitespace after an initial '+' is invalid.
|
|
if token.type == 'ident':
|
|
ident = token.lower_value
|
|
if ident == 'n':
|
|
return parse_b(tokens, 1)
|
|
elif ident == 'n-':
|
|
return parse_signless_b(tokens, 1, -1)
|
|
else:
|
|
match = N_DASH_DIGITS_RE.match(ident)
|
|
if match:
|
|
return parse_end(tokens, 1, int(match.group(1)))
|
|
|
|
|
|
def parse_b(tokens, a):
|
|
token = _next_significant(tokens)
|
|
if token is None:
|
|
return (a, 0)
|
|
elif token == '+':
|
|
return parse_signless_b(tokens, a, 1)
|
|
elif token == '-':
|
|
return parse_signless_b(tokens, a, -1)
|
|
elif (token.type == 'number' and token.is_integer and
|
|
token.representation[0] in '-+'):
|
|
return parse_end(tokens, a, token.int_value)
|
|
|
|
|
|
def parse_signless_b(tokens, a, b_sign):
|
|
token = _next_significant(tokens)
|
|
if (token.type == 'number' and token.is_integer and
|
|
not token.representation[0] in '-+'):
|
|
return parse_end(tokens, a, b_sign * token.int_value)
|
|
|
|
|
|
def parse_end(tokens, a, b):
|
|
if _next_significant(tokens) is None:
|
|
return (a, b)
|
|
|
|
|
|
N_DASH_DIGITS_RE = re.compile('^n(-[0-9]+)$')
|