mirror of
https://github.com/mikf/gallery-dl.git
synced 2024-11-24 03:32:33 +01:00
968d3e8465
'/?&#' -> '/?#' and '?&#' -> '?#' According to https://www.ietf.org/rfc/rfc3986.txt, URLs are "organized hierarchically" by using "the slash ("/"), question mark ("?"), and number sign ("#") characters to delimit components"
210 lines
7.6 KiB
Python
210 lines
7.6 KiB
Python
# -*- coding: utf-8 -*-
|
|
|
|
# Copyright 2016-2020 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.
|
|
|
|
"""Extractors for https://www.imagefap.com/"""
|
|
|
|
from .common import Extractor, Message
|
|
from .. import text
|
|
import json
|
|
|
|
|
|
BASE_PATTERN = r"(?:https?://)?(?:www\.|beta\.)?imagefap\.com"
|
|
|
|
|
|
class ImagefapExtractor(Extractor):
|
|
"""Base class for imagefap extractors"""
|
|
category = "imagefap"
|
|
directory_fmt = ("{category}", "{gallery_id} {title}")
|
|
filename_fmt = "{category}_{gallery_id}_{filename}.{extension}"
|
|
archive_fmt = "{gallery_id}_{image_id}"
|
|
root = "https://www.imagefap.com"
|
|
|
|
def __init__(self, match):
|
|
Extractor.__init__(self, match)
|
|
self.session.headers["Referer"] = self.root
|
|
|
|
|
|
class ImagefapGalleryExtractor(ImagefapExtractor):
|
|
"""Extractor for image galleries from imagefap.com"""
|
|
subcategory = "gallery"
|
|
pattern = BASE_PATTERN + r"/(?:gallery\.php\?gid=|gallery/|pictures/)(\d+)"
|
|
|
|
test = (
|
|
("https://www.imagefap.com/pictures/7102714", {
|
|
"pattern": r"https://cdn.imagefap.com/images/full/\d+/\d+/\d+.jpg",
|
|
"keyword": "2ba96e84c2952c4750e9fa94a3f2b1f965cec2f3",
|
|
"content": "694a0a57385980a6f90fbc296cadcd6c11ba2dab",
|
|
}),
|
|
("https://www.imagefap.com/gallery/5486966", {
|
|
"pattern": r"https://cdn.imagefap.com/images/full/\d+/\d+/\d+.jpg",
|
|
"keyword": "3e24eace5b09639b881ebd393165862feb46adde",
|
|
}),
|
|
("https://www.imagefap.com/gallery.php?gid=7102714"),
|
|
("https://beta.imagefap.com/gallery.php?gid=7102714"),
|
|
)
|
|
|
|
def __init__(self, match):
|
|
ImagefapExtractor.__init__(self, match)
|
|
self.gid = match.group(1)
|
|
self.image_id = ""
|
|
|
|
def items(self):
|
|
url = "{}/pictures/{}/".format(self.root, self.gid)
|
|
page = self.request(url).text
|
|
data = self.get_job_metadata(page)
|
|
yield Message.Version, 1
|
|
yield Message.Directory, data
|
|
for url, image in self.get_images():
|
|
data.update(image)
|
|
yield Message.Url, url, data
|
|
|
|
def get_job_metadata(self, page):
|
|
"""Collect metadata for extractor-job"""
|
|
descr, pos = text.extract(
|
|
page, '<meta name="description" content="Browse ', '"')
|
|
count, pos = text.extract(page, ' 1 of ', ' pics"', pos)
|
|
self.image_id = text.extract(page, 'id="img_ed_', '"', pos)[0]
|
|
|
|
title, _, descr = descr.partition(" porn picture gallery by ")
|
|
uploader, _, tags = descr.partition(" to see hottest ")
|
|
return {
|
|
"gallery_id": text.parse_int(self.gid),
|
|
"title": text.unescape(title),
|
|
"uploader": uploader,
|
|
"tags": tags[:-11].split(", "),
|
|
"count": text.parse_int(count),
|
|
}
|
|
|
|
def get_images(self):
|
|
"""Collect image-urls and -metadata"""
|
|
num = 0
|
|
url = "{}/photo/{}/".format(self.root, self.image_id)
|
|
params = {"gid": self.gid, "idx": 0, "partial": "true"}
|
|
while True:
|
|
pos = 0
|
|
page = self.request(url, params=params).text
|
|
for _ in range(24):
|
|
imgurl, pos = text.extract(page, '<a href="', '"', pos)
|
|
if not imgurl:
|
|
return
|
|
num += 1
|
|
data = text.nameext_from_url(imgurl)
|
|
data["num"] = num
|
|
data["image_id"] = text.parse_int(data["filename"])
|
|
yield imgurl, data
|
|
params["idx"] += 24
|
|
|
|
|
|
class ImagefapImageExtractor(ImagefapExtractor):
|
|
"""Extractor for single images from imagefap.com"""
|
|
subcategory = "image"
|
|
pattern = BASE_PATTERN + r"/photo/(\d+)"
|
|
test = (
|
|
("https://www.imagefap.com/photo/1369341772/", {
|
|
"pattern": r"https://cdn.imagefap.com/images/full/\d+/\d+/\d+.jpg",
|
|
"keyword": "8894e45f7262020d8d66ce59917315def1fc475b",
|
|
}),
|
|
("https://beta.imagefap.com/photo/1369341772/"),
|
|
)
|
|
|
|
def __init__(self, match):
|
|
ImagefapExtractor.__init__(self, match)
|
|
self.image_id = match.group(1)
|
|
|
|
def items(self):
|
|
url, data = self.get_image()
|
|
yield Message.Version, 1
|
|
yield Message.Directory, data
|
|
yield Message.Url, url, data
|
|
|
|
def get_image(self):
|
|
url = "{}/photo/{}/".format(self.root, self.image_id)
|
|
page = self.request(url).text
|
|
|
|
info, pos = text.extract(
|
|
page, '<script type="application/ld+json">', '</script>')
|
|
image_id, pos = text.extract(
|
|
page, 'id="imageid_input" value="', '"', pos)
|
|
gallery_id, pos = text.extract(
|
|
page, 'id="galleryid_input" value="', '"', pos)
|
|
info = json.loads(info)
|
|
url = info["contentUrl"]
|
|
|
|
return url, text.nameext_from_url(url, {
|
|
"title": text.unescape(info["name"]),
|
|
"uploader": info["author"],
|
|
"date": info["datePublished"],
|
|
"width": text.parse_int(info["width"]),
|
|
"height": text.parse_int(info["height"]),
|
|
"gallery_id": text.parse_int(gallery_id),
|
|
"image_id": text.parse_int(image_id),
|
|
})
|
|
|
|
|
|
class ImagefapUserExtractor(ImagefapExtractor):
|
|
"""Extractor for all galleries from a user at imagefap.com"""
|
|
subcategory = "user"
|
|
categorytransfer = True
|
|
pattern = (BASE_PATTERN +
|
|
r"/(?:profile(?:\.php\?user=|/)([^/?#]+)"
|
|
r"|usergallery\.php\?userid=(\d+))")
|
|
test = (
|
|
("https://www.imagefap.com/profile/LucyRae/galleries", {
|
|
"url": "d941aa906f56a75972a7a5283030eb9a8d27a4fd",
|
|
}),
|
|
("https://www.imagefap.com/usergallery.php?userid=1862791", {
|
|
"url": "d941aa906f56a75972a7a5283030eb9a8d27a4fd",
|
|
}),
|
|
("https://www.imagefap.com/profile.php?user=LucyRae"),
|
|
("https://beta.imagefap.com/profile.php?user=LucyRae"),
|
|
)
|
|
|
|
def __init__(self, match):
|
|
ImagefapExtractor.__init__(self, match)
|
|
self.user, self.user_id = match.groups()
|
|
|
|
def items(self):
|
|
yield Message.Version, 1
|
|
for gid, name in self.get_gallery_data():
|
|
url = "{}/gallery/{}".format(self.root, gid)
|
|
data = {
|
|
"gallery_id": text.parse_int(gid),
|
|
"title": text.unescape(name),
|
|
"_extractor": ImagefapGalleryExtractor,
|
|
}
|
|
yield Message.Queue, url, data
|
|
|
|
def get_gallery_data(self):
|
|
"""Yield all gallery_ids of a specific user"""
|
|
folders = self.get_gallery_folders()
|
|
url = "{}/ajax_usergallery_folder.php".format(self.root)
|
|
params = {"userid": self.user_id}
|
|
for folder_id in folders:
|
|
params["id"] = folder_id
|
|
page = self.request(url, params=params).text
|
|
|
|
pos = 0
|
|
while True:
|
|
gid, pos = text.extract(page, '<a href="/gallery/', '"', pos)
|
|
if not gid:
|
|
break
|
|
name, pos = text.extract(page, "<b>", "<", pos)
|
|
yield gid, name
|
|
|
|
def get_gallery_folders(self):
|
|
"""Create a list of all folder_ids of a specific user"""
|
|
if self.user:
|
|
url = "{}/profile/{}/galleries".format(self.root, self.user)
|
|
else:
|
|
url = "{}/usergallery.php?userid={}".format(
|
|
self.root, self.user_id)
|
|
page = self.request(url).text
|
|
self.user_id, pos = text.extract(page, '?userid=', '"')
|
|
folders, pos = text.extract(page, ' id="tgl_all" value="', '"', pos)
|
|
return folders.split("|")[:-1]
|