# -*- coding: utf-8 -*-
from django.conf import settings
-from lucene import SimpleFSDirectory, IndexWriter, CheckIndex, \
+from django.dispatch import Signal
+from lucene import SimpleFSDirectory, NIOFSDirectory, IndexWriter, IndexReader, IndexWriterConfig, CheckIndex, \
File, Field, Integer, \
NumericField, Version, Document, JavaError, IndexSearcher, \
QueryParser, PerFieldAnalyzerWrapper, \
from librarian.parser import WLDocument
from lxml import etree
import catalogue.models
-from pdcounter.models import Author as PDCounterAuthor
+from pdcounter.models import Author as PDCounterAuthor, BookStub as PDCounterBook
from multiprocessing.pool import ThreadPool
from threading import current_thread
import atexit
"""
def __init__(self):
self.make_index_dir()
- self.store = SimpleFSDirectory(File(settings.SEARCH_INDEX))
+ self.store = NIOFSDirectory(File(settings.SEARCH_INDEX))
def make_index_dir(self):
try:
pass
else: raise
+ def close(self):
+ self.store.close()
+
class IndexChecker(IndexStore):
def __init__(self):
"""
SNIPPET_DIR = "snippets"
- def __init__(self, book_id):
+ def __init__(self, book_id, revision=None):
try:
os.makedirs(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR))
except OSError as exc:
pass
else: raise
self.book_id = book_id
+ self.revision = revision
self.file = None
+ @property
+ def path(self):
+ if self.revision: fn = "%d.%d" % (self.book_id, self.revision)
+ else: fn = "%d" % self.book_id
+
+ return os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR, fn)
+
def open(self, mode='r'):
"""
Open the snippet file. Call .close() afterwards.
"""
if not 'b' in mode:
mode += 'b'
- self.file = open(os.path.join(settings.SEARCH_INDEX, self.SNIPPET_DIR, str(self.book_id)), mode)
+
+ if 'w' in mode:
+ if os.path.exists(self.path):
+ self.revision = 1
+ while True:
+ if not os.path.exists(self.path):
+ break
+ self.revision += 1
+ print "using %s" % self.path
+
+ self.file = open(self.path, mode)
self.position = 0
return self
"""Close snippet file"""
self.file.close()
+ def remove(self):
+ self.revision = None
+ try:
+ os.unlink(self.path)
+ self.revision = 0
+ while True:
+ self.revision += 1
+ os.unlink(self.path)
+ except OSError:
+ pass
+
class BaseIndex(IndexStore):
"""
analyzer = WLAnalyzer()
self.analyzer = analyzer
- def open(self, analyzer=None):
+ def open(self, timeout=None):
if self.index:
raise Exception("Index is already opened")
- self.index = IndexWriter(self.store, self.analyzer,\
- IndexWriter.MaxFieldLength.LIMITED)
+ conf = IndexWriterConfig(Version.LUCENE_34, self.analyzer)
+ if timeout:
+ conf.setWriteLockTimeout(long(timeout))
+ self.index = IndexWriter(self.store, conf)
return self.index
def optimize(self):
self.index.close()
self.index = None
+ index_changed.send_robust(self)
+
+ super(BaseIndex, self).close()
+
def __enter__(self):
self.open()
return self
self.close()
+index_changed = Signal()
+
+
class Index(BaseIndex):
"""
Class indexing books.
def __init__(self, analyzer=None):
super(Index, self).__init__(analyzer)
- def index_tags(self):
+ def index_tags(self, *tags, **kw):
"""
Re-index global tag list.
Removes all tags from index, then index them again.
Indexed fields include: id, name (with and without polish stems), category
"""
- q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True)
- self.index.deleteDocuments(q)
+ remove_only = kw.get('remove_only', False)
+ # first, remove tags from index.
+ if tags:
+ q = BooleanQuery()
+ for tag in tags:
+ b_id_cat = BooleanQuery()
+
+ q_id = NumericRangeQuery.newIntRange("tag_id", tag.id, tag.id, True, True)
+ b_id_cat.add(q_id, BooleanClause.Occur.MUST)
- for tag in catalogue.models.Tag.objects.all():
- doc = Document()
- doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id)))
- doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED))
- doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED))
- doc.add(Field("tag_category", tag.category, Field.Store.NO, Field.Index.NOT_ANALYZED))
- self.index.addDocument(doc)
-
- for pdtag in PDCounterAuthor.objects.all():
- doc = Document()
- doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(pdtag.id)))
- doc.add(Field("tag_name", pdtag.name, Field.Store.NO, Field.Index.ANALYZED))
- doc.add(Field("tag_name_pl", pdtag.name, Field.Store.NO, Field.Index.ANALYZED))
- doc.add(Field("tag_category", 'pdcounter', Field.Store.NO, Field.Index.NOT_ANALYZED))
- doc.add(Field("is_pdcounter", 'true', Field.Store.YES, Field.Index.NOT_ANALYZED))
- self.index.addDocument(doc)
+ if isinstance(tag, PDCounterAuthor):
+ q_cat = TermQuery(Term('tag_category', 'pd_author'))
+ elif isinstance(tag, PDCounterBook):
+ q_cat = TermQuery(Term('tag_category', 'pd_book'))
+ else:
+ q_cat = TermQuery(Term('tag_category', tag.category))
+ b_id_cat.add(q_cat, BooleanClause.Occur.MUST)
+
+ q.add(b_id_cat, BooleanClause.Occur.SHOULD)
+ else: # all
+ q = NumericRangeQuery.newIntRange("tag_id", 0, Integer.MAX_VALUE, True, True)
+ self.index.deleteDocuments(q)
+
+ if not remove_only:
+ # then add them [all or just one passed]
+ if not tags:
+ tags = catalogue.models.Tag.objects.exclude(category='set') + \
+ PDCounterAuthor.objects.all() + \
+ PDCounterBook.objects.all()
+
+ for tag in tags:
+ if isinstance(tag, PDCounterAuthor):
+ doc = Document()
+ doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id)))
+ doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_category", 'pd_author', Field.Store.YES, Field.Index.NOT_ANALYZED))
+ doc.add(Field("is_pdcounter", 'true', Field.Store.YES, Field.Index.NOT_ANALYZED))
+ self.index.addDocument(doc)
+ elif isinstance(tag, PDCounterBook):
+ doc = Document()
+ doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id)))
+ doc.add(Field("tag_name", tag.title, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_name_pl", tag.title, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_category", 'pd_book', Field.Store.YES, Field.Index.NOT_ANALYZED))
+ doc.add(Field("is_pdcounter", 'true', Field.Store.YES, Field.Index.NOT_ANALYZED))
+ self.index.addDocument(doc)
+ else:
+ doc = Document()
+ doc.add(NumericField("tag_id", Field.Store.YES, True).setIntValue(int(tag.id)))
+ doc.add(Field("tag_name", tag.name, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_name_pl", tag.name, Field.Store.NO, Field.Index.ANALYZED))
+ doc.add(Field("tag_category", tag.category, Field.Store.NO, Field.Index.NOT_ANALYZED))
+ self.index.addDocument(doc)
def create_book_doc(self, book):
"""
doc.add(NumericField("parent_id", Field.Store.YES, True).setIntValue(int(book.parent.id)))
return doc
- def remove_book(self, book):
+ def remove_book(self, book, remove_snippets=True):
"""Removes a book from search index.
book - Book instance."""
q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True, True)
self.index.deleteDocuments(q)
+ if remove_snippets:
+ snippets = Snippets(book.id)
+ snippets.remove()
+
def index_book(self, book, book_info=None, overwrite=True):
"""
Indexes the book.
and calls self.index_content() to index the contents of the book.
"""
if overwrite:
- self.remove_book(book)
+ # we don't remove snippets, since they might be still needed by
+ # threads using not reopened index
+ self.remove_book(book, remove_snippets=False)
book_doc = self.create_book_doc(book)
meta_fields = self.extract_metadata(book, book_info)
book_doc.add(elem)
else:
book_doc.add(f)
-
self.index.addDocument(book_doc)
del book_doc
(dt.year, dt.month, dt.day), Field.Store.NO, Field.Index.NOT_ANALYZED)
# get published date
- source = book_info.source_name
- match = self.published_date_re.search(source)
- if match is not None:
- fields["published_date"] = Field("published_date", str(match.groups()[0]), Field.Store.YES, Field.Index.NOT_ANALYZED)
+ pd = None
+ if hasattr(book_info, 'source_name') and book_info.source_name:
+ match = self.published_date_re.search(book_info.source_name)
+ if match is not None:
+ pd = str(match.groups()[0])
+ if not pd: pd = ""
+ fields["published_date"] = Field("published_date", pd, Field.Store.YES, Field.Index.NOT_ANALYZED)
return fields
return []
def walker(node, ignore_tags=[]):
- yield node, None
- for child in filter(lambda n: n.tag not in ignore_tags, list(node)):
- for b, e in walker(child):
- yield b, e
- yield None, node
+
+ if node.tag not in ignore_tags:
+ yield node, None, None
+ if node.text is not None:
+ yield None, node.text, None
+ for child in list(node):
+ for b, t, e in walker(child):
+ yield b, t, e
+ yield None, None, node
+
+ if node.tail is not None:
+ yield None, node.tail, None
return
def fix_format(text):
.setIntValue('header_span' in fields and fields['header_span'] or 1))
doc.add(Field('header_type', fields["header_type"], Field.Store.YES, Field.Index.NOT_ANALYZED))
+ print ">>[%s]>%s<<<" % (fields.get('fragment_anchor', ''), fields['content'])
+
doc.add(Field('content', fields["content"], Field.Store.NO, Field.Index.ANALYZED, \
Field.TermVector.WITH_POSITIONS_OFFSETS))
snip_pos = snippets.add(fields["content"])
doc.add(NumericField("snippets_position", Field.Store.YES, True).setIntValue(snip_pos[0]))
doc.add(NumericField("snippets_length", Field.Store.YES, True).setIntValue(snip_pos[1]))
+ if snippets.revision:
+ doc.add(NumericField("snippets_revision", Field.Store.YES, True).setIntValue(snippets.revision))
if 'fragment_anchor' in fields:
doc.add(Field("fragment_anchor", fields['fragment_anchor'],
# section content
content = []
- footnote = None
-
- for start, end in walker(header, ignore_tags=self.ignore_content_tags):
- # handle footnotes
- # if start is not None and start.tag in self.footnote_tags:
- # footnote = ' '.join(start.itertext())
- # elif end is not None and footnote is not None and end.tag in self.footnote_tags:
- # doc = add_part(snippets, header_index=position, header_type=header.tag,
- # content=footnote)
+ footnote = []
- # self.index.addDocument(doc)
+ def all_content(text):
+ for frag in fragments.values():
+ frag['content'].append(text)
+ content.append(text)
+ handle_text = [all_content]
- # footnote = None
+ for start, text, end in walker(header, ignore_tags=self.ignore_content_tags):
+ # handle footnotes
+ if start is not None and start.tag in self.footnote_tags:
+ footnote = []
+ def collect_footnote(t):
+ footnote.append(t)
+ handle_text.append(collect_footnote)
+ elif end is not None and footnote is not [] and end.tag in self.footnote_tags:
+ handle_text.pop()
+ doc = add_part(snippets, header_index=position, header_type=header.tag,
+ content=u''.join(footnote),
+ is_footnote=Field("is_footnote", 'true', Field.Store.NO, Field.Index.NOT_ANALYZED))
+
+ self.index.addDocument(doc)
+ #print "@ footnote text: %s" % footnote
+ footnote = []
+
# handle fragments and themes.
if start is not None and start.tag == 'begin':
fid = start.attrib['id'][1:]
fragments[fid] = {'content': [], 'themes': [], 'start_section': position, 'start_header': header.tag}
+ # themes for this fragment
elif start is not None and start.tag == 'motyw':
fid = start.attrib['id'][1:]
+ handle_text.append(None)
if start.text is not None:
fragments[fid]['themes'] += map(str.strip, map(give_me_utf8, start.text.split(',')))
+ elif end is not None and end.tag == 'motyw':
+ handle_text.pop()
elif start is not None and start.tag == 'end':
fid = start.attrib['id'][1:]
if fid not in fragments:
continue # a broken <end> node, skip it
- # import pdb; pdb.set_trace()
frag = fragments[fid]
if frag['themes'] == []:
continue # empty themes list.
fragment_anchor=fid,
content=fix_format(frag['content']),
themes=frag['themes'])
-
+ #print '@ FRAG %s' % frag['content']
self.index.addDocument(doc)
# Collect content.
- elif start is not None:
- for frag in fragments.values():
- frag['content'].append(start.text)
- content.append(start.text)
- elif end is not None:
- for frag in fragments.values():
- frag['content'].append(end.tail)
- content.append(end.tail)
+
+ if text is not None and handle_text is not []:
+ hdl = handle_text[-1]
+ if hdl is not None:
+ hdl(text)
# in the end, add a section text.
doc = add_part(snippets, header_index=position, header_type=header.tag,
content=fix_format(content))
+ #print '@ CONTENT: %s' % fix_format(content)
self.index.addDocument(doc)
"""
index = None
- def open(self, analyzer=None, threads=4):
- if ReusableIndex.index is not None:
+ def open(self, analyzer=None, **kw):
+ if ReusableIndex.index:
self.index = ReusableIndex.index
else:
print("opening index")
- Index.open(self, analyzer)
+ Index.open(self, analyzer, **kw)
ReusableIndex.index = self.index
atexit.register(ReusableIndex.close_reusable)
@staticmethod
def close_reusable():
- if ReusableIndex.index is not None:
+ if ReusableIndex.index:
+ print("closing index")
ReusableIndex.index.optimize()
ReusableIndex.index.close()
ReusableIndex.index = None
+ index_changed.send_robust(None)
+
def close(self):
- pass
+ if ReusableIndex.index:
+ ReusableIndex.index.commit()
class JoinSearch(object):
self.book_id = int(stored.get("book_id"))
pd = stored.get("published_date")
- if pd is None:
- pd = 0
- self.published_date = int(pd)
+ try:
+ self.published_date = int(pd)
+ except ValueError:
+ self.published_date = 0
header_type = stored.get("header_type")
# we have a content hit in some header of fragment
return self
def get_book(self):
+ if hasattr(self, '_book'):
+ return self._book
return catalogue.models.Book.objects.get(id=self.book_id)
book = property(get_book)
# to sections and fragments
frags = filter(lambda r: r[FRAGMENT] is not None, self._hits)
+
sect = filter(lambda r: r[FRAGMENT] is None, self._hits)
+
+ # sections not covered by fragments
sect = filter(lambda s: 0 == len(filter(
lambda f: s[POSITION][POSITION_INDEX] >= f[POSITION][POSITION_INDEX]
and s[POSITION][POSITION_INDEX] < f[POSITION][POSITION_INDEX] + f[POSITION][POSITION_SPAN],
hits = []
- # remove duplicate fragments
- fragments = {}
- for f in frags:
- fid = f[FRAGMENT]
- if fid in fragments:
- if fragments[fid][SCORE] >= f[SCORE]:
- continue
- fragments[fid] = f
- frags = fragments.values()
+ def remove_duplicates(lst, keyfn, compare):
+ els = {}
+ for e in lst:
+ eif = keyfn(e)
+ if eif in els:
+ if compare(els[eif], e) >= 1:
+ continue
+ els[eif] = e
+ return els.values()
+
+ # remove fragments with duplicated fid's and duplicated snippets
+ frags = remove_duplicates(frags, lambda f: f[FRAGMENT], lambda a, b: cmp(a[SCORE], b[SCORE]))
+ frags = remove_duplicates(frags, lambda f: f[OTHER]['snippets'] and f[OTHER]['snippets'][0] or hash(f),
+ lambda a, b: cmp(a[SCORE], b[SCORE]))
# remove duplicate sections
sections = {}
for f in frags:
try:
- frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT])
+ frag = catalogue.models.Fragment.objects.get(anchor=f[FRAGMENT], book__id=self.book_id)
except catalogue.models.Fragment.DoesNotExist:
# stale index
continue
def __cmp__(self, other):
c = cmp(self.score, other.score)
if c == 0:
- if not hasattr(other,'published_date') or not hasattr(self, 'published_date'):
- import pdb; pdb.set_trace()
# this is inverted, because earlier date is better
return cmp(other.published_date, self.published_date)
else:
IndexStore.__init__(self)
self.analyzer = WLAnalyzer() # PolishAnalyzer(Version.LUCENE_34)
# self.analyzer = WLAnalyzer()
- self.searcher = IndexSearcher(self.store, True)
+ reader = IndexReader.open(self.store, True)
+ self.searcher = IndexSearcher(reader)
self.parser = QueryParser(Version.LUCENE_34, default_field,
self.analyzer)
self.parent_filter = TermsFilter()
self.parent_filter.addTerm(Term("is_book", "true"))
+ index_changed.connect(self.reopen)
+
+ def close(self):
+ reader = self.searcher.getIndexReader()
+ self.searcher.close()
+ reader.close()
+ super(Search, self).close()
+ index_changed.disconnect(self.reopen)
+
+ def reopen(self, **unused):
+ reader = self.searcher.getIndexReader()
+ rdr = reader.reopen()
+ print "got signal to reopen index"
+ if not rdr.equals(reader):
+ print "will reopen index"
+ oldsearch = self.searcher
+ self.searcher = IndexSearcher(rdr)
+ oldsearch.close()
+ reader.close()
def query(self, query):
"""Parse query in default Lucene Syntax. (for humans)
length = stored.get('snippets_length')
if position is None or length is None:
return None
+ revision = stored.get('snippets_revision')
+ if revision: revision = int(revision)
# locate content.
- snippets = Snippets(stored.get('book_id')).open()
+ book_id = int(stored.get('book_id'))
+ snippets = Snippets(book_id, revision=revision).open()
try:
- text = snippets.get((int(position),
- int(length)))
- finally:
- snippets.close()
+ try:
+ text = snippets.get((int(position),
+ int(length)))
+ finally:
+ snippets.close()
- tokenStream = TokenSources.getAnyTokenStream(self.searcher.getIndexReader(), scoreDoc.doc, field, self.analyzer)
- # highlighter.getBestTextFragments(tokenStream, text, False, 10)
- snip = highlighter.getBestFragments(tokenStream, text, 3, "...")
+ tokenStream = TokenSources.getAnyTokenStream(self.searcher.getIndexReader(), scoreDoc.doc, field, self.analyzer)
+ # highlighter.getBestTextFragments(tokenStream, text, False, 10)
+ snip = highlighter.getBestFragments(tokenStream, text, 3, "...")
+ except Exception, e:
+ e2 = e
+ if hasattr(e, 'getJavaException'):
+ e2 = unicode(e.getJavaException())
+ raise Exception("Problem fetching snippets for book %d, @%d len=%d" % (book_id, int(position), int(length)),
+ e2)
return snip
@staticmethod
for found in tops.scoreDocs:
doc = self.searcher.doc(found.doc)
is_pdcounter = doc.get('is_pdcounter')
- if is_pdcounter:
- tag = PDCounterAuthor.objects.get(id=doc.get('tag_id'))
- else:
- tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id"))
- # don't add the pdcounter tag if same tag already exists
- if not (is_pdcounter and filter(lambda t: tag.slug == t.slug, tags)):
- tags.append(tag)
+ category = doc.get('tag_category')
+ try:
+ if is_pdcounter == 'true':
+ if category == 'pd_author':
+ tag = PDCounterAuthor.objects.get(id=doc.get('tag_id'))
+ elif category == 'pd_book':
+ tag = PDCounterBook.objects.get(id=doc.get('tag_id'))
+ tag.category = 'pd_book' # make it look more lik a tag.
+ else:
+ print "Warning. cannot get pdcounter tag_id=%d from db; cat=%s" % (int(doc.get('tag_id')), category)
+ else:
+ tag = catalogue.models.Tag.objects.get(id=doc.get("tag_id"))
+ # don't add the pdcounter tag if same tag already exists
+ if not (is_pdcounter and filter(lambda t: tag.slug == t.slug, tags)):
+ tags.append(tag)
+ except catalogue.models.Tag.DoesNotExist: pass
+ except PDCounterAuthor.DoesNotExist: pass
+ except PDCounterBook.DoesNotExist: pass
+
# print "%s (%d) -> %f" % (tag, tag.id, found.score)
print 'returning %s' % tags
return tags
tops = self.searcher.search(query, filter, max_results)
for found in tops.scoreDocs:
doc = self.searcher.doc(found.doc)
- bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
+ try:
+ bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
+ except catalogue.models.Book.DoesNotExist: pass
return bks
def make_prefix_phrase(self, toks, field):
return only_term
- def hint_tags(self, string, max_results=50, pdcounter=True, prefix=True):
+ def hint_tags(self, string, max_results=50, pdcounter=True, prefix=True, fuzzy=False):
"""
Return auto-complete hints for tags
using prefix search.
if prefix:
q = self.make_prefix_phrase(toks, field)
else:
- q = self.make_term_query(toks, field)
+ q = self.make_term_query(toks, field, fuzzy=fuzzy)
top.add(BooleanClause(q, BooleanClause.Occur.SHOULD))
no_book_cat = self.term_filter(Term("tag_category", "book"), inverse=True)
return self.search_tags(top, no_book_cat, max_results=max_results, pdcounter=pdcounter)
- def hint_books(self, string, max_results=50, prefix=True):
+ def hint_books(self, string, max_results=50, prefix=True, fuzzy=False):
"""
Returns auto-complete hints for book titles
Because we do not index 'pseudo' title-tags.
if prefix:
q = self.make_prefix_phrase(toks, 'title')
else:
- q = self.make_term_query(toks, 'title')
+ q = self.make_term_query(toks, 'title', fuzzy=fuzzy)
return self.search_books(q, self.term_filter(Term("is_book", "true")), max_results=max_results)
fnp / wolnelektury.git / blobdiff