# -*- mode: python; indent-tabs-mode: nil; -*- coding: iso-8859-1 -*- """ BaseSearcher.py Copyright 2009-2014 by Marcello Perathoner Distributable under the GNU General Public License Version 3 or newer. Project Gutenberg Catalog Search Base class """ from __future__ import unicode_literals from __future__ import division import datetime from six.moves import urllib import logging import threading from collections import defaultdict import cherrypy import routes import babel import regex # module re does not support \p{L} import six from libgutenberg.MediaTypes import mediatypes as mt from libgutenberg.GutenbergDatabase import xl, DatabaseError from libgutenberg import DublinCore from libgutenberg import GutenbergDatabaseDublinCore from libgutenberg import GutenbergGlobals as gg from i18n_tool import ugettext as _ from i18n_tool import ungettext as __ import DublinCoreI18n from SupportedLocales import FB_LANGS, TWITTER_LANGS, GOOGLE_LANGS, PAYPAL_LANGS, FLATTR_LANGS VALID_PROTOCOLS = ('http', 'https') MEDIATYPE_TO_FORMAT = { 'text/html' : 'html', mt.mobile : 'mobile', mt.opds : 'opds', mt.json : 'json', } USER_FORMATS = 'html mobile print opds stanza json'.split () # max no. of results returned by search MAX_RESULTS = 1000 # sort orders available to the user USER_SORT_ORDERS = 'downloads release_date title alpha quantity nentry random'.split () # internally used sort orders SORT_ORDERS = USER_SORT_ORDERS + 'nentry'.split () # fk_categories of sound files AUDIOBOOK_CATEGORIES = set ([1, 2, 3, 6]) # updated by cron thread books_in_archive = 0 class ClassAttr (object): """ Holds an XML class attribute. """ __slots__ = 'value' def __init__ (self, v = None): self.value = set () self.__iadd__ (v) def __len__ (self): return len (self.value) def __unicode__ (self): return ' '.join (self.value) if self.value else '' def __str__ (self): return ' '.join (self.value) if self.value else '' def __iadd__ (self, v): """ Implements operator += """ if not v: return self if isinstance (v, six.string_types): for i in six.text_type (v).split (): self.value.add (i) return self if isinstance (v, ClassAttr): self.value |= v.value return self def __contains__ (self, b): return b in self.value class Accumulator (object): """ Thread-safe master/slave counter for statistics collection. Many serving threads can increment the master counter while the statistic thread safely reads the slave counter. """ def __init__ (self): self.master = defaultdict (int) self.slave = defaultdict (int) self.lock = threading.Lock () def increment (self, key): """ Thread-safe increment the master counter. """ try: self.lock.acquire () self.master[key] += 1 finally: self.lock.release () def reset (self): """ Copy values to slave counter and reset the master counter. """ try: self.lock.acquire () self.slave = self.master.copy () self.master.clear () finally: self.lock.release () def __getitem__ (self, k): """ Read value from the slave counter. """ return self.slave[k] def iter_results (self): """ Iterate over the slave counters. """ total_hits = float (sum (self.slave.values ())) if total_hits > 0: for k, v in sorted (self.slave.items (), key = lambda x: -x[1]): yield (k, v, v / total_hits) formats_acc = Accumulator () formats_sum_acc = Accumulator () class DC (GutenbergDatabaseDublinCore.GutenbergDatabaseDublinCore, DublinCoreI18n.DublinCoreI18nMixin): """ A localized DublinCore. """ def __init__ (self, pool): GutenbergDatabaseDublinCore.GutenbergDatabaseDublinCore.__init__ (self, pool) DublinCoreI18n.DublinCoreI18nMixin.__init__ (self) class Cat (object): """ Hold data of one list item in output. """ def __init__ (self): self.type = None # use default self.header = '' self.class_ = ClassAttr () self.downloads = 1 self.rel = None self.order = 0 self.charset = None self.title = None self.subtitle = None self.extra = None self.icon = None self.icon2 = None self.url = None self.thumb_url = None class SearchUrlFormatter (object): """ Callable to format a search url. """ def __init__ (self, action): self.action = action def __call__ (self, row): os = cherrypy.request.os return os.url ( self.action, format = os.format, id = row.pk) class SQLStatement (object): """ Class implementing an SQL statement. """ prefix_to_prefix = { 'a.': 'ax', 't.': 'tx', 's.': 'sx', 'bs.': 'bsx', 'l.': 'l0', '#': 'no.', 'n.': 'no.', 'type.': 'y0', 'lcn.': 'lcnx', 'lcc.': 'lcc0', 'cat.': 'cat0', } """Dict of user-visible prefixes to translate. User-visible prefixes must be easy to type. The dot is on the lowercase keyboard of most phones, so you need no shifting to type these. Internal prefixes exploit the quirks of the tsvec stemmer. Words containing numbers do not get stemmed, so any '*0' prefix searches for the unstemmed word. All other words get stemmed, so any '*x' prefix searches for the stem of the word. 'x' was selected because it is a rare character that will cause few false positives. """ regex_cache = {} """ Cache of compiled regexes. """ def __init__ (self): self.query = '' self.params = {} self.from_ = [] self.where = [] self.groupby = [] self.sort_order = None self.start_index = 1 self.items_per_page = -1 @classmethod def sub (cls, regex_, replace, query): """ Like re.sub but also compile and cache the regex. """ cregex = cls.regex_cache.setdefault ( regex_, regex.compile (regex_, regex.UNICODE | regex.VERSION1)) return cregex.sub (replace, query) @classmethod def preprocess_query (cls, query): """ Preprocess query. The preprocessed query might get echoed to the user. """ sub = cls.sub # strip most not (letter or digit) query = sub (r'[\p{Z}\p{P}\p{S}\p{M}\p{C}--.!|()#]', ' ', query) # strip operators adjacent to non-whitespace # if you want grouping you have to add space on both sides of the parens query = sub (r'\b[!)]', ' ', query) query = sub (r'[(]\b', ' ', query) # insert spaces around operators query = sub (r'\s*[|!()]\s*', r' \g<0> ', query) return ' '.join (query.split ()) @classmethod def translate_query (cls, query): """ Translate query from user syntax to postgres tsvec syntax. """ sub = cls.sub def prefix_sub (match_object): """ Translate from user-visible prefix to internal prefix. """ s = match_object.group (0) return cls.prefix_to_prefix.get (s, s) def balance(query): """ Balance parens. """ def scan(query, up, down): scan = '' depth = 0 for char in query: if char == up: depth += 1 scan += char elif char == down: depth += -1 if depth < 0: depth = 0 else: scan += char else: scan += char return scan, depth balanced, depth = scan(query, '(', ')') if depth: balanced, depth = scan(balanced[::-1], ')', '(') balanced = balanced[::-1] return balanced # Replace the user-visible prefixes with the internally used prefixes. query = sub (r'(\b\w+\.|#)(?=\w)', prefix_sub, query) # add wildcards to all words query = sub (r'\b(\p{L}+)(\s|$)', r'\1:*\2', query) query = query.replace ('. ', ' ') # if parens aren't balanced, remove them query = balance (query) # if ! or | are at the wrong ends, remove them query = sub (r'(^[ \|]+|[ \|\!]+$)', '', query) # replace ' ' with ' & ' query = ' '.join (query.split ()) query = sub (r'(? 1: # opensearch is 1-based, SQL is 0-based params['offset'] = self.start_index - 1 query += " OFFSET %(offset)s" if self.items_per_page > -1: # need one more to know when to display 'next' link params['limit'] = self.items_per_page + 1 query += " LIMIT %(limit)s" return query, params def split (self, field, query): """ Split multiple-term query for sql consumption. """ terms = [] n = len (self.params) for i, q in enumerate (query.split ()): q = q.strip ('.,:;') terms.append ("%s ~* %%(p%d)s" % (field, n + i)) # self.params['p%d' % (n + i)] = r'\m' + q self.params['p%d' % (n + i)] = '(^| )' + q return terms def split_and_append (self, field, query): """ Split multiple-term query for sql consumption and append terms to query. """ for term in self.split (field, query): self.where.append (term) def fulltext (self, field, query, stemmer = 'english'): """ Perform fulltext search on query words. """ query = query.strip () if len (query) == 0: return query = self.translate_query (query) self.where.append ("%s @@ to_tsquery ('%s', %%(p%d)s)" % (field, stemmer, len (self.params))) self.params['p%d' % len (self.params)] = query class OpenSearch (object): """ Hold search results and lots of other stuff. We use this to pass everything we know around and into the templating engine. """ lang_to_default_locale = { 'en': 'en_US', 'de': 'de_DE', 'fr': 'fr_FR', 'es': 'es_ES', 'it': 'it_IT', 'pt': 'pt_BR', 'ru': 'ru_RU', } def __init__ (self): self.format = None self.page = None self.template = None self.query = None self.id = None self.sort_order = None self.search_terms = None self.start_index = 1 self.items_per_page = 1 self.total_results = -1 self.page_mode = 'screen' self.user_dialog = ('', '') self.opensearch_support = 0 # 0 = none, 1 = full, 2 = fake (Stanza, Aldiko, ...) self.books_in_archive = babel.numbers.format_number ( books_in_archive, locale = str (cherrypy.response.i18n.locale)) self.breadcrumbs = [ (_('Project Gutenberg'), _('Go to the Main page.'), '/'), (__('1 free ebook', '{count} free ebooks', books_in_archive).format ( count = self.books_in_archive), _('Start a new search.'), '/ebooks/'), ] # default output formatting functions self.f_format_title = self.format_title self.f_format_subtitle = self.format_author self.f_format_extra = self.format_none # depends on sort order, set in fix_sortorder () self.f_format_url = self.format_bibrec_url self.f_format_thumb_url = self.format_thumb_url self.f_format_icon = self.format_icon # icon class self.user_agent = cherrypy.request.headers.get ('User-Agent', '') cherrypy.request.os = self s = cherrypy.session k = cherrypy.request.params host = cherrypy.request.headers.get ('X-Forwarded-Host', cherrypy.config['host']) self.host = host.split (',')[-1].strip () # keep only the last hub # turns out X-Forwarded-Protocol (X-Forwarded-Proto is the defacto standaard) # is not a thing and has to be set in HAProxy self.protocol = cherrypy.request.headers.get ('X-Forwarded-Protocol', 'https') # sanity check if self.host not in (cherrypy.config['all_hosts']): self.host = cherrypy.config['host'] if self.protocol not in VALID_PROTOCOLS: self.protocol = 'https' self.urlgen = routes.URLGenerator ( cherrypy.routes_mapper, { 'HTTP_HOST': self.host, 'HTTPS': 1 , } ) self.set_format (k.get ('format')) # query: this param is set when an actual query is requested self.query = '' if 'query' in k: self.query = SQLStatement.preprocess_query (k['query']) # search_terms: this is used to carry the last query # to display in the search input box self.search_terms = self.query or s.get ('search_terms', '') self.sort_order = k.get ('sort_order') or s.get ('sort_order') or USER_SORT_ORDERS[0] if self.sort_order not in USER_SORT_ORDERS: raise cherrypy.HTTPError (400, 'Bad Request. Unknown sort order.') s['sort_order'] = self.sort_order try: self.id = int (k.get ('id') or '0') self.start_index = int (k.get ('start_index') or '1') self.items_per_page = min (100, int (k.get ('items_per_page') or '25')) except ValueError as what: raise cherrypy.HTTPError (400, 'Bad Request. ' + str (what)) formats_acc.increment (self.format) if self.format in ('mobile', 'opds', 'stanza'): formats_sum_acc.increment ('mobile') if self.format == 'html': formats_sum_acc.increment ('html') self.file_host = cherrypy.config['file_host'] self.now = datetime.datetime.utcnow ().replace (microsecond = 0).isoformat () + 'Z' self.do_animations = 'Kindle/' not in self.user_agent # no animations on e-ink self.ip = cherrypy.request.remote.ip self.type_opds = 'application/atom+xml;profile=opds-catalog' self.base_url = None self.canonical_url = None self.entries = [] # NOTE: For page titles etc. self.pg = self.title = _('Project Gutenberg') # NOTE: The tagline at the top of every page. self.tagline = _('Project Gutenberg offers {count} free ebooks to download.').format ( count = self.books_in_archive) # NOTE: The site's description in the html meta tags. self.description = _('Project Gutenberg offers {count} free ebooks for ' 'Kindle, iPad, Nook, Android, and iPhone.').format ( count = self.books_in_archive) # NOTE: The placeholder inside an empty search box. self.placeholder = _('Search Project Gutenberg.') # these need to be here because they have to be localized # NOTE: Msg to user indicating the order of the search results. self.sorted_msgs = { 'downloads': _("sorted by popularity"), 'release_date': _("sorted by release date"), 'quantity': _("sorted by quantity of books"), 'title': _("sorted alphabetically"), 'alpha': _("sorted alphabetically"), 'nentry': _("sorted by relevance"), 'random': _("in random order"), } self.snippet_image_url = self.url ('/pics/logo-144x144.png', host = self.file_host) self.og_type = 'website' self.class_ = ClassAttr () self.title_icon = None self.icon = None self.sort_orders = [] self.alternate_sort_orders = [] lang = self.lang = s.get ('_lang_', 'en_US') if len (lang) == 2: lang = self.lang_to_default_locale.get (lang, 'en_US') lang2 = self.lang[:2] self.fb_lang = lang if lang in FB_LANGS else 'en_US' self.paypal_lang = lang if lang in PAYPAL_LANGS else 'en_US' self.flattr_lang = lang if lang in FLATTR_LANGS else 'en_US' lang = lang.replace ('_', '-') self.google_lang = lang if lang in GOOGLE_LANGS else ( lang2 if lang2 in GOOGLE_LANGS else 'en-US') lang = lang.lower () self.twitter_lang = lang if lang in TWITTER_LANGS else ( lang2 if lang2 in TWITTER_LANGS else 'en') self.viewport = "width=device-width" # , initial-scale=1.0" self.touch_icon = '/pics/apple-touch-icon' self.touch_icon_precomposed = None # not yet used if 'user_dialog' in s: self.user_dialog = s['user_dialog'] del s['user_dialog'] msg = k.get ('msg') if msg is not None: if msg == 'welcome_stranger': self.user_dialog = ( _("Welcome to Project Gutenberg. " "You'll find here {count} ebooks completely free of charge.") .format (count = self.books_in_archive), _('Welcome')) def finalize (self): """ Calculate fields that depend on start_index, items_per_page and total_results. start_index, etc. must be set before calling this. """ # FIXME: android browser crashes on XHR with # meta name=viewport or link rel=apple-touch-icon # see: # http://code.google.com/p/android/issues/detail?id=6593 # http://code.google.com/p/android/issues/detail?id=9261 # # remove this from all browsers because we are caching responses if self.start_index > 1: self.touch_icon = None self.viewport = None self.desktop_host = cherrypy.config['host'] self.mobile_host = cherrypy.config['host_mobile'] last_page = max ((self.total_results - 1) // self.items_per_page, 0) # 0-based self.end_index = min (self.start_index + self.items_per_page - 1, self.total_results) self.prev_page_index = max (self.start_index - self.items_per_page, 1) self.next_page_index = min (self.start_index + self.items_per_page, self.total_results) self.last_page_index = last_page * self.items_per_page + 1 self.show_prev_page_link = self.start_index > 1 self.show_next_page_link = (self.end_index < self.total_results) self.desktop_search = self.url ('search', format = None) self.mobile_search = self.url ('search', format = 'mobile') self.json_search = self.url ('suggest', format = None) self.base_url = self.protocol_relative (self.url (host = self.host)) # for google, fb etc. self.canonical_url = self.url_carry (host = self.file_host, format = None) self.desktop_url = self.protocol_relative (self.url_carry (host = self.desktop_host, format = None)) self.mobile_url = self.protocol_relative (self.url_carry (host = self.mobile_host, format = 'mobile')) self.osd_url = self.qualify ('/catalog/osd-books.xml') s = cherrypy.session # write this late so pages can change it s['search_terms'] = self.search_terms def url (self, *args, **params): """ Generate url carrying the 'format' parameter from self. See: http://tools.cherrypy.org/wiki/RoutesUrlGeneration """ # We need to explicitly carry the parameters in the query # string (eg. those not matched by routes) because routes has # no memory for those. # # Also route memory is not used when generating named routes # eg. url ('search'). params.setdefault ('format', str (self.format)) route_name = args[0] if args else str (cherrypy.request.params['route_name']) rn = cherrypy.routes_mapper._routenames # pylint: disable=protected-access if route_name in rn: route_obj = rn[route_name] if 'id' in route_obj.minkeys: params.setdefault ('id', str (self.id)) # Eliminate null and superflous params. for k, v in list (params.items ()): if v is None or (k == 'start_index' and int (v) < 2) or (k == 'format' and v == 'html'): del params[k] return self.urlgen (route_name, **params) @staticmethod def params (**kwargs): """ Get dict of current params with override option. """ d = cherrypy.request.params.copy () # del d['action'] # del d['controller'] # del d['route_name'] try: del d['fb_locale'] except KeyError: pass d.update (kwargs) return d def url_carry (self, *args, **params): """ Generate url carrying most params from self. """ return self.url (*args, **self.params (**params)) @staticmethod def add_amp (url): """ Add ? or & to url. """ if '?' in url: return url + '&' return url + '?' def qualify (self, url): """ Append host part. """ return urllib.parse.urljoin (self.base_url, url) def protocol_relative (self, url): """ Make absolute url protocol relative. """ offset = url.find ('//') if offset > -1: return url[offset:] return url def set_format (self, format_): """ Sanity check and set the parameter we got from the user. Calc format and mediatype to send to the client. """ if format_ and format_ not in USER_FORMATS: raise cherrypy.HTTPError (400, 'Bad Request. Unknown format.') # fold print into html if format_ == 'print': format_ = 'html' self.page_mode = 'print' # user explicitly requested format if format_: self.format = format_ self.mediatype = mt[format_] self.opensearch_support = 1 if format_ == 'opds' else 2 return # no specific format requested ua = self.user_agent # user accessed the www site if self.host == cherrypy.config['host']: # but might want the mobile site ... if 'Kindle/' in ua: self.format = 'mobile' self.mediatype = mt.mobile elif 'tolino' in ua: self.format = 'mobile' self.mediatype = mt.mobile elif ua.startswith ('W3C-mobileOK/DDC-1.0'): self.format = 'mobile' self.mediatype = mt.mobile else: self.format = 'html' self.mediatype = 'text/html' return # user accessed the mobile site mediatype = None if self.host == cherrypy.config['host_mobile']: format_ = 'mobile' mediatype = mt.mobile # known OPDS consumers # 'stanza' is the older opds-ish format supported by stanza et al. opensearch_support = 0 if ua: if ua.startswith ('Stanza/'): # Stanza/2.1.1 iPhone OS/3.1.3/iPod touch catalog/2.1.1 # Stanza/3.0 iPhone OS/3.1.3/iPod touch catalog/3.0 format_ = 'stanza' mediatype = mt.opds opensearch_support = 2 elif ua.startswith ('FBReader/'): # FBReader/0.6.6(java) format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif 'Aldiko/' in ua: format_ = 'opds' mediatype = mt.opds opensearch_support = 2 elif ua.startswith ('Ibis-Reader/'): # Ibis-Reader/0.1 format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif ua.startswith ('ouiivo'): # ouiivo format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif (ua.startswith ('QuickR') or ua.startswith ('Young Reader') or ua.startswith ('MegaRead') or ua.startswith ('eBook Search')): # MegaReadLite 1.0 (iPhone Simulator; iPhone OS 4.2; en_US) # QuickReader 2.1.0 (iPhone; iPhone OS 3.1.3; en_US) # QuickRdrLite 3.0.1 (iPhone Simulator; iPhone OS 4.2; en_US) # eBook Search1.0(iPhone Simulator; iPhone OS 4.2; en_US) format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif ua.startswith ('CoolReader/'): # CoolReader/3(Android) format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif 'Freda' in ua: format_ = 'opds' mediatype = mt.opds opensearch_support = 1 elif 'Duokan' in ua: format_ = 'opds' mediatype = mt.opds opensearch_support = 1 # still nothing? if not format_: mediatype = 'text/html' format_ = 'html' self.format = format_ self.mediatype = mediatype self.opensearch_support = opensearch_support def log_request (self, page): """ Log the request params. Now a dummy. """ pass def fix_sortorder (self): """ Check selected sort order against available sort orders. """ if self.sort_orders: if not self.sort_order or self.sort_order not in self.sort_orders: self.sort_order = self.sort_orders [0] self.alternate_sort_orders = [x for x in self.sort_orders if x != self.sort_order] self.sorted_by = self.sorted_msgs [self.sort_order] self.title += " (%s)" % self.sorted_by # content of extra field depends on sorting self.f_format_extra = { 'alpha': self.format_none, 'title': self.format_none, 'downloads': self.format_downloads, 'quantity': self.format_quantity, 'release_date': self.format_release_date, 'random': self.format_none, }[self.sort_order] if self.sort_order == 'title': self.f_format_title = self.format_title_filing @staticmethod def format_title (row): """ Format a book title for display in results. """ title = gg.cut_at_newline (row.get ('title') or 'No Title') for lang_id in row.get ('fk_langs') or []: if lang_id != 'en': title += " (%s)" % cherrypy.response.i18n.locale.languages.get (lang_id, lang_id) return title @staticmethod def format_title_filing (row): """ Format a book title for display in results. """ title = gg.cut_at_newline (row.get ('filing') or 'No Title') for lang_id in row.get ('fk_langs') or []: if lang_id != 'en': title += " (%s)" % cherrypy.response.i18n.locale.languages.get (lang_id, lang_id) return title @staticmethod def format_author (row): """ Format an author name for display in results. """ authors = row.get ('author') if authors is None: return None authors = [ DublinCore.DublinCore.make_pretty_name (a) for a in authors ] return DublinCore.DublinCore.strunk (authors) @staticmethod def format_language (row): """ Format a language name for display in results. """ if row.pk in cherrypy.response.i18n.locale.languages: return cherrypy.response.i18n.locale.languages[row.pk] return row.title @staticmethod def format_none (dummy_row): """ Output nothing on results. """ return None @staticmethod def format_subtitle (row): """ Format a book subtitle for display in results. """ return row.get ('subtitle') @staticmethod def format_downloads (row): """ Format the no. of download for display in results. """ downloads = int (row.get ('downloads', 0)) # NOTE: No. of times a book was downloaded return __('1 download', '{0} downloads', downloads).format (downloads) @staticmethod def format_quantity (row): """ Format the quantity of books for display in results. """ count = int (row.get ('quantity', 0)) # NOTE: No. of books by some author, on a subject, etc. return __('1 book', '{0} books', count).format (count) @staticmethod def format_release_date (row): """ Format the release date for display in results. """ return babel.dates.format_date (row.get ('release_date'), locale = str (cherrypy.response.i18n.locale)) def format_suggestion (self, row): """ Format a suggestion for display in results. """ query = ' '.join (self.query.split ()[0:-1]) if query: query += ' ' return query + gg.cut_at_newline (row.get ('title') or '') @staticmethod def format_no_url (dummy_row): """ Show no url in results. """ return None def format_bibrec_url (self, row): """ Generate a bibrec url """ return self.url ('bibrec', id = row.pk) def format_canonical_bibrec_url (self, row): """ Generate the rel=canonical bibrec url for a book. """ return self.url ('bibrec', host = self.file_host, protocol = 'https', id = row.pk, format = None) def format_thumb_url (self, row): """ Generate the thumb url in results. """ if row.coverpages: return urllib.parse.urljoin ('//' + self.file_host, row.coverpages[0]) # return self.url ('bibrec', host = self.file_host, id = row.pk, # format = 'cover.small') return None def format_icon (self, dummy_row): """ Show a book icon in results. """ return self.icon def format_icon_titles (self, row): """ Show a book icon or audio icon in results. """ # for 'title' listings, replace book icon with audio icon if row.fk_categories and AUDIOBOOK_CATEGORIES.intersection (row.fk_categories): return 'audiobook' return self.icon def sql_get (query, **params): """ Quick and dirty SQL query returning one value. """ conn = cherrypy.engine.pool.connect () try: c = conn.cursor () c.execute (query, params) row = c.fetchone () if row: return row[0] return None except DatabaseError as what: cherrypy.log ("SQL Error: %s\n" % what, context = 'REQUEST', severity = logging.ERROR) cherrypy.log ("Query was: %s\n" % c.mogrify (query, params), context = 'REQUEST', severity = logging.ERROR) conn.detach () raise class SQLSearcher (object): """ An SQL searcher. """ def search (self, os, sql): """ Perform the SQL query and format rows into `Cat´s . Use plugin functions to format rows. """ sql.sort_order = os.sort_order sql.start_index = os.start_index sql.items_per_page = os.items_per_page query, params = sql.build () query += ' -- ' + os.ip rows = self.execute (query, params) # this is not necessarily the size of the result set. # if the result set is bigger than this page can show # total_results will be last item on page + 1 os.total_results = min (os.start_index - 1 + len (rows), MAX_RESULTS) for i in range (0, min (len (rows), os.items_per_page)): row = rows[i] cat = Cat () cat.title = os.f_format_title (row) cat.subtitle = os.f_format_subtitle (row) cat.extra = os.f_format_extra (row) cat.url = os.f_format_url (row) cat.thumb_url = os.f_format_thumb_url (row) cat.icon = os.f_format_icon (row) cat.header = row.get ('header', '') cat.class_ += os.class_ cat.order = 10 os.entries.append (cat) return os @staticmethod def mogrify (dummy_os, sql): """ Format a query and return it as string without executing it. """ conn = cherrypy.engine.pool.connect () c = conn.cursor () query, params = sql.build () return c.mogrify (query, params).decode ('utf-8') @staticmethod def execute (query, params): """ Execute a query and return an array of rows. """ conn = cherrypy.engine.pool.connect () try: c = conn.cursor () #cherrypy.log ("SQL Query: %s\n" % c.mogrify (query, params), # context = 'REQUEST', severity = logging.ERROR) c.execute (query, params) return [xl (c, row) for row in c.fetchall ()] except DatabaseError as what: cherrypy.log ("SQL Error: %s\n" % what, context = 'REQUEST', severity = logging.ERROR) cherrypy.log ("Query was: %s\n" % c.mogrify (query, params), context = 'REQUEST', severity = logging.ERROR) conn.detach () raise