diff options
Diffstat (limited to 'gallery_dl/extractor/mangahere.py')
| -rw-r--r-- | gallery_dl/extractor/mangahere.py | 18 |
1 files changed, 8 insertions, 10 deletions
diff --git a/gallery_dl/extractor/mangahere.py b/gallery_dl/extractor/mangahere.py index 8c94f04..151e809 100644 --- a/gallery_dl/extractor/mangahere.py +++ b/gallery_dl/extractor/mangahere.py @@ -1,6 +1,6 @@ # -*- coding: utf-8 -*- -# Copyright 2015-2023 Mike Fährmann +# Copyright 2015-2025 Mike Fährmann # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License version 2 as @@ -9,8 +9,7 @@ """Extractors for https://www.mangahere.cc/""" from .common import ChapterExtractor, MangaExtractor -from .. import text -import re +from .. import text, util class MangahereBase(): @@ -18,7 +17,6 @@ class MangahereBase(): category = "mangahere" root = "https://www.mangahere.cc" root_mobile = "https://m.mangahere.cc" - url_fmt = root_mobile + "/manga/{}/{}.html" class MangahereChapterExtractor(MangahereBase, ChapterExtractor): @@ -29,8 +27,8 @@ class MangahereChapterExtractor(MangahereBase, ChapterExtractor): def __init__(self, match): self.part, self.volume, self.chapter = match.groups() - url = self.url_fmt.format(self.part, 1) - ChapterExtractor.__init__(self, match, url) + self.base = f"{self.root_mobile}/manga/{self.part}/" + ChapterExtractor.__init__(self, match, f"{self.base}1.html") def _init(self): self.session.headers["Referer"] = self.root_mobile + "/" @@ -65,10 +63,10 @@ class MangahereChapterExtractor(MangahereBase, ChapterExtractor): url, pos = text.extract(page, ' src="', '"', pos) yield text.ensure_http_scheme(text.unescape(url)), None pnum += 2 - page = self.request(self.url_fmt.format(self.part, pnum)).text + page = self.request(f"{self.base}{pnum}.html").text def _get_title(self): - url = "{}/manga/{}/".format(self.root, self.part) + url = f"{self.root}/manga/{self.part}/" page = self.request(url).text try: @@ -104,8 +102,8 @@ class MangahereMangaExtractor(MangahereBase, MangaExtractor): info, pos = text.extract(page, 'class="title3">', '<', pos) date, pos = text.extract(page, 'class="title2">', '<', pos) - match = re.match( - r"(?:Vol\.0*(\d+) )?Ch\.0*(\d+)(\S*)(?: - (.*))?", info) + match = util.re( + r"(?:Vol\.0*(\d+) )?Ch\.0*(\d+)(\S*)(?: - (.*))?").match(info) if match: volume, chapter, minor, title = match.groups() else: |
