X-Git-Url: https://git.mdrn.pl/redakcja.git/blobdiff_plain/f3339efe0d7ffa1199a113108c0e2f6f6b07afe2..180e82abec8f380595152c2f51d2967ff2d6d42e:/apps/cover/views.py diff --git a/apps/cover/views.py b/apps/cover/views.py index 19d81c22..0f341ad7 100644 --- a/apps/cover/views.py +++ b/apps/cover/views.py @@ -1,7 +1,7 @@ # Create your views here. import os.path from django.conf import settings -from django.http import HttpResponse, Http404 +from django.http import HttpResponse, HttpResponseRedirect, Http404 from catalogue.models import Chunk from django.views.decorators.http import require_POST from django.views.decorators.csrf import csrf_exempt @@ -10,6 +10,7 @@ from django.shortcuts import render PREVIEW_SIZE = (216, 300) + def preview(request, book, chunk=None, rev=None): """Creates a cover image. @@ -31,8 +32,11 @@ def preview(request, book, chunk=None, rev=None): if revision is None: raise Http404 xml = revision.materialize().encode('utf-8') - - info = BookInfo.from_string(xml) + + try: + info = BookInfo.from_string(xml) + except: + return HttpResponseRedirect(os.path.join(settings.STATIC_URL, "img/sample_cover.png")) cover = WLCover(info) response = HttpResponse(mimetype=cover.mime_type()) image = cover.image().resize(PREVIEW_SIZE, Image.ANTIALIAS) @@ -51,7 +55,10 @@ def preview_from_xml(request): from librarian.dcparser import BookInfo xml = request.POST['xml'] - info = BookInfo.from_string(xml.encode('utf-8')) + try: + info = BookInfo.from_string(xml.encode('utf-8')) + except: + return HttpResponse(os.path.join(settings.STATIC_URL, "img/sample_cover.png")) coverid = sha1(etree.tostring(info.to_etree())).hexdigest() cover = WLCover(info) @@ -74,17 +81,21 @@ def flickr(request): html = urlopen(url).read() match = re.search(r'Some rights reserved', html) - if match: + try: + assert match license_url = match.group(1) - - re_license = re.compile(r'http://creativecommons.org/licenses/([^/]*)/([^/]*)/.*') - m = re_license.match(license_url) - if m: + re_license = re.compile(r'http://creativecommons.org/licenses/([^/]*)/([^/]*)/.*') + m = re_license.match(license_url) + assert m license_name = 'CC %s %s' % (m.group(1).upper(), m.group(2)) + except AssertionError: + license_name = 'NIEZNANA LICENCJA' m = re.search(r'By ([^<]*)', html) if m: author = m.group(1) + else: + author = "NIEZNANY AUTOR" url_size = url.rstrip('/') + '/sizes/o/' html = urlopen(url_size).read()