OLD | NEW |
| (Empty) |
1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. | |
2 // Use of this source code is governed by a BSD-style license that can be | |
3 // found in the LICENSE file. | |
4 | |
5 #include <limits> | |
6 #include <set> | |
7 #include <string> | |
8 | |
9 #include "chrome/browser/history/text_database.h" | |
10 | |
11 #include "base/file_util.h" | |
12 #include "base/logging.h" | |
13 #include "base/metrics/histogram.h" | |
14 #include "base/strings/string_number_conversions.h" | |
15 #include "base/strings/stringprintf.h" | |
16 #include "base/strings/utf_string_conversions.h" | |
17 #include "sql/statement.h" | |
18 #include "sql/transaction.h" | |
19 | |
20 // There are two tables in each database, one full-text search (FTS) table which | |
21 // indexes the contents and title of the pages. The other is a regular SQLITE | |
22 // table which contains non-indexed information about the page. All columns of | |
23 // a FTS table are indexed using the text search algorithm, which isn't what we | |
24 // want for things like times. If this were in the FTS table, there would be | |
25 // different words in the index for each time number. | |
26 // | |
27 // "pages" FTS table: | |
28 // url URL of the page so searches will match the URL. | |
29 // title Title of the page. | |
30 // body Body of the page. | |
31 // | |
32 // "info" regular table: | |
33 // time Time the corresponding FTS entry was visited. | |
34 // | |
35 // We do joins across these two tables by using their internal rowids, which we | |
36 // keep in sync between the two tables. The internal rowid is the only part of | |
37 // an FTS table that is indexed like a normal table, and the index over it is | |
38 // free since sqlite always indexes the internal rowid. | |
39 | |
40 namespace history { | |
41 | |
42 namespace { | |
43 | |
44 // Version 1 uses FTS2 for index files. | |
45 // Version 2 uses FTS3. | |
46 static const int kCurrentVersionNumber = 2; | |
47 static const int kCompatibleVersionNumber = 2; | |
48 | |
49 // Snippet computation relies on the index of the columns in the original | |
50 // create statement. These are the 0-based indices (as strings) of the | |
51 // corresponding columns. | |
52 const char kTitleColumnIndex[] = "1"; | |
53 const char kBodyColumnIndex[] = "2"; | |
54 | |
55 // The string prepended to the database identifier to generate the filename. | |
56 const base::FilePath::CharType kFilePrefix[] = | |
57 FILE_PATH_LITERAL("History Index "); | |
58 | |
59 } // namespace | |
60 | |
61 TextDatabase::Match::Match() {} | |
62 | |
63 TextDatabase::Match::~Match() {} | |
64 | |
65 TextDatabase::TextDatabase(const base::FilePath& path, | |
66 DBIdent id, | |
67 bool allow_create) | |
68 : path_(path), | |
69 ident_(id), | |
70 allow_create_(allow_create) { | |
71 // Compute the file name. | |
72 file_name_ = path_.Append(IDToFileName(ident_)); | |
73 } | |
74 | |
75 TextDatabase::~TextDatabase() { | |
76 } | |
77 | |
78 // static | |
79 const base::FilePath::CharType* TextDatabase::file_base() { | |
80 return kFilePrefix; | |
81 } | |
82 | |
83 // static | |
84 base::FilePath TextDatabase::IDToFileName(DBIdent id) { | |
85 // Identifiers are intended to be a combination of the year and month, for | |
86 // example, 200801 for January 2008. We convert this to | |
87 // "History Index 2008-01". However, we don't make assumptions about this | |
88 // scheme: the caller should assign IDs as it feels fit with the knowledge | |
89 // that they will apppear on disk in this form. | |
90 base::FilePath::StringType filename(file_base()); | |
91 base::StringAppendF(&filename, FILE_PATH_LITERAL("%d-%02d"), | |
92 id / 100, id % 100); | |
93 return base::FilePath(filename); | |
94 } | |
95 | |
96 // static | |
97 TextDatabase::DBIdent TextDatabase::FileNameToID( | |
98 const base::FilePath& file_path) { | |
99 base::FilePath::StringType file_name = file_path.BaseName().value(); | |
100 | |
101 // We don't actually check the prefix here. Since the file system could | |
102 // be case insensitive in ways we can't predict (NTFS), checking could | |
103 // potentially be the wrong thing to do. Instead, we just look for a suffix. | |
104 static const size_t kIDStringLength = 7; // Room for "xxxx-xx". | |
105 if (file_name.length() < kIDStringLength) | |
106 return 0; | |
107 const base::FilePath::StringType suffix( | |
108 &file_name[file_name.length() - kIDStringLength]); | |
109 | |
110 if (suffix.length() != kIDStringLength || | |
111 suffix[4] != FILE_PATH_LITERAL('-')) { | |
112 return 0; | |
113 } | |
114 | |
115 // TODO: Once StringPiece supports a templated interface over the | |
116 // underlying string type, use it here instead of substr, since that | |
117 // will avoid needless string copies. StringPiece cannot be used | |
118 // right now because base::FilePath::StringType could use either 8 or 16 bit | |
119 // characters, depending on the OS. | |
120 int year, month; | |
121 base::StringToInt(suffix.substr(0, 4), &year); | |
122 base::StringToInt(suffix.substr(5, 2), &month); | |
123 | |
124 return year * 100 + month; | |
125 } | |
126 | |
127 bool TextDatabase::Init() { | |
128 // Make sure, if we're not allowed to create the file, that it exists. | |
129 if (!allow_create_) { | |
130 if (!base::PathExists(file_name_)) | |
131 return false; | |
132 } | |
133 | |
134 db_.set_histogram_tag("Text"); | |
135 | |
136 // Set the database page size to something a little larger to give us | |
137 // better performance (we're typically seek rather than bandwidth limited). | |
138 // This only has an effect before any tables have been created, otherwise | |
139 // this is a NOP. Must be a power of 2 and a max of 8192. | |
140 db_.set_page_size(4096); | |
141 | |
142 // The default cache size is 2000 which give >8MB of data. Since we will often | |
143 // have 2-3 of these objects, each with their own 8MB, this adds up very fast. | |
144 // We therefore reduce the size so when there are multiple objects, we're not | |
145 // too big. | |
146 db_.set_cache_size(512); | |
147 | |
148 // Run the database in exclusive mode. Nobody else should be accessing the | |
149 // database while we're running, and this will give somewhat improved perf. | |
150 db_.set_exclusive_locking(); | |
151 | |
152 // Attach the database to our index file. | |
153 if (!db_.Open(file_name_)) | |
154 return false; | |
155 | |
156 // Meta table tracking version information. | |
157 if (!meta_table_.Init(&db_, kCurrentVersionNumber, kCompatibleVersionNumber)) | |
158 return false; | |
159 if (meta_table_.GetCompatibleVersionNumber() > kCurrentVersionNumber) { | |
160 // This version is too new. We don't bother notifying the user on this | |
161 // error, and just fail to use the file. Normally if they have version skew, | |
162 // they will get it for the main history file and it won't be necessary | |
163 // here. If that's not the case, since this is only indexed data, it's | |
164 // probably better to just not give FTS results than strange errors when | |
165 // everything else is working OK. | |
166 LOG(WARNING) << "Text database is too new."; | |
167 return false; | |
168 } | |
169 | |
170 return CreateTables(); | |
171 } | |
172 | |
173 void TextDatabase::BeginTransaction() { | |
174 db_.BeginTransaction(); | |
175 } | |
176 | |
177 void TextDatabase::CommitTransaction() { | |
178 db_.CommitTransaction(); | |
179 } | |
180 | |
181 bool TextDatabase::CreateTables() { | |
182 // FTS table of page contents. | |
183 if (!db_.DoesTableExist("pages")) { | |
184 if (!db_.Execute("CREATE VIRTUAL TABLE pages USING fts3(" | |
185 "TOKENIZE icu," | |
186 "url LONGVARCHAR," | |
187 "title LONGVARCHAR," | |
188 "body LONGVARCHAR)")) | |
189 return false; | |
190 } | |
191 | |
192 // Non-FTS table containing URLs and times so we can efficiently find them | |
193 // using a regular index (all FTS columns are special and are treated as | |
194 // full-text-search, which is not what we want when retrieving this data). | |
195 if (!db_.DoesTableExist("info")) { | |
196 // Note that there is no point in creating an index over time. Since | |
197 // we must always query the entire FTS table (it can not efficiently do | |
198 // subsets), we will always end up doing that first, and joining the info | |
199 // table off of that. | |
200 if (!db_.Execute("CREATE TABLE info(time INTEGER NOT NULL)")) | |
201 return false; | |
202 } | |
203 | |
204 // Create the index. | |
205 return db_.Execute("CREATE INDEX IF NOT EXISTS info_time ON info(time)"); | |
206 } | |
207 | |
208 bool TextDatabase::AddPageData(base::Time time, | |
209 const std::string& url, | |
210 const std::string& title, | |
211 const std::string& contents) { | |
212 sql::Transaction committer(&db_); | |
213 if (!committer.Begin()) | |
214 return false; | |
215 | |
216 // Add to the pages table. | |
217 sql::Statement add_to_pages(db_.GetCachedStatement(SQL_FROM_HERE, | |
218 "INSERT INTO pages (url, title, body) VALUES (?,?,?)")); | |
219 add_to_pages.BindString(0, url); | |
220 add_to_pages.BindString(1, title); | |
221 add_to_pages.BindString(2, contents); | |
222 if (!add_to_pages.Run()) | |
223 return false; | |
224 | |
225 int64 rowid = db_.GetLastInsertRowId(); | |
226 | |
227 // Add to the info table with the same rowid. | |
228 sql::Statement add_to_info(db_.GetCachedStatement(SQL_FROM_HERE, | |
229 "INSERT INTO info (rowid, time) VALUES (?,?)")); | |
230 add_to_info.BindInt64(0, rowid); | |
231 add_to_info.BindInt64(1, time.ToInternalValue()); | |
232 | |
233 if (!add_to_info.Run()) | |
234 return false; | |
235 | |
236 return committer.Commit(); | |
237 } | |
238 | |
239 void TextDatabase::DeletePageData(base::Time time, const std::string& url) { | |
240 // First get all rows that match. Selecing on time (which has an index) allows | |
241 // us to avoid brute-force searches on the full-text-index table (there will | |
242 // generally be only one match per time). | |
243 sql::Statement select_ids(db_.GetCachedStatement(SQL_FROM_HERE, | |
244 "SELECT info.rowid " | |
245 "FROM info JOIN pages ON info.rowid = pages.rowid " | |
246 "WHERE info.time=? AND pages.url=?")); | |
247 select_ids.BindInt64(0, time.ToInternalValue()); | |
248 select_ids.BindString(1, url); | |
249 | |
250 std::set<int64> rows_to_delete; | |
251 while (select_ids.Step()) | |
252 rows_to_delete.insert(select_ids.ColumnInt64(0)); | |
253 | |
254 // Delete from the pages table. | |
255 sql::Statement delete_page(db_.GetCachedStatement(SQL_FROM_HERE, | |
256 "DELETE FROM pages WHERE rowid=?")); | |
257 | |
258 for (std::set<int64>::const_iterator i = rows_to_delete.begin(); | |
259 i != rows_to_delete.end(); ++i) { | |
260 delete_page.BindInt64(0, *i); | |
261 if (!delete_page.Run()) | |
262 return; | |
263 delete_page.Reset(true); | |
264 } | |
265 | |
266 // Delete from the info table. | |
267 sql::Statement delete_info(db_.GetCachedStatement(SQL_FROM_HERE, | |
268 "DELETE FROM info WHERE rowid=?")); | |
269 | |
270 for (std::set<int64>::const_iterator i = rows_to_delete.begin(); | |
271 i != rows_to_delete.end(); ++i) { | |
272 delete_info.BindInt64(0, *i); | |
273 if (!delete_info.Run()) | |
274 return; | |
275 delete_info.Reset(true); | |
276 } | |
277 } | |
278 | |
279 void TextDatabase::Optimize() { | |
280 sql::Statement statement(db_.GetCachedStatement(SQL_FROM_HERE, | |
281 "SELECT OPTIMIZE(pages) FROM pages LIMIT 1")); | |
282 statement.Run(); | |
283 } | |
284 | |
285 bool TextDatabase::GetTextMatches(const std::string& query, | |
286 const QueryOptions& options, | |
287 std::vector<Match>* results, | |
288 URLSet* found_urls) { | |
289 std::string sql = "SELECT url, title, time, offsets(pages), body FROM pages " | |
290 "LEFT OUTER JOIN info ON pages.rowid = info.rowid WHERE "; | |
291 sql += options.body_only ? "body " : "pages "; | |
292 sql += "MATCH ? AND time >= ? AND time < ? "; | |
293 // Times may not be unique, so also sort by rowid to ensure a stable order. | |
294 sql += "ORDER BY time DESC, info.rowid DESC"; | |
295 | |
296 // Generate unique IDs for the two possible variations of the statement, | |
297 // so they don't share the same cached prepared statement. | |
298 sql::StatementID body_only_id = SQL_FROM_HERE; | |
299 sql::StatementID pages_id = SQL_FROM_HERE; | |
300 | |
301 sql::Statement statement(db_.GetCachedStatement( | |
302 (options.body_only ? body_only_id : pages_id), sql.c_str())); | |
303 | |
304 statement.BindString(0, query); | |
305 statement.BindInt64(1, options.EffectiveBeginTime()); | |
306 statement.BindInt64(2, options.EffectiveEndTime()); | |
307 | |
308 // |results| may not be initially empty, so keep track of how many were added | |
309 // by this call. | |
310 int result_count = 0; | |
311 | |
312 while (statement.Step()) { | |
313 // TODO(brettw) allow canceling the query in the middle. | |
314 // if (canceled_or_something) | |
315 // break; | |
316 | |
317 GURL url(statement.ColumnString(0)); | |
318 URLSet::const_iterator found_url = found_urls->find(url); | |
319 if (found_url != found_urls->end()) | |
320 continue; // Don't add this duplicate. | |
321 | |
322 if (++result_count > options.EffectiveMaxCount()) | |
323 break; | |
324 | |
325 // Fill the results into the vector (avoid copying the URL with Swap()). | |
326 results->resize(results->size() + 1); | |
327 Match& match = results->at(results->size() - 1); | |
328 match.url.Swap(&url); | |
329 | |
330 match.title = statement.ColumnString16(1); | |
331 match.time = base::Time::FromInternalValue(statement.ColumnInt64(2)); | |
332 | |
333 // Extract any matches in the title. | |
334 std::string offsets_str = statement.ColumnString(3); | |
335 Snippet::ExtractMatchPositions(offsets_str, kTitleColumnIndex, | |
336 &match.title_match_positions); | |
337 Snippet::ConvertMatchPositionsToWide(statement.ColumnString(1), | |
338 &match.title_match_positions); | |
339 | |
340 // Extract the matches in the body. | |
341 Snippet::MatchPositions match_positions; | |
342 Snippet::ExtractMatchPositions(offsets_str, kBodyColumnIndex, | |
343 &match_positions); | |
344 | |
345 // Compute the snippet based on those matches. | |
346 std::string body = statement.ColumnString(4); | |
347 match.snippet.ComputeSnippet(match_positions, body); | |
348 } | |
349 statement.Reset(true); | |
350 return result_count > options.EffectiveMaxCount(); | |
351 } | |
352 | |
353 } // namespace history | |
OLD | NEW |