summaryrefslogtreecommitdiffstats
path: root/gallery_dl/extractor/mangadex.py
diff options
context:
space:
mode:
Diffstat (limited to 'gallery_dl/extractor/mangadex.py')
-rw-r--r--gallery_dl/extractor/mangadex.py12
1 files changed, 11 insertions, 1 deletions
diff --git a/gallery_dl/extractor/mangadex.py b/gallery_dl/extractor/mangadex.py
index c91e9a8..96c81c7 100644
--- a/gallery_dl/extractor/mangadex.py
+++ b/gallery_dl/extractor/mangadex.py
@@ -9,7 +9,7 @@
"""Extractors for https://mangadex.org/"""
from .common import Extractor, Message
-from .. import text, util
+from .. import text, util, exception
from ..cache import memcache
@@ -74,6 +74,10 @@ class MangadexChapterExtractor(MangadexExtractor):
"count": 64,
"keyword": "c53a0e4c12250578a4e630281085875e59532c03",
}),
+ # MANGA Plus (#1154)
+ ("https://mangadex.org/chapter/1122815", {
+ "excepion": exception.StopExtraction,
+ }),
)
def __init__(self, match):
@@ -82,6 +86,12 @@ class MangadexChapterExtractor(MangadexExtractor):
def items(self):
cdata = self.chapter_data(self.chapter_id)
+ if "server" not in cdata:
+ if cdata["status"] == "external":
+ raise exception.StopExtraction(
+ "Chapter is not available on MangaDex and can be read on "
+ "the official publisher's website at %s.", cdata["pages"])
+ raise exception.StopExtraction("No download server available.")
mdata = self.manga_data(cdata["mangaId"])
chapter, sep, minor = cdata["chapter"].partition(".")