dosage/dosagelib/plugins/rhjunior.py

38 lines
1.4 KiB
Python
Raw Permalink Normal View History

# SPDX-License-Identifier: MIT
# Copyright (C) 2019-2022 Tobias Gruetzmacher
# Copyright (C) 2019-2020 Daniel Ring
from ..scraper import ParserScraper
class RHJunior(ParserScraper):
stripUrl = 'https://www.rhjunior.com/%s/'
imageSearch = '//div[contains(@class, "entry-content")]//img'
multipleImagesPerStrip = True
def __init__(self, name, sub, prev, first, last=None):
super().__init__('RHJunior/' + name)
self.prevSearch = ('//a[@rel="prev"]', '//a[@title="' + prev + '"]')
self.url = self.stripUrl % ('comics/' + sub)
self.firstStripUrl = self.stripUrl % (sub + '-' + first)
if last:
self.url = self.stripUrl % (sub + '-' + last)
self.endOfLife = True
@classmethod
def getmodules(cls):
return (
cls('GoblinHollow', 'goblin-hollow',
'', '0001', last='7'),
cls('NipAndTuck', 'nip-and-tuck',
'Nip and Tuck', '0000'),
cls('QuentynQuinnSpaceRanger', 'quentyn-quinn-space-ranger',
'Quentyn Quinn, Space Ranger', '0001'),
cls('TalesOfTheQuestor', 'tales-of-the-questor',
'Tales of the Questor', 'cover'),
cls('TheJournalOfEnniasLongscript', 'the-journal-of-ennias-longscript',
'', '0001', last='0111'),
cls('TheProbabilityBomb', 'the-probability-bomb',
'the Probability Bomb', 'kickstarter'),
)