aboutsummaryrefslogtreecommitdiffstats
path: root/gallery_dl/extractor/s3ndpics.py
diff options
context:
space:
mode:
Diffstat (limited to 'gallery_dl/extractor/s3ndpics.py')
-rw-r--r--gallery_dl/extractor/s3ndpics.py101
1 files changed, 101 insertions, 0 deletions
diff --git a/gallery_dl/extractor/s3ndpics.py b/gallery_dl/extractor/s3ndpics.py
new file mode 100644
index 0000000..215f160
--- /dev/null
+++ b/gallery_dl/extractor/s3ndpics.py
@@ -0,0 +1,101 @@
+# -*- coding: utf-8 -*-
+
+# Copyright 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
+# published by the Free Software Foundation.
+
+"""Extractors for https://s3nd.pics/"""
+
+from .common import Extractor, Message
+from .. import text
+
+BASE_PATTERN = r"(?:https?://)?(?:www\.)?s3nd\.pics"
+
+
+class S3ndpicsExtractor(Extractor):
+ """Base class for s3ndpics extractors"""
+ category = "s3ndpics"
+ root = "https://s3nd.pics"
+ root_api = f"{root}/api"
+ directory_fmt = ("{category}", "{user[username]}",
+ "{date} {title:?/ /}({id})")
+ filename_fmt = "{num:>02}.{extension}"
+ archive_fmt = "{id}_{num}"
+
+ def items(self):
+ base = "https://s3.s3nd.pics/s3nd-pics/"
+
+ for post in self.posts():
+ post["id"] = post.pop("_id", None)
+ post["user"] = post.pop("userId", None)
+ post["date"] = text.parse_datetime(
+ post["createdAt"], "%Y-%m-%dT%H:%M:%S.%fZ")
+ post["date_updated"] = text.parse_datetime(
+ post["updatedAt"], "%Y-%m-%dT%H:%M:%S.%fZ")
+
+ files = post.pop("files", ())
+ post["count"] = len(files)
+
+ yield Message.Directory, post
+ for post["num"], file in enumerate(files, 1):
+ post["type"] = file["type"]
+ path = file["url"]
+ text.nameext_from_url(path, post)
+ yield Message.Url, f"{base}{path}", post
+
+ def _pagination(self, url, params):
+ params["page"] = 1
+
+ while True:
+ data = self.request_json(url, params=params)
+
+ self.kwdict["total"] = data["pagination"]["total"]
+ yield from data["posts"]
+
+ if params["page"] >= data["pagination"]["pages"]:
+ return
+ params["page"] += 1
+
+
+class S3ndpicsPostExtractor(S3ndpicsExtractor):
+ subcategory = "post"
+ pattern = rf"{BASE_PATTERN}/post/([0-9a-f]+)"
+ example = "https://s3nd.pics/post/0123456789abcdef01234567"
+
+ def posts(self):
+ url = f"{self.root_api}/posts/{self.groups[0]}"
+ return (self.request_json(url)["post"],)
+
+
+class S3ndpicsUserExtractor(S3ndpicsExtractor):
+ subcategory = "user"
+ pattern = rf"{BASE_PATTERN}/user/(\w+)"
+ example = "https://s3nd.pics/user/USER"
+
+ def posts(self):
+ url = f"{self.root_api}/users/username/{self.groups[0]}"
+ self.kwdict["user"] = user = self.request_json(url)["user"]
+
+ url = f"{self.root_api}/posts"
+ params = {
+ "userId": user["_id"],
+ "limit" : "12",
+ "sortBy": "newest",
+ }
+ return self._pagination(url, params)
+
+
+class S3ndpicsSearchExtractor(S3ndpicsExtractor):
+ subcategory = "search"
+ pattern = rf"{BASE_PATTERN}/search/?\?([^#]+)"
+ example = "https://s3nd.pics/search?QUERY"
+
+ def posts(self):
+ url = f"{self.root_api}/posts"
+ params = text.parse_query(self.groups[0])
+ params.setdefault("limit", "20")
+ self.kwdict["search_tags"] = \
+ params.get("tag") or params.get("tags") or params.get("q")
+ return self._pagination(url, params)