# Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
#
from django.conf import settings
-from django.http.response import HttpResponseRedirect
from django.shortcuts import render
from django.views.decorators import cache
from django.http import HttpResponse, JsonResponse
from catalogue.models import Book, Tag
-from pdcounter.models import Author
-from picture.models import Picture
-from search.index import Search, SearchResult, PictureResult
-from suggest.forms import PublishingSuggestForm
+from .forms import SearchFilters
import re
import json
from wolnelektury.utils import re_escape
-def match_word_re(word):
- if 'sqlite' in settings.DATABASES['default']['ENGINE']:
- return r"\b%s\b" % word
- elif 'mysql' in settings.DATABASES['default']['ENGINE']:
- return "[[:<:]]%s[[:>:]]" % word
-
-
query_syntax_chars = re.compile(r"[\\/*:(){}?.[\]+]")
return query_syntax_chars.sub(replace, query)
-def did_you_mean(query, tokens):
- return query
- # change = {}
- # for t in tokens:
- # authors = Tag.objects.filter(category='author', name__iregex=match_word_re(t))
- # if len(authors) > 0:
- # continue
-
- # if False:
- # if not dictionary.check(t):
- # try:
- # change_to = dictionary.suggest(t)[0].lower()
- # if change_to != t.lower():
- # change[t] = change_to
- # except IndexError:
- # pass
-
- # if change == {}:
- # return None
-
- # for frm, to in change.items():
- # query = query.replace(frm, to)
-
- # return query
-
-
@cache.never_cache
-def hint(request):
- prefix = request.GET.get('term', '')
+def hint(request, mozhint=False, param='term'):
+ prefix = request.GET.get(param, '')
if len(prefix) < 2:
return JsonResponse([], safe=False)
for author in authors[:limit]
]
if len(data) < limit:
- data += [
- {
- 'label': b.title,
- 'author': b.author_unicode(),
- 'id': b.id,
- 'url': b.get_absolute_url()
- }
- for b in Book.objects.filter(findable=True, title__iregex='\m' + prefix)[:limit-len(data)]
+ for b in Book.objects.filter(findable=True, title__iregex='\m' + prefix)[:limit-len(data)]:
+ author_str = b.author_unicode()
+ translator = b.translator()
+ if translator:
+ author_str += ' (tłum. ' + translator + ')'
+ data.append(
+ {
+ 'label': b.title,
+ 'author': author_str,
+ 'id': b.id,
+ 'url': b.get_absolute_url()
+ }
+ )
+
+ if mozhint:
+ data = [
+ prefix,
+ [
+ item['label']
+ for item in data
+ ]
]
+
callback = request.GET.get('callback', None)
if callback:
return HttpResponse("%s(%s);" % (callback, json.dumps(data)),
return JsonResponse(data, safe=False)
+
@cache.never_cache
-def main(request):
- query = request.GET.get('q', '')
- if len(query) < 2:
- return render(
- request, 'catalogue/search_too_short.html',
- {'prefix': query})
- elif len(query) > 256:
- return render(
- request, 'catalogue/search_too_long.html',
- {'prefix': query})
-
- query = prepare_query(query)
- pd_authors = search_pd_authors(query)
- books = search_books(query)
- pictures = search_pictures(query)
- suggestion = ''
-
- if not (books or pictures or pd_authors):
- form = PublishingSuggestForm(initial={"books": query + ", "})
- return render(
- request,
- 'catalogue/search_no_hits.html',
- {
- 'form': form,
- 'did_you_mean': suggestion
- })
-
- if not (books or pictures) and len(pd_authors) == 1:
- return HttpResponseRedirect(pd_authors[0].get_absolute_url())
-
- return render(
- request,
- 'catalogue/search_multiple_hits.html',
- {
- 'pd_authors': pd_authors,
- 'books': books,
- 'pictures': pictures,
- 'did_you_mean': suggestion
- })
-
-def search_books(query):
- search = Search()
- results_parts = []
- search_fields = []
- words = query.split()
- fieldsets = (
- (['authors', 'authors_nonstem'], True),
- (['title', 'title_nonstem'], True),
- (['metadata', 'metadata_nonstem'], True),
- (['text', 'text_nonstem', 'themes_pl', 'themes_pl_nonstem'], False),
- )
- for fields, is_book in fieldsets:
- search_fields += fields
- results_parts.append(search.search_words(words, search_fields, required=fields, book=is_book))
- results = []
- ids_results = {}
- for results_part in results_parts:
- for result in sorted(SearchResult.aggregate(results_part), reverse=True):
- book_id = result.book_id
- if book_id in ids_results:
- ids_results[book_id].merge(result)
- else:
- results.append(result)
- ids_results[book_id] = result
- descendant_ids = set(
- Book.objects.filter(id__in=ids_results, ancestor__in=ids_results).values_list('id', flat=True))
- results = [result for result in results if result.book_id not in descendant_ids]
- for result in results:
- search.get_snippets(result, query, num=3)
-
- def ensure_exists(r):
- try:
- return r.book
- except Book.DoesNotExist:
- return False
-
- results = [r for r in results if ensure_exists(r)]
- return results
-
-
-def search_pictures(query):
- search = Search()
- results_parts = []
- search_fields = []
- words = query.split()
- fieldsets = (
- (['authors', 'authors_nonstem'], True),
- (['title', 'title_nonstem'], True),
- (['metadata', 'metadata_nonstem'], True),
- (['themes_pl', 'themes_pl_nonstem'], False),
- )
- for fields, is_book in fieldsets:
- search_fields += fields
- results_parts.append(search.search_words(words, search_fields, required=fields, book=is_book, picture=True))
- results = []
- ids_results = {}
- for results_part in results_parts:
- for result in sorted(PictureResult.aggregate(results_part), reverse=True):
- picture_id = result.picture_id
- if picture_id in ids_results:
- ids_results[picture_id].merge(result)
- else:
- results.append(result)
- ids_results[picture_id] = result
-
- def ensure_exists(r):
- try:
- return r.picture
- except Picture.DoesNotExist:
- return False
-
- results = [r for r in results if ensure_exists(r)]
- return results
-
-
-def search_pd_authors(query):
- pd_authors = Author.objects.filter(name__icontains=query)
- existing_slugs = Tag.objects.filter(
- category='author', slug__in=list(pd_authors.values_list('slug', flat=True))) \
- .values_list('slug', flat=True)
- pd_authors = pd_authors.exclude(slug__in=existing_slugs)
- return pd_authors
-
-
-def prepare_query(query):
- query = ' '.join(query.split())
- # filter out private use characters
- import unicodedata
- query = ''.join(ch for ch in query if unicodedata.category(ch) != 'Co')
- query = remove_query_syntax_chars(query)
-
- words = query.split()
- if len(words) > 10:
- query = ' '.join(words[:10])
- return query
+def search(request):
+ filters = SearchFilters(request.GET)
+ ctx = {
+ 'title': 'Wynik wyszukiwania',
+ 'query': request.GET.get('q', ''),
+ 'filters': filters,
+ }
+ if filters.is_valid():
+ ctx['results'] = filters.results()
+ for k, v in ctx['results'].items():
+ if v:
+ ctx['hasresults'] = True
+ break
+ return render(request, 'search/results.html', ctx)