1 #
2 # This module is derived from the module described at:
3 # http://gnosis.cx/publish/programming/charming_python_15.txt
4 #
5 # Author: David Mertz (mertz@gnosis.cx)
6 # Thanks to: Pat Knight (p.knight@ktgroup.co.uk)
7 # Gregory Popovitch (greg@gpy.com)
8 #
9 # The original module was released under this license, and remains under
10 # it:
11 #
12 # This file is released to the public domain. I (dqm) would
13 # appreciate it if you choose to keep derived works under terms
14 # that promote freedom, but obviously am giving up any rights
15 # to compel such.
16 #
17 #$Id: indexer.py,v 1.7 2002-07-09 21:38:43 richard Exp $
18 '''
19 This module provides an indexer class, RoundupIndexer, that stores text
20 indices in a roundup instance. This class makes searching the content of
21 messages and text files possible.
22 '''
23 import os, shutil, re, mimetypes, marshal, zlib, errno
25 class Indexer:
26 ''' Indexes information from roundup's hyperdb to allow efficient
27 searching.
29 Three structures are created by the indexer:
30 files {identifier: (fileid, wordcount)}
31 words {word: {fileid: count}}
32 fileids {fileid: identifier}
33 '''
34 def __init__(self, db_path):
35 self.indexdb_path = os.path.join(db_path, 'indexes')
36 self.indexdb = os.path.join(self.indexdb_path, 'index.db')
37 self.reindex = 0
38 self.quiet = 9
39 self.changed = 0
41 # see if we need to reindex because of a change in code
42 if (not os.path.exists(self.indexdb_path) or
43 not os.path.exists(os.path.join(self.indexdb_path, 'version'))):
44 # TODO: if the version file exists (in the future) we'll want to
45 # check the value in it - for now the file itself is a flag
46 self.force_reindex()
48 def force_reindex(self):
49 '''Force a reindex condition
50 '''
51 if os.path.exists(self.indexdb_path):
52 shutil.rmtree(self.indexdb_path)
53 os.makedirs(self.indexdb_path)
54 os.chmod(self.indexdb_path, 0775)
55 open(os.path.join(self.indexdb_path, 'version'), 'w').write('1\n')
56 self.reindex = 1
57 self.changed = 1
59 def should_reindex(self):
60 '''Should we reindex?
61 '''
62 return self.reindex
64 def add_text(self, identifier, text, mime_type='text/plain'):
65 ''' Add some text associated with the (classname, nodeid, property)
66 identifier.
67 '''
68 # make sure the index is loaded
69 self.load_index()
71 # remove old entries for this identifier
72 if self.files.has_key(identifier):
73 self.purge_entry(identifier)
75 # split into words
76 words = self.splitter(text, mime_type)
78 # Find new file index, and assign it to identifier
79 # (_TOP uses trick of negative to avoid conflict with file index)
80 self.files['_TOP'] = (self.files['_TOP'][0]-1, None)
81 file_index = abs(self.files['_TOP'][0])
82 self.files[identifier] = (file_index, len(words))
83 self.fileids[file_index] = identifier
85 # find the unique words
86 filedict = {}
87 for word in words:
88 if filedict.has_key(word):
89 filedict[word] = filedict[word]+1
90 else:
91 filedict[word] = 1
93 # now add to the totals
94 for word in filedict.keys():
95 # each word has a dict of {identifier: count}
96 if self.words.has_key(word):
97 entry = self.words[word]
98 else:
99 # new word
100 entry = {}
101 self.words[word] = entry
103 # make a reference to the file for this word
104 entry[file_index] = filedict[word]
106 # save needed
107 self.changed = 1
109 def splitter(self, text, ftype):
110 ''' Split the contents of a text string into a list of 'words'
111 '''
112 if ftype == 'text/plain':
113 words = self.text_splitter(text)
114 else:
115 return []
116 return words
118 def text_splitter(self, text):
119 """Split text/plain string into a list of words
120 """
121 # case insensitive
122 text = text.upper()
124 # Split the raw text, losing anything longer than 25 characters
125 # since that'll be gibberish (encoded text or somesuch) or shorter
126 # than 3 characters since those short words appear all over the
127 # place
128 return re.findall(r'\b\w{2,25}\b', text)
130 def search(self, search_terms, klass, ignore={},
131 dre=re.compile(r'([^\d]+)(\d+)')):
132 ''' Display search results looking for [search, terms] associated
133 with the hyperdb Class "klass". Ignore hits on {class: property}.
135 "dre" is a helper, not an argument.
136 '''
137 # do the index lookup
138 hits = self.find(search_terms)
139 if not hits:
140 return {}
142 # this is specific to "issue" klass ... eugh
143 designator_propname = {'msg': 'messages', 'file': 'files'}
145 # build a dictionary of nodes and their associated messages
146 # and files
147 nodeids = {}
148 for classname, nodeid, property in hits.values():
149 # skip this result if we don't care about this class/property
150 if ignore.has_key((classname, property)):
151 continue
153 # if it's a property on klass, it's easy
154 if classname == klass.classname:
155 if not nodeids.has_key(nodeid):
156 nodeids[nodeid] = {}
157 continue
159 # it's a linked class - find the klass entries that are
160 # linked to it
161 linkprop = designator_propname[classname]
162 for resid in klass.find(**{linkprop: nodeid}):
163 resid = str(resid)
164 if not nodeids.has_key(id):
165 nodeids[resid] = {}
167 # update the links for this klass nodeid
168 node_dict = nodeids[resid]
169 if not node_dict.has_key(linkprop):
170 node_dict[linkprop] = [nodeid]
171 elif node_dict.has_key(linkprop):
172 node_dict[linkprop].append(nodeid)
173 return nodeids
175 # we override this to ignore not 2 < word < 25 and also to fix a bug -
176 # the (fail) case.
177 def find(self, wordlist):
178 ''' Locate files that match ALL the words in wordlist
179 '''
180 if not hasattr(self, 'words'):
181 self.load_index()
182 self.load_index(wordlist=wordlist)
183 entries = {}
184 hits = None
185 for word in wordlist:
186 if not 2 < len(word) < 25:
187 # word outside the bounds of what we index - ignore
188 continue
189 word = word.upper()
190 entry = self.words.get(word) # For each word, get index
191 entries[word] = entry # of matching files
192 if not entry: # Nothing for this one word (fail)
193 return {}
194 if hits is None:
195 hits = {}
196 for k in entry.keys():
197 hits[k] = self.fileids[k]
198 else:
199 # Eliminate hits for every non-match
200 for fileid in hits.keys():
201 if not entry.has_key(fileid):
202 del hits[fileid]
203 if hits is None:
204 return {}
205 return hits
207 segments = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ#_-!"
208 def load_index(self, reload=0, wordlist=None):
209 # Unless reload is indicated, do not load twice
210 if self.index_loaded() and not reload:
211 return 0
213 # Ok, now let's actually load it
214 db = {'WORDS': {}, 'FILES': {'_TOP':(0,None)}, 'FILEIDS': {}}
216 # Identify the relevant word-dictionary segments
217 if not wordlist:
218 segments = self.segments
219 else:
220 segments = ['-','#']
221 for word in wordlist:
222 segments.append(word[0].upper())
224 # Load the segments
225 for segment in segments:
226 try:
227 f = open(self.indexdb + segment, 'rb')
228 except IOError, error:
229 if error.errno != errno.ENOENT:
230 raise
231 else:
232 pickle_str = zlib.decompress(f.read())
233 f.close()
234 dbslice = marshal.loads(pickle_str)
235 if dbslice.get('WORDS'):
236 # if it has some words, add them
237 for word, entry in dbslice['WORDS'].items():
238 db['WORDS'][word] = entry
239 if dbslice.get('FILES'):
240 # if it has some files, add them
241 db['FILES'] = dbslice['FILES']
242 if dbslice.get('FILEIDS'):
243 # if it has fileids, add them
244 db['FILEIDS'] = dbslice['FILEIDS']
246 self.words = db['WORDS']
247 self.files = db['FILES']
248 self.fileids = db['FILEIDS']
249 self.changed = 0
251 def save_index(self):
252 # only save if the index is loaded and changed
253 if not self.index_loaded() or not self.changed:
254 return
256 # brutal space saver... delete all the small segments
257 for segment in self.segments:
258 try:
259 os.remove(self.indexdb + segment)
260 except OSError:
261 # probably just nonexistent segment index file
262 # TODO: make sure it's an EEXIST
263 pass
265 # First write the much simpler filename/fileid dictionaries
266 dbfil = {'WORDS':None, 'FILES':self.files, 'FILEIDS':self.fileids}
267 open(self.indexdb+'-','wb').write(zlib.compress(marshal.dumps(dbfil)))
269 # The hard part is splitting the word dictionary up, of course
270 letters = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ#_"
271 segdicts = {} # Need batch of empty dicts
272 for segment in letters:
273 segdicts[segment] = {}
274 for word, entry in self.words.items(): # Split into segment dicts
275 initchar = word[0].upper()
276 segdicts[initchar][word] = entry
278 # save
279 for initchar in letters:
280 db = {'WORDS':segdicts[initchar], 'FILES':None, 'FILEIDS':None}
281 pickle_str = marshal.dumps(db)
282 filename = self.indexdb + initchar
283 pickle_fh = open(filename, 'wb')
284 pickle_fh.write(zlib.compress(pickle_str))
285 os.chmod(filename, 0664)
287 # save done
288 self.changed = 0
290 def purge_entry(self, identifier):
291 ''' Remove a file from file index and word index
292 '''
293 if not self.files.has_key(identifier):
294 return
296 file_index = self.files[identifier][0]
297 del self.files[identifier]
298 del self.fileids[file_index]
300 # The much harder part, cleanup the word index
301 for key, occurs in self.words.items():
302 if occurs.has_key(file_index):
303 del occurs[file_index]
305 # save needed
306 self.changed = 1
308 def index_loaded(self):
309 return (hasattr(self,'fileids') and hasattr(self,'files') and
310 hasattr(self,'words'))
312 #
313 #$Log: not supported by cvs2svn $
314 #Revision 1.6 2002/07/09 04:26:44 richard
315 #We're indexing numbers now, and _underscore words
316 #
317 #Revision 1.5 2002/07/09 04:19:09 richard
318 #Added reindex command to roundup-admin.
319 #Fixed reindex on first access.
320 #Also fixed reindexing of entries that change.
321 #
322 #Revision 1.4 2002/07/09 03:02:52 richard
323 #More indexer work:
324 #- all String properties may now be indexed too. Currently there's a bit of
325 # "issue" specific code in the actual searching which needs to be
326 # addressed. In a nutshell:
327 # + pass 'indexme="yes"' as a String() property initialisation arg, eg:
328 # file = FileClass(db, "file", name=String(), type=String(),
329 # comment=String(indexme="yes"))
330 # + the comment will then be indexed and be searchable, with the results
331 # related back to the issue that the file is linked to
332 #- as a result of this work, the FileClass has a default MIME type that may
333 # be overridden in a subclass, or by the use of a "type" property as is
334 # done in the default templates.
335 #- the regeneration of the indexes (if necessary) is done once the schema is
336 # set up in the dbinit.
337 #
338 #Revision 1.3 2002/07/08 06:58:15 richard
339 #cleaned up the indexer code:
340 # - it splits more words out (much simpler, faster splitter)
341 # - removed code we'll never use (roundup.roundup_indexer has the full
342 # implementation, and replaces roundup.indexer)
343 # - only index text/plain and rfc822/message (ideas for other text formats to
344 # index are welcome)
345 # - added simple unit test for indexer. Needs more tests for regression.
346 #
347 #Revision 1.2 2002/05/25 07:16:24 rochecompaan
348 #Merged search_indexing-branch with HEAD
349 #
350 #Revision 1.1.2.3 2002/05/02 11:52:12 rochecompaan
351 #Fixed small bug that prevented indexes from being generated.
352 #
353 #Revision 1.1.2.2 2002/04/19 19:54:42 rochecompaan
354 #cgi_client.py
355 # removed search link for the time being
356 # moved rendering of matches to htmltemplate
357 #hyperdb.py
358 # filtering of nodes on full text search incorporated in filter method
359 #roundupdb.py
360 # added paramater to call of filter method
361 #roundup_indexer.py
362 # added search method to RoundupIndexer class
363 #
364 #Revision 1.1.2.1 2002/04/03 11:55:57 rochecompaan
365 # . Added feature #526730 - search for messages capability
366 #