# -*- coding: utf-8 -*- # Copyright 2015-2019 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 # published by the Free Software Foundation. """Extract images from https://nhentai.net/""" from .common import Extractor, Message from .. import text import json class NHentaiExtractor(Extractor): """Base class for nhentai extractors""" category = "nhentai" root = "https://nhentai.net" media_url = "https://i.nhentai.net" @staticmethod def transform_to_metadata(ginfo): """Transform an nhentai API response into compatible metadata""" title_en = ginfo["title"].get("english", "") title_ja = ginfo["title"].get("japanese", "") return { "gallery_id": ginfo["id"], "upload_date": ginfo["upload_date"], "media_id": ginfo["media_id"], "scanlator": ginfo["scanlator"], "count": ginfo["num_pages"], "title": title_en or title_ja, "title_en": title_en, "title_ja": title_ja, } class NhentaiGalleryExtractor(NHentaiExtractor): """Extractor for image galleries from nhentai.net""" subcategory = "gallery" directory_fmt = ["{category}", "{gallery_id} {title}"] filename_fmt = "{category}_{gallery_id}_{num:>03}.{extension}" archive_fmt = "{gallery_id}_{num}" pattern = [r"(?:https?://)?nhentai\.net/g/(\d+)"] test = [("https://nhentai.net/g/147850/", { "url": "5179dbf0f96af44005a0ff705a0ad64ac26547d0", "keyword": "2f94976e657f3043a89997e22f4de8e1b22d9175", })] def __init__(self, match): NHentaiExtractor.__init__(self) self.gid = match.group(1) def items(self): ginfo = self.get_gallery_info(self.gid) data = self.transform_to_metadata(ginfo) urlfmt = "{}/galleries/{}/{{}}.{{}}".format( self.media_url, data["media_id"]) extdict = {"j": "jpg", "p": "png", "g": "gif"} yield Message.Version, 1 yield Message.Directory, data for data["num"], image in enumerate(ginfo["images"]["pages"], 1): ext = extdict.get(image["t"], "jpg") data["width"] = image["w"] data["height"] = image["h"] data["extension"] = ext yield Message.Url, urlfmt.format(data["num"], ext), data def get_gallery_info(self, gallery_id): """Extract and return info about a gallery by ID""" url = "{}/g/{}".format(self.root, gallery_id) page = self.request(url).text return json.loads(text.extract(page, "N.gallery(", ");")[0]) class NhentaiSearchExtractor(NHentaiExtractor): """Extractor for nhentai search results""" category = "nhentai" subcategory = "search" pattern = [r"(?:https?://)?nhentai\.net/search/?\?([^#]+)"] test = [("https://nhentai.net/search/?q=touhou", { "pattern": NhentaiGalleryExtractor.pattern[0], "count": 30, "range": "1-30", })] def __init__(self, match): NHentaiExtractor.__init__(self) self.params = text.parse_query(match.group(1)) def items(self): yield Message.Version, 1 for gid in self._pagination(self.params): url = "{}/g/{}/".format(self.root, gid) yield Message.Queue, url, {} def _pagination(self, params): url = "{}/search/".format(self.root) params["page"] = text.parse_int(params.get("page"), 1) while True: page = self.request(url, params=params).text yield from text.extract_iter(page, 'href="/g/', '/') if 'class="next"' not in page: return params["page"] += 1