Minor fixes.
[wolnelektury.git] / src / search / custom.py
index b3b704d..9337157 100644 (file)
-# -*- coding: utf-8 -*-
 # This file is part of Wolnelektury, licensed under GNU Affero GPLv3 or later.
 # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
 #
-from sunburnt import sunburnt
-from lxml import etree
-import urllib
+import re
+from urllib.parse import urlencode
 import warnings
-from sunburnt import search
-import copy
 from httplib2 import socket
-import re
+from lxml import etree
+from scorched import connection, exc, search
 
 
-class TermVectorOptions(search.Options):
-    def __init__(self, schema, original=None):
-        self.schema = schema
-        if original is None:
-            self.fields = set()
-            self.positions = False
-        else:
-            self.fields = copy.copy(original.fields)
-            self.positions = copy.copy(original.positions)
-
-    def update(self, positions=False, fields=None):
-        if fields is None:
-            fields = []
-        if isinstance(fields, basestring):
-            fields = [fields]
-        self.schema.check_fields(fields, {"stored": True})
-        self.fields.update(fields)
-        self.positions = positions
-
-    def options(self):
-        opts = {}
-        if self.positions or self.fields:
-            opts['tv'] = 'true'
-        if self.positions:
-            opts['tv.positions'] = 'true'
-        if self.fields:
-            opts['tv.fl'] = ','.join(sorted(self.fields))
-        return opts
-
-
-class CustomSolrConnection(sunburnt.SolrConnection):
+class CustomSolrConnection(connection.SolrConnection):
     def __init__(self, *args, **kw):
         super(CustomSolrConnection, self).__init__(*args, **kw)
         self.analysis_url = self.url + "analysis/field/"
 
     def analyze(self, params):
-        qs = urllib.urlencode(params)
+        qs = urlencode(params)
         url = "%s?%s" % (self.analysis_url, qs)
         if len(url) > self.max_length_get_url:
-            warnings.warn("Long query URL encountered - POSTing instead of "
-                "GETting. This query will not be cached at the HTTP layer")
+            warnings.warn("Long query URL encountered - POSTing instead of GETting. "
+                          "This query will not be cached at the HTTP layer")
             url = self.analysis_url
             kwargs = dict(
                 method="POST",
-                body=qs,
+                data=qs,
                 headers={"Content-Type": "application/x-www-form-urlencoded"},
             )
         else:
             kwargs = dict(method="GET")
-        r, c = self.request(url, **kwargs)
-        if r.status != 200:
-            raise sunburnt.SolrError(r, c)
-        return c
-
-
-# monkey patching sunburnt SolrSearch
-search.SolrSearch.option_modules += ('term_vectorer',)
+        response = self.request(url=url, **kwargs)
+        if response.status_code != 200:
+            raise exc.SolrError(response)
+        return response.content
 
 
-def __term_vector(self, positions=False, fields=None):
-    newself = self.clone()
-    newself.term_vectorer.update(positions, fields)
-    return newself
-setattr(search.SolrSearch, 'term_vector', __term_vector)
-
-
-def __patched__init_common_modules(self):
-    __original__init_common_modules(self)
-    self.term_vectorer = TermVectorOptions(self.schema)
-__original__init_common_modules = search.SolrSearch._init_common_modules
-setattr(search.SolrSearch, '_init_common_modules', __patched__init_common_modules)
-
-
-class CustomSolrInterface(sunburnt.SolrInterface):
+class CustomSolrInterface(connection.SolrInterface):
     # just copied from parent and SolrConnection -> CustomSolrConnection
-    def __init__(self, url, schemadoc=None, http_connection=None, mode='', retry_timeout=-1, max_length_get_url=sunburnt.MAX_LENGTH_GET_URL):
-        self.conn = CustomSolrConnection(url, http_connection, retry_timeout, max_length_get_url)
-        self.schemadoc = schemadoc
-        if 'w' not in mode:
-            self.writeable = False
-        elif 'r' not in mode:
-            self.readable = False
-        try:
-            self.init_schema()
-        except socket.error, e:
-            raise socket.error, "Cannot connect to Solr server, and search indexing is enabled (%s)" % str(e)
+    def __init__(self, url, http_connection=None, mode='',
+                 retry_timeout=-1, max_length_get_url=connection.MAX_LENGTH_GET_URL,
+                 search_timeout=()):
+        """
+        :param url: url to Solr
+        :type url: str
+        :param http_connection: optional -- already existing connection
+        :type http_connection: requests connection
+        :param mode: optional -- mode (readable, writable) Solr
+        :type mode: str
+        :param retry_timeout: optional -- timeout until retry
+        :type retry_timeout: int
+        :param max_length_get_url: optional -- max length until switch to post
+        :type max_length_get_url: int
+        :param search_timeout: (optional) How long to wait for the server to
+                               send data before giving up, as a float, or a
+                               (connect timeout, read timeout) tuple.
+        :type search_timeout: float or tuple
+        """
+
+        self.conn = CustomSolrConnection(
+            url, http_connection, mode, retry_timeout, max_length_get_url)
+        self.schema = self.init_schema()
+        self._datefields = self._extract_datefields(self.schema)
+
 
     def _analyze(self, **kwargs):
-        if not self.readable:
+        if not self.conn.readable:
             raise TypeError("This Solr instance is only for writing")
         args = {
             'analysis_showmatch': True
             }
-        if 'field' in kwargs: args['analysis_fieldname'] = kwargs['field']
-        if 'text' in kwargs: args['analysis_fieldvalue'] = kwargs['text']
-        if 'q' in kwargs: args['q'] = kwargs['q']
-        if 'query' in kwargs: args['q'] = kwargs['q']
-
-        params = map(lambda (k, v): (k.replace('_', '.'), v), sunburnt.params_from_dict(**args))
+        if 'field' in kwargs:
+            args['analysis_fieldname'] = kwargs['field']
+        if 'text' in kwargs:
+            args['analysis_fieldvalue'] = kwargs['text']
+        if 'q' in kwargs:
+            args['q'] = kwargs['q']
+        if 'query' in kwargs:
+            args['q'] = kwargs['q']
+
+        params = [
+            (k.replace('_', '.'), v)
+            for (k, v) in search.params_from_dict(**args)
+        ]
 
         content = self.conn.analyze(params)
         doc = etree.fromstring(content)
@@ -126,16 +96,15 @@ class CustomSolrInterface(sunburnt.SolrInterface):
             matches.add((start, end))
 
         if matches:
-            return self.substring(kwargs['text'], matches,
-                margins=kwargs.get('margins', 30),
-                mark=kwargs.get('mark', ("<b>", "</b>")))
+            return self.substring(
+                kwargs['text'], matches, margins=kwargs.get('margins', 30), mark=kwargs.get('mark', ("<b>", "</b>")))
         else:
             return None
 
     def analyze(self, **kwargs):
         doc = self._analyze(**kwargs)
         terms = doc.xpath("//lst[@name='index']/arr[last()]/lst/str[1]")
-        terms = map(lambda n: unicode(n.text), terms)
+        terms = map(lambda n: str(n.text), terms)
         return terms
 
     def expand_margins(self, text, start, end):
@@ -155,37 +124,31 @@ class CustomSolrInterface(sunburnt.SolrInterface):
                 break
             end += 1
 
-        return (start, end)
+        return start, end
 
     def substring(self, text, matches, margins=30, mark=("<b>", "</b>")):
-        start = None
-        end = None
         totlen = len(text)
-        matches_margins = map(lambda (s, e):
-                              ((s, e),
-                               (max(0, s - margins), min(totlen, e + margins))),
-                                  matches)
-        matches_margins = map(lambda (m, (s, e)):
-                              (m, self.expand_margins(text, s, e)),
-            matches_margins)
-
-            # lets start with first match
+        matches_margins = [
+            ((s, e), self.expand_margins(text, max(0, s - margins), min(totlen, e + margins))) for s, e in matches]
+
+        # lets start with first match
         (start, end) = matches_margins[0][1]
-        matches = [matches_margins[0][0]]
+        new_matches = [matches_margins[0][0]]
 
         for (m, (s, e)) in matches_margins[1:]:
             if end < s or start > e:
                 continue
             start = min(start, s)
             end = max(end, e)
-            matches.append(m)
+            new_matches.append(m)
 
         snip = text[start:end]
-        matches.sort(lambda a, b: cmp(b[0], a[0]))
+        new_matches.sort(key=lambda a: -a[0])
 
-        for (s, e) in matches:
-            off = - start
+        for (s, e) in new_matches:
+            off = -start
             snip = snip[:e + off] + mark[1] + snip[e + off:]
             snip = snip[:s + off] + mark[0] + snip[s + off:]
+        snip = re.sub('%s[ \t\n]+%s' % (mark[1], mark[0]), " ", snip)
 
         return snip