regluit/api/opds.py

226 lines
7.7 KiB
Python

from itertools import islice
from lxml import etree
import datetime
import urlparse
import pytz
from regluit.core import models
import regluit.core.cc as cc
licenses = cc.LICENSE_LIST
FORMAT_TO_MIMETYPE = {'pdf':"application/pdf",
'epub':"application/epub+zip",
'mobi':"application/x-mobipocket-ebook",
'html':"text/html",
'text':"text/html"}
facets = ["creative_commons","active_campaigns"]
UNGLUEIT_URL= 'https://unglue.it'
NAVIGATION = "application/atom+xml;profile=opds-catalog;kind=navigation"
def feeds():
for facet in facets:
yield globals()[facet]()
def text_node(tag, text):
node = etree.Element(tag)
node.text = text
return node
def add_query_component(url, qc):
"""
add component qc to the querystring of url
"""
m = list(urlparse.urlparse(url))
if len(m[4]):
m[4] = "&".join([m[4],qc])
else:
m[4] = qc
return urlparse.urlunparse(m)
def isbn_node(isbn):
node = etree.Element("{http://purl.org/dc/terms/}identifier")
node.attrib.update({"{http://www.w3.org/2001/XMLSchema-instance}type":'dcterms:URI'})
node.text = 'urn:ISBN:'+ isbn
return node
def work_node(work):
node = etree.Element("entry")
# title
node.append(text_node("title", work.title))
# id
node.append(text_node('id', "{base}{url}".format(base=UNGLUEIT_URL,url=work.get_absolute_url())))
# updated -- using creation date
node.append(text_node('updated', work.created.isoformat()))
# links for all ebooks
for ebook in work.ebooks():
link_node = etree.Element("link")
# ebook.download_url is an absolute URL with the protocol, domain, and path baked in
link_node.attrib.update({"href":add_query_component(ebook.download_url, "feed=opds"),
"type":FORMAT_TO_MIMETYPE.get(ebook.format, ""),
"rel":"http://opds-spec.org/acquisition"})
node.append(link_node)
# get the cover -- assume jpg?
cover_node = etree.Element("link")
cover_node.attrib.update({"href":work.cover_image_small(),
"type":"image/"+work.cover_filetype(),
"rel":"http://opds-spec.org/image/thumbnail"})
node.append(cover_node)
cover_node = etree.Element("link")
cover_node.attrib.update({"href":work.cover_image_thumbnail(),
"type":"image/"+work.cover_filetype(),
"rel":"http://opds-spec.org/image"})
node.append(cover_node)
# <dcterms:issued>2012</dcterms:issued>
node.append(text_node("{http://purl.org/dc/terms/}issued", work.publication_date_year))
# author
# TO DO: include all authors?
author_node = etree.Element("author")
author_node.append(text_node("name", work.author()))
node.append(author_node)
# publisher
#<dcterms:publisher>Open Book Publishers</dcterms:publisher>
if len(work.publishers()):
for publisher in work.publishers():
node.append(text_node("{http://purl.org/dc/terms/}issued", publisher.name.name))
# language
#<dcterms:language>en</dcterms:language>
node.append(text_node("{http://purl.org/dc/terms/}language", work.language))
# identifiers
if work.identifiers.filter(type='isbn'):
for isbn in work.identifiers.filter(type='isbn')[0:9]: #10 should be more than enough
node.append(isbn_node(isbn.value))
# subject tags
# [[subject.name for subject in work.subjects.all()] for work in ccworks if work.subjects.all()]
if work.subjects.all():
for subject in work.subjects.all():
category_node = etree.Element("category")
category_node.attrib["term"] = subject.name
node.append(category_node)
return node
class Facet:
title = ''
works = None
feed_path = ''
description = ''
def feed(self, page=None):
return opds_feed_for_works(self.works, self.feed_path, title=self.title, page=page)
def updated(self):
# return the creation date for most recently added item
if not self.works:
return pytz.utc.localize(datetime.datetime.utcnow()).isoformat()
else:
return pytz.utc.localize(self.works[0].created).isoformat()
class creative_commons(Facet):
title = "Unglue.it Catalog: Creative Commons Books"
feed_path = "creative_commons"
works = models.Work.objects.filter(editions__ebooks__isnull=False,
editions__ebooks__rights__in=cc.LICENSE_LIST).distinct().order_by('-created')
description= "These Creative Commons licensed ebooks are ready to read - the people who created them want you to read and share them."
class active_campaigns(Facet):
"""
return opds feed for works associated with active campaigns
"""
title = "Unglue.it Catalog: Books under Active Campaign"
feed_path = "active_campaigns"
works = models.Work.objects.filter(campaigns__status='ACTIVE',
editions__ebooks__isnull=False).distinct().order_by('-created')
description= "With your help we're raising money to make these books free to the world."
def opds_feed_for_works(works, feed_path, title="Unglue.it Catalog", page=None):
feed_xml = """<feed xmlns:dcterms="http://purl.org/dc/terms/"
xmlns:opds="http://opds-spec.org/"
xmlns="http://www.w3.org/2005/Atom"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:noNamespaceSchemaLocation="http://www.kbcafe.com/rss/atom.xsd.xml"
xsi:schemaLocation="http://purl.org/dc/elements/1.1/ http://dublincore.org/schemas/xmls/qdc/2008/02/11/dc.xsd http://purl.org/dc/terms/ http://dublincore.org/schemas/xmls/qdc/2008/02/11/dcterms.xsd"/>"""
feed = etree.fromstring(feed_xml)
# add title
# TO DO: will need to calculate the number items and where in the feed we are
feed.append(text_node('title', title))
# id
feed.append(text_node('id', "{url}/api/opds/{feed_path}".format(url=UNGLUEIT_URL,
feed_path=feed_path)))
# updated
# TO DO: fix time zone?
# also use our wrapped datetime code
feed.append(text_node('updated',
pytz.utc.localize(datetime.datetime.utcnow()).isoformat()))
# author
author_node = etree.Element("author")
author_node.append(text_node('name', 'unglue.it'))
author_node.append(text_node('uri', UNGLUEIT_URL))
feed.append(author_node)
# links: start, self, next/prev (depending what's necessary -- to start with put all CC books)
# start link
append_navlink(feed, 'start', feed_path, None )
# next link
if not page:
page =0
else:
try:
page=int(page)
except TypeError:
page=0
try:
works[10 * page + 10]
append_navlink(feed, 'next', feed_path, page+1 )
except IndexError:
pass
works = islice(works, 10 * page, 10 * page + 10)
if page > 0:
append_navlink(feed, 'previous', feed_path, page-1)
for work in works:
node = work_node(work)
feed.append(node)
return etree.tostring(feed, pretty_print=True)
def append_navlink(feed, rel, path, page):
link = etree.Element("link")
link.attrib.update({"rel":rel,
"href": UNGLUEIT_URL + "/api/opds/" + path + ('/?page=' + unicode(page) if page!=None else '/'),
"type": NAVIGATION,
})
feed.append(link)