summaryrefslogtreecommitdiffstats
path: root/gallery_dl/extractor/2chen.py
diff options
context:
space:
mode:
Diffstat (limited to 'gallery_dl/extractor/2chen.py')
-rw-r--r--gallery_dl/extractor/2chen.py99
1 files changed, 99 insertions, 0 deletions
diff --git a/gallery_dl/extractor/2chen.py b/gallery_dl/extractor/2chen.py
new file mode 100644
index 0000000..8fffeb0
--- /dev/null
+++ b/gallery_dl/extractor/2chen.py
@@ -0,0 +1,99 @@
+# -*- coding: utf-8 -*-
+
+# 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://2chen.moe/"""
+
+from .common import Extractor, Message
+from .. import text
+
+
+class _2chenThreadExtractor(Extractor):
+ """Extractor for 2chen threads"""
+ category = "2chen"
+ subcategory = "thread"
+ directory_fmt = ("{category}", "{board}", "{thread} {title}")
+ filename_fmt = "{time} {filename}.{extension}"
+ archive_fmt = "{board}_{thread}_{hash}"
+ root = "https://2chen.moe"
+ pattern = r"(?:https?://)?2chen\.moe/([^/?#]+)/(\d+)"
+ test = (
+ ("https://2chen.moe/jp/303786", {
+ "count": ">= 10",
+ }),
+ )
+
+ def __init__(self, match):
+ Extractor.__init__(self, match)
+ self.board, self.thread = match.groups()
+
+ def items(self):
+ url = "{}/{}/{}".format(self.root, self.board, self.thread)
+ page = self.request(url, encoding="utf-8").text
+ data = self.metadata(page)
+ yield Message.Directory, data
+ for post in self.posts(page):
+ if not post["url"]:
+ continue
+ post.update(data)
+ post["url"] = self.root + post["url"]
+ post["time"] = text.parse_int(post["date"].timestamp())
+ yield Message.Url, post["url"], text.nameext_from_url(
+ post["filename"], post)
+
+ def metadata(self, page):
+ board, pos = text.extract(page, 'class="board">/', '/<')
+ title = text.extract(page, "<h3>", "</h3>", pos)[0]
+ return {
+ "board" : board,
+ "thread": self.thread,
+ "title" : text.unescape(title),
+ }
+
+ def posts(self, page):
+ """Return iterable with relevant posts"""
+ return map(self.parse, text.extract_iter(
+ page, 'class="glass media', '</article>'))
+
+ def parse(self, post):
+ extr = text.extract_from(post)
+ return {
+ "name" : text.unescape(extr("<span>", "</span>")),
+ "date" : text.parse_datetime(
+ extr("<time", "<").partition(">")[2],
+ "%d %b %Y (%a) %H:%M:%S"
+ ),
+ "no" : extr('href="#p', '"'),
+ "url" : extr('</span><a href="', '"'),
+ "filename": text.unescape(extr('download="', '"')),
+ "hash" : extr('data-hash="', '"'),
+ }
+
+
+class _2chenBoardExtractor(Extractor):
+ """Extractor for 2chen boards"""
+ category = "2chen"
+ subcategory = "board"
+ root = "https://2chen.moe"
+ pattern = r"(?:https?://)?2chen\.moe/([^/?#]+)(?:/catalog)?/?$"
+ test = (
+ ("https://2chen.moe/co/", {
+ "pattern": _2chenThreadExtractor.pattern
+ }),
+ ("https://2chen.moe/co"),
+ ("https://2chen.moe/co/catalog")
+ )
+
+ def __init__(self, match):
+ Extractor.__init__(self, match)
+ self.board = match.group(1)
+
+ def items(self):
+ url = "{}/{}/catalog".format(self.root, self.board)
+ page = self.request(url).text
+ data = {"_extractor": _2chenThreadExtractor}
+ for thread in text.extract_iter(
+ page, '<figure><a href="', '"'):
+ yield Message.Queue, self.root + thread, data