from basicauth import logged_in_or_basicauth, factory_decorator
from catalogue.models import Book, Tag
-from search import Search, SearchResult, JVM
+from search.views import get_search, SearchResult, JVM
from lucene import Term, QueryWrapperFilter, TermQuery
+import logging
import re
+log = logging.getLogger('opds')
+
from stats.utils import piwik_track
_root_feeds = (
def items(self, tag):
books = Book.tagged.with_any([tag])
- l_tags = Tag.objects.filter(category='book', slug__in=[book.book_tag_slug() for book in books])
+ l_tags = Tag.objects.filter(category='book', slug__in=[book.book_tag_slug() for book in books.iterator()])
descendants_keys = [book.pk for book in Book.tagged.with_any(l_tags)]
if descendants_keys:
books = books.exclude(pk__in=descendants_keys)
JVM.attachCurrentThread()
query = request.GET.get('q', '')
-
+
inline_criteria = re.findall(self.INLINE_QUERY_RE, query)
if inline_criteria:
def get_criteria(criteria, name, position):
e = filter(lambda el: el[0][0:len(name)] == name, criteria)
- print e
+ log.info("get_criteria: %s" % e)
if not e:
return None
c = e[0][position]
- print c
+ log.info("get_criteria: %s" % c)
if c[0] == '"' and c[-1] == '"':
c = c[1:-1]
c = c.replace('+', ' ')
return c
- #import pdb; pdb.set_trace()
author = get_criteria(inline_criteria, 'author', 1)
title = get_criteria(inline_criteria, 'title', 2)
translator = None
author = request.GET.get('author', '')
title = request.GET.get('title', '')
translator = request.GET.get('translator', '')
+
+ # Our client didn't handle the opds placeholders
+ if author == '{atom:author}': author = ''
+ if title == '{atom:title}': title = ''
+ if translator == '{atom:contributor}': translator = ''
categories = None
fuzzy = False
-
- srch = Search()
+ srch = get_search()
hint = srch.hint()
# Scenario 1: full search terms provided.
filters = []
if author:
- print "narrow to author %s" % author
+ log.info( "narrow to author %s" % author)
hint.tags(srch.search_tags(author, filt=srch.term_filter(Term('tag_category', 'author'))))
if translator:
- print "filter by translator %s" % translator
+ log.info( "filter by translator %s" % translator)
filters.append(QueryWrapperFilter(
srch.make_phrase(srch.get_tokens(translator, field='translators'),
field='translators')))
flt = srch.chain_filters(filters)
if title:
- print "hint by book title %s" % title
+ log.info( "hint by book title %s" % title)
q = srch.make_phrase(srch.get_tokens(title, field='title'), field='title')
hint.books(*srch.search_books(q, filt=flt))
toks = srch.get_tokens(query)
- print "tokens: %s" % toks
- # import pdb; pdb.set_trace()
+ log.info("tokens for query: %s" % toks)
+
results = SearchResult.aggregate(srch.search_perfect_book(toks, fuzzy=fuzzy, hint=hint),
srch.search_perfect_parts(toks, fuzzy=fuzzy, hint=hint),
srch.search_everywhere(toks, fuzzy=fuzzy, hint=hint))
results.sort(reverse=True)
- return [r.book for r in results]
+ books = []
+ for r in results:
+ try:
+ books.append(r.book)
+ except Book.DoesNotExist:
+ pass
+ log.info("books: %s" % books)
+ return books
else:
# Scenario 2: since we no longer have to figure out what the query term means to the user,
# we can just use filters and not the Hint class.