if not dictionary.check(t):
try:
- change[t] = dictionary.suggest(t)[0]
+ change_to = dictionary.suggest(t)[0].lower()
+ if change_to != t.lower():
+ change[t] = change_to
except IndexError:
pass
return query
+JVM.attachCurrentThread()
+search = Search()
def hint(request):
prefix = request.GET.get('term', '')
if len(prefix) < 2:
return JSONResponse([])
JVM.attachCurrentThread()
- s = Search()
- hint = s.hint()
+ hint = search.hint()
try:
tags = request.GET.get('tags', '')
hint.tags(Tag.get_tag_list(tags))
# jezeli tagi dot tylko ksiazki, to wazne zeby te nowe byly w tej samej ksiazce
# jesli zas dotycza themes, to wazne, zeby byly w tym samym fragmencie.
- tags = s.hint_tags(prefix, pdcounter=True)
- books = s.hint_books(prefix)
+ tags = search.hint_tags(prefix, pdcounter=True)
+ books = search.hint_books(prefix)
- # TODO DODAC TU HINTY
+ def category_name(c):
+ if c.startswith('pd_'):
+ c=c[len('pd_'):]
+ return _(c)
return JSONResponse(
[{'label': t.name,
- 'category': _(t.category),
+ 'category': category_name(t.category),
'id': t.id,
'url': t.get_absolute_url()}
for t in tags] + \
def main(request):
results = {}
JVM.attachCurrentThread() # where to put this?
- srch = Search()
results = None
query = None
- fuzzy = False
+ fuzzy = False #0.8
if 'q' in request.GET:
# tags = request.GET.get('tags', '')
# if book_id is not None:
# book = get_object_or_404(Book, id=book_id)
- # hint = srch.hint()
+ # hint = search.hint()
# try:
# tag_list = Tag.get_tag_list(tags)
# except:
# hint.tags(tag_list)
# if book:
# hint.books(book)
- tags = srch.hint_tags(query, pdcounter=True, prefix=False)
+ tags = search.hint_tags(query, pdcounter=True, prefix=False, fuzzy=fuzzy)
tags = split_tags(tags)
toks = StringReader(query)
tokens_cache = {}
- fuzzy = 'fuzzy' in request.GET
- if fuzzy:
- fuzzy = 0.7
- author_results = srch.search_phrase(toks, 'authors', fuzzy=fuzzy, tokens_cache=tokens_cache)
- title_results = srch.search_phrase(toks, 'title', fuzzy=fuzzy, tokens_cache=tokens_cache)
+ author_results = search.search_phrase(toks, 'authors', fuzzy=fuzzy, tokens_cache=tokens_cache)
+ title_results = search.search_phrase(toks, 'title', fuzzy=fuzzy, tokens_cache=tokens_cache)
# Boost main author/title results with mixed search, and save some of its results for end of list.
# boost author, title results
- author_title_mixed = srch.search_some(toks, ['authors', 'title', 'tags'], fuzzy=fuzzy, tokens_cache=tokens_cache)
+ author_title_mixed = search.search_some(toks, ['authors', 'title', 'tags'], fuzzy=fuzzy, tokens_cache=tokens_cache)
author_title_rest = []
for b in author_title_mixed:
bks = filter(lambda ba: ba.book_id == b.book_id, author_results + title_results)
# Do a phrase search but a term search as well - this can give us better snippets then search_everywhere,
# Because the query is using only one field.
text_phrase = SearchResult.aggregate(
- srch.search_phrase(toks, 'content', fuzzy=fuzzy, tokens_cache=tokens_cache, snippets=True, book=False, slop=4),
- srch.search_some(toks, ['content'], tokens_cache=tokens_cache, snippets=True, book=False))
+ search.search_phrase(toks, 'content', fuzzy=fuzzy, tokens_cache=tokens_cache, snippets=True, book=False, slop=4),
+ search.search_some(toks, ['content'], tokens_cache=tokens_cache, snippets=True, book=False))
- everywhere = srch.search_everywhere(toks, fuzzy=fuzzy, tokens_cache=tokens_cache)
+ everywhere = search.search_everywhere(toks, fuzzy=fuzzy, tokens_cache=tokens_cache)
def already_found(results):
def f(e):
re.subn(r"(^[ \t\n]+|[ \t\n]+$)", u"",
re.subn(r"[ \t\n]*\n[ \t\n]*", u"\n", s)[0])[0], h['snippets'])
- suggestion = did_you_mean(query, srch.get_tokens(toks, field="SIMPLE"))
+ suggestion = did_you_mean(query, search.get_tokens(toks, field="SIMPLE"))
print "dym? %s" % repr(suggestion).encode('utf-8')
results = author_results + title_results + text_phrase + everywhere
if len(results) == 1:
fragment_hits = filter(lambda h: 'fragment' in h, results[0].hits)
if len(fragment_hits) == 1:
- anchor = fragment_hits[0]['fragment']
- frag = Fragment.objects.get(anchor=anchor)
- return HttpResponseRedirect(frag.get_absolute_url())
+ #anchor = fragment_hits[0]['fragment']
+ #frag = Fragment.objects.get(anchor=anchor)
+ return HttpResponseRedirect(fragment_hits[0]['fragment'].get_absolute_url())
return HttpResponseRedirect(results[0].book.get_absolute_url())
elif len(results) == 0:
form = PublishingSuggestForm(initial={"books": query + ", "})