OLD | NEW |
(Empty) | |
| 1 /* |
| 2 ** 2009 Oct 23 |
| 3 ** |
| 4 ** The author disclaims copyright to this source code. In place of |
| 5 ** a legal notice, here is a blessing: |
| 6 ** |
| 7 ** May you do good and not evil. |
| 8 ** May you find forgiveness for yourself and forgive others. |
| 9 ** May you share freely, never taking more than you give. |
| 10 ** |
| 11 ****************************************************************************** |
| 12 ** |
| 13 ** This file is part of the SQLite FTS3 extension module. Specifically, |
| 14 ** this file contains code to insert, update and delete rows from FTS3 |
| 15 ** tables. It also contains code to merge FTS3 b-tree segments. Some |
| 16 ** of the sub-routines used to merge segments are also used by the query |
| 17 ** code in fts3.c. |
| 18 */ |
| 19 |
| 20 #include "fts3Int.h" |
| 21 #if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) |
| 22 |
| 23 #include <string.h> |
| 24 #include <assert.h> |
| 25 #include <stdlib.h> |
| 26 |
| 27 |
| 28 #define FTS_MAX_APPENDABLE_HEIGHT 16 |
| 29 |
| 30 /* |
| 31 ** When full-text index nodes are loaded from disk, the buffer that they |
| 32 ** are loaded into has the following number of bytes of padding at the end |
| 33 ** of it. i.e. if a full-text index node is 900 bytes in size, then a buffer |
| 34 ** of 920 bytes is allocated for it. |
| 35 ** |
| 36 ** This means that if we have a pointer into a buffer containing node data, |
| 37 ** it is always safe to read up to two varints from it without risking an |
| 38 ** overread, even if the node data is corrupted. |
| 39 */ |
| 40 #define FTS3_NODE_PADDING (FTS3_VARINT_MAX*2) |
| 41 |
| 42 /* |
| 43 ** Under certain circumstances, b-tree nodes (doclists) can be loaded into |
| 44 ** memory incrementally instead of all at once. This can be a big performance |
| 45 ** win (reduced IO and CPU) if SQLite stops calling the virtual table xNext() |
| 46 ** method before retrieving all query results (as may happen, for example, |
| 47 ** if a query has a LIMIT clause). |
| 48 ** |
| 49 ** Incremental loading is used for b-tree nodes FTS3_NODE_CHUNK_THRESHOLD |
| 50 ** bytes and larger. Nodes are loaded in chunks of FTS3_NODE_CHUNKSIZE bytes. |
| 51 ** The code is written so that the hard lower-limit for each of these values |
| 52 ** is 1. Clearly such small values would be inefficient, but can be useful |
| 53 ** for testing purposes. |
| 54 ** |
| 55 ** If this module is built with SQLITE_TEST defined, these constants may |
| 56 ** be overridden at runtime for testing purposes. File fts3_test.c contains |
| 57 ** a Tcl interface to read and write the values. |
| 58 */ |
| 59 #ifdef SQLITE_TEST |
| 60 int test_fts3_node_chunksize = (4*1024); |
| 61 int test_fts3_node_chunk_threshold = (4*1024)*4; |
| 62 # define FTS3_NODE_CHUNKSIZE test_fts3_node_chunksize |
| 63 # define FTS3_NODE_CHUNK_THRESHOLD test_fts3_node_chunk_threshold |
| 64 #else |
| 65 # define FTS3_NODE_CHUNKSIZE (4*1024) |
| 66 # define FTS3_NODE_CHUNK_THRESHOLD (FTS3_NODE_CHUNKSIZE*4) |
| 67 #endif |
| 68 |
| 69 /* |
| 70 ** The two values that may be meaningfully bound to the :1 parameter in |
| 71 ** statements SQL_REPLACE_STAT and SQL_SELECT_STAT. |
| 72 */ |
| 73 #define FTS_STAT_DOCTOTAL 0 |
| 74 #define FTS_STAT_INCRMERGEHINT 1 |
| 75 #define FTS_STAT_AUTOINCRMERGE 2 |
| 76 |
| 77 /* |
| 78 ** If FTS_LOG_MERGES is defined, call sqlite3_log() to report each automatic |
| 79 ** and incremental merge operation that takes place. This is used for |
| 80 ** debugging FTS only, it should not usually be turned on in production |
| 81 ** systems. |
| 82 */ |
| 83 #ifdef FTS3_LOG_MERGES |
| 84 static void fts3LogMerge(int nMerge, sqlite3_int64 iAbsLevel){ |
| 85 sqlite3_log(SQLITE_OK, "%d-way merge from level %d", nMerge, (int)iAbsLevel); |
| 86 } |
| 87 #else |
| 88 #define fts3LogMerge(x, y) |
| 89 #endif |
| 90 |
| 91 |
| 92 typedef struct PendingList PendingList; |
| 93 typedef struct SegmentNode SegmentNode; |
| 94 typedef struct SegmentWriter SegmentWriter; |
| 95 |
| 96 /* |
| 97 ** An instance of the following data structure is used to build doclists |
| 98 ** incrementally. See function fts3PendingListAppend() for details. |
| 99 */ |
| 100 struct PendingList { |
| 101 int nData; |
| 102 char *aData; |
| 103 int nSpace; |
| 104 sqlite3_int64 iLastDocid; |
| 105 sqlite3_int64 iLastCol; |
| 106 sqlite3_int64 iLastPos; |
| 107 }; |
| 108 |
| 109 |
| 110 /* |
| 111 ** Each cursor has a (possibly empty) linked list of the following objects. |
| 112 */ |
| 113 struct Fts3DeferredToken { |
| 114 Fts3PhraseToken *pToken; /* Pointer to corresponding expr token */ |
| 115 int iCol; /* Column token must occur in */ |
| 116 Fts3DeferredToken *pNext; /* Next in list of deferred tokens */ |
| 117 PendingList *pList; /* Doclist is assembled here */ |
| 118 }; |
| 119 |
| 120 /* |
| 121 ** An instance of this structure is used to iterate through the terms on |
| 122 ** a contiguous set of segment b-tree leaf nodes. Although the details of |
| 123 ** this structure are only manipulated by code in this file, opaque handles |
| 124 ** of type Fts3SegReader* are also used by code in fts3.c to iterate through |
| 125 ** terms when querying the full-text index. See functions: |
| 126 ** |
| 127 ** sqlite3Fts3SegReaderNew() |
| 128 ** sqlite3Fts3SegReaderFree() |
| 129 ** sqlite3Fts3SegReaderIterate() |
| 130 ** |
| 131 ** Methods used to manipulate Fts3SegReader structures: |
| 132 ** |
| 133 ** fts3SegReaderNext() |
| 134 ** fts3SegReaderFirstDocid() |
| 135 ** fts3SegReaderNextDocid() |
| 136 */ |
| 137 struct Fts3SegReader { |
| 138 int iIdx; /* Index within level, or 0x7FFFFFFF for PT */ |
| 139 u8 bLookup; /* True for a lookup only */ |
| 140 u8 rootOnly; /* True for a root-only reader */ |
| 141 |
| 142 sqlite3_int64 iStartBlock; /* Rowid of first leaf block to traverse */ |
| 143 sqlite3_int64 iLeafEndBlock; /* Rowid of final leaf block to traverse */ |
| 144 sqlite3_int64 iEndBlock; /* Rowid of final block in segment (or 0) */ |
| 145 sqlite3_int64 iCurrentBlock; /* Current leaf block (or 0) */ |
| 146 |
| 147 char *aNode; /* Pointer to node data (or NULL) */ |
| 148 int nNode; /* Size of buffer at aNode (or 0) */ |
| 149 int nPopulate; /* If >0, bytes of buffer aNode[] loaded */ |
| 150 sqlite3_blob *pBlob; /* If not NULL, blob handle to read node */ |
| 151 |
| 152 Fts3HashElem **ppNextElem; |
| 153 |
| 154 /* Variables set by fts3SegReaderNext(). These may be read directly |
| 155 ** by the caller. They are valid from the time SegmentReaderNew() returns |
| 156 ** until SegmentReaderNext() returns something other than SQLITE_OK |
| 157 ** (i.e. SQLITE_DONE). |
| 158 */ |
| 159 int nTerm; /* Number of bytes in current term */ |
| 160 char *zTerm; /* Pointer to current term */ |
| 161 int nTermAlloc; /* Allocated size of zTerm buffer */ |
| 162 char *aDoclist; /* Pointer to doclist of current entry */ |
| 163 int nDoclist; /* Size of doclist in current entry */ |
| 164 |
| 165 /* The following variables are used by fts3SegReaderNextDocid() to iterate |
| 166 ** through the current doclist (aDoclist/nDoclist). |
| 167 */ |
| 168 char *pOffsetList; |
| 169 int nOffsetList; /* For descending pending seg-readers only */ |
| 170 sqlite3_int64 iDocid; |
| 171 }; |
| 172 |
| 173 #define fts3SegReaderIsPending(p) ((p)->ppNextElem!=0) |
| 174 #define fts3SegReaderIsRootOnly(p) ((p)->rootOnly!=0) |
| 175 |
| 176 /* |
| 177 ** An instance of this structure is used to create a segment b-tree in the |
| 178 ** database. The internal details of this type are only accessed by the |
| 179 ** following functions: |
| 180 ** |
| 181 ** fts3SegWriterAdd() |
| 182 ** fts3SegWriterFlush() |
| 183 ** fts3SegWriterFree() |
| 184 */ |
| 185 struct SegmentWriter { |
| 186 SegmentNode *pTree; /* Pointer to interior tree structure */ |
| 187 sqlite3_int64 iFirst; /* First slot in %_segments written */ |
| 188 sqlite3_int64 iFree; /* Next free slot in %_segments */ |
| 189 char *zTerm; /* Pointer to previous term buffer */ |
| 190 int nTerm; /* Number of bytes in zTerm */ |
| 191 int nMalloc; /* Size of malloc'd buffer at zMalloc */ |
| 192 char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ |
| 193 int nSize; /* Size of allocation at aData */ |
| 194 int nData; /* Bytes of data in aData */ |
| 195 char *aData; /* Pointer to block from malloc() */ |
| 196 i64 nLeafData; /* Number of bytes of leaf data written */ |
| 197 }; |
| 198 |
| 199 /* |
| 200 ** Type SegmentNode is used by the following three functions to create |
| 201 ** the interior part of the segment b+-tree structures (everything except |
| 202 ** the leaf nodes). These functions and type are only ever used by code |
| 203 ** within the fts3SegWriterXXX() family of functions described above. |
| 204 ** |
| 205 ** fts3NodeAddTerm() |
| 206 ** fts3NodeWrite() |
| 207 ** fts3NodeFree() |
| 208 ** |
| 209 ** When a b+tree is written to the database (either as a result of a merge |
| 210 ** or the pending-terms table being flushed), leaves are written into the |
| 211 ** database file as soon as they are completely populated. The interior of |
| 212 ** the tree is assembled in memory and written out only once all leaves have |
| 213 ** been populated and stored. This is Ok, as the b+-tree fanout is usually |
| 214 ** very large, meaning that the interior of the tree consumes relatively |
| 215 ** little memory. |
| 216 */ |
| 217 struct SegmentNode { |
| 218 SegmentNode *pParent; /* Parent node (or NULL for root node) */ |
| 219 SegmentNode *pRight; /* Pointer to right-sibling */ |
| 220 SegmentNode *pLeftmost; /* Pointer to left-most node of this depth */ |
| 221 int nEntry; /* Number of terms written to node so far */ |
| 222 char *zTerm; /* Pointer to previous term buffer */ |
| 223 int nTerm; /* Number of bytes in zTerm */ |
| 224 int nMalloc; /* Size of malloc'd buffer at zMalloc */ |
| 225 char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ |
| 226 int nData; /* Bytes of valid data so far */ |
| 227 char *aData; /* Node data */ |
| 228 }; |
| 229 |
| 230 /* |
| 231 ** Valid values for the second argument to fts3SqlStmt(). |
| 232 */ |
| 233 #define SQL_DELETE_CONTENT 0 |
| 234 #define SQL_IS_EMPTY 1 |
| 235 #define SQL_DELETE_ALL_CONTENT 2 |
| 236 #define SQL_DELETE_ALL_SEGMENTS 3 |
| 237 #define SQL_DELETE_ALL_SEGDIR 4 |
| 238 #define SQL_DELETE_ALL_DOCSIZE 5 |
| 239 #define SQL_DELETE_ALL_STAT 6 |
| 240 #define SQL_SELECT_CONTENT_BY_ROWID 7 |
| 241 #define SQL_NEXT_SEGMENT_INDEX 8 |
| 242 #define SQL_INSERT_SEGMENTS 9 |
| 243 #define SQL_NEXT_SEGMENTS_ID 10 |
| 244 #define SQL_INSERT_SEGDIR 11 |
| 245 #define SQL_SELECT_LEVEL 12 |
| 246 #define SQL_SELECT_LEVEL_RANGE 13 |
| 247 #define SQL_SELECT_LEVEL_COUNT 14 |
| 248 #define SQL_SELECT_SEGDIR_MAX_LEVEL 15 |
| 249 #define SQL_DELETE_SEGDIR_LEVEL 16 |
| 250 #define SQL_DELETE_SEGMENTS_RANGE 17 |
| 251 #define SQL_CONTENT_INSERT 18 |
| 252 #define SQL_DELETE_DOCSIZE 19 |
| 253 #define SQL_REPLACE_DOCSIZE 20 |
| 254 #define SQL_SELECT_DOCSIZE 21 |
| 255 #define SQL_SELECT_STAT 22 |
| 256 #define SQL_REPLACE_STAT 23 |
| 257 |
| 258 #define SQL_SELECT_ALL_PREFIX_LEVEL 24 |
| 259 #define SQL_DELETE_ALL_TERMS_SEGDIR 25 |
| 260 #define SQL_DELETE_SEGDIR_RANGE 26 |
| 261 #define SQL_SELECT_ALL_LANGID 27 |
| 262 #define SQL_FIND_MERGE_LEVEL 28 |
| 263 #define SQL_MAX_LEAF_NODE_ESTIMATE 29 |
| 264 #define SQL_DELETE_SEGDIR_ENTRY 30 |
| 265 #define SQL_SHIFT_SEGDIR_ENTRY 31 |
| 266 #define SQL_SELECT_SEGDIR 32 |
| 267 #define SQL_CHOMP_SEGDIR 33 |
| 268 #define SQL_SEGMENT_IS_APPENDABLE 34 |
| 269 #define SQL_SELECT_INDEXES 35 |
| 270 #define SQL_SELECT_MXLEVEL 36 |
| 271 |
| 272 #define SQL_SELECT_LEVEL_RANGE2 37 |
| 273 #define SQL_UPDATE_LEVEL_IDX 38 |
| 274 #define SQL_UPDATE_LEVEL 39 |
| 275 |
| 276 /* |
| 277 ** This function is used to obtain an SQLite prepared statement handle |
| 278 ** for the statement identified by the second argument. If successful, |
| 279 ** *pp is set to the requested statement handle and SQLITE_OK returned. |
| 280 ** Otherwise, an SQLite error code is returned and *pp is set to 0. |
| 281 ** |
| 282 ** If argument apVal is not NULL, then it must point to an array with |
| 283 ** at least as many entries as the requested statement has bound |
| 284 ** parameters. The values are bound to the statements parameters before |
| 285 ** returning. |
| 286 */ |
| 287 static int fts3SqlStmt( |
| 288 Fts3Table *p, /* Virtual table handle */ |
| 289 int eStmt, /* One of the SQL_XXX constants above */ |
| 290 sqlite3_stmt **pp, /* OUT: Statement handle */ |
| 291 sqlite3_value **apVal /* Values to bind to statement */ |
| 292 ){ |
| 293 const char *azSql[] = { |
| 294 /* 0 */ "DELETE FROM %Q.'%q_content' WHERE rowid = ?", |
| 295 /* 1 */ "SELECT NOT EXISTS(SELECT docid FROM %Q.'%q_content' WHERE rowid!=?)", |
| 296 /* 2 */ "DELETE FROM %Q.'%q_content'", |
| 297 /* 3 */ "DELETE FROM %Q.'%q_segments'", |
| 298 /* 4 */ "DELETE FROM %Q.'%q_segdir'", |
| 299 /* 5 */ "DELETE FROM %Q.'%q_docsize'", |
| 300 /* 6 */ "DELETE FROM %Q.'%q_stat'", |
| 301 /* 7 */ "SELECT %s WHERE rowid=?", |
| 302 /* 8 */ "SELECT (SELECT max(idx) FROM %Q.'%q_segdir' WHERE level = ?) + 1", |
| 303 /* 9 */ "REPLACE INTO %Q.'%q_segments'(blockid, block) VALUES(?, ?)", |
| 304 /* 10 */ "SELECT coalesce((SELECT max(blockid) FROM %Q.'%q_segments') + 1, 1)", |
| 305 /* 11 */ "REPLACE INTO %Q.'%q_segdir' VALUES(?,?,?,?,?,?)", |
| 306 |
| 307 /* Return segments in order from oldest to newest.*/ |
| 308 /* 12 */ "SELECT idx, start_block, leaves_end_block, end_block, root " |
| 309 "FROM %Q.'%q_segdir' WHERE level = ? ORDER BY idx ASC", |
| 310 /* 13 */ "SELECT idx, start_block, leaves_end_block, end_block, root " |
| 311 "FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?" |
| 312 "ORDER BY level DESC, idx ASC", |
| 313 |
| 314 /* 14 */ "SELECT count(*) FROM %Q.'%q_segdir' WHERE level = ?", |
| 315 /* 15 */ "SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", |
| 316 |
| 317 /* 16 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ?", |
| 318 /* 17 */ "DELETE FROM %Q.'%q_segments' WHERE blockid BETWEEN ? AND ?", |
| 319 /* 18 */ "INSERT INTO %Q.'%q_content' VALUES(%s)", |
| 320 /* 19 */ "DELETE FROM %Q.'%q_docsize' WHERE docid = ?", |
| 321 /* 20 */ "REPLACE INTO %Q.'%q_docsize' VALUES(?,?)", |
| 322 /* 21 */ "SELECT size FROM %Q.'%q_docsize' WHERE docid=?", |
| 323 /* 22 */ "SELECT value FROM %Q.'%q_stat' WHERE id=?", |
| 324 /* 23 */ "REPLACE INTO %Q.'%q_stat' VALUES(?,?)", |
| 325 /* 24 */ "", |
| 326 /* 25 */ "", |
| 327 |
| 328 /* 26 */ "DELETE FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", |
| 329 /* 27 */ "SELECT DISTINCT level / (1024 * ?) FROM %Q.'%q_segdir'", |
| 330 |
| 331 /* This statement is used to determine which level to read the input from |
| 332 ** when performing an incremental merge. It returns the absolute level number |
| 333 ** of the oldest level in the db that contains at least ? segments. Or, |
| 334 ** if no level in the FTS index contains more than ? segments, the statement |
| 335 ** returns zero rows. */ |
| 336 /* 28 */ "SELECT level FROM %Q.'%q_segdir' GROUP BY level HAVING count(*)>=?" |
| 337 " ORDER BY (level %% 1024) ASC LIMIT 1", |
| 338 |
| 339 /* Estimate the upper limit on the number of leaf nodes in a new segment |
| 340 ** created by merging the oldest :2 segments from absolute level :1. See |
| 341 ** function sqlite3Fts3Incrmerge() for details. */ |
| 342 /* 29 */ "SELECT 2 * total(1 + leaves_end_block - start_block) " |
| 343 " FROM %Q.'%q_segdir' WHERE level = ? AND idx < ?", |
| 344 |
| 345 /* SQL_DELETE_SEGDIR_ENTRY |
| 346 ** Delete the %_segdir entry on absolute level :1 with index :2. */ |
| 347 /* 30 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ? AND idx = ?", |
| 348 |
| 349 /* SQL_SHIFT_SEGDIR_ENTRY |
| 350 ** Modify the idx value for the segment with idx=:3 on absolute level :2 |
| 351 ** to :1. */ |
| 352 /* 31 */ "UPDATE %Q.'%q_segdir' SET idx = ? WHERE level=? AND idx=?", |
| 353 |
| 354 /* SQL_SELECT_SEGDIR |
| 355 ** Read a single entry from the %_segdir table. The entry from absolute |
| 356 ** level :1 with index value :2. */ |
| 357 /* 32 */ "SELECT idx, start_block, leaves_end_block, end_block, root " |
| 358 "FROM %Q.'%q_segdir' WHERE level = ? AND idx = ?", |
| 359 |
| 360 /* SQL_CHOMP_SEGDIR |
| 361 ** Update the start_block (:1) and root (:2) fields of the %_segdir |
| 362 ** entry located on absolute level :3 with index :4. */ |
| 363 /* 33 */ "UPDATE %Q.'%q_segdir' SET start_block = ?, root = ?" |
| 364 "WHERE level = ? AND idx = ?", |
| 365 |
| 366 /* SQL_SEGMENT_IS_APPENDABLE |
| 367 ** Return a single row if the segment with end_block=? is appendable. Or |
| 368 ** no rows otherwise. */ |
| 369 /* 34 */ "SELECT 1 FROM %Q.'%q_segments' WHERE blockid=? AND block IS NULL", |
| 370 |
| 371 /* SQL_SELECT_INDEXES |
| 372 ** Return the list of valid segment indexes for absolute level ? */ |
| 373 /* 35 */ "SELECT idx FROM %Q.'%q_segdir' WHERE level=? ORDER BY 1 ASC", |
| 374 |
| 375 /* SQL_SELECT_MXLEVEL |
| 376 ** Return the largest relative level in the FTS index or indexes. */ |
| 377 /* 36 */ "SELECT max( level %% 1024 ) FROM %Q.'%q_segdir'", |
| 378 |
| 379 /* Return segments in order from oldest to newest.*/ |
| 380 /* 37 */ "SELECT level, idx, end_block " |
| 381 "FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? " |
| 382 "ORDER BY level DESC, idx ASC", |
| 383 |
| 384 /* Update statements used while promoting segments */ |
| 385 /* 38 */ "UPDATE OR FAIL %Q.'%q_segdir' SET level=-1,idx=? " |
| 386 "WHERE level=? AND idx=?", |
| 387 /* 39 */ "UPDATE OR FAIL %Q.'%q_segdir' SET level=? WHERE level=-1" |
| 388 |
| 389 }; |
| 390 int rc = SQLITE_OK; |
| 391 sqlite3_stmt *pStmt; |
| 392 |
| 393 assert( SizeofArray(azSql)==SizeofArray(p->aStmt) ); |
| 394 assert( eStmt<SizeofArray(azSql) && eStmt>=0 ); |
| 395 |
| 396 pStmt = p->aStmt[eStmt]; |
| 397 if( !pStmt ){ |
| 398 char *zSql; |
| 399 if( eStmt==SQL_CONTENT_INSERT ){ |
| 400 zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName, p->zWriteExprlist); |
| 401 }else if( eStmt==SQL_SELECT_CONTENT_BY_ROWID ){ |
| 402 zSql = sqlite3_mprintf(azSql[eStmt], p->zReadExprlist); |
| 403 }else{ |
| 404 zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName); |
| 405 } |
| 406 if( !zSql ){ |
| 407 rc = SQLITE_NOMEM; |
| 408 }else{ |
| 409 rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, NULL); |
| 410 sqlite3_free(zSql); |
| 411 assert( rc==SQLITE_OK || pStmt==0 ); |
| 412 p->aStmt[eStmt] = pStmt; |
| 413 } |
| 414 } |
| 415 if( apVal ){ |
| 416 int i; |
| 417 int nParam = sqlite3_bind_parameter_count(pStmt); |
| 418 for(i=0; rc==SQLITE_OK && i<nParam; i++){ |
| 419 rc = sqlite3_bind_value(pStmt, i+1, apVal[i]); |
| 420 } |
| 421 } |
| 422 *pp = pStmt; |
| 423 return rc; |
| 424 } |
| 425 |
| 426 |
| 427 static int fts3SelectDocsize( |
| 428 Fts3Table *pTab, /* FTS3 table handle */ |
| 429 sqlite3_int64 iDocid, /* Docid to bind for SQL_SELECT_DOCSIZE */ |
| 430 sqlite3_stmt **ppStmt /* OUT: Statement handle */ |
| 431 ){ |
| 432 sqlite3_stmt *pStmt = 0; /* Statement requested from fts3SqlStmt() */ |
| 433 int rc; /* Return code */ |
| 434 |
| 435 rc = fts3SqlStmt(pTab, SQL_SELECT_DOCSIZE, &pStmt, 0); |
| 436 if( rc==SQLITE_OK ){ |
| 437 sqlite3_bind_int64(pStmt, 1, iDocid); |
| 438 rc = sqlite3_step(pStmt); |
| 439 if( rc!=SQLITE_ROW || sqlite3_column_type(pStmt, 0)!=SQLITE_BLOB ){ |
| 440 rc = sqlite3_reset(pStmt); |
| 441 if( rc==SQLITE_OK ) rc = FTS_CORRUPT_VTAB; |
| 442 pStmt = 0; |
| 443 }else{ |
| 444 rc = SQLITE_OK; |
| 445 } |
| 446 } |
| 447 |
| 448 *ppStmt = pStmt; |
| 449 return rc; |
| 450 } |
| 451 |
| 452 int sqlite3Fts3SelectDoctotal( |
| 453 Fts3Table *pTab, /* Fts3 table handle */ |
| 454 sqlite3_stmt **ppStmt /* OUT: Statement handle */ |
| 455 ){ |
| 456 sqlite3_stmt *pStmt = 0; |
| 457 int rc; |
| 458 rc = fts3SqlStmt(pTab, SQL_SELECT_STAT, &pStmt, 0); |
| 459 if( rc==SQLITE_OK ){ |
| 460 sqlite3_bind_int(pStmt, 1, FTS_STAT_DOCTOTAL); |
| 461 if( sqlite3_step(pStmt)!=SQLITE_ROW |
| 462 || sqlite3_column_type(pStmt, 0)!=SQLITE_BLOB |
| 463 ){ |
| 464 rc = sqlite3_reset(pStmt); |
| 465 if( rc==SQLITE_OK ) rc = FTS_CORRUPT_VTAB; |
| 466 pStmt = 0; |
| 467 } |
| 468 } |
| 469 *ppStmt = pStmt; |
| 470 return rc; |
| 471 } |
| 472 |
| 473 int sqlite3Fts3SelectDocsize( |
| 474 Fts3Table *pTab, /* Fts3 table handle */ |
| 475 sqlite3_int64 iDocid, /* Docid to read size data for */ |
| 476 sqlite3_stmt **ppStmt /* OUT: Statement handle */ |
| 477 ){ |
| 478 return fts3SelectDocsize(pTab, iDocid, ppStmt); |
| 479 } |
| 480 |
| 481 /* |
| 482 ** Similar to fts3SqlStmt(). Except, after binding the parameters in |
| 483 ** array apVal[] to the SQL statement identified by eStmt, the statement |
| 484 ** is executed. |
| 485 ** |
| 486 ** Returns SQLITE_OK if the statement is successfully executed, or an |
| 487 ** SQLite error code otherwise. |
| 488 */ |
| 489 static void fts3SqlExec( |
| 490 int *pRC, /* Result code */ |
| 491 Fts3Table *p, /* The FTS3 table */ |
| 492 int eStmt, /* Index of statement to evaluate */ |
| 493 sqlite3_value **apVal /* Parameters to bind */ |
| 494 ){ |
| 495 sqlite3_stmt *pStmt; |
| 496 int rc; |
| 497 if( *pRC ) return; |
| 498 rc = fts3SqlStmt(p, eStmt, &pStmt, apVal); |
| 499 if( rc==SQLITE_OK ){ |
| 500 sqlite3_step(pStmt); |
| 501 rc = sqlite3_reset(pStmt); |
| 502 } |
| 503 *pRC = rc; |
| 504 } |
| 505 |
| 506 |
| 507 /* |
| 508 ** This function ensures that the caller has obtained an exclusive |
| 509 ** shared-cache table-lock on the %_segdir table. This is required before |
| 510 ** writing data to the fts3 table. If this lock is not acquired first, then |
| 511 ** the caller may end up attempting to take this lock as part of committing |
| 512 ** a transaction, causing SQLite to return SQLITE_LOCKED or |
| 513 ** LOCKED_SHAREDCACHEto a COMMIT command. |
| 514 ** |
| 515 ** It is best to avoid this because if FTS3 returns any error when |
| 516 ** committing a transaction, the whole transaction will be rolled back. |
| 517 ** And this is not what users expect when they get SQLITE_LOCKED_SHAREDCACHE. |
| 518 ** It can still happen if the user locks the underlying tables directly |
| 519 ** instead of accessing them via FTS. |
| 520 */ |
| 521 static int fts3Writelock(Fts3Table *p){ |
| 522 int rc = SQLITE_OK; |
| 523 |
| 524 if( p->nPendingData==0 ){ |
| 525 sqlite3_stmt *pStmt; |
| 526 rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_LEVEL, &pStmt, 0); |
| 527 if( rc==SQLITE_OK ){ |
| 528 sqlite3_bind_null(pStmt, 1); |
| 529 sqlite3_step(pStmt); |
| 530 rc = sqlite3_reset(pStmt); |
| 531 } |
| 532 } |
| 533 |
| 534 return rc; |
| 535 } |
| 536 |
| 537 /* |
| 538 ** FTS maintains a separate indexes for each language-id (a 32-bit integer). |
| 539 ** Within each language id, a separate index is maintained to store the |
| 540 ** document terms, and each configured prefix size (configured the FTS |
| 541 ** "prefix=" option). And each index consists of multiple levels ("relative |
| 542 ** levels"). |
| 543 ** |
| 544 ** All three of these values (the language id, the specific index and the |
| 545 ** level within the index) are encoded in 64-bit integer values stored |
| 546 ** in the %_segdir table on disk. This function is used to convert three |
| 547 ** separate component values into the single 64-bit integer value that |
| 548 ** can be used to query the %_segdir table. |
| 549 ** |
| 550 ** Specifically, each language-id/index combination is allocated 1024 |
| 551 ** 64-bit integer level values ("absolute levels"). The main terms index |
| 552 ** for language-id 0 is allocate values 0-1023. The first prefix index |
| 553 ** (if any) for language-id 0 is allocated values 1024-2047. And so on. |
| 554 ** Language 1 indexes are allocated immediately following language 0. |
| 555 ** |
| 556 ** So, for a system with nPrefix prefix indexes configured, the block of |
| 557 ** absolute levels that corresponds to language-id iLangid and index |
| 558 ** iIndex starts at absolute level ((iLangid * (nPrefix+1) + iIndex) * 1024). |
| 559 */ |
| 560 static sqlite3_int64 getAbsoluteLevel( |
| 561 Fts3Table *p, /* FTS3 table handle */ |
| 562 int iLangid, /* Language id */ |
| 563 int iIndex, /* Index in p->aIndex[] */ |
| 564 int iLevel /* Level of segments */ |
| 565 ){ |
| 566 sqlite3_int64 iBase; /* First absolute level for iLangid/iIndex */ |
| 567 assert( iLangid>=0 ); |
| 568 assert( p->nIndex>0 ); |
| 569 assert( iIndex>=0 && iIndex<p->nIndex ); |
| 570 |
| 571 iBase = ((sqlite3_int64)iLangid * p->nIndex + iIndex) * FTS3_SEGDIR_MAXLEVEL; |
| 572 return iBase + iLevel; |
| 573 } |
| 574 |
| 575 /* |
| 576 ** Set *ppStmt to a statement handle that may be used to iterate through |
| 577 ** all rows in the %_segdir table, from oldest to newest. If successful, |
| 578 ** return SQLITE_OK. If an error occurs while preparing the statement, |
| 579 ** return an SQLite error code. |
| 580 ** |
| 581 ** There is only ever one instance of this SQL statement compiled for |
| 582 ** each FTS3 table. |
| 583 ** |
| 584 ** The statement returns the following columns from the %_segdir table: |
| 585 ** |
| 586 ** 0: idx |
| 587 ** 1: start_block |
| 588 ** 2: leaves_end_block |
| 589 ** 3: end_block |
| 590 ** 4: root |
| 591 */ |
| 592 int sqlite3Fts3AllSegdirs( |
| 593 Fts3Table *p, /* FTS3 table */ |
| 594 int iLangid, /* Language being queried */ |
| 595 int iIndex, /* Index for p->aIndex[] */ |
| 596 int iLevel, /* Level to select (relative level) */ |
| 597 sqlite3_stmt **ppStmt /* OUT: Compiled statement */ |
| 598 ){ |
| 599 int rc; |
| 600 sqlite3_stmt *pStmt = 0; |
| 601 |
| 602 assert( iLevel==FTS3_SEGCURSOR_ALL || iLevel>=0 ); |
| 603 assert( iLevel<FTS3_SEGDIR_MAXLEVEL ); |
| 604 assert( iIndex>=0 && iIndex<p->nIndex ); |
| 605 |
| 606 if( iLevel<0 ){ |
| 607 /* "SELECT * FROM %_segdir WHERE level BETWEEN ? AND ? ORDER BY ..." */ |
| 608 rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_RANGE, &pStmt, 0); |
| 609 if( rc==SQLITE_OK ){ |
| 610 sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); |
| 611 sqlite3_bind_int64(pStmt, 2, |
| 612 getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) |
| 613 ); |
| 614 } |
| 615 }else{ |
| 616 /* "SELECT * FROM %_segdir WHERE level = ? ORDER BY ..." */ |
| 617 rc = fts3SqlStmt(p, SQL_SELECT_LEVEL, &pStmt, 0); |
| 618 if( rc==SQLITE_OK ){ |
| 619 sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex,iLevel)); |
| 620 } |
| 621 } |
| 622 *ppStmt = pStmt; |
| 623 return rc; |
| 624 } |
| 625 |
| 626 |
| 627 /* |
| 628 ** Append a single varint to a PendingList buffer. SQLITE_OK is returned |
| 629 ** if successful, or an SQLite error code otherwise. |
| 630 ** |
| 631 ** This function also serves to allocate the PendingList structure itself. |
| 632 ** For example, to create a new PendingList structure containing two |
| 633 ** varints: |
| 634 ** |
| 635 ** PendingList *p = 0; |
| 636 ** fts3PendingListAppendVarint(&p, 1); |
| 637 ** fts3PendingListAppendVarint(&p, 2); |
| 638 */ |
| 639 static int fts3PendingListAppendVarint( |
| 640 PendingList **pp, /* IN/OUT: Pointer to PendingList struct */ |
| 641 sqlite3_int64 i /* Value to append to data */ |
| 642 ){ |
| 643 PendingList *p = *pp; |
| 644 |
| 645 /* Allocate or grow the PendingList as required. */ |
| 646 if( !p ){ |
| 647 p = sqlite3_malloc(sizeof(*p) + 100); |
| 648 if( !p ){ |
| 649 return SQLITE_NOMEM; |
| 650 } |
| 651 p->nSpace = 100; |
| 652 p->aData = (char *)&p[1]; |
| 653 p->nData = 0; |
| 654 } |
| 655 else if( p->nData+FTS3_VARINT_MAX+1>p->nSpace ){ |
| 656 int nNew = p->nSpace * 2; |
| 657 p = sqlite3_realloc(p, sizeof(*p) + nNew); |
| 658 if( !p ){ |
| 659 sqlite3_free(*pp); |
| 660 *pp = 0; |
| 661 return SQLITE_NOMEM; |
| 662 } |
| 663 p->nSpace = nNew; |
| 664 p->aData = (char *)&p[1]; |
| 665 } |
| 666 |
| 667 /* Append the new serialized varint to the end of the list. */ |
| 668 p->nData += sqlite3Fts3PutVarint(&p->aData[p->nData], i); |
| 669 p->aData[p->nData] = '\0'; |
| 670 *pp = p; |
| 671 return SQLITE_OK; |
| 672 } |
| 673 |
| 674 /* |
| 675 ** Add a docid/column/position entry to a PendingList structure. Non-zero |
| 676 ** is returned if the structure is sqlite3_realloced as part of adding |
| 677 ** the entry. Otherwise, zero. |
| 678 ** |
| 679 ** If an OOM error occurs, *pRc is set to SQLITE_NOMEM before returning. |
| 680 ** Zero is always returned in this case. Otherwise, if no OOM error occurs, |
| 681 ** it is set to SQLITE_OK. |
| 682 */ |
| 683 static int fts3PendingListAppend( |
| 684 PendingList **pp, /* IN/OUT: PendingList structure */ |
| 685 sqlite3_int64 iDocid, /* Docid for entry to add */ |
| 686 sqlite3_int64 iCol, /* Column for entry to add */ |
| 687 sqlite3_int64 iPos, /* Position of term for entry to add */ |
| 688 int *pRc /* OUT: Return code */ |
| 689 ){ |
| 690 PendingList *p = *pp; |
| 691 int rc = SQLITE_OK; |
| 692 |
| 693 assert( !p || p->iLastDocid<=iDocid ); |
| 694 |
| 695 if( !p || p->iLastDocid!=iDocid ){ |
| 696 sqlite3_int64 iDelta = iDocid - (p ? p->iLastDocid : 0); |
| 697 if( p ){ |
| 698 assert( p->nData<p->nSpace ); |
| 699 assert( p->aData[p->nData]==0 ); |
| 700 p->nData++; |
| 701 } |
| 702 if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iDelta)) ){ |
| 703 goto pendinglistappend_out; |
| 704 } |
| 705 p->iLastCol = -1; |
| 706 p->iLastPos = 0; |
| 707 p->iLastDocid = iDocid; |
| 708 } |
| 709 if( iCol>0 && p->iLastCol!=iCol ){ |
| 710 if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, 1)) |
| 711 || SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iCol)) |
| 712 ){ |
| 713 goto pendinglistappend_out; |
| 714 } |
| 715 p->iLastCol = iCol; |
| 716 p->iLastPos = 0; |
| 717 } |
| 718 if( iCol>=0 ){ |
| 719 assert( iPos>p->iLastPos || (iPos==0 && p->iLastPos==0) ); |
| 720 rc = fts3PendingListAppendVarint(&p, 2+iPos-p->iLastPos); |
| 721 if( rc==SQLITE_OK ){ |
| 722 p->iLastPos = iPos; |
| 723 } |
| 724 } |
| 725 |
| 726 pendinglistappend_out: |
| 727 *pRc = rc; |
| 728 if( p!=*pp ){ |
| 729 *pp = p; |
| 730 return 1; |
| 731 } |
| 732 return 0; |
| 733 } |
| 734 |
| 735 /* |
| 736 ** Free a PendingList object allocated by fts3PendingListAppend(). |
| 737 */ |
| 738 static void fts3PendingListDelete(PendingList *pList){ |
| 739 sqlite3_free(pList); |
| 740 } |
| 741 |
| 742 /* |
| 743 ** Add an entry to one of the pending-terms hash tables. |
| 744 */ |
| 745 static int fts3PendingTermsAddOne( |
| 746 Fts3Table *p, |
| 747 int iCol, |
| 748 int iPos, |
| 749 Fts3Hash *pHash, /* Pending terms hash table to add entry to */ |
| 750 const char *zToken, |
| 751 int nToken |
| 752 ){ |
| 753 PendingList *pList; |
| 754 int rc = SQLITE_OK; |
| 755 |
| 756 pList = (PendingList *)fts3HashFind(pHash, zToken, nToken); |
| 757 if( pList ){ |
| 758 p->nPendingData -= (pList->nData + nToken + sizeof(Fts3HashElem)); |
| 759 } |
| 760 if( fts3PendingListAppend(&pList, p->iPrevDocid, iCol, iPos, &rc) ){ |
| 761 if( pList==fts3HashInsert(pHash, zToken, nToken, pList) ){ |
| 762 /* Malloc failed while inserting the new entry. This can only |
| 763 ** happen if there was no previous entry for this token. |
| 764 */ |
| 765 assert( 0==fts3HashFind(pHash, zToken, nToken) ); |
| 766 sqlite3_free(pList); |
| 767 rc = SQLITE_NOMEM; |
| 768 } |
| 769 } |
| 770 if( rc==SQLITE_OK ){ |
| 771 p->nPendingData += (pList->nData + nToken + sizeof(Fts3HashElem)); |
| 772 } |
| 773 return rc; |
| 774 } |
| 775 |
| 776 /* |
| 777 ** Tokenize the nul-terminated string zText and add all tokens to the |
| 778 ** pending-terms hash-table. The docid used is that currently stored in |
| 779 ** p->iPrevDocid, and the column is specified by argument iCol. |
| 780 ** |
| 781 ** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. |
| 782 */ |
| 783 static int fts3PendingTermsAdd( |
| 784 Fts3Table *p, /* Table into which text will be inserted */ |
| 785 int iLangid, /* Language id to use */ |
| 786 const char *zText, /* Text of document to be inserted */ |
| 787 int iCol, /* Column into which text is being inserted */ |
| 788 u32 *pnWord /* IN/OUT: Incr. by number tokens inserted */ |
| 789 ){ |
| 790 int rc; |
| 791 int iStart = 0; |
| 792 int iEnd = 0; |
| 793 int iPos = 0; |
| 794 int nWord = 0; |
| 795 |
| 796 char const *zToken; |
| 797 int nToken = 0; |
| 798 |
| 799 sqlite3_tokenizer *pTokenizer = p->pTokenizer; |
| 800 sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; |
| 801 sqlite3_tokenizer_cursor *pCsr; |
| 802 int (*xNext)(sqlite3_tokenizer_cursor *pCursor, |
| 803 const char**,int*,int*,int*,int*); |
| 804 |
| 805 assert( pTokenizer && pModule ); |
| 806 |
| 807 /* If the user has inserted a NULL value, this function may be called with |
| 808 ** zText==0. In this case, add zero token entries to the hash table and |
| 809 ** return early. */ |
| 810 if( zText==0 ){ |
| 811 *pnWord = 0; |
| 812 return SQLITE_OK; |
| 813 } |
| 814 |
| 815 rc = sqlite3Fts3OpenTokenizer(pTokenizer, iLangid, zText, -1, &pCsr); |
| 816 if( rc!=SQLITE_OK ){ |
| 817 return rc; |
| 818 } |
| 819 |
| 820 xNext = pModule->xNext; |
| 821 while( SQLITE_OK==rc |
| 822 && SQLITE_OK==(rc = xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos)) |
| 823 ){ |
| 824 int i; |
| 825 if( iPos>=nWord ) nWord = iPos+1; |
| 826 |
| 827 /* Positions cannot be negative; we use -1 as a terminator internally. |
| 828 ** Tokens must have a non-zero length. |
| 829 */ |
| 830 if( iPos<0 || !zToken || nToken<=0 ){ |
| 831 rc = SQLITE_ERROR; |
| 832 break; |
| 833 } |
| 834 |
| 835 /* Add the term to the terms index */ |
| 836 rc = fts3PendingTermsAddOne( |
| 837 p, iCol, iPos, &p->aIndex[0].hPending, zToken, nToken |
| 838 ); |
| 839 |
| 840 /* Add the term to each of the prefix indexes that it is not too |
| 841 ** short for. */ |
| 842 for(i=1; rc==SQLITE_OK && i<p->nIndex; i++){ |
| 843 struct Fts3Index *pIndex = &p->aIndex[i]; |
| 844 if( nToken<pIndex->nPrefix ) continue; |
| 845 rc = fts3PendingTermsAddOne( |
| 846 p, iCol, iPos, &pIndex->hPending, zToken, pIndex->nPrefix |
| 847 ); |
| 848 } |
| 849 } |
| 850 |
| 851 pModule->xClose(pCsr); |
| 852 *pnWord += nWord; |
| 853 return (rc==SQLITE_DONE ? SQLITE_OK : rc); |
| 854 } |
| 855 |
| 856 /* |
| 857 ** Calling this function indicates that subsequent calls to |
| 858 ** fts3PendingTermsAdd() are to add term/position-list pairs for the |
| 859 ** contents of the document with docid iDocid. |
| 860 */ |
| 861 static int fts3PendingTermsDocid( |
| 862 Fts3Table *p, /* Full-text table handle */ |
| 863 int iLangid, /* Language id of row being written */ |
| 864 sqlite_int64 iDocid /* Docid of row being written */ |
| 865 ){ |
| 866 assert( iLangid>=0 ); |
| 867 |
| 868 /* TODO(shess) Explore whether partially flushing the buffer on |
| 869 ** forced-flush would provide better performance. I suspect that if |
| 870 ** we ordered the doclists by size and flushed the largest until the |
| 871 ** buffer was half empty, that would let the less frequent terms |
| 872 ** generate longer doclists. |
| 873 */ |
| 874 if( iDocid<=p->iPrevDocid |
| 875 || p->iPrevLangid!=iLangid |
| 876 || p->nPendingData>p->nMaxPendingData |
| 877 ){ |
| 878 int rc = sqlite3Fts3PendingTermsFlush(p); |
| 879 if( rc!=SQLITE_OK ) return rc; |
| 880 } |
| 881 p->iPrevDocid = iDocid; |
| 882 p->iPrevLangid = iLangid; |
| 883 return SQLITE_OK; |
| 884 } |
| 885 |
| 886 /* |
| 887 ** Discard the contents of the pending-terms hash tables. |
| 888 */ |
| 889 void sqlite3Fts3PendingTermsClear(Fts3Table *p){ |
| 890 int i; |
| 891 for(i=0; i<p->nIndex; i++){ |
| 892 Fts3HashElem *pElem; |
| 893 Fts3Hash *pHash = &p->aIndex[i].hPending; |
| 894 for(pElem=fts3HashFirst(pHash); pElem; pElem=fts3HashNext(pElem)){ |
| 895 PendingList *pList = (PendingList *)fts3HashData(pElem); |
| 896 fts3PendingListDelete(pList); |
| 897 } |
| 898 fts3HashClear(pHash); |
| 899 } |
| 900 p->nPendingData = 0; |
| 901 } |
| 902 |
| 903 /* |
| 904 ** This function is called by the xUpdate() method as part of an INSERT |
| 905 ** operation. It adds entries for each term in the new record to the |
| 906 ** pendingTerms hash table. |
| 907 ** |
| 908 ** Argument apVal is the same as the similarly named argument passed to |
| 909 ** fts3InsertData(). Parameter iDocid is the docid of the new row. |
| 910 */ |
| 911 static int fts3InsertTerms( |
| 912 Fts3Table *p, |
| 913 int iLangid, |
| 914 sqlite3_value **apVal, |
| 915 u32 *aSz |
| 916 ){ |
| 917 int i; /* Iterator variable */ |
| 918 for(i=2; i<p->nColumn+2; i++){ |
| 919 int iCol = i-2; |
| 920 if( p->abNotindexed[iCol]==0 ){ |
| 921 const char *zText = (const char *)sqlite3_value_text(apVal[i]); |
| 922 int rc = fts3PendingTermsAdd(p, iLangid, zText, iCol, &aSz[iCol]); |
| 923 if( rc!=SQLITE_OK ){ |
| 924 return rc; |
| 925 } |
| 926 aSz[p->nColumn] += sqlite3_value_bytes(apVal[i]); |
| 927 } |
| 928 } |
| 929 return SQLITE_OK; |
| 930 } |
| 931 |
| 932 /* |
| 933 ** This function is called by the xUpdate() method for an INSERT operation. |
| 934 ** The apVal parameter is passed a copy of the apVal argument passed by |
| 935 ** SQLite to the xUpdate() method. i.e: |
| 936 ** |
| 937 ** apVal[0] Not used for INSERT. |
| 938 ** apVal[1] rowid |
| 939 ** apVal[2] Left-most user-defined column |
| 940 ** ... |
| 941 ** apVal[p->nColumn+1] Right-most user-defined column |
| 942 ** apVal[p->nColumn+2] Hidden column with same name as table |
| 943 ** apVal[p->nColumn+3] Hidden "docid" column (alias for rowid) |
| 944 ** apVal[p->nColumn+4] Hidden languageid column |
| 945 */ |
| 946 static int fts3InsertData( |
| 947 Fts3Table *p, /* Full-text table */ |
| 948 sqlite3_value **apVal, /* Array of values to insert */ |
| 949 sqlite3_int64 *piDocid /* OUT: Docid for row just inserted */ |
| 950 ){ |
| 951 int rc; /* Return code */ |
| 952 sqlite3_stmt *pContentInsert; /* INSERT INTO %_content VALUES(...) */ |
| 953 |
| 954 if( p->zContentTbl ){ |
| 955 sqlite3_value *pRowid = apVal[p->nColumn+3]; |
| 956 if( sqlite3_value_type(pRowid)==SQLITE_NULL ){ |
| 957 pRowid = apVal[1]; |
| 958 } |
| 959 if( sqlite3_value_type(pRowid)!=SQLITE_INTEGER ){ |
| 960 return SQLITE_CONSTRAINT; |
| 961 } |
| 962 *piDocid = sqlite3_value_int64(pRowid); |
| 963 return SQLITE_OK; |
| 964 } |
| 965 |
| 966 /* Locate the statement handle used to insert data into the %_content |
| 967 ** table. The SQL for this statement is: |
| 968 ** |
| 969 ** INSERT INTO %_content VALUES(?, ?, ?, ...) |
| 970 ** |
| 971 ** The statement features N '?' variables, where N is the number of user |
| 972 ** defined columns in the FTS3 table, plus one for the docid field. |
| 973 */ |
| 974 rc = fts3SqlStmt(p, SQL_CONTENT_INSERT, &pContentInsert, &apVal[1]); |
| 975 if( rc==SQLITE_OK && p->zLanguageid ){ |
| 976 rc = sqlite3_bind_int( |
| 977 pContentInsert, p->nColumn+2, |
| 978 sqlite3_value_int(apVal[p->nColumn+4]) |
| 979 ); |
| 980 } |
| 981 if( rc!=SQLITE_OK ) return rc; |
| 982 |
| 983 /* There is a quirk here. The users INSERT statement may have specified |
| 984 ** a value for the "rowid" field, for the "docid" field, or for both. |
| 985 ** Which is a problem, since "rowid" and "docid" are aliases for the |
| 986 ** same value. For example: |
| 987 ** |
| 988 ** INSERT INTO fts3tbl(rowid, docid) VALUES(1, 2); |
| 989 ** |
| 990 ** In FTS3, this is an error. It is an error to specify non-NULL values |
| 991 ** for both docid and some other rowid alias. |
| 992 */ |
| 993 if( SQLITE_NULL!=sqlite3_value_type(apVal[3+p->nColumn]) ){ |
| 994 if( SQLITE_NULL==sqlite3_value_type(apVal[0]) |
| 995 && SQLITE_NULL!=sqlite3_value_type(apVal[1]) |
| 996 ){ |
| 997 /* A rowid/docid conflict. */ |
| 998 return SQLITE_ERROR; |
| 999 } |
| 1000 rc = sqlite3_bind_value(pContentInsert, 1, apVal[3+p->nColumn]); |
| 1001 if( rc!=SQLITE_OK ) return rc; |
| 1002 } |
| 1003 |
| 1004 /* Execute the statement to insert the record. Set *piDocid to the |
| 1005 ** new docid value. |
| 1006 */ |
| 1007 sqlite3_step(pContentInsert); |
| 1008 rc = sqlite3_reset(pContentInsert); |
| 1009 |
| 1010 *piDocid = sqlite3_last_insert_rowid(p->db); |
| 1011 return rc; |
| 1012 } |
| 1013 |
| 1014 |
| 1015 |
| 1016 /* |
| 1017 ** Remove all data from the FTS3 table. Clear the hash table containing |
| 1018 ** pending terms. |
| 1019 */ |
| 1020 static int fts3DeleteAll(Fts3Table *p, int bContent){ |
| 1021 int rc = SQLITE_OK; /* Return code */ |
| 1022 |
| 1023 /* Discard the contents of the pending-terms hash table. */ |
| 1024 sqlite3Fts3PendingTermsClear(p); |
| 1025 |
| 1026 /* Delete everything from the shadow tables. Except, leave %_content as |
| 1027 ** is if bContent is false. */ |
| 1028 assert( p->zContentTbl==0 || bContent==0 ); |
| 1029 if( bContent ) fts3SqlExec(&rc, p, SQL_DELETE_ALL_CONTENT, 0); |
| 1030 fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGMENTS, 0); |
| 1031 fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGDIR, 0); |
| 1032 if( p->bHasDocsize ){ |
| 1033 fts3SqlExec(&rc, p, SQL_DELETE_ALL_DOCSIZE, 0); |
| 1034 } |
| 1035 if( p->bHasStat ){ |
| 1036 fts3SqlExec(&rc, p, SQL_DELETE_ALL_STAT, 0); |
| 1037 } |
| 1038 return rc; |
| 1039 } |
| 1040 |
| 1041 /* |
| 1042 ** |
| 1043 */ |
| 1044 static int langidFromSelect(Fts3Table *p, sqlite3_stmt *pSelect){ |
| 1045 int iLangid = 0; |
| 1046 if( p->zLanguageid ) iLangid = sqlite3_column_int(pSelect, p->nColumn+1); |
| 1047 return iLangid; |
| 1048 } |
| 1049 |
| 1050 /* |
| 1051 ** The first element in the apVal[] array is assumed to contain the docid |
| 1052 ** (an integer) of a row about to be deleted. Remove all terms from the |
| 1053 ** full-text index. |
| 1054 */ |
| 1055 static void fts3DeleteTerms( |
| 1056 int *pRC, /* Result code */ |
| 1057 Fts3Table *p, /* The FTS table to delete from */ |
| 1058 sqlite3_value *pRowid, /* The docid to be deleted */ |
| 1059 u32 *aSz, /* Sizes of deleted document written here */ |
| 1060 int *pbFound /* OUT: Set to true if row really does exist */ |
| 1061 ){ |
| 1062 int rc; |
| 1063 sqlite3_stmt *pSelect; |
| 1064 |
| 1065 assert( *pbFound==0 ); |
| 1066 if( *pRC ) return; |
| 1067 rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pSelect, &pRowid); |
| 1068 if( rc==SQLITE_OK ){ |
| 1069 if( SQLITE_ROW==sqlite3_step(pSelect) ){ |
| 1070 int i; |
| 1071 int iLangid = langidFromSelect(p, pSelect); |
| 1072 rc = fts3PendingTermsDocid(p, iLangid, sqlite3_column_int64(pSelect, 0)); |
| 1073 for(i=1; rc==SQLITE_OK && i<=p->nColumn; i++){ |
| 1074 int iCol = i-1; |
| 1075 if( p->abNotindexed[iCol]==0 ){ |
| 1076 const char *zText = (const char *)sqlite3_column_text(pSelect, i); |
| 1077 rc = fts3PendingTermsAdd(p, iLangid, zText, -1, &aSz[iCol]); |
| 1078 aSz[p->nColumn] += sqlite3_column_bytes(pSelect, i); |
| 1079 } |
| 1080 } |
| 1081 if( rc!=SQLITE_OK ){ |
| 1082 sqlite3_reset(pSelect); |
| 1083 *pRC = rc; |
| 1084 return; |
| 1085 } |
| 1086 *pbFound = 1; |
| 1087 } |
| 1088 rc = sqlite3_reset(pSelect); |
| 1089 }else{ |
| 1090 sqlite3_reset(pSelect); |
| 1091 } |
| 1092 *pRC = rc; |
| 1093 } |
| 1094 |
| 1095 /* |
| 1096 ** Forward declaration to account for the circular dependency between |
| 1097 ** functions fts3SegmentMerge() and fts3AllocateSegdirIdx(). |
| 1098 */ |
| 1099 static int fts3SegmentMerge(Fts3Table *, int, int, int); |
| 1100 |
| 1101 /* |
| 1102 ** This function allocates a new level iLevel index in the segdir table. |
| 1103 ** Usually, indexes are allocated within a level sequentially starting |
| 1104 ** with 0, so the allocated index is one greater than the value returned |
| 1105 ** by: |
| 1106 ** |
| 1107 ** SELECT max(idx) FROM %_segdir WHERE level = :iLevel |
| 1108 ** |
| 1109 ** However, if there are already FTS3_MERGE_COUNT indexes at the requested |
| 1110 ** level, they are merged into a single level (iLevel+1) segment and the |
| 1111 ** allocated index is 0. |
| 1112 ** |
| 1113 ** If successful, *piIdx is set to the allocated index slot and SQLITE_OK |
| 1114 ** returned. Otherwise, an SQLite error code is returned. |
| 1115 */ |
| 1116 static int fts3AllocateSegdirIdx( |
| 1117 Fts3Table *p, |
| 1118 int iLangid, /* Language id */ |
| 1119 int iIndex, /* Index for p->aIndex */ |
| 1120 int iLevel, |
| 1121 int *piIdx |
| 1122 ){ |
| 1123 int rc; /* Return Code */ |
| 1124 sqlite3_stmt *pNextIdx; /* Query for next idx at level iLevel */ |
| 1125 int iNext = 0; /* Result of query pNextIdx */ |
| 1126 |
| 1127 assert( iLangid>=0 ); |
| 1128 assert( p->nIndex>=1 ); |
| 1129 |
| 1130 /* Set variable iNext to the next available segdir index at level iLevel. */ |
| 1131 rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pNextIdx, 0); |
| 1132 if( rc==SQLITE_OK ){ |
| 1133 sqlite3_bind_int64( |
| 1134 pNextIdx, 1, getAbsoluteLevel(p, iLangid, iIndex, iLevel) |
| 1135 ); |
| 1136 if( SQLITE_ROW==sqlite3_step(pNextIdx) ){ |
| 1137 iNext = sqlite3_column_int(pNextIdx, 0); |
| 1138 } |
| 1139 rc = sqlite3_reset(pNextIdx); |
| 1140 } |
| 1141 |
| 1142 if( rc==SQLITE_OK ){ |
| 1143 /* If iNext is FTS3_MERGE_COUNT, indicating that level iLevel is already |
| 1144 ** full, merge all segments in level iLevel into a single iLevel+1 |
| 1145 ** segment and allocate (newly freed) index 0 at level iLevel. Otherwise, |
| 1146 ** if iNext is less than FTS3_MERGE_COUNT, allocate index iNext. |
| 1147 */ |
| 1148 if( iNext>=FTS3_MERGE_COUNT ){ |
| 1149 fts3LogMerge(16, getAbsoluteLevel(p, iLangid, iIndex, iLevel)); |
| 1150 rc = fts3SegmentMerge(p, iLangid, iIndex, iLevel); |
| 1151 *piIdx = 0; |
| 1152 }else{ |
| 1153 *piIdx = iNext; |
| 1154 } |
| 1155 } |
| 1156 |
| 1157 return rc; |
| 1158 } |
| 1159 |
| 1160 /* |
| 1161 ** The %_segments table is declared as follows: |
| 1162 ** |
| 1163 ** CREATE TABLE %_segments(blockid INTEGER PRIMARY KEY, block BLOB) |
| 1164 ** |
| 1165 ** This function reads data from a single row of the %_segments table. The |
| 1166 ** specific row is identified by the iBlockid parameter. If paBlob is not |
| 1167 ** NULL, then a buffer is allocated using sqlite3_malloc() and populated |
| 1168 ** with the contents of the blob stored in the "block" column of the |
| 1169 ** identified table row is. Whether or not paBlob is NULL, *pnBlob is set |
| 1170 ** to the size of the blob in bytes before returning. |
| 1171 ** |
| 1172 ** If an error occurs, or the table does not contain the specified row, |
| 1173 ** an SQLite error code is returned. Otherwise, SQLITE_OK is returned. If |
| 1174 ** paBlob is non-NULL, then it is the responsibility of the caller to |
| 1175 ** eventually free the returned buffer. |
| 1176 ** |
| 1177 ** This function may leave an open sqlite3_blob* handle in the |
| 1178 ** Fts3Table.pSegments variable. This handle is reused by subsequent calls |
| 1179 ** to this function. The handle may be closed by calling the |
| 1180 ** sqlite3Fts3SegmentsClose() function. Reusing a blob handle is a handy |
| 1181 ** performance improvement, but the blob handle should always be closed |
| 1182 ** before control is returned to the user (to prevent a lock being held |
| 1183 ** on the database file for longer than necessary). Thus, any virtual table |
| 1184 ** method (xFilter etc.) that may directly or indirectly call this function |
| 1185 ** must call sqlite3Fts3SegmentsClose() before returning. |
| 1186 */ |
| 1187 int sqlite3Fts3ReadBlock( |
| 1188 Fts3Table *p, /* FTS3 table handle */ |
| 1189 sqlite3_int64 iBlockid, /* Access the row with blockid=$iBlockid */ |
| 1190 char **paBlob, /* OUT: Blob data in malloc'd buffer */ |
| 1191 int *pnBlob, /* OUT: Size of blob data */ |
| 1192 int *pnLoad /* OUT: Bytes actually loaded */ |
| 1193 ){ |
| 1194 int rc; /* Return code */ |
| 1195 |
| 1196 /* pnBlob must be non-NULL. paBlob may be NULL or non-NULL. */ |
| 1197 assert( pnBlob ); |
| 1198 |
| 1199 if( p->pSegments ){ |
| 1200 rc = sqlite3_blob_reopen(p->pSegments, iBlockid); |
| 1201 }else{ |
| 1202 if( 0==p->zSegmentsTbl ){ |
| 1203 p->zSegmentsTbl = sqlite3_mprintf("%s_segments", p->zName); |
| 1204 if( 0==p->zSegmentsTbl ) return SQLITE_NOMEM; |
| 1205 } |
| 1206 rc = sqlite3_blob_open( |
| 1207 p->db, p->zDb, p->zSegmentsTbl, "block", iBlockid, 0, &p->pSegments |
| 1208 ); |
| 1209 } |
| 1210 |
| 1211 if( rc==SQLITE_OK ){ |
| 1212 int nByte = sqlite3_blob_bytes(p->pSegments); |
| 1213 *pnBlob = nByte; |
| 1214 if( paBlob ){ |
| 1215 char *aByte = sqlite3_malloc(nByte + FTS3_NODE_PADDING); |
| 1216 if( !aByte ){ |
| 1217 rc = SQLITE_NOMEM; |
| 1218 }else{ |
| 1219 if( pnLoad && nByte>(FTS3_NODE_CHUNK_THRESHOLD) ){ |
| 1220 nByte = FTS3_NODE_CHUNKSIZE; |
| 1221 *pnLoad = nByte; |
| 1222 } |
| 1223 rc = sqlite3_blob_read(p->pSegments, aByte, nByte, 0); |
| 1224 memset(&aByte[nByte], 0, FTS3_NODE_PADDING); |
| 1225 if( rc!=SQLITE_OK ){ |
| 1226 sqlite3_free(aByte); |
| 1227 aByte = 0; |
| 1228 } |
| 1229 } |
| 1230 *paBlob = aByte; |
| 1231 } |
| 1232 } |
| 1233 |
| 1234 return rc; |
| 1235 } |
| 1236 |
| 1237 /* |
| 1238 ** Close the blob handle at p->pSegments, if it is open. See comments above |
| 1239 ** the sqlite3Fts3ReadBlock() function for details. |
| 1240 */ |
| 1241 void sqlite3Fts3SegmentsClose(Fts3Table *p){ |
| 1242 sqlite3_blob_close(p->pSegments); |
| 1243 p->pSegments = 0; |
| 1244 } |
| 1245 |
| 1246 static int fts3SegReaderIncrRead(Fts3SegReader *pReader){ |
| 1247 int nRead; /* Number of bytes to read */ |
| 1248 int rc; /* Return code */ |
| 1249 |
| 1250 nRead = MIN(pReader->nNode - pReader->nPopulate, FTS3_NODE_CHUNKSIZE); |
| 1251 rc = sqlite3_blob_read( |
| 1252 pReader->pBlob, |
| 1253 &pReader->aNode[pReader->nPopulate], |
| 1254 nRead, |
| 1255 pReader->nPopulate |
| 1256 ); |
| 1257 |
| 1258 if( rc==SQLITE_OK ){ |
| 1259 pReader->nPopulate += nRead; |
| 1260 memset(&pReader->aNode[pReader->nPopulate], 0, FTS3_NODE_PADDING); |
| 1261 if( pReader->nPopulate==pReader->nNode ){ |
| 1262 sqlite3_blob_close(pReader->pBlob); |
| 1263 pReader->pBlob = 0; |
| 1264 pReader->nPopulate = 0; |
| 1265 } |
| 1266 } |
| 1267 return rc; |
| 1268 } |
| 1269 |
| 1270 static int fts3SegReaderRequire(Fts3SegReader *pReader, char *pFrom, int nByte){ |
| 1271 int rc = SQLITE_OK; |
| 1272 assert( !pReader->pBlob |
| 1273 || (pFrom>=pReader->aNode && pFrom<&pReader->aNode[pReader->nNode]) |
| 1274 ); |
| 1275 while( pReader->pBlob && rc==SQLITE_OK |
| 1276 && (pFrom - pReader->aNode + nByte)>pReader->nPopulate |
| 1277 ){ |
| 1278 rc = fts3SegReaderIncrRead(pReader); |
| 1279 } |
| 1280 return rc; |
| 1281 } |
| 1282 |
| 1283 /* |
| 1284 ** Set an Fts3SegReader cursor to point at EOF. |
| 1285 */ |
| 1286 static void fts3SegReaderSetEof(Fts3SegReader *pSeg){ |
| 1287 if( !fts3SegReaderIsRootOnly(pSeg) ){ |
| 1288 sqlite3_free(pSeg->aNode); |
| 1289 sqlite3_blob_close(pSeg->pBlob); |
| 1290 pSeg->pBlob = 0; |
| 1291 } |
| 1292 pSeg->aNode = 0; |
| 1293 } |
| 1294 |
| 1295 /* |
| 1296 ** Move the iterator passed as the first argument to the next term in the |
| 1297 ** segment. If successful, SQLITE_OK is returned. If there is no next term, |
| 1298 ** SQLITE_DONE. Otherwise, an SQLite error code. |
| 1299 */ |
| 1300 static int fts3SegReaderNext( |
| 1301 Fts3Table *p, |
| 1302 Fts3SegReader *pReader, |
| 1303 int bIncr |
| 1304 ){ |
| 1305 int rc; /* Return code of various sub-routines */ |
| 1306 char *pNext; /* Cursor variable */ |
| 1307 int nPrefix; /* Number of bytes in term prefix */ |
| 1308 int nSuffix; /* Number of bytes in term suffix */ |
| 1309 |
| 1310 if( !pReader->aDoclist ){ |
| 1311 pNext = pReader->aNode; |
| 1312 }else{ |
| 1313 pNext = &pReader->aDoclist[pReader->nDoclist]; |
| 1314 } |
| 1315 |
| 1316 if( !pNext || pNext>=&pReader->aNode[pReader->nNode] ){ |
| 1317 |
| 1318 if( fts3SegReaderIsPending(pReader) ){ |
| 1319 Fts3HashElem *pElem = *(pReader->ppNextElem); |
| 1320 if( pElem==0 ){ |
| 1321 pReader->aNode = 0; |
| 1322 }else{ |
| 1323 PendingList *pList = (PendingList *)fts3HashData(pElem); |
| 1324 pReader->zTerm = (char *)fts3HashKey(pElem); |
| 1325 pReader->nTerm = fts3HashKeysize(pElem); |
| 1326 pReader->nNode = pReader->nDoclist = pList->nData + 1; |
| 1327 pReader->aNode = pReader->aDoclist = pList->aData; |
| 1328 pReader->ppNextElem++; |
| 1329 assert( pReader->aNode ); |
| 1330 } |
| 1331 return SQLITE_OK; |
| 1332 } |
| 1333 |
| 1334 fts3SegReaderSetEof(pReader); |
| 1335 |
| 1336 /* If iCurrentBlock>=iLeafEndBlock, this is an EOF condition. All leaf |
| 1337 ** blocks have already been traversed. */ |
| 1338 assert( pReader->iCurrentBlock<=pReader->iLeafEndBlock ); |
| 1339 if( pReader->iCurrentBlock>=pReader->iLeafEndBlock ){ |
| 1340 return SQLITE_OK; |
| 1341 } |
| 1342 |
| 1343 rc = sqlite3Fts3ReadBlock( |
| 1344 p, ++pReader->iCurrentBlock, &pReader->aNode, &pReader->nNode, |
| 1345 (bIncr ? &pReader->nPopulate : 0) |
| 1346 ); |
| 1347 if( rc!=SQLITE_OK ) return rc; |
| 1348 assert( pReader->pBlob==0 ); |
| 1349 if( bIncr && pReader->nPopulate<pReader->nNode ){ |
| 1350 pReader->pBlob = p->pSegments; |
| 1351 p->pSegments = 0; |
| 1352 } |
| 1353 pNext = pReader->aNode; |
| 1354 } |
| 1355 |
| 1356 assert( !fts3SegReaderIsPending(pReader) ); |
| 1357 |
| 1358 rc = fts3SegReaderRequire(pReader, pNext, FTS3_VARINT_MAX*2); |
| 1359 if( rc!=SQLITE_OK ) return rc; |
| 1360 |
| 1361 /* Because of the FTS3_NODE_PADDING bytes of padding, the following is |
| 1362 ** safe (no risk of overread) even if the node data is corrupted. */ |
| 1363 pNext += fts3GetVarint32(pNext, &nPrefix); |
| 1364 pNext += fts3GetVarint32(pNext, &nSuffix); |
| 1365 if( nPrefix<0 || nSuffix<=0 |
| 1366 || &pNext[nSuffix]>&pReader->aNode[pReader->nNode] |
| 1367 ){ |
| 1368 return FTS_CORRUPT_VTAB; |
| 1369 } |
| 1370 |
| 1371 if( nPrefix+nSuffix>pReader->nTermAlloc ){ |
| 1372 int nNew = (nPrefix+nSuffix)*2; |
| 1373 char *zNew = sqlite3_realloc(pReader->zTerm, nNew); |
| 1374 if( !zNew ){ |
| 1375 return SQLITE_NOMEM; |
| 1376 } |
| 1377 pReader->zTerm = zNew; |
| 1378 pReader->nTermAlloc = nNew; |
| 1379 } |
| 1380 |
| 1381 rc = fts3SegReaderRequire(pReader, pNext, nSuffix+FTS3_VARINT_MAX); |
| 1382 if( rc!=SQLITE_OK ) return rc; |
| 1383 |
| 1384 memcpy(&pReader->zTerm[nPrefix], pNext, nSuffix); |
| 1385 pReader->nTerm = nPrefix+nSuffix; |
| 1386 pNext += nSuffix; |
| 1387 pNext += fts3GetVarint32(pNext, &pReader->nDoclist); |
| 1388 pReader->aDoclist = pNext; |
| 1389 pReader->pOffsetList = 0; |
| 1390 |
| 1391 /* Check that the doclist does not appear to extend past the end of the |
| 1392 ** b-tree node. And that the final byte of the doclist is 0x00. If either |
| 1393 ** of these statements is untrue, then the data structure is corrupt. |
| 1394 */ |
| 1395 if( &pReader->aDoclist[pReader->nDoclist]>&pReader->aNode[pReader->nNode] |
| 1396 || (pReader->nPopulate==0 && pReader->aDoclist[pReader->nDoclist-1]) |
| 1397 ){ |
| 1398 return FTS_CORRUPT_VTAB; |
| 1399 } |
| 1400 return SQLITE_OK; |
| 1401 } |
| 1402 |
| 1403 /* |
| 1404 ** Set the SegReader to point to the first docid in the doclist associated |
| 1405 ** with the current term. |
| 1406 */ |
| 1407 static int fts3SegReaderFirstDocid(Fts3Table *pTab, Fts3SegReader *pReader){ |
| 1408 int rc = SQLITE_OK; |
| 1409 assert( pReader->aDoclist ); |
| 1410 assert( !pReader->pOffsetList ); |
| 1411 if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ |
| 1412 u8 bEof = 0; |
| 1413 pReader->iDocid = 0; |
| 1414 pReader->nOffsetList = 0; |
| 1415 sqlite3Fts3DoclistPrev(0, |
| 1416 pReader->aDoclist, pReader->nDoclist, &pReader->pOffsetList, |
| 1417 &pReader->iDocid, &pReader->nOffsetList, &bEof |
| 1418 ); |
| 1419 }else{ |
| 1420 rc = fts3SegReaderRequire(pReader, pReader->aDoclist, FTS3_VARINT_MAX); |
| 1421 if( rc==SQLITE_OK ){ |
| 1422 int n = sqlite3Fts3GetVarint(pReader->aDoclist, &pReader->iDocid); |
| 1423 pReader->pOffsetList = &pReader->aDoclist[n]; |
| 1424 } |
| 1425 } |
| 1426 return rc; |
| 1427 } |
| 1428 |
| 1429 /* |
| 1430 ** Advance the SegReader to point to the next docid in the doclist |
| 1431 ** associated with the current term. |
| 1432 ** |
| 1433 ** If arguments ppOffsetList and pnOffsetList are not NULL, then |
| 1434 ** *ppOffsetList is set to point to the first column-offset list |
| 1435 ** in the doclist entry (i.e. immediately past the docid varint). |
| 1436 ** *pnOffsetList is set to the length of the set of column-offset |
| 1437 ** lists, not including the nul-terminator byte. For example: |
| 1438 */ |
| 1439 static int fts3SegReaderNextDocid( |
| 1440 Fts3Table *pTab, |
| 1441 Fts3SegReader *pReader, /* Reader to advance to next docid */ |
| 1442 char **ppOffsetList, /* OUT: Pointer to current position-list */ |
| 1443 int *pnOffsetList /* OUT: Length of *ppOffsetList in bytes */ |
| 1444 ){ |
| 1445 int rc = SQLITE_OK; |
| 1446 char *p = pReader->pOffsetList; |
| 1447 char c = 0; |
| 1448 |
| 1449 assert( p ); |
| 1450 |
| 1451 if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ |
| 1452 /* A pending-terms seg-reader for an FTS4 table that uses order=desc. |
| 1453 ** Pending-terms doclists are always built up in ascending order, so |
| 1454 ** we have to iterate through them backwards here. */ |
| 1455 u8 bEof = 0; |
| 1456 if( ppOffsetList ){ |
| 1457 *ppOffsetList = pReader->pOffsetList; |
| 1458 *pnOffsetList = pReader->nOffsetList - 1; |
| 1459 } |
| 1460 sqlite3Fts3DoclistPrev(0, |
| 1461 pReader->aDoclist, pReader->nDoclist, &p, &pReader->iDocid, |
| 1462 &pReader->nOffsetList, &bEof |
| 1463 ); |
| 1464 if( bEof ){ |
| 1465 pReader->pOffsetList = 0; |
| 1466 }else{ |
| 1467 pReader->pOffsetList = p; |
| 1468 } |
| 1469 }else{ |
| 1470 char *pEnd = &pReader->aDoclist[pReader->nDoclist]; |
| 1471 |
| 1472 /* Pointer p currently points at the first byte of an offset list. The |
| 1473 ** following block advances it to point one byte past the end of |
| 1474 ** the same offset list. */ |
| 1475 while( 1 ){ |
| 1476 |
| 1477 /* The following line of code (and the "p++" below the while() loop) is |
| 1478 ** normally all that is required to move pointer p to the desired |
| 1479 ** position. The exception is if this node is being loaded from disk |
| 1480 ** incrementally and pointer "p" now points to the first byte past |
| 1481 ** the populated part of pReader->aNode[]. |
| 1482 */ |
| 1483 while( *p | c ) c = *p++ & 0x80; |
| 1484 assert( *p==0 ); |
| 1485 |
| 1486 if( pReader->pBlob==0 || p<&pReader->aNode[pReader->nPopulate] ) break; |
| 1487 rc = fts3SegReaderIncrRead(pReader); |
| 1488 if( rc!=SQLITE_OK ) return rc; |
| 1489 } |
| 1490 p++; |
| 1491 |
| 1492 /* If required, populate the output variables with a pointer to and the |
| 1493 ** size of the previous offset-list. |
| 1494 */ |
| 1495 if( ppOffsetList ){ |
| 1496 *ppOffsetList = pReader->pOffsetList; |
| 1497 *pnOffsetList = (int)(p - pReader->pOffsetList - 1); |
| 1498 } |
| 1499 |
| 1500 /* List may have been edited in place by fts3EvalNearTrim() */ |
| 1501 while( p<pEnd && *p==0 ) p++; |
| 1502 |
| 1503 /* If there are no more entries in the doclist, set pOffsetList to |
| 1504 ** NULL. Otherwise, set Fts3SegReader.iDocid to the next docid and |
| 1505 ** Fts3SegReader.pOffsetList to point to the next offset list before |
| 1506 ** returning. |
| 1507 */ |
| 1508 if( p>=pEnd ){ |
| 1509 pReader->pOffsetList = 0; |
| 1510 }else{ |
| 1511 rc = fts3SegReaderRequire(pReader, p, FTS3_VARINT_MAX); |
| 1512 if( rc==SQLITE_OK ){ |
| 1513 sqlite3_int64 iDelta; |
| 1514 pReader->pOffsetList = p + sqlite3Fts3GetVarint(p, &iDelta); |
| 1515 if( pTab->bDescIdx ){ |
| 1516 pReader->iDocid -= iDelta; |
| 1517 }else{ |
| 1518 pReader->iDocid += iDelta; |
| 1519 } |
| 1520 } |
| 1521 } |
| 1522 } |
| 1523 |
| 1524 return SQLITE_OK; |
| 1525 } |
| 1526 |
| 1527 |
| 1528 int sqlite3Fts3MsrOvfl( |
| 1529 Fts3Cursor *pCsr, |
| 1530 Fts3MultiSegReader *pMsr, |
| 1531 int *pnOvfl |
| 1532 ){ |
| 1533 Fts3Table *p = (Fts3Table*)pCsr->base.pVtab; |
| 1534 int nOvfl = 0; |
| 1535 int ii; |
| 1536 int rc = SQLITE_OK; |
| 1537 int pgsz = p->nPgsz; |
| 1538 |
| 1539 assert( p->bFts4 ); |
| 1540 assert( pgsz>0 ); |
| 1541 |
| 1542 for(ii=0; rc==SQLITE_OK && ii<pMsr->nSegment; ii++){ |
| 1543 Fts3SegReader *pReader = pMsr->apSegment[ii]; |
| 1544 if( !fts3SegReaderIsPending(pReader) |
| 1545 && !fts3SegReaderIsRootOnly(pReader) |
| 1546 ){ |
| 1547 sqlite3_int64 jj; |
| 1548 for(jj=pReader->iStartBlock; jj<=pReader->iLeafEndBlock; jj++){ |
| 1549 int nBlob; |
| 1550 rc = sqlite3Fts3ReadBlock(p, jj, 0, &nBlob, 0); |
| 1551 if( rc!=SQLITE_OK ) break; |
| 1552 if( (nBlob+35)>pgsz ){ |
| 1553 nOvfl += (nBlob + 34)/pgsz; |
| 1554 } |
| 1555 } |
| 1556 } |
| 1557 } |
| 1558 *pnOvfl = nOvfl; |
| 1559 return rc; |
| 1560 } |
| 1561 |
| 1562 /* |
| 1563 ** Free all allocations associated with the iterator passed as the |
| 1564 ** second argument. |
| 1565 */ |
| 1566 void sqlite3Fts3SegReaderFree(Fts3SegReader *pReader){ |
| 1567 if( pReader && !fts3SegReaderIsPending(pReader) ){ |
| 1568 sqlite3_free(pReader->zTerm); |
| 1569 if( !fts3SegReaderIsRootOnly(pReader) ){ |
| 1570 sqlite3_free(pReader->aNode); |
| 1571 sqlite3_blob_close(pReader->pBlob); |
| 1572 } |
| 1573 } |
| 1574 sqlite3_free(pReader); |
| 1575 } |
| 1576 |
| 1577 /* |
| 1578 ** Allocate a new SegReader object. |
| 1579 */ |
| 1580 int sqlite3Fts3SegReaderNew( |
| 1581 int iAge, /* Segment "age". */ |
| 1582 int bLookup, /* True for a lookup only */ |
| 1583 sqlite3_int64 iStartLeaf, /* First leaf to traverse */ |
| 1584 sqlite3_int64 iEndLeaf, /* Final leaf to traverse */ |
| 1585 sqlite3_int64 iEndBlock, /* Final block of segment */ |
| 1586 const char *zRoot, /* Buffer containing root node */ |
| 1587 int nRoot, /* Size of buffer containing root node */ |
| 1588 Fts3SegReader **ppReader /* OUT: Allocated Fts3SegReader */ |
| 1589 ){ |
| 1590 Fts3SegReader *pReader; /* Newly allocated SegReader object */ |
| 1591 int nExtra = 0; /* Bytes to allocate segment root node */ |
| 1592 |
| 1593 assert( iStartLeaf<=iEndLeaf ); |
| 1594 if( iStartLeaf==0 ){ |
| 1595 nExtra = nRoot + FTS3_NODE_PADDING; |
| 1596 } |
| 1597 |
| 1598 pReader = (Fts3SegReader *)sqlite3_malloc(sizeof(Fts3SegReader) + nExtra); |
| 1599 if( !pReader ){ |
| 1600 return SQLITE_NOMEM; |
| 1601 } |
| 1602 memset(pReader, 0, sizeof(Fts3SegReader)); |
| 1603 pReader->iIdx = iAge; |
| 1604 pReader->bLookup = bLookup!=0; |
| 1605 pReader->iStartBlock = iStartLeaf; |
| 1606 pReader->iLeafEndBlock = iEndLeaf; |
| 1607 pReader->iEndBlock = iEndBlock; |
| 1608 |
| 1609 if( nExtra ){ |
| 1610 /* The entire segment is stored in the root node. */ |
| 1611 pReader->aNode = (char *)&pReader[1]; |
| 1612 pReader->rootOnly = 1; |
| 1613 pReader->nNode = nRoot; |
| 1614 memcpy(pReader->aNode, zRoot, nRoot); |
| 1615 memset(&pReader->aNode[nRoot], 0, FTS3_NODE_PADDING); |
| 1616 }else{ |
| 1617 pReader->iCurrentBlock = iStartLeaf-1; |
| 1618 } |
| 1619 *ppReader = pReader; |
| 1620 return SQLITE_OK; |
| 1621 } |
| 1622 |
| 1623 /* |
| 1624 ** This is a comparison function used as a qsort() callback when sorting |
| 1625 ** an array of pending terms by term. This occurs as part of flushing |
| 1626 ** the contents of the pending-terms hash table to the database. |
| 1627 */ |
| 1628 static int fts3CompareElemByTerm(const void *lhs, const void *rhs){ |
| 1629 char *z1 = fts3HashKey(*(Fts3HashElem **)lhs); |
| 1630 char *z2 = fts3HashKey(*(Fts3HashElem **)rhs); |
| 1631 int n1 = fts3HashKeysize(*(Fts3HashElem **)lhs); |
| 1632 int n2 = fts3HashKeysize(*(Fts3HashElem **)rhs); |
| 1633 |
| 1634 int n = (n1<n2 ? n1 : n2); |
| 1635 int c = memcmp(z1, z2, n); |
| 1636 if( c==0 ){ |
| 1637 c = n1 - n2; |
| 1638 } |
| 1639 return c; |
| 1640 } |
| 1641 |
| 1642 /* |
| 1643 ** This function is used to allocate an Fts3SegReader that iterates through |
| 1644 ** a subset of the terms stored in the Fts3Table.pendingTerms array. |
| 1645 ** |
| 1646 ** If the isPrefixIter parameter is zero, then the returned SegReader iterates |
| 1647 ** through each term in the pending-terms table. Or, if isPrefixIter is |
| 1648 ** non-zero, it iterates through each term and its prefixes. For example, if |
| 1649 ** the pending terms hash table contains the terms "sqlite", "mysql" and |
| 1650 ** "firebird", then the iterator visits the following 'terms' (in the order |
| 1651 ** shown): |
| 1652 ** |
| 1653 ** f fi fir fire fireb firebi firebir firebird |
| 1654 ** m my mys mysq mysql |
| 1655 ** s sq sql sqli sqlit sqlite |
| 1656 ** |
| 1657 ** Whereas if isPrefixIter is zero, the terms visited are: |
| 1658 ** |
| 1659 ** firebird mysql sqlite |
| 1660 */ |
| 1661 int sqlite3Fts3SegReaderPending( |
| 1662 Fts3Table *p, /* Virtual table handle */ |
| 1663 int iIndex, /* Index for p->aIndex */ |
| 1664 const char *zTerm, /* Term to search for */ |
| 1665 int nTerm, /* Size of buffer zTerm */ |
| 1666 int bPrefix, /* True for a prefix iterator */ |
| 1667 Fts3SegReader **ppReader /* OUT: SegReader for pending-terms */ |
| 1668 ){ |
| 1669 Fts3SegReader *pReader = 0; /* Fts3SegReader object to return */ |
| 1670 Fts3HashElem *pE; /* Iterator variable */ |
| 1671 Fts3HashElem **aElem = 0; /* Array of term hash entries to scan */ |
| 1672 int nElem = 0; /* Size of array at aElem */ |
| 1673 int rc = SQLITE_OK; /* Return Code */ |
| 1674 Fts3Hash *pHash; |
| 1675 |
| 1676 pHash = &p->aIndex[iIndex].hPending; |
| 1677 if( bPrefix ){ |
| 1678 int nAlloc = 0; /* Size of allocated array at aElem */ |
| 1679 |
| 1680 for(pE=fts3HashFirst(pHash); pE; pE=fts3HashNext(pE)){ |
| 1681 char *zKey = (char *)fts3HashKey(pE); |
| 1682 int nKey = fts3HashKeysize(pE); |
| 1683 if( nTerm==0 || (nKey>=nTerm && 0==memcmp(zKey, zTerm, nTerm)) ){ |
| 1684 if( nElem==nAlloc ){ |
| 1685 Fts3HashElem **aElem2; |
| 1686 nAlloc += 16; |
| 1687 aElem2 = (Fts3HashElem **)sqlite3_realloc( |
| 1688 aElem, nAlloc*sizeof(Fts3HashElem *) |
| 1689 ); |
| 1690 if( !aElem2 ){ |
| 1691 rc = SQLITE_NOMEM; |
| 1692 nElem = 0; |
| 1693 break; |
| 1694 } |
| 1695 aElem = aElem2; |
| 1696 } |
| 1697 |
| 1698 aElem[nElem++] = pE; |
| 1699 } |
| 1700 } |
| 1701 |
| 1702 /* If more than one term matches the prefix, sort the Fts3HashElem |
| 1703 ** objects in term order using qsort(). This uses the same comparison |
| 1704 ** callback as is used when flushing terms to disk. |
| 1705 */ |
| 1706 if( nElem>1 ){ |
| 1707 qsort(aElem, nElem, sizeof(Fts3HashElem *), fts3CompareElemByTerm); |
| 1708 } |
| 1709 |
| 1710 }else{ |
| 1711 /* The query is a simple term lookup that matches at most one term in |
| 1712 ** the index. All that is required is a straight hash-lookup. |
| 1713 ** |
| 1714 ** Because the stack address of pE may be accessed via the aElem pointer |
| 1715 ** below, the "Fts3HashElem *pE" must be declared so that it is valid |
| 1716 ** within this entire function, not just this "else{...}" block. |
| 1717 */ |
| 1718 pE = fts3HashFindElem(pHash, zTerm, nTerm); |
| 1719 if( pE ){ |
| 1720 aElem = &pE; |
| 1721 nElem = 1; |
| 1722 } |
| 1723 } |
| 1724 |
| 1725 if( nElem>0 ){ |
| 1726 int nByte = sizeof(Fts3SegReader) + (nElem+1)*sizeof(Fts3HashElem *); |
| 1727 pReader = (Fts3SegReader *)sqlite3_malloc(nByte); |
| 1728 if( !pReader ){ |
| 1729 rc = SQLITE_NOMEM; |
| 1730 }else{ |
| 1731 memset(pReader, 0, nByte); |
| 1732 pReader->iIdx = 0x7FFFFFFF; |
| 1733 pReader->ppNextElem = (Fts3HashElem **)&pReader[1]; |
| 1734 memcpy(pReader->ppNextElem, aElem, nElem*sizeof(Fts3HashElem *)); |
| 1735 } |
| 1736 } |
| 1737 |
| 1738 if( bPrefix ){ |
| 1739 sqlite3_free(aElem); |
| 1740 } |
| 1741 *ppReader = pReader; |
| 1742 return rc; |
| 1743 } |
| 1744 |
| 1745 /* |
| 1746 ** Compare the entries pointed to by two Fts3SegReader structures. |
| 1747 ** Comparison is as follows: |
| 1748 ** |
| 1749 ** 1) EOF is greater than not EOF. |
| 1750 ** |
| 1751 ** 2) The current terms (if any) are compared using memcmp(). If one |
| 1752 ** term is a prefix of another, the longer term is considered the |
| 1753 ** larger. |
| 1754 ** |
| 1755 ** 3) By segment age. An older segment is considered larger. |
| 1756 */ |
| 1757 static int fts3SegReaderCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ |
| 1758 int rc; |
| 1759 if( pLhs->aNode && pRhs->aNode ){ |
| 1760 int rc2 = pLhs->nTerm - pRhs->nTerm; |
| 1761 if( rc2<0 ){ |
| 1762 rc = memcmp(pLhs->zTerm, pRhs->zTerm, pLhs->nTerm); |
| 1763 }else{ |
| 1764 rc = memcmp(pLhs->zTerm, pRhs->zTerm, pRhs->nTerm); |
| 1765 } |
| 1766 if( rc==0 ){ |
| 1767 rc = rc2; |
| 1768 } |
| 1769 }else{ |
| 1770 rc = (pLhs->aNode==0) - (pRhs->aNode==0); |
| 1771 } |
| 1772 if( rc==0 ){ |
| 1773 rc = pRhs->iIdx - pLhs->iIdx; |
| 1774 } |
| 1775 assert( rc!=0 ); |
| 1776 return rc; |
| 1777 } |
| 1778 |
| 1779 /* |
| 1780 ** A different comparison function for SegReader structures. In this |
| 1781 ** version, it is assumed that each SegReader points to an entry in |
| 1782 ** a doclist for identical terms. Comparison is made as follows: |
| 1783 ** |
| 1784 ** 1) EOF (end of doclist in this case) is greater than not EOF. |
| 1785 ** |
| 1786 ** 2) By current docid. |
| 1787 ** |
| 1788 ** 3) By segment age. An older segment is considered larger. |
| 1789 */ |
| 1790 static int fts3SegReaderDoclistCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ |
| 1791 int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); |
| 1792 if( rc==0 ){ |
| 1793 if( pLhs->iDocid==pRhs->iDocid ){ |
| 1794 rc = pRhs->iIdx - pLhs->iIdx; |
| 1795 }else{ |
| 1796 rc = (pLhs->iDocid > pRhs->iDocid) ? 1 : -1; |
| 1797 } |
| 1798 } |
| 1799 assert( pLhs->aNode && pRhs->aNode ); |
| 1800 return rc; |
| 1801 } |
| 1802 static int fts3SegReaderDoclistCmpRev(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ |
| 1803 int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); |
| 1804 if( rc==0 ){ |
| 1805 if( pLhs->iDocid==pRhs->iDocid ){ |
| 1806 rc = pRhs->iIdx - pLhs->iIdx; |
| 1807 }else{ |
| 1808 rc = (pLhs->iDocid < pRhs->iDocid) ? 1 : -1; |
| 1809 } |
| 1810 } |
| 1811 assert( pLhs->aNode && pRhs->aNode ); |
| 1812 return rc; |
| 1813 } |
| 1814 |
| 1815 /* |
| 1816 ** Compare the term that the Fts3SegReader object passed as the first argument |
| 1817 ** points to with the term specified by arguments zTerm and nTerm. |
| 1818 ** |
| 1819 ** If the pSeg iterator is already at EOF, return 0. Otherwise, return |
| 1820 ** -ve if the pSeg term is less than zTerm/nTerm, 0 if the two terms are |
| 1821 ** equal, or +ve if the pSeg term is greater than zTerm/nTerm. |
| 1822 */ |
| 1823 static int fts3SegReaderTermCmp( |
| 1824 Fts3SegReader *pSeg, /* Segment reader object */ |
| 1825 const char *zTerm, /* Term to compare to */ |
| 1826 int nTerm /* Size of term zTerm in bytes */ |
| 1827 ){ |
| 1828 int res = 0; |
| 1829 if( pSeg->aNode ){ |
| 1830 if( pSeg->nTerm>nTerm ){ |
| 1831 res = memcmp(pSeg->zTerm, zTerm, nTerm); |
| 1832 }else{ |
| 1833 res = memcmp(pSeg->zTerm, zTerm, pSeg->nTerm); |
| 1834 } |
| 1835 if( res==0 ){ |
| 1836 res = pSeg->nTerm-nTerm; |
| 1837 } |
| 1838 } |
| 1839 return res; |
| 1840 } |
| 1841 |
| 1842 /* |
| 1843 ** Argument apSegment is an array of nSegment elements. It is known that |
| 1844 ** the final (nSegment-nSuspect) members are already in sorted order |
| 1845 ** (according to the comparison function provided). This function shuffles |
| 1846 ** the array around until all entries are in sorted order. |
| 1847 */ |
| 1848 static void fts3SegReaderSort( |
| 1849 Fts3SegReader **apSegment, /* Array to sort entries of */ |
| 1850 int nSegment, /* Size of apSegment array */ |
| 1851 int nSuspect, /* Unsorted entry count */ |
| 1852 int (*xCmp)(Fts3SegReader *, Fts3SegReader *) /* Comparison function */ |
| 1853 ){ |
| 1854 int i; /* Iterator variable */ |
| 1855 |
| 1856 assert( nSuspect<=nSegment ); |
| 1857 |
| 1858 if( nSuspect==nSegment ) nSuspect--; |
| 1859 for(i=nSuspect-1; i>=0; i--){ |
| 1860 int j; |
| 1861 for(j=i; j<(nSegment-1); j++){ |
| 1862 Fts3SegReader *pTmp; |
| 1863 if( xCmp(apSegment[j], apSegment[j+1])<0 ) break; |
| 1864 pTmp = apSegment[j+1]; |
| 1865 apSegment[j+1] = apSegment[j]; |
| 1866 apSegment[j] = pTmp; |
| 1867 } |
| 1868 } |
| 1869 |
| 1870 #ifndef NDEBUG |
| 1871 /* Check that the list really is sorted now. */ |
| 1872 for(i=0; i<(nSuspect-1); i++){ |
| 1873 assert( xCmp(apSegment[i], apSegment[i+1])<0 ); |
| 1874 } |
| 1875 #endif |
| 1876 } |
| 1877 |
| 1878 /* |
| 1879 ** Insert a record into the %_segments table. |
| 1880 */ |
| 1881 static int fts3WriteSegment( |
| 1882 Fts3Table *p, /* Virtual table handle */ |
| 1883 sqlite3_int64 iBlock, /* Block id for new block */ |
| 1884 char *z, /* Pointer to buffer containing block data */ |
| 1885 int n /* Size of buffer z in bytes */ |
| 1886 ){ |
| 1887 sqlite3_stmt *pStmt; |
| 1888 int rc = fts3SqlStmt(p, SQL_INSERT_SEGMENTS, &pStmt, 0); |
| 1889 if( rc==SQLITE_OK ){ |
| 1890 sqlite3_bind_int64(pStmt, 1, iBlock); |
| 1891 sqlite3_bind_blob(pStmt, 2, z, n, SQLITE_STATIC); |
| 1892 sqlite3_step(pStmt); |
| 1893 rc = sqlite3_reset(pStmt); |
| 1894 } |
| 1895 return rc; |
| 1896 } |
| 1897 |
| 1898 /* |
| 1899 ** Find the largest relative level number in the table. If successful, set |
| 1900 ** *pnMax to this value and return SQLITE_OK. Otherwise, if an error occurs, |
| 1901 ** set *pnMax to zero and return an SQLite error code. |
| 1902 */ |
| 1903 int sqlite3Fts3MaxLevel(Fts3Table *p, int *pnMax){ |
| 1904 int rc; |
| 1905 int mxLevel = 0; |
| 1906 sqlite3_stmt *pStmt = 0; |
| 1907 |
| 1908 rc = fts3SqlStmt(p, SQL_SELECT_MXLEVEL, &pStmt, 0); |
| 1909 if( rc==SQLITE_OK ){ |
| 1910 if( SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 1911 mxLevel = sqlite3_column_int(pStmt, 0); |
| 1912 } |
| 1913 rc = sqlite3_reset(pStmt); |
| 1914 } |
| 1915 *pnMax = mxLevel; |
| 1916 return rc; |
| 1917 } |
| 1918 |
| 1919 /* |
| 1920 ** Insert a record into the %_segdir table. |
| 1921 */ |
| 1922 static int fts3WriteSegdir( |
| 1923 Fts3Table *p, /* Virtual table handle */ |
| 1924 sqlite3_int64 iLevel, /* Value for "level" field (absolute level) */ |
| 1925 int iIdx, /* Value for "idx" field */ |
| 1926 sqlite3_int64 iStartBlock, /* Value for "start_block" field */ |
| 1927 sqlite3_int64 iLeafEndBlock, /* Value for "leaves_end_block" field */ |
| 1928 sqlite3_int64 iEndBlock, /* Value for "end_block" field */ |
| 1929 sqlite3_int64 nLeafData, /* Bytes of leaf data in segment */ |
| 1930 char *zRoot, /* Blob value for "root" field */ |
| 1931 int nRoot /* Number of bytes in buffer zRoot */ |
| 1932 ){ |
| 1933 sqlite3_stmt *pStmt; |
| 1934 int rc = fts3SqlStmt(p, SQL_INSERT_SEGDIR, &pStmt, 0); |
| 1935 if( rc==SQLITE_OK ){ |
| 1936 sqlite3_bind_int64(pStmt, 1, iLevel); |
| 1937 sqlite3_bind_int(pStmt, 2, iIdx); |
| 1938 sqlite3_bind_int64(pStmt, 3, iStartBlock); |
| 1939 sqlite3_bind_int64(pStmt, 4, iLeafEndBlock); |
| 1940 if( nLeafData==0 ){ |
| 1941 sqlite3_bind_int64(pStmt, 5, iEndBlock); |
| 1942 }else{ |
| 1943 char *zEnd = sqlite3_mprintf("%lld %lld", iEndBlock, nLeafData); |
| 1944 if( !zEnd ) return SQLITE_NOMEM; |
| 1945 sqlite3_bind_text(pStmt, 5, zEnd, -1, sqlite3_free); |
| 1946 } |
| 1947 sqlite3_bind_blob(pStmt, 6, zRoot, nRoot, SQLITE_STATIC); |
| 1948 sqlite3_step(pStmt); |
| 1949 rc = sqlite3_reset(pStmt); |
| 1950 } |
| 1951 return rc; |
| 1952 } |
| 1953 |
| 1954 /* |
| 1955 ** Return the size of the common prefix (if any) shared by zPrev and |
| 1956 ** zNext, in bytes. For example, |
| 1957 ** |
| 1958 ** fts3PrefixCompress("abc", 3, "abcdef", 6) // returns 3 |
| 1959 ** fts3PrefixCompress("abX", 3, "abcdef", 6) // returns 2 |
| 1960 ** fts3PrefixCompress("abX", 3, "Xbcdef", 6) // returns 0 |
| 1961 */ |
| 1962 static int fts3PrefixCompress( |
| 1963 const char *zPrev, /* Buffer containing previous term */ |
| 1964 int nPrev, /* Size of buffer zPrev in bytes */ |
| 1965 const char *zNext, /* Buffer containing next term */ |
| 1966 int nNext /* Size of buffer zNext in bytes */ |
| 1967 ){ |
| 1968 int n; |
| 1969 UNUSED_PARAMETER(nNext); |
| 1970 for(n=0; n<nPrev && zPrev[n]==zNext[n]; n++); |
| 1971 return n; |
| 1972 } |
| 1973 |
| 1974 /* |
| 1975 ** Add term zTerm to the SegmentNode. It is guaranteed that zTerm is larger |
| 1976 ** (according to memcmp) than the previous term. |
| 1977 */ |
| 1978 static int fts3NodeAddTerm( |
| 1979 Fts3Table *p, /* Virtual table handle */ |
| 1980 SegmentNode **ppTree, /* IN/OUT: SegmentNode handle */ |
| 1981 int isCopyTerm, /* True if zTerm/nTerm is transient */ |
| 1982 const char *zTerm, /* Pointer to buffer containing term */ |
| 1983 int nTerm /* Size of term in bytes */ |
| 1984 ){ |
| 1985 SegmentNode *pTree = *ppTree; |
| 1986 int rc; |
| 1987 SegmentNode *pNew; |
| 1988 |
| 1989 /* First try to append the term to the current node. Return early if |
| 1990 ** this is possible. |
| 1991 */ |
| 1992 if( pTree ){ |
| 1993 int nData = pTree->nData; /* Current size of node in bytes */ |
| 1994 int nReq = nData; /* Required space after adding zTerm */ |
| 1995 int nPrefix; /* Number of bytes of prefix compression */ |
| 1996 int nSuffix; /* Suffix length */ |
| 1997 |
| 1998 nPrefix = fts3PrefixCompress(pTree->zTerm, pTree->nTerm, zTerm, nTerm); |
| 1999 nSuffix = nTerm-nPrefix; |
| 2000 |
| 2001 nReq += sqlite3Fts3VarintLen(nPrefix)+sqlite3Fts3VarintLen(nSuffix)+nSuffix; |
| 2002 if( nReq<=p->nNodeSize || !pTree->zTerm ){ |
| 2003 |
| 2004 if( nReq>p->nNodeSize ){ |
| 2005 /* An unusual case: this is the first term to be added to the node |
| 2006 ** and the static node buffer (p->nNodeSize bytes) is not large |
| 2007 ** enough. Use a separately malloced buffer instead This wastes |
| 2008 ** p->nNodeSize bytes, but since this scenario only comes about when |
| 2009 ** the database contain two terms that share a prefix of almost 2KB, |
| 2010 ** this is not expected to be a serious problem. |
| 2011 */ |
| 2012 assert( pTree->aData==(char *)&pTree[1] ); |
| 2013 pTree->aData = (char *)sqlite3_malloc(nReq); |
| 2014 if( !pTree->aData ){ |
| 2015 return SQLITE_NOMEM; |
| 2016 } |
| 2017 } |
| 2018 |
| 2019 if( pTree->zTerm ){ |
| 2020 /* There is no prefix-length field for first term in a node */ |
| 2021 nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nPrefix); |
| 2022 } |
| 2023 |
| 2024 nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nSuffix); |
| 2025 memcpy(&pTree->aData[nData], &zTerm[nPrefix], nSuffix); |
| 2026 pTree->nData = nData + nSuffix; |
| 2027 pTree->nEntry++; |
| 2028 |
| 2029 if( isCopyTerm ){ |
| 2030 if( pTree->nMalloc<nTerm ){ |
| 2031 char *zNew = sqlite3_realloc(pTree->zMalloc, nTerm*2); |
| 2032 if( !zNew ){ |
| 2033 return SQLITE_NOMEM; |
| 2034 } |
| 2035 pTree->nMalloc = nTerm*2; |
| 2036 pTree->zMalloc = zNew; |
| 2037 } |
| 2038 pTree->zTerm = pTree->zMalloc; |
| 2039 memcpy(pTree->zTerm, zTerm, nTerm); |
| 2040 pTree->nTerm = nTerm; |
| 2041 }else{ |
| 2042 pTree->zTerm = (char *)zTerm; |
| 2043 pTree->nTerm = nTerm; |
| 2044 } |
| 2045 return SQLITE_OK; |
| 2046 } |
| 2047 } |
| 2048 |
| 2049 /* If control flows to here, it was not possible to append zTerm to the |
| 2050 ** current node. Create a new node (a right-sibling of the current node). |
| 2051 ** If this is the first node in the tree, the term is added to it. |
| 2052 ** |
| 2053 ** Otherwise, the term is not added to the new node, it is left empty for |
| 2054 ** now. Instead, the term is inserted into the parent of pTree. If pTree |
| 2055 ** has no parent, one is created here. |
| 2056 */ |
| 2057 pNew = (SegmentNode *)sqlite3_malloc(sizeof(SegmentNode) + p->nNodeSize); |
| 2058 if( !pNew ){ |
| 2059 return SQLITE_NOMEM; |
| 2060 } |
| 2061 memset(pNew, 0, sizeof(SegmentNode)); |
| 2062 pNew->nData = 1 + FTS3_VARINT_MAX; |
| 2063 pNew->aData = (char *)&pNew[1]; |
| 2064 |
| 2065 if( pTree ){ |
| 2066 SegmentNode *pParent = pTree->pParent; |
| 2067 rc = fts3NodeAddTerm(p, &pParent, isCopyTerm, zTerm, nTerm); |
| 2068 if( pTree->pParent==0 ){ |
| 2069 pTree->pParent = pParent; |
| 2070 } |
| 2071 pTree->pRight = pNew; |
| 2072 pNew->pLeftmost = pTree->pLeftmost; |
| 2073 pNew->pParent = pParent; |
| 2074 pNew->zMalloc = pTree->zMalloc; |
| 2075 pNew->nMalloc = pTree->nMalloc; |
| 2076 pTree->zMalloc = 0; |
| 2077 }else{ |
| 2078 pNew->pLeftmost = pNew; |
| 2079 rc = fts3NodeAddTerm(p, &pNew, isCopyTerm, zTerm, nTerm); |
| 2080 } |
| 2081 |
| 2082 *ppTree = pNew; |
| 2083 return rc; |
| 2084 } |
| 2085 |
| 2086 /* |
| 2087 ** Helper function for fts3NodeWrite(). |
| 2088 */ |
| 2089 static int fts3TreeFinishNode( |
| 2090 SegmentNode *pTree, |
| 2091 int iHeight, |
| 2092 sqlite3_int64 iLeftChild |
| 2093 ){ |
| 2094 int nStart; |
| 2095 assert( iHeight>=1 && iHeight<128 ); |
| 2096 nStart = FTS3_VARINT_MAX - sqlite3Fts3VarintLen(iLeftChild); |
| 2097 pTree->aData[nStart] = (char)iHeight; |
| 2098 sqlite3Fts3PutVarint(&pTree->aData[nStart+1], iLeftChild); |
| 2099 return nStart; |
| 2100 } |
| 2101 |
| 2102 /* |
| 2103 ** Write the buffer for the segment node pTree and all of its peers to the |
| 2104 ** database. Then call this function recursively to write the parent of |
| 2105 ** pTree and its peers to the database. |
| 2106 ** |
| 2107 ** Except, if pTree is a root node, do not write it to the database. Instead, |
| 2108 ** set output variables *paRoot and *pnRoot to contain the root node. |
| 2109 ** |
| 2110 ** If successful, SQLITE_OK is returned and output variable *piLast is |
| 2111 ** set to the largest blockid written to the database (or zero if no |
| 2112 ** blocks were written to the db). Otherwise, an SQLite error code is |
| 2113 ** returned. |
| 2114 */ |
| 2115 static int fts3NodeWrite( |
| 2116 Fts3Table *p, /* Virtual table handle */ |
| 2117 SegmentNode *pTree, /* SegmentNode handle */ |
| 2118 int iHeight, /* Height of this node in tree */ |
| 2119 sqlite3_int64 iLeaf, /* Block id of first leaf node */ |
| 2120 sqlite3_int64 iFree, /* Block id of next free slot in %_segments */ |
| 2121 sqlite3_int64 *piLast, /* OUT: Block id of last entry written */ |
| 2122 char **paRoot, /* OUT: Data for root node */ |
| 2123 int *pnRoot /* OUT: Size of root node in bytes */ |
| 2124 ){ |
| 2125 int rc = SQLITE_OK; |
| 2126 |
| 2127 if( !pTree->pParent ){ |
| 2128 /* Root node of the tree. */ |
| 2129 int nStart = fts3TreeFinishNode(pTree, iHeight, iLeaf); |
| 2130 *piLast = iFree-1; |
| 2131 *pnRoot = pTree->nData - nStart; |
| 2132 *paRoot = &pTree->aData[nStart]; |
| 2133 }else{ |
| 2134 SegmentNode *pIter; |
| 2135 sqlite3_int64 iNextFree = iFree; |
| 2136 sqlite3_int64 iNextLeaf = iLeaf; |
| 2137 for(pIter=pTree->pLeftmost; pIter && rc==SQLITE_OK; pIter=pIter->pRight){ |
| 2138 int nStart = fts3TreeFinishNode(pIter, iHeight, iNextLeaf); |
| 2139 int nWrite = pIter->nData - nStart; |
| 2140 |
| 2141 rc = fts3WriteSegment(p, iNextFree, &pIter->aData[nStart], nWrite); |
| 2142 iNextFree++; |
| 2143 iNextLeaf += (pIter->nEntry+1); |
| 2144 } |
| 2145 if( rc==SQLITE_OK ){ |
| 2146 assert( iNextLeaf==iFree ); |
| 2147 rc = fts3NodeWrite( |
| 2148 p, pTree->pParent, iHeight+1, iFree, iNextFree, piLast, paRoot, pnRoot |
| 2149 ); |
| 2150 } |
| 2151 } |
| 2152 |
| 2153 return rc; |
| 2154 } |
| 2155 |
| 2156 /* |
| 2157 ** Free all memory allocations associated with the tree pTree. |
| 2158 */ |
| 2159 static void fts3NodeFree(SegmentNode *pTree){ |
| 2160 if( pTree ){ |
| 2161 SegmentNode *p = pTree->pLeftmost; |
| 2162 fts3NodeFree(p->pParent); |
| 2163 while( p ){ |
| 2164 SegmentNode *pRight = p->pRight; |
| 2165 if( p->aData!=(char *)&p[1] ){ |
| 2166 sqlite3_free(p->aData); |
| 2167 } |
| 2168 assert( pRight==0 || p->zMalloc==0 ); |
| 2169 sqlite3_free(p->zMalloc); |
| 2170 sqlite3_free(p); |
| 2171 p = pRight; |
| 2172 } |
| 2173 } |
| 2174 } |
| 2175 |
| 2176 /* |
| 2177 ** Add a term to the segment being constructed by the SegmentWriter object |
| 2178 ** *ppWriter. When adding the first term to a segment, *ppWriter should |
| 2179 ** be passed NULL. This function will allocate a new SegmentWriter object |
| 2180 ** and return it via the input/output variable *ppWriter in this case. |
| 2181 ** |
| 2182 ** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. |
| 2183 */ |
| 2184 static int fts3SegWriterAdd( |
| 2185 Fts3Table *p, /* Virtual table handle */ |
| 2186 SegmentWriter **ppWriter, /* IN/OUT: SegmentWriter handle */ |
| 2187 int isCopyTerm, /* True if buffer zTerm must be copied */ |
| 2188 const char *zTerm, /* Pointer to buffer containing term */ |
| 2189 int nTerm, /* Size of term in bytes */ |
| 2190 const char *aDoclist, /* Pointer to buffer containing doclist */ |
| 2191 int nDoclist /* Size of doclist in bytes */ |
| 2192 ){ |
| 2193 int nPrefix; /* Size of term prefix in bytes */ |
| 2194 int nSuffix; /* Size of term suffix in bytes */ |
| 2195 int nReq; /* Number of bytes required on leaf page */ |
| 2196 int nData; |
| 2197 SegmentWriter *pWriter = *ppWriter; |
| 2198 |
| 2199 if( !pWriter ){ |
| 2200 int rc; |
| 2201 sqlite3_stmt *pStmt; |
| 2202 |
| 2203 /* Allocate the SegmentWriter structure */ |
| 2204 pWriter = (SegmentWriter *)sqlite3_malloc(sizeof(SegmentWriter)); |
| 2205 if( !pWriter ) return SQLITE_NOMEM; |
| 2206 memset(pWriter, 0, sizeof(SegmentWriter)); |
| 2207 *ppWriter = pWriter; |
| 2208 |
| 2209 /* Allocate a buffer in which to accumulate data */ |
| 2210 pWriter->aData = (char *)sqlite3_malloc(p->nNodeSize); |
| 2211 if( !pWriter->aData ) return SQLITE_NOMEM; |
| 2212 pWriter->nSize = p->nNodeSize; |
| 2213 |
| 2214 /* Find the next free blockid in the %_segments table */ |
| 2215 rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pStmt, 0); |
| 2216 if( rc!=SQLITE_OK ) return rc; |
| 2217 if( SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 2218 pWriter->iFree = sqlite3_column_int64(pStmt, 0); |
| 2219 pWriter->iFirst = pWriter->iFree; |
| 2220 } |
| 2221 rc = sqlite3_reset(pStmt); |
| 2222 if( rc!=SQLITE_OK ) return rc; |
| 2223 } |
| 2224 nData = pWriter->nData; |
| 2225 |
| 2226 nPrefix = fts3PrefixCompress(pWriter->zTerm, pWriter->nTerm, zTerm, nTerm); |
| 2227 nSuffix = nTerm-nPrefix; |
| 2228 |
| 2229 /* Figure out how many bytes are required by this new entry */ |
| 2230 nReq = sqlite3Fts3VarintLen(nPrefix) + /* varint containing prefix size */ |
| 2231 sqlite3Fts3VarintLen(nSuffix) + /* varint containing suffix size */ |
| 2232 nSuffix + /* Term suffix */ |
| 2233 sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ |
| 2234 nDoclist; /* Doclist data */ |
| 2235 |
| 2236 if( nData>0 && nData+nReq>p->nNodeSize ){ |
| 2237 int rc; |
| 2238 |
| 2239 /* The current leaf node is full. Write it out to the database. */ |
| 2240 rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, nData); |
| 2241 if( rc!=SQLITE_OK ) return rc; |
| 2242 p->nLeafAdd++; |
| 2243 |
| 2244 /* Add the current term to the interior node tree. The term added to |
| 2245 ** the interior tree must: |
| 2246 ** |
| 2247 ** a) be greater than the largest term on the leaf node just written |
| 2248 ** to the database (still available in pWriter->zTerm), and |
| 2249 ** |
| 2250 ** b) be less than or equal to the term about to be added to the new |
| 2251 ** leaf node (zTerm/nTerm). |
| 2252 ** |
| 2253 ** In other words, it must be the prefix of zTerm 1 byte longer than |
| 2254 ** the common prefix (if any) of zTerm and pWriter->zTerm. |
| 2255 */ |
| 2256 assert( nPrefix<nTerm ); |
| 2257 rc = fts3NodeAddTerm(p, &pWriter->pTree, isCopyTerm, zTerm, nPrefix+1); |
| 2258 if( rc!=SQLITE_OK ) return rc; |
| 2259 |
| 2260 nData = 0; |
| 2261 pWriter->nTerm = 0; |
| 2262 |
| 2263 nPrefix = 0; |
| 2264 nSuffix = nTerm; |
| 2265 nReq = 1 + /* varint containing prefix size */ |
| 2266 sqlite3Fts3VarintLen(nTerm) + /* varint containing suffix size */ |
| 2267 nTerm + /* Term suffix */ |
| 2268 sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ |
| 2269 nDoclist; /* Doclist data */ |
| 2270 } |
| 2271 |
| 2272 /* Increase the total number of bytes written to account for the new entry. */ |
| 2273 pWriter->nLeafData += nReq; |
| 2274 |
| 2275 /* If the buffer currently allocated is too small for this entry, realloc |
| 2276 ** the buffer to make it large enough. |
| 2277 */ |
| 2278 if( nReq>pWriter->nSize ){ |
| 2279 char *aNew = sqlite3_realloc(pWriter->aData, nReq); |
| 2280 if( !aNew ) return SQLITE_NOMEM; |
| 2281 pWriter->aData = aNew; |
| 2282 pWriter->nSize = nReq; |
| 2283 } |
| 2284 assert( nData+nReq<=pWriter->nSize ); |
| 2285 |
| 2286 /* Append the prefix-compressed term and doclist to the buffer. */ |
| 2287 nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nPrefix); |
| 2288 nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nSuffix); |
| 2289 memcpy(&pWriter->aData[nData], &zTerm[nPrefix], nSuffix); |
| 2290 nData += nSuffix; |
| 2291 nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nDoclist); |
| 2292 memcpy(&pWriter->aData[nData], aDoclist, nDoclist); |
| 2293 pWriter->nData = nData + nDoclist; |
| 2294 |
| 2295 /* Save the current term so that it can be used to prefix-compress the next. |
| 2296 ** If the isCopyTerm parameter is true, then the buffer pointed to by |
| 2297 ** zTerm is transient, so take a copy of the term data. Otherwise, just |
| 2298 ** store a copy of the pointer. |
| 2299 */ |
| 2300 if( isCopyTerm ){ |
| 2301 if( nTerm>pWriter->nMalloc ){ |
| 2302 char *zNew = sqlite3_realloc(pWriter->zMalloc, nTerm*2); |
| 2303 if( !zNew ){ |
| 2304 return SQLITE_NOMEM; |
| 2305 } |
| 2306 pWriter->nMalloc = nTerm*2; |
| 2307 pWriter->zMalloc = zNew; |
| 2308 pWriter->zTerm = zNew; |
| 2309 } |
| 2310 assert( pWriter->zTerm==pWriter->zMalloc ); |
| 2311 memcpy(pWriter->zTerm, zTerm, nTerm); |
| 2312 }else{ |
| 2313 pWriter->zTerm = (char *)zTerm; |
| 2314 } |
| 2315 pWriter->nTerm = nTerm; |
| 2316 |
| 2317 return SQLITE_OK; |
| 2318 } |
| 2319 |
| 2320 /* |
| 2321 ** Flush all data associated with the SegmentWriter object pWriter to the |
| 2322 ** database. This function must be called after all terms have been added |
| 2323 ** to the segment using fts3SegWriterAdd(). If successful, SQLITE_OK is |
| 2324 ** returned. Otherwise, an SQLite error code. |
| 2325 */ |
| 2326 static int fts3SegWriterFlush( |
| 2327 Fts3Table *p, /* Virtual table handle */ |
| 2328 SegmentWriter *pWriter, /* SegmentWriter to flush to the db */ |
| 2329 sqlite3_int64 iLevel, /* Value for 'level' column of %_segdir */ |
| 2330 int iIdx /* Value for 'idx' column of %_segdir */ |
| 2331 ){ |
| 2332 int rc; /* Return code */ |
| 2333 if( pWriter->pTree ){ |
| 2334 sqlite3_int64 iLast = 0; /* Largest block id written to database */ |
| 2335 sqlite3_int64 iLastLeaf; /* Largest leaf block id written to db */ |
| 2336 char *zRoot = NULL; /* Pointer to buffer containing root node */ |
| 2337 int nRoot = 0; /* Size of buffer zRoot */ |
| 2338 |
| 2339 iLastLeaf = pWriter->iFree; |
| 2340 rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, pWriter->nData); |
| 2341 if( rc==SQLITE_OK ){ |
| 2342 rc = fts3NodeWrite(p, pWriter->pTree, 1, |
| 2343 pWriter->iFirst, pWriter->iFree, &iLast, &zRoot, &nRoot); |
| 2344 } |
| 2345 if( rc==SQLITE_OK ){ |
| 2346 rc = fts3WriteSegdir(p, iLevel, iIdx, |
| 2347 pWriter->iFirst, iLastLeaf, iLast, pWriter->nLeafData, zRoot, nRoot); |
| 2348 } |
| 2349 }else{ |
| 2350 /* The entire tree fits on the root node. Write it to the segdir table. */ |
| 2351 rc = fts3WriteSegdir(p, iLevel, iIdx, |
| 2352 0, 0, 0, pWriter->nLeafData, pWriter->aData, pWriter->nData); |
| 2353 } |
| 2354 p->nLeafAdd++; |
| 2355 return rc; |
| 2356 } |
| 2357 |
| 2358 /* |
| 2359 ** Release all memory held by the SegmentWriter object passed as the |
| 2360 ** first argument. |
| 2361 */ |
| 2362 static void fts3SegWriterFree(SegmentWriter *pWriter){ |
| 2363 if( pWriter ){ |
| 2364 sqlite3_free(pWriter->aData); |
| 2365 sqlite3_free(pWriter->zMalloc); |
| 2366 fts3NodeFree(pWriter->pTree); |
| 2367 sqlite3_free(pWriter); |
| 2368 } |
| 2369 } |
| 2370 |
| 2371 /* |
| 2372 ** The first value in the apVal[] array is assumed to contain an integer. |
| 2373 ** This function tests if there exist any documents with docid values that |
| 2374 ** are different from that integer. i.e. if deleting the document with docid |
| 2375 ** pRowid would mean the FTS3 table were empty. |
| 2376 ** |
| 2377 ** If successful, *pisEmpty is set to true if the table is empty except for |
| 2378 ** document pRowid, or false otherwise, and SQLITE_OK is returned. If an |
| 2379 ** error occurs, an SQLite error code is returned. |
| 2380 */ |
| 2381 static int fts3IsEmpty(Fts3Table *p, sqlite3_value *pRowid, int *pisEmpty){ |
| 2382 sqlite3_stmt *pStmt; |
| 2383 int rc; |
| 2384 if( p->zContentTbl ){ |
| 2385 /* If using the content=xxx option, assume the table is never empty */ |
| 2386 *pisEmpty = 0; |
| 2387 rc = SQLITE_OK; |
| 2388 }else{ |
| 2389 rc = fts3SqlStmt(p, SQL_IS_EMPTY, &pStmt, &pRowid); |
| 2390 if( rc==SQLITE_OK ){ |
| 2391 if( SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 2392 *pisEmpty = sqlite3_column_int(pStmt, 0); |
| 2393 } |
| 2394 rc = sqlite3_reset(pStmt); |
| 2395 } |
| 2396 } |
| 2397 return rc; |
| 2398 } |
| 2399 |
| 2400 /* |
| 2401 ** Set *pnMax to the largest segment level in the database for the index |
| 2402 ** iIndex. |
| 2403 ** |
| 2404 ** Segment levels are stored in the 'level' column of the %_segdir table. |
| 2405 ** |
| 2406 ** Return SQLITE_OK if successful, or an SQLite error code if not. |
| 2407 */ |
| 2408 static int fts3SegmentMaxLevel( |
| 2409 Fts3Table *p, |
| 2410 int iLangid, |
| 2411 int iIndex, |
| 2412 sqlite3_int64 *pnMax |
| 2413 ){ |
| 2414 sqlite3_stmt *pStmt; |
| 2415 int rc; |
| 2416 assert( iIndex>=0 && iIndex<p->nIndex ); |
| 2417 |
| 2418 /* Set pStmt to the compiled version of: |
| 2419 ** |
| 2420 ** SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? |
| 2421 ** |
| 2422 ** (1024 is actually the value of macro FTS3_SEGDIR_PREFIXLEVEL_STR). |
| 2423 */ |
| 2424 rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_MAX_LEVEL, &pStmt, 0); |
| 2425 if( rc!=SQLITE_OK ) return rc; |
| 2426 sqlite3_bind_int64(pStmt, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); |
| 2427 sqlite3_bind_int64(pStmt, 2, |
| 2428 getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) |
| 2429 ); |
| 2430 if( SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 2431 *pnMax = sqlite3_column_int64(pStmt, 0); |
| 2432 } |
| 2433 return sqlite3_reset(pStmt); |
| 2434 } |
| 2435 |
| 2436 /* |
| 2437 ** iAbsLevel is an absolute level that may be assumed to exist within |
| 2438 ** the database. This function checks if it is the largest level number |
| 2439 ** within its index. Assuming no error occurs, *pbMax is set to 1 if |
| 2440 ** iAbsLevel is indeed the largest level, or 0 otherwise, and SQLITE_OK |
| 2441 ** is returned. If an error occurs, an error code is returned and the |
| 2442 ** final value of *pbMax is undefined. |
| 2443 */ |
| 2444 static int fts3SegmentIsMaxLevel(Fts3Table *p, i64 iAbsLevel, int *pbMax){ |
| 2445 |
| 2446 /* Set pStmt to the compiled version of: |
| 2447 ** |
| 2448 ** SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? |
| 2449 ** |
| 2450 ** (1024 is actually the value of macro FTS3_SEGDIR_PREFIXLEVEL_STR). |
| 2451 */ |
| 2452 sqlite3_stmt *pStmt; |
| 2453 int rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_MAX_LEVEL, &pStmt, 0); |
| 2454 if( rc!=SQLITE_OK ) return rc; |
| 2455 sqlite3_bind_int64(pStmt, 1, iAbsLevel+1); |
| 2456 sqlite3_bind_int64(pStmt, 2, |
| 2457 ((iAbsLevel/FTS3_SEGDIR_MAXLEVEL)+1) * FTS3_SEGDIR_MAXLEVEL |
| 2458 ); |
| 2459 |
| 2460 *pbMax = 0; |
| 2461 if( SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 2462 *pbMax = sqlite3_column_type(pStmt, 0)==SQLITE_NULL; |
| 2463 } |
| 2464 return sqlite3_reset(pStmt); |
| 2465 } |
| 2466 |
| 2467 /* |
| 2468 ** Delete all entries in the %_segments table associated with the segment |
| 2469 ** opened with seg-reader pSeg. This function does not affect the contents |
| 2470 ** of the %_segdir table. |
| 2471 */ |
| 2472 static int fts3DeleteSegment( |
| 2473 Fts3Table *p, /* FTS table handle */ |
| 2474 Fts3SegReader *pSeg /* Segment to delete */ |
| 2475 ){ |
| 2476 int rc = SQLITE_OK; /* Return code */ |
| 2477 if( pSeg->iStartBlock ){ |
| 2478 sqlite3_stmt *pDelete; /* SQL statement to delete rows */ |
| 2479 rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDelete, 0); |
| 2480 if( rc==SQLITE_OK ){ |
| 2481 sqlite3_bind_int64(pDelete, 1, pSeg->iStartBlock); |
| 2482 sqlite3_bind_int64(pDelete, 2, pSeg->iEndBlock); |
| 2483 sqlite3_step(pDelete); |
| 2484 rc = sqlite3_reset(pDelete); |
| 2485 } |
| 2486 } |
| 2487 return rc; |
| 2488 } |
| 2489 |
| 2490 /* |
| 2491 ** This function is used after merging multiple segments into a single large |
| 2492 ** segment to delete the old, now redundant, segment b-trees. Specifically, |
| 2493 ** it: |
| 2494 ** |
| 2495 ** 1) Deletes all %_segments entries for the segments associated with |
| 2496 ** each of the SegReader objects in the array passed as the third |
| 2497 ** argument, and |
| 2498 ** |
| 2499 ** 2) deletes all %_segdir entries with level iLevel, or all %_segdir |
| 2500 ** entries regardless of level if (iLevel<0). |
| 2501 ** |
| 2502 ** SQLITE_OK is returned if successful, otherwise an SQLite error code. |
| 2503 */ |
| 2504 static int fts3DeleteSegdir( |
| 2505 Fts3Table *p, /* Virtual table handle */ |
| 2506 int iLangid, /* Language id */ |
| 2507 int iIndex, /* Index for p->aIndex */ |
| 2508 int iLevel, /* Level of %_segdir entries to delete */ |
| 2509 Fts3SegReader **apSegment, /* Array of SegReader objects */ |
| 2510 int nReader /* Size of array apSegment */ |
| 2511 ){ |
| 2512 int rc = SQLITE_OK; /* Return Code */ |
| 2513 int i; /* Iterator variable */ |
| 2514 sqlite3_stmt *pDelete = 0; /* SQL statement to delete rows */ |
| 2515 |
| 2516 for(i=0; rc==SQLITE_OK && i<nReader; i++){ |
| 2517 rc = fts3DeleteSegment(p, apSegment[i]); |
| 2518 } |
| 2519 if( rc!=SQLITE_OK ){ |
| 2520 return rc; |
| 2521 } |
| 2522 |
| 2523 assert( iLevel>=0 || iLevel==FTS3_SEGCURSOR_ALL ); |
| 2524 if( iLevel==FTS3_SEGCURSOR_ALL ){ |
| 2525 rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_RANGE, &pDelete, 0); |
| 2526 if( rc==SQLITE_OK ){ |
| 2527 sqlite3_bind_int64(pDelete, 1, getAbsoluteLevel(p, iLangid, iIndex, 0)); |
| 2528 sqlite3_bind_int64(pDelete, 2, |
| 2529 getAbsoluteLevel(p, iLangid, iIndex, FTS3_SEGDIR_MAXLEVEL-1) |
| 2530 ); |
| 2531 } |
| 2532 }else{ |
| 2533 rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_LEVEL, &pDelete, 0); |
| 2534 if( rc==SQLITE_OK ){ |
| 2535 sqlite3_bind_int64( |
| 2536 pDelete, 1, getAbsoluteLevel(p, iLangid, iIndex, iLevel) |
| 2537 ); |
| 2538 } |
| 2539 } |
| 2540 |
| 2541 if( rc==SQLITE_OK ){ |
| 2542 sqlite3_step(pDelete); |
| 2543 rc = sqlite3_reset(pDelete); |
| 2544 } |
| 2545 |
| 2546 return rc; |
| 2547 } |
| 2548 |
| 2549 /* |
| 2550 ** When this function is called, buffer *ppList (size *pnList bytes) contains |
| 2551 ** a position list that may (or may not) feature multiple columns. This |
| 2552 ** function adjusts the pointer *ppList and the length *pnList so that they |
| 2553 ** identify the subset of the position list that corresponds to column iCol. |
| 2554 ** |
| 2555 ** If there are no entries in the input position list for column iCol, then |
| 2556 ** *pnList is set to zero before returning. |
| 2557 ** |
| 2558 ** If parameter bZero is non-zero, then any part of the input list following |
| 2559 ** the end of the output list is zeroed before returning. |
| 2560 */ |
| 2561 static void fts3ColumnFilter( |
| 2562 int iCol, /* Column to filter on */ |
| 2563 int bZero, /* Zero out anything following *ppList */ |
| 2564 char **ppList, /* IN/OUT: Pointer to position list */ |
| 2565 int *pnList /* IN/OUT: Size of buffer *ppList in bytes */ |
| 2566 ){ |
| 2567 char *pList = *ppList; |
| 2568 int nList = *pnList; |
| 2569 char *pEnd = &pList[nList]; |
| 2570 int iCurrent = 0; |
| 2571 char *p = pList; |
| 2572 |
| 2573 assert( iCol>=0 ); |
| 2574 while( 1 ){ |
| 2575 char c = 0; |
| 2576 while( p<pEnd && (c | *p)&0xFE ) c = *p++ & 0x80; |
| 2577 |
| 2578 if( iCol==iCurrent ){ |
| 2579 nList = (int)(p - pList); |
| 2580 break; |
| 2581 } |
| 2582 |
| 2583 nList -= (int)(p - pList); |
| 2584 pList = p; |
| 2585 if( nList==0 ){ |
| 2586 break; |
| 2587 } |
| 2588 p = &pList[1]; |
| 2589 p += fts3GetVarint32(p, &iCurrent); |
| 2590 } |
| 2591 |
| 2592 if( bZero && &pList[nList]!=pEnd ){ |
| 2593 memset(&pList[nList], 0, pEnd - &pList[nList]); |
| 2594 } |
| 2595 *ppList = pList; |
| 2596 *pnList = nList; |
| 2597 } |
| 2598 |
| 2599 /* |
| 2600 ** Cache data in the Fts3MultiSegReader.aBuffer[] buffer (overwriting any |
| 2601 ** existing data). Grow the buffer if required. |
| 2602 ** |
| 2603 ** If successful, return SQLITE_OK. Otherwise, if an OOM error is encountered |
| 2604 ** trying to resize the buffer, return SQLITE_NOMEM. |
| 2605 */ |
| 2606 static int fts3MsrBufferData( |
| 2607 Fts3MultiSegReader *pMsr, /* Multi-segment-reader handle */ |
| 2608 char *pList, |
| 2609 int nList |
| 2610 ){ |
| 2611 if( nList>pMsr->nBuffer ){ |
| 2612 char *pNew; |
| 2613 pMsr->nBuffer = nList*2; |
| 2614 pNew = (char *)sqlite3_realloc(pMsr->aBuffer, pMsr->nBuffer); |
| 2615 if( !pNew ) return SQLITE_NOMEM; |
| 2616 pMsr->aBuffer = pNew; |
| 2617 } |
| 2618 |
| 2619 memcpy(pMsr->aBuffer, pList, nList); |
| 2620 return SQLITE_OK; |
| 2621 } |
| 2622 |
| 2623 int sqlite3Fts3MsrIncrNext( |
| 2624 Fts3Table *p, /* Virtual table handle */ |
| 2625 Fts3MultiSegReader *pMsr, /* Multi-segment-reader handle */ |
| 2626 sqlite3_int64 *piDocid, /* OUT: Docid value */ |
| 2627 char **paPoslist, /* OUT: Pointer to position list */ |
| 2628 int *pnPoslist /* OUT: Size of position list in bytes */ |
| 2629 ){ |
| 2630 int nMerge = pMsr->nAdvance; |
| 2631 Fts3SegReader **apSegment = pMsr->apSegment; |
| 2632 int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( |
| 2633 p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp |
| 2634 ); |
| 2635 |
| 2636 if( nMerge==0 ){ |
| 2637 *paPoslist = 0; |
| 2638 return SQLITE_OK; |
| 2639 } |
| 2640 |
| 2641 while( 1 ){ |
| 2642 Fts3SegReader *pSeg; |
| 2643 pSeg = pMsr->apSegment[0]; |
| 2644 |
| 2645 if( pSeg->pOffsetList==0 ){ |
| 2646 *paPoslist = 0; |
| 2647 break; |
| 2648 }else{ |
| 2649 int rc; |
| 2650 char *pList; |
| 2651 int nList; |
| 2652 int j; |
| 2653 sqlite3_int64 iDocid = apSegment[0]->iDocid; |
| 2654 |
| 2655 rc = fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); |
| 2656 j = 1; |
| 2657 while( rc==SQLITE_OK |
| 2658 && j<nMerge |
| 2659 && apSegment[j]->pOffsetList |
| 2660 && apSegment[j]->iDocid==iDocid |
| 2661 ){ |
| 2662 rc = fts3SegReaderNextDocid(p, apSegment[j], 0, 0); |
| 2663 j++; |
| 2664 } |
| 2665 if( rc!=SQLITE_OK ) return rc; |
| 2666 fts3SegReaderSort(pMsr->apSegment, nMerge, j, xCmp); |
| 2667 |
| 2668 if( nList>0 && fts3SegReaderIsPending(apSegment[0]) ){ |
| 2669 rc = fts3MsrBufferData(pMsr, pList, nList+1); |
| 2670 if( rc!=SQLITE_OK ) return rc; |
| 2671 assert( (pMsr->aBuffer[nList] & 0xFE)==0x00 ); |
| 2672 pList = pMsr->aBuffer; |
| 2673 } |
| 2674 |
| 2675 if( pMsr->iColFilter>=0 ){ |
| 2676 fts3ColumnFilter(pMsr->iColFilter, 1, &pList, &nList); |
| 2677 } |
| 2678 |
| 2679 if( nList>0 ){ |
| 2680 *paPoslist = pList; |
| 2681 *piDocid = iDocid; |
| 2682 *pnPoslist = nList; |
| 2683 break; |
| 2684 } |
| 2685 } |
| 2686 } |
| 2687 |
| 2688 return SQLITE_OK; |
| 2689 } |
| 2690 |
| 2691 static int fts3SegReaderStart( |
| 2692 Fts3Table *p, /* Virtual table handle */ |
| 2693 Fts3MultiSegReader *pCsr, /* Cursor object */ |
| 2694 const char *zTerm, /* Term searched for (or NULL) */ |
| 2695 int nTerm /* Length of zTerm in bytes */ |
| 2696 ){ |
| 2697 int i; |
| 2698 int nSeg = pCsr->nSegment; |
| 2699 |
| 2700 /* If the Fts3SegFilter defines a specific term (or term prefix) to search |
| 2701 ** for, then advance each segment iterator until it points to a term of |
| 2702 ** equal or greater value than the specified term. This prevents many |
| 2703 ** unnecessary merge/sort operations for the case where single segment |
| 2704 ** b-tree leaf nodes contain more than one term. |
| 2705 */ |
| 2706 for(i=0; pCsr->bRestart==0 && i<pCsr->nSegment; i++){ |
| 2707 int res = 0; |
| 2708 Fts3SegReader *pSeg = pCsr->apSegment[i]; |
| 2709 do { |
| 2710 int rc = fts3SegReaderNext(p, pSeg, 0); |
| 2711 if( rc!=SQLITE_OK ) return rc; |
| 2712 }while( zTerm && (res = fts3SegReaderTermCmp(pSeg, zTerm, nTerm))<0 ); |
| 2713 |
| 2714 if( pSeg->bLookup && res!=0 ){ |
| 2715 fts3SegReaderSetEof(pSeg); |
| 2716 } |
| 2717 } |
| 2718 fts3SegReaderSort(pCsr->apSegment, nSeg, nSeg, fts3SegReaderCmp); |
| 2719 |
| 2720 return SQLITE_OK; |
| 2721 } |
| 2722 |
| 2723 int sqlite3Fts3SegReaderStart( |
| 2724 Fts3Table *p, /* Virtual table handle */ |
| 2725 Fts3MultiSegReader *pCsr, /* Cursor object */ |
| 2726 Fts3SegFilter *pFilter /* Restrictions on range of iteration */ |
| 2727 ){ |
| 2728 pCsr->pFilter = pFilter; |
| 2729 return fts3SegReaderStart(p, pCsr, pFilter->zTerm, pFilter->nTerm); |
| 2730 } |
| 2731 |
| 2732 int sqlite3Fts3MsrIncrStart( |
| 2733 Fts3Table *p, /* Virtual table handle */ |
| 2734 Fts3MultiSegReader *pCsr, /* Cursor object */ |
| 2735 int iCol, /* Column to match on. */ |
| 2736 const char *zTerm, /* Term to iterate through a doclist for */ |
| 2737 int nTerm /* Number of bytes in zTerm */ |
| 2738 ){ |
| 2739 int i; |
| 2740 int rc; |
| 2741 int nSegment = pCsr->nSegment; |
| 2742 int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( |
| 2743 p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp |
| 2744 ); |
| 2745 |
| 2746 assert( pCsr->pFilter==0 ); |
| 2747 assert( zTerm && nTerm>0 ); |
| 2748 |
| 2749 /* Advance each segment iterator until it points to the term zTerm/nTerm. */ |
| 2750 rc = fts3SegReaderStart(p, pCsr, zTerm, nTerm); |
| 2751 if( rc!=SQLITE_OK ) return rc; |
| 2752 |
| 2753 /* Determine how many of the segments actually point to zTerm/nTerm. */ |
| 2754 for(i=0; i<nSegment; i++){ |
| 2755 Fts3SegReader *pSeg = pCsr->apSegment[i]; |
| 2756 if( !pSeg->aNode || fts3SegReaderTermCmp(pSeg, zTerm, nTerm) ){ |
| 2757 break; |
| 2758 } |
| 2759 } |
| 2760 pCsr->nAdvance = i; |
| 2761 |
| 2762 /* Advance each of the segments to point to the first docid. */ |
| 2763 for(i=0; i<pCsr->nAdvance; i++){ |
| 2764 rc = fts3SegReaderFirstDocid(p, pCsr->apSegment[i]); |
| 2765 if( rc!=SQLITE_OK ) return rc; |
| 2766 } |
| 2767 fts3SegReaderSort(pCsr->apSegment, i, i, xCmp); |
| 2768 |
| 2769 assert( iCol<0 || iCol<p->nColumn ); |
| 2770 pCsr->iColFilter = iCol; |
| 2771 |
| 2772 return SQLITE_OK; |
| 2773 } |
| 2774 |
| 2775 /* |
| 2776 ** This function is called on a MultiSegReader that has been started using |
| 2777 ** sqlite3Fts3MsrIncrStart(). One or more calls to MsrIncrNext() may also |
| 2778 ** have been made. Calling this function puts the MultiSegReader in such |
| 2779 ** a state that if the next two calls are: |
| 2780 ** |
| 2781 ** sqlite3Fts3SegReaderStart() |
| 2782 ** sqlite3Fts3SegReaderStep() |
| 2783 ** |
| 2784 ** then the entire doclist for the term is available in |
| 2785 ** MultiSegReader.aDoclist/nDoclist. |
| 2786 */ |
| 2787 int sqlite3Fts3MsrIncrRestart(Fts3MultiSegReader *pCsr){ |
| 2788 int i; /* Used to iterate through segment-readers */ |
| 2789 |
| 2790 assert( pCsr->zTerm==0 ); |
| 2791 assert( pCsr->nTerm==0 ); |
| 2792 assert( pCsr->aDoclist==0 ); |
| 2793 assert( pCsr->nDoclist==0 ); |
| 2794 |
| 2795 pCsr->nAdvance = 0; |
| 2796 pCsr->bRestart = 1; |
| 2797 for(i=0; i<pCsr->nSegment; i++){ |
| 2798 pCsr->apSegment[i]->pOffsetList = 0; |
| 2799 pCsr->apSegment[i]->nOffsetList = 0; |
| 2800 pCsr->apSegment[i]->iDocid = 0; |
| 2801 } |
| 2802 |
| 2803 return SQLITE_OK; |
| 2804 } |
| 2805 |
| 2806 |
| 2807 int sqlite3Fts3SegReaderStep( |
| 2808 Fts3Table *p, /* Virtual table handle */ |
| 2809 Fts3MultiSegReader *pCsr /* Cursor object */ |
| 2810 ){ |
| 2811 int rc = SQLITE_OK; |
| 2812 |
| 2813 int isIgnoreEmpty = (pCsr->pFilter->flags & FTS3_SEGMENT_IGNORE_EMPTY); |
| 2814 int isRequirePos = (pCsr->pFilter->flags & FTS3_SEGMENT_REQUIRE_POS); |
| 2815 int isColFilter = (pCsr->pFilter->flags & FTS3_SEGMENT_COLUMN_FILTER); |
| 2816 int isPrefix = (pCsr->pFilter->flags & FTS3_SEGMENT_PREFIX); |
| 2817 int isScan = (pCsr->pFilter->flags & FTS3_SEGMENT_SCAN); |
| 2818 int isFirst = (pCsr->pFilter->flags & FTS3_SEGMENT_FIRST); |
| 2819 |
| 2820 Fts3SegReader **apSegment = pCsr->apSegment; |
| 2821 int nSegment = pCsr->nSegment; |
| 2822 Fts3SegFilter *pFilter = pCsr->pFilter; |
| 2823 int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( |
| 2824 p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp |
| 2825 ); |
| 2826 |
| 2827 if( pCsr->nSegment==0 ) return SQLITE_OK; |
| 2828 |
| 2829 do { |
| 2830 int nMerge; |
| 2831 int i; |
| 2832 |
| 2833 /* Advance the first pCsr->nAdvance entries in the apSegment[] array |
| 2834 ** forward. Then sort the list in order of current term again. |
| 2835 */ |
| 2836 for(i=0; i<pCsr->nAdvance; i++){ |
| 2837 Fts3SegReader *pSeg = apSegment[i]; |
| 2838 if( pSeg->bLookup ){ |
| 2839 fts3SegReaderSetEof(pSeg); |
| 2840 }else{ |
| 2841 rc = fts3SegReaderNext(p, pSeg, 0); |
| 2842 } |
| 2843 if( rc!=SQLITE_OK ) return rc; |
| 2844 } |
| 2845 fts3SegReaderSort(apSegment, nSegment, pCsr->nAdvance, fts3SegReaderCmp); |
| 2846 pCsr->nAdvance = 0; |
| 2847 |
| 2848 /* If all the seg-readers are at EOF, we're finished. return SQLITE_OK. */ |
| 2849 assert( rc==SQLITE_OK ); |
| 2850 if( apSegment[0]->aNode==0 ) break; |
| 2851 |
| 2852 pCsr->nTerm = apSegment[0]->nTerm; |
| 2853 pCsr->zTerm = apSegment[0]->zTerm; |
| 2854 |
| 2855 /* If this is a prefix-search, and if the term that apSegment[0] points |
| 2856 ** to does not share a suffix with pFilter->zTerm/nTerm, then all |
| 2857 ** required callbacks have been made. In this case exit early. |
| 2858 ** |
| 2859 ** Similarly, if this is a search for an exact match, and the first term |
| 2860 ** of segment apSegment[0] is not a match, exit early. |
| 2861 */ |
| 2862 if( pFilter->zTerm && !isScan ){ |
| 2863 if( pCsr->nTerm<pFilter->nTerm |
| 2864 || (!isPrefix && pCsr->nTerm>pFilter->nTerm) |
| 2865 || memcmp(pCsr->zTerm, pFilter->zTerm, pFilter->nTerm) |
| 2866 ){ |
| 2867 break; |
| 2868 } |
| 2869 } |
| 2870 |
| 2871 nMerge = 1; |
| 2872 while( nMerge<nSegment |
| 2873 && apSegment[nMerge]->aNode |
| 2874 && apSegment[nMerge]->nTerm==pCsr->nTerm |
| 2875 && 0==memcmp(pCsr->zTerm, apSegment[nMerge]->zTerm, pCsr->nTerm) |
| 2876 ){ |
| 2877 nMerge++; |
| 2878 } |
| 2879 |
| 2880 assert( isIgnoreEmpty || (isRequirePos && !isColFilter) ); |
| 2881 if( nMerge==1 |
| 2882 && !isIgnoreEmpty |
| 2883 && !isFirst |
| 2884 && (p->bDescIdx==0 || fts3SegReaderIsPending(apSegment[0])==0) |
| 2885 ){ |
| 2886 pCsr->nDoclist = apSegment[0]->nDoclist; |
| 2887 if( fts3SegReaderIsPending(apSegment[0]) ){ |
| 2888 rc = fts3MsrBufferData(pCsr, apSegment[0]->aDoclist, pCsr->nDoclist); |
| 2889 pCsr->aDoclist = pCsr->aBuffer; |
| 2890 }else{ |
| 2891 pCsr->aDoclist = apSegment[0]->aDoclist; |
| 2892 } |
| 2893 if( rc==SQLITE_OK ) rc = SQLITE_ROW; |
| 2894 }else{ |
| 2895 int nDoclist = 0; /* Size of doclist */ |
| 2896 sqlite3_int64 iPrev = 0; /* Previous docid stored in doclist */ |
| 2897 |
| 2898 /* The current term of the first nMerge entries in the array |
| 2899 ** of Fts3SegReader objects is the same. The doclists must be merged |
| 2900 ** and a single term returned with the merged doclist. |
| 2901 */ |
| 2902 for(i=0; i<nMerge; i++){ |
| 2903 fts3SegReaderFirstDocid(p, apSegment[i]); |
| 2904 } |
| 2905 fts3SegReaderSort(apSegment, nMerge, nMerge, xCmp); |
| 2906 while( apSegment[0]->pOffsetList ){ |
| 2907 int j; /* Number of segments that share a docid */ |
| 2908 char *pList = 0; |
| 2909 int nList = 0; |
| 2910 int nByte; |
| 2911 sqlite3_int64 iDocid = apSegment[0]->iDocid; |
| 2912 fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); |
| 2913 j = 1; |
| 2914 while( j<nMerge |
| 2915 && apSegment[j]->pOffsetList |
| 2916 && apSegment[j]->iDocid==iDocid |
| 2917 ){ |
| 2918 fts3SegReaderNextDocid(p, apSegment[j], 0, 0); |
| 2919 j++; |
| 2920 } |
| 2921 |
| 2922 if( isColFilter ){ |
| 2923 fts3ColumnFilter(pFilter->iCol, 0, &pList, &nList); |
| 2924 } |
| 2925 |
| 2926 if( !isIgnoreEmpty || nList>0 ){ |
| 2927 |
| 2928 /* Calculate the 'docid' delta value to write into the merged |
| 2929 ** doclist. */ |
| 2930 sqlite3_int64 iDelta; |
| 2931 if( p->bDescIdx && nDoclist>0 ){ |
| 2932 iDelta = iPrev - iDocid; |
| 2933 }else{ |
| 2934 iDelta = iDocid - iPrev; |
| 2935 } |
| 2936 assert( iDelta>0 || (nDoclist==0 && iDelta==iDocid) ); |
| 2937 assert( nDoclist>0 || iDelta==iDocid ); |
| 2938 |
| 2939 nByte = sqlite3Fts3VarintLen(iDelta) + (isRequirePos?nList+1:0); |
| 2940 if( nDoclist+nByte>pCsr->nBuffer ){ |
| 2941 char *aNew; |
| 2942 pCsr->nBuffer = (nDoclist+nByte)*2; |
| 2943 aNew = sqlite3_realloc(pCsr->aBuffer, pCsr->nBuffer); |
| 2944 if( !aNew ){ |
| 2945 return SQLITE_NOMEM; |
| 2946 } |
| 2947 pCsr->aBuffer = aNew; |
| 2948 } |
| 2949 |
| 2950 if( isFirst ){ |
| 2951 char *a = &pCsr->aBuffer[nDoclist]; |
| 2952 int nWrite; |
| 2953 |
| 2954 nWrite = sqlite3Fts3FirstFilter(iDelta, pList, nList, a); |
| 2955 if( nWrite ){ |
| 2956 iPrev = iDocid; |
| 2957 nDoclist += nWrite; |
| 2958 } |
| 2959 }else{ |
| 2960 nDoclist += sqlite3Fts3PutVarint(&pCsr->aBuffer[nDoclist], iDelta); |
| 2961 iPrev = iDocid; |
| 2962 if( isRequirePos ){ |
| 2963 memcpy(&pCsr->aBuffer[nDoclist], pList, nList); |
| 2964 nDoclist += nList; |
| 2965 pCsr->aBuffer[nDoclist++] = '\0'; |
| 2966 } |
| 2967 } |
| 2968 } |
| 2969 |
| 2970 fts3SegReaderSort(apSegment, nMerge, j, xCmp); |
| 2971 } |
| 2972 if( nDoclist>0 ){ |
| 2973 pCsr->aDoclist = pCsr->aBuffer; |
| 2974 pCsr->nDoclist = nDoclist; |
| 2975 rc = SQLITE_ROW; |
| 2976 } |
| 2977 } |
| 2978 pCsr->nAdvance = nMerge; |
| 2979 }while( rc==SQLITE_OK ); |
| 2980 |
| 2981 return rc; |
| 2982 } |
| 2983 |
| 2984 |
| 2985 void sqlite3Fts3SegReaderFinish( |
| 2986 Fts3MultiSegReader *pCsr /* Cursor object */ |
| 2987 ){ |
| 2988 if( pCsr ){ |
| 2989 int i; |
| 2990 for(i=0; i<pCsr->nSegment; i++){ |
| 2991 sqlite3Fts3SegReaderFree(pCsr->apSegment[i]); |
| 2992 } |
| 2993 sqlite3_free(pCsr->apSegment); |
| 2994 sqlite3_free(pCsr->aBuffer); |
| 2995 |
| 2996 pCsr->nSegment = 0; |
| 2997 pCsr->apSegment = 0; |
| 2998 pCsr->aBuffer = 0; |
| 2999 } |
| 3000 } |
| 3001 |
| 3002 /* |
| 3003 ** Decode the "end_block" field, selected by column iCol of the SELECT |
| 3004 ** statement passed as the first argument. |
| 3005 ** |
| 3006 ** The "end_block" field may contain either an integer, or a text field |
| 3007 ** containing the text representation of two non-negative integers separated |
| 3008 ** by one or more space (0x20) characters. In the first case, set *piEndBlock |
| 3009 ** to the integer value and *pnByte to zero before returning. In the second, |
| 3010 ** set *piEndBlock to the first value and *pnByte to the second. |
| 3011 */ |
| 3012 static void fts3ReadEndBlockField( |
| 3013 sqlite3_stmt *pStmt, |
| 3014 int iCol, |
| 3015 i64 *piEndBlock, |
| 3016 i64 *pnByte |
| 3017 ){ |
| 3018 const unsigned char *zText = sqlite3_column_text(pStmt, iCol); |
| 3019 if( zText ){ |
| 3020 int i; |
| 3021 int iMul = 1; |
| 3022 i64 iVal = 0; |
| 3023 for(i=0; zText[i]>='0' && zText[i]<='9'; i++){ |
| 3024 iVal = iVal*10 + (zText[i] - '0'); |
| 3025 } |
| 3026 *piEndBlock = iVal; |
| 3027 while( zText[i]==' ' ) i++; |
| 3028 iVal = 0; |
| 3029 if( zText[i]=='-' ){ |
| 3030 i++; |
| 3031 iMul = -1; |
| 3032 } |
| 3033 for(/* no-op */; zText[i]>='0' && zText[i]<='9'; i++){ |
| 3034 iVal = iVal*10 + (zText[i] - '0'); |
| 3035 } |
| 3036 *pnByte = (iVal * (i64)iMul); |
| 3037 } |
| 3038 } |
| 3039 |
| 3040 |
| 3041 /* |
| 3042 ** A segment of size nByte bytes has just been written to absolute level |
| 3043 ** iAbsLevel. Promote any segments that should be promoted as a result. |
| 3044 */ |
| 3045 static int fts3PromoteSegments( |
| 3046 Fts3Table *p, /* FTS table handle */ |
| 3047 sqlite3_int64 iAbsLevel, /* Absolute level just updated */ |
| 3048 sqlite3_int64 nByte /* Size of new segment at iAbsLevel */ |
| 3049 ){ |
| 3050 int rc = SQLITE_OK; |
| 3051 sqlite3_stmt *pRange; |
| 3052 |
| 3053 rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_RANGE2, &pRange, 0); |
| 3054 |
| 3055 if( rc==SQLITE_OK ){ |
| 3056 int bOk = 0; |
| 3057 i64 iLast = (iAbsLevel/FTS3_SEGDIR_MAXLEVEL + 1) * FTS3_SEGDIR_MAXLEVEL - 1; |
| 3058 i64 nLimit = (nByte*3)/2; |
| 3059 |
| 3060 /* Loop through all entries in the %_segdir table corresponding to |
| 3061 ** segments in this index on levels greater than iAbsLevel. If there is |
| 3062 ** at least one such segment, and it is possible to determine that all |
| 3063 ** such segments are smaller than nLimit bytes in size, they will be |
| 3064 ** promoted to level iAbsLevel. */ |
| 3065 sqlite3_bind_int64(pRange, 1, iAbsLevel+1); |
| 3066 sqlite3_bind_int64(pRange, 2, iLast); |
| 3067 while( SQLITE_ROW==sqlite3_step(pRange) ){ |
| 3068 i64 nSize = 0, dummy; |
| 3069 fts3ReadEndBlockField(pRange, 2, &dummy, &nSize); |
| 3070 if( nSize<=0 || nSize>nLimit ){ |
| 3071 /* If nSize==0, then the %_segdir.end_block field does not not |
| 3072 ** contain a size value. This happens if it was written by an |
| 3073 ** old version of FTS. In this case it is not possible to determine |
| 3074 ** the size of the segment, and so segment promotion does not |
| 3075 ** take place. */ |
| 3076 bOk = 0; |
| 3077 break; |
| 3078 } |
| 3079 bOk = 1; |
| 3080 } |
| 3081 rc = sqlite3_reset(pRange); |
| 3082 |
| 3083 if( bOk ){ |
| 3084 int iIdx = 0; |
| 3085 sqlite3_stmt *pUpdate1; |
| 3086 sqlite3_stmt *pUpdate2; |
| 3087 |
| 3088 if( rc==SQLITE_OK ){ |
| 3089 rc = fts3SqlStmt(p, SQL_UPDATE_LEVEL_IDX, &pUpdate1, 0); |
| 3090 } |
| 3091 if( rc==SQLITE_OK ){ |
| 3092 rc = fts3SqlStmt(p, SQL_UPDATE_LEVEL, &pUpdate2, 0); |
| 3093 } |
| 3094 |
| 3095 if( rc==SQLITE_OK ){ |
| 3096 |
| 3097 /* Loop through all %_segdir entries for segments in this index with |
| 3098 ** levels equal to or greater than iAbsLevel. As each entry is visited, |
| 3099 ** updated it to set (level = -1) and (idx = N), where N is 0 for the |
| 3100 ** oldest segment in the range, 1 for the next oldest, and so on. |
| 3101 ** |
| 3102 ** In other words, move all segments being promoted to level -1, |
| 3103 ** setting the "idx" fields as appropriate to keep them in the same |
| 3104 ** order. The contents of level -1 (which is never used, except |
| 3105 ** transiently here), will be moved back to level iAbsLevel below. */ |
| 3106 sqlite3_bind_int64(pRange, 1, iAbsLevel); |
| 3107 while( SQLITE_ROW==sqlite3_step(pRange) ){ |
| 3108 sqlite3_bind_int(pUpdate1, 1, iIdx++); |
| 3109 sqlite3_bind_int(pUpdate1, 2, sqlite3_column_int(pRange, 0)); |
| 3110 sqlite3_bind_int(pUpdate1, 3, sqlite3_column_int(pRange, 1)); |
| 3111 sqlite3_step(pUpdate1); |
| 3112 rc = sqlite3_reset(pUpdate1); |
| 3113 if( rc!=SQLITE_OK ){ |
| 3114 sqlite3_reset(pRange); |
| 3115 break; |
| 3116 } |
| 3117 } |
| 3118 } |
| 3119 if( rc==SQLITE_OK ){ |
| 3120 rc = sqlite3_reset(pRange); |
| 3121 } |
| 3122 |
| 3123 /* Move level -1 to level iAbsLevel */ |
| 3124 if( rc==SQLITE_OK ){ |
| 3125 sqlite3_bind_int64(pUpdate2, 1, iAbsLevel); |
| 3126 sqlite3_step(pUpdate2); |
| 3127 rc = sqlite3_reset(pUpdate2); |
| 3128 } |
| 3129 } |
| 3130 } |
| 3131 |
| 3132 |
| 3133 return rc; |
| 3134 } |
| 3135 |
| 3136 /* |
| 3137 ** Merge all level iLevel segments in the database into a single |
| 3138 ** iLevel+1 segment. Or, if iLevel<0, merge all segments into a |
| 3139 ** single segment with a level equal to the numerically largest level |
| 3140 ** currently present in the database. |
| 3141 ** |
| 3142 ** If this function is called with iLevel<0, but there is only one |
| 3143 ** segment in the database, SQLITE_DONE is returned immediately. |
| 3144 ** Otherwise, if successful, SQLITE_OK is returned. If an error occurs, |
| 3145 ** an SQLite error code is returned. |
| 3146 */ |
| 3147 static int fts3SegmentMerge( |
| 3148 Fts3Table *p, |
| 3149 int iLangid, /* Language id to merge */ |
| 3150 int iIndex, /* Index in p->aIndex[] to merge */ |
| 3151 int iLevel /* Level to merge */ |
| 3152 ){ |
| 3153 int rc; /* Return code */ |
| 3154 int iIdx = 0; /* Index of new segment */ |
| 3155 sqlite3_int64 iNewLevel = 0; /* Level/index to create new segment at */ |
| 3156 SegmentWriter *pWriter = 0; /* Used to write the new, merged, segment */ |
| 3157 Fts3SegFilter filter; /* Segment term filter condition */ |
| 3158 Fts3MultiSegReader csr; /* Cursor to iterate through level(s) */ |
| 3159 int bIgnoreEmpty = 0; /* True to ignore empty segments */ |
| 3160 i64 iMaxLevel = 0; /* Max level number for this index/langid */ |
| 3161 |
| 3162 assert( iLevel==FTS3_SEGCURSOR_ALL |
| 3163 || iLevel==FTS3_SEGCURSOR_PENDING |
| 3164 || iLevel>=0 |
| 3165 ); |
| 3166 assert( iLevel<FTS3_SEGDIR_MAXLEVEL ); |
| 3167 assert( iIndex>=0 && iIndex<p->nIndex ); |
| 3168 |
| 3169 rc = sqlite3Fts3SegReaderCursor(p, iLangid, iIndex, iLevel, 0, 0, 1, 0, &csr); |
| 3170 if( rc!=SQLITE_OK || csr.nSegment==0 ) goto finished; |
| 3171 |
| 3172 if( iLevel!=FTS3_SEGCURSOR_PENDING ){ |
| 3173 rc = fts3SegmentMaxLevel(p, iLangid, iIndex, &iMaxLevel); |
| 3174 if( rc!=SQLITE_OK ) goto finished; |
| 3175 } |
| 3176 |
| 3177 if( iLevel==FTS3_SEGCURSOR_ALL ){ |
| 3178 /* This call is to merge all segments in the database to a single |
| 3179 ** segment. The level of the new segment is equal to the numerically |
| 3180 ** greatest segment level currently present in the database for this |
| 3181 ** index. The idx of the new segment is always 0. */ |
| 3182 if( csr.nSegment==1 ){ |
| 3183 rc = SQLITE_DONE; |
| 3184 goto finished; |
| 3185 } |
| 3186 iNewLevel = iMaxLevel; |
| 3187 bIgnoreEmpty = 1; |
| 3188 |
| 3189 }else{ |
| 3190 /* This call is to merge all segments at level iLevel. find the next |
| 3191 ** available segment index at level iLevel+1. The call to |
| 3192 ** fts3AllocateSegdirIdx() will merge the segments at level iLevel+1 to |
| 3193 ** a single iLevel+2 segment if necessary. */ |
| 3194 assert( FTS3_SEGCURSOR_PENDING==-1 ); |
| 3195 iNewLevel = getAbsoluteLevel(p, iLangid, iIndex, iLevel+1); |
| 3196 rc = fts3AllocateSegdirIdx(p, iLangid, iIndex, iLevel+1, &iIdx); |
| 3197 bIgnoreEmpty = (iLevel!=FTS3_SEGCURSOR_PENDING) && (iNewLevel>iMaxLevel); |
| 3198 } |
| 3199 if( rc!=SQLITE_OK ) goto finished; |
| 3200 |
| 3201 assert( csr.nSegment>0 ); |
| 3202 assert( iNewLevel>=getAbsoluteLevel(p, iLangid, iIndex, 0) ); |
| 3203 assert( iNewLevel<getAbsoluteLevel(p, iLangid, iIndex,FTS3_SEGDIR_MAXLEVEL) ); |
| 3204 |
| 3205 memset(&filter, 0, sizeof(Fts3SegFilter)); |
| 3206 filter.flags = FTS3_SEGMENT_REQUIRE_POS; |
| 3207 filter.flags |= (bIgnoreEmpty ? FTS3_SEGMENT_IGNORE_EMPTY : 0); |
| 3208 |
| 3209 rc = sqlite3Fts3SegReaderStart(p, &csr, &filter); |
| 3210 while( SQLITE_OK==rc ){ |
| 3211 rc = sqlite3Fts3SegReaderStep(p, &csr); |
| 3212 if( rc!=SQLITE_ROW ) break; |
| 3213 rc = fts3SegWriterAdd(p, &pWriter, 1, |
| 3214 csr.zTerm, csr.nTerm, csr.aDoclist, csr.nDoclist); |
| 3215 } |
| 3216 if( rc!=SQLITE_OK ) goto finished; |
| 3217 assert( pWriter || bIgnoreEmpty ); |
| 3218 |
| 3219 if( iLevel!=FTS3_SEGCURSOR_PENDING ){ |
| 3220 rc = fts3DeleteSegdir( |
| 3221 p, iLangid, iIndex, iLevel, csr.apSegment, csr.nSegment |
| 3222 ); |
| 3223 if( rc!=SQLITE_OK ) goto finished; |
| 3224 } |
| 3225 if( pWriter ){ |
| 3226 rc = fts3SegWriterFlush(p, pWriter, iNewLevel, iIdx); |
| 3227 if( rc==SQLITE_OK ){ |
| 3228 if( iLevel==FTS3_SEGCURSOR_PENDING || iNewLevel<iMaxLevel ){ |
| 3229 rc = fts3PromoteSegments(p, iNewLevel, pWriter->nLeafData); |
| 3230 } |
| 3231 } |
| 3232 } |
| 3233 |
| 3234 finished: |
| 3235 fts3SegWriterFree(pWriter); |
| 3236 sqlite3Fts3SegReaderFinish(&csr); |
| 3237 return rc; |
| 3238 } |
| 3239 |
| 3240 |
| 3241 /* |
| 3242 ** Flush the contents of pendingTerms to level 0 segments. |
| 3243 */ |
| 3244 int sqlite3Fts3PendingTermsFlush(Fts3Table *p){ |
| 3245 int rc = SQLITE_OK; |
| 3246 int i; |
| 3247 |
| 3248 for(i=0; rc==SQLITE_OK && i<p->nIndex; i++){ |
| 3249 rc = fts3SegmentMerge(p, p->iPrevLangid, i, FTS3_SEGCURSOR_PENDING); |
| 3250 if( rc==SQLITE_DONE ) rc = SQLITE_OK; |
| 3251 } |
| 3252 sqlite3Fts3PendingTermsClear(p); |
| 3253 |
| 3254 /* Determine the auto-incr-merge setting if unknown. If enabled, |
| 3255 ** estimate the number of leaf blocks of content to be written |
| 3256 */ |
| 3257 if( rc==SQLITE_OK && p->bHasStat |
| 3258 && p->nAutoincrmerge==0xff && p->nLeafAdd>0 |
| 3259 ){ |
| 3260 sqlite3_stmt *pStmt = 0; |
| 3261 rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pStmt, 0); |
| 3262 if( rc==SQLITE_OK ){ |
| 3263 sqlite3_bind_int(pStmt, 1, FTS_STAT_AUTOINCRMERGE); |
| 3264 rc = sqlite3_step(pStmt); |
| 3265 if( rc==SQLITE_ROW ){ |
| 3266 p->nAutoincrmerge = sqlite3_column_int(pStmt, 0); |
| 3267 if( p->nAutoincrmerge==1 ) p->nAutoincrmerge = 8; |
| 3268 }else if( rc==SQLITE_DONE ){ |
| 3269 p->nAutoincrmerge = 0; |
| 3270 } |
| 3271 rc = sqlite3_reset(pStmt); |
| 3272 } |
| 3273 } |
| 3274 return rc; |
| 3275 } |
| 3276 |
| 3277 /* |
| 3278 ** Encode N integers as varints into a blob. |
| 3279 */ |
| 3280 static void fts3EncodeIntArray( |
| 3281 int N, /* The number of integers to encode */ |
| 3282 u32 *a, /* The integer values */ |
| 3283 char *zBuf, /* Write the BLOB here */ |
| 3284 int *pNBuf /* Write number of bytes if zBuf[] used here */ |
| 3285 ){ |
| 3286 int i, j; |
| 3287 for(i=j=0; i<N; i++){ |
| 3288 j += sqlite3Fts3PutVarint(&zBuf[j], (sqlite3_int64)a[i]); |
| 3289 } |
| 3290 *pNBuf = j; |
| 3291 } |
| 3292 |
| 3293 /* |
| 3294 ** Decode a blob of varints into N integers |
| 3295 */ |
| 3296 static void fts3DecodeIntArray( |
| 3297 int N, /* The number of integers to decode */ |
| 3298 u32 *a, /* Write the integer values */ |
| 3299 const char *zBuf, /* The BLOB containing the varints */ |
| 3300 int nBuf /* size of the BLOB */ |
| 3301 ){ |
| 3302 int i, j; |
| 3303 UNUSED_PARAMETER(nBuf); |
| 3304 for(i=j=0; i<N; i++){ |
| 3305 sqlite3_int64 x; |
| 3306 j += sqlite3Fts3GetVarint(&zBuf[j], &x); |
| 3307 assert(j<=nBuf); |
| 3308 a[i] = (u32)(x & 0xffffffff); |
| 3309 } |
| 3310 } |
| 3311 |
| 3312 /* |
| 3313 ** Insert the sizes (in tokens) for each column of the document |
| 3314 ** with docid equal to p->iPrevDocid. The sizes are encoded as |
| 3315 ** a blob of varints. |
| 3316 */ |
| 3317 static void fts3InsertDocsize( |
| 3318 int *pRC, /* Result code */ |
| 3319 Fts3Table *p, /* Table into which to insert */ |
| 3320 u32 *aSz /* Sizes of each column, in tokens */ |
| 3321 ){ |
| 3322 char *pBlob; /* The BLOB encoding of the document size */ |
| 3323 int nBlob; /* Number of bytes in the BLOB */ |
| 3324 sqlite3_stmt *pStmt; /* Statement used to insert the encoding */ |
| 3325 int rc; /* Result code from subfunctions */ |
| 3326 |
| 3327 if( *pRC ) return; |
| 3328 pBlob = sqlite3_malloc( 10*p->nColumn ); |
| 3329 if( pBlob==0 ){ |
| 3330 *pRC = SQLITE_NOMEM; |
| 3331 return; |
| 3332 } |
| 3333 fts3EncodeIntArray(p->nColumn, aSz, pBlob, &nBlob); |
| 3334 rc = fts3SqlStmt(p, SQL_REPLACE_DOCSIZE, &pStmt, 0); |
| 3335 if( rc ){ |
| 3336 sqlite3_free(pBlob); |
| 3337 *pRC = rc; |
| 3338 return; |
| 3339 } |
| 3340 sqlite3_bind_int64(pStmt, 1, p->iPrevDocid); |
| 3341 sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, sqlite3_free); |
| 3342 sqlite3_step(pStmt); |
| 3343 *pRC = sqlite3_reset(pStmt); |
| 3344 } |
| 3345 |
| 3346 /* |
| 3347 ** Record 0 of the %_stat table contains a blob consisting of N varints, |
| 3348 ** where N is the number of user defined columns in the fts3 table plus |
| 3349 ** two. If nCol is the number of user defined columns, then values of the |
| 3350 ** varints are set as follows: |
| 3351 ** |
| 3352 ** Varint 0: Total number of rows in the table. |
| 3353 ** |
| 3354 ** Varint 1..nCol: For each column, the total number of tokens stored in |
| 3355 ** the column for all rows of the table. |
| 3356 ** |
| 3357 ** Varint 1+nCol: The total size, in bytes, of all text values in all |
| 3358 ** columns of all rows of the table. |
| 3359 ** |
| 3360 */ |
| 3361 static void fts3UpdateDocTotals( |
| 3362 int *pRC, /* The result code */ |
| 3363 Fts3Table *p, /* Table being updated */ |
| 3364 u32 *aSzIns, /* Size increases */ |
| 3365 u32 *aSzDel, /* Size decreases */ |
| 3366 int nChng /* Change in the number of documents */ |
| 3367 ){ |
| 3368 char *pBlob; /* Storage for BLOB written into %_stat */ |
| 3369 int nBlob; /* Size of BLOB written into %_stat */ |
| 3370 u32 *a; /* Array of integers that becomes the BLOB */ |
| 3371 sqlite3_stmt *pStmt; /* Statement for reading and writing */ |
| 3372 int i; /* Loop counter */ |
| 3373 int rc; /* Result code from subfunctions */ |
| 3374 |
| 3375 const int nStat = p->nColumn+2; |
| 3376 |
| 3377 if( *pRC ) return; |
| 3378 a = sqlite3_malloc( (sizeof(u32)+10)*nStat ); |
| 3379 if( a==0 ){ |
| 3380 *pRC = SQLITE_NOMEM; |
| 3381 return; |
| 3382 } |
| 3383 pBlob = (char*)&a[nStat]; |
| 3384 rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pStmt, 0); |
| 3385 if( rc ){ |
| 3386 sqlite3_free(a); |
| 3387 *pRC = rc; |
| 3388 return; |
| 3389 } |
| 3390 sqlite3_bind_int(pStmt, 1, FTS_STAT_DOCTOTAL); |
| 3391 if( sqlite3_step(pStmt)==SQLITE_ROW ){ |
| 3392 fts3DecodeIntArray(nStat, a, |
| 3393 sqlite3_column_blob(pStmt, 0), |
| 3394 sqlite3_column_bytes(pStmt, 0)); |
| 3395 }else{ |
| 3396 memset(a, 0, sizeof(u32)*(nStat) ); |
| 3397 } |
| 3398 rc = sqlite3_reset(pStmt); |
| 3399 if( rc!=SQLITE_OK ){ |
| 3400 sqlite3_free(a); |
| 3401 *pRC = rc; |
| 3402 return; |
| 3403 } |
| 3404 if( nChng<0 && a[0]<(u32)(-nChng) ){ |
| 3405 a[0] = 0; |
| 3406 }else{ |
| 3407 a[0] += nChng; |
| 3408 } |
| 3409 for(i=0; i<p->nColumn+1; i++){ |
| 3410 u32 x = a[i+1]; |
| 3411 if( x+aSzIns[i] < aSzDel[i] ){ |
| 3412 x = 0; |
| 3413 }else{ |
| 3414 x = x + aSzIns[i] - aSzDel[i]; |
| 3415 } |
| 3416 a[i+1] = x; |
| 3417 } |
| 3418 fts3EncodeIntArray(nStat, a, pBlob, &nBlob); |
| 3419 rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pStmt, 0); |
| 3420 if( rc ){ |
| 3421 sqlite3_free(a); |
| 3422 *pRC = rc; |
| 3423 return; |
| 3424 } |
| 3425 sqlite3_bind_int(pStmt, 1, FTS_STAT_DOCTOTAL); |
| 3426 sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, SQLITE_STATIC); |
| 3427 sqlite3_step(pStmt); |
| 3428 *pRC = sqlite3_reset(pStmt); |
| 3429 sqlite3_free(a); |
| 3430 } |
| 3431 |
| 3432 /* |
| 3433 ** Merge the entire database so that there is one segment for each |
| 3434 ** iIndex/iLangid combination. |
| 3435 */ |
| 3436 static int fts3DoOptimize(Fts3Table *p, int bReturnDone){ |
| 3437 int bSeenDone = 0; |
| 3438 int rc; |
| 3439 sqlite3_stmt *pAllLangid = 0; |
| 3440 |
| 3441 rc = fts3SqlStmt(p, SQL_SELECT_ALL_LANGID, &pAllLangid, 0); |
| 3442 if( rc==SQLITE_OK ){ |
| 3443 int rc2; |
| 3444 sqlite3_bind_int(pAllLangid, 1, p->nIndex); |
| 3445 while( sqlite3_step(pAllLangid)==SQLITE_ROW ){ |
| 3446 int i; |
| 3447 int iLangid = sqlite3_column_int(pAllLangid, 0); |
| 3448 for(i=0; rc==SQLITE_OK && i<p->nIndex; i++){ |
| 3449 rc = fts3SegmentMerge(p, iLangid, i, FTS3_SEGCURSOR_ALL); |
| 3450 if( rc==SQLITE_DONE ){ |
| 3451 bSeenDone = 1; |
| 3452 rc = SQLITE_OK; |
| 3453 } |
| 3454 } |
| 3455 } |
| 3456 rc2 = sqlite3_reset(pAllLangid); |
| 3457 if( rc==SQLITE_OK ) rc = rc2; |
| 3458 } |
| 3459 |
| 3460 sqlite3Fts3SegmentsClose(p); |
| 3461 sqlite3Fts3PendingTermsClear(p); |
| 3462 |
| 3463 return (rc==SQLITE_OK && bReturnDone && bSeenDone) ? SQLITE_DONE : rc; |
| 3464 } |
| 3465 |
| 3466 /* |
| 3467 ** This function is called when the user executes the following statement: |
| 3468 ** |
| 3469 ** INSERT INTO <tbl>(<tbl>) VALUES('rebuild'); |
| 3470 ** |
| 3471 ** The entire FTS index is discarded and rebuilt. If the table is one |
| 3472 ** created using the content=xxx option, then the new index is based on |
| 3473 ** the current contents of the xxx table. Otherwise, it is rebuilt based |
| 3474 ** on the contents of the %_content table. |
| 3475 */ |
| 3476 static int fts3DoRebuild(Fts3Table *p){ |
| 3477 int rc; /* Return Code */ |
| 3478 |
| 3479 rc = fts3DeleteAll(p, 0); |
| 3480 if( rc==SQLITE_OK ){ |
| 3481 u32 *aSz = 0; |
| 3482 u32 *aSzIns = 0; |
| 3483 u32 *aSzDel = 0; |
| 3484 sqlite3_stmt *pStmt = 0; |
| 3485 int nEntry = 0; |
| 3486 |
| 3487 /* Compose and prepare an SQL statement to loop through the content table */ |
| 3488 char *zSql = sqlite3_mprintf("SELECT %s" , p->zReadExprlist); |
| 3489 if( !zSql ){ |
| 3490 rc = SQLITE_NOMEM; |
| 3491 }else{ |
| 3492 rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); |
| 3493 sqlite3_free(zSql); |
| 3494 } |
| 3495 |
| 3496 if( rc==SQLITE_OK ){ |
| 3497 int nByte = sizeof(u32) * (p->nColumn+1)*3; |
| 3498 aSz = (u32 *)sqlite3_malloc(nByte); |
| 3499 if( aSz==0 ){ |
| 3500 rc = SQLITE_NOMEM; |
| 3501 }else{ |
| 3502 memset(aSz, 0, nByte); |
| 3503 aSzIns = &aSz[p->nColumn+1]; |
| 3504 aSzDel = &aSzIns[p->nColumn+1]; |
| 3505 } |
| 3506 } |
| 3507 |
| 3508 while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 3509 int iCol; |
| 3510 int iLangid = langidFromSelect(p, pStmt); |
| 3511 rc = fts3PendingTermsDocid(p, iLangid, sqlite3_column_int64(pStmt, 0)); |
| 3512 memset(aSz, 0, sizeof(aSz[0]) * (p->nColumn+1)); |
| 3513 for(iCol=0; rc==SQLITE_OK && iCol<p->nColumn; iCol++){ |
| 3514 if( p->abNotindexed[iCol]==0 ){ |
| 3515 const char *z = (const char *) sqlite3_column_text(pStmt, iCol+1); |
| 3516 rc = fts3PendingTermsAdd(p, iLangid, z, iCol, &aSz[iCol]); |
| 3517 aSz[p->nColumn] += sqlite3_column_bytes(pStmt, iCol+1); |
| 3518 } |
| 3519 } |
| 3520 if( p->bHasDocsize ){ |
| 3521 fts3InsertDocsize(&rc, p, aSz); |
| 3522 } |
| 3523 if( rc!=SQLITE_OK ){ |
| 3524 sqlite3_finalize(pStmt); |
| 3525 pStmt = 0; |
| 3526 }else{ |
| 3527 nEntry++; |
| 3528 for(iCol=0; iCol<=p->nColumn; iCol++){ |
| 3529 aSzIns[iCol] += aSz[iCol]; |
| 3530 } |
| 3531 } |
| 3532 } |
| 3533 if( p->bFts4 ){ |
| 3534 fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nEntry); |
| 3535 } |
| 3536 sqlite3_free(aSz); |
| 3537 |
| 3538 if( pStmt ){ |
| 3539 int rc2 = sqlite3_finalize(pStmt); |
| 3540 if( rc==SQLITE_OK ){ |
| 3541 rc = rc2; |
| 3542 } |
| 3543 } |
| 3544 } |
| 3545 |
| 3546 return rc; |
| 3547 } |
| 3548 |
| 3549 |
| 3550 /* |
| 3551 ** This function opens a cursor used to read the input data for an |
| 3552 ** incremental merge operation. Specifically, it opens a cursor to scan |
| 3553 ** the oldest nSeg segments (idx=0 through idx=(nSeg-1)) in absolute |
| 3554 ** level iAbsLevel. |
| 3555 */ |
| 3556 static int fts3IncrmergeCsr( |
| 3557 Fts3Table *p, /* FTS3 table handle */ |
| 3558 sqlite3_int64 iAbsLevel, /* Absolute level to open */ |
| 3559 int nSeg, /* Number of segments to merge */ |
| 3560 Fts3MultiSegReader *pCsr /* Cursor object to populate */ |
| 3561 ){ |
| 3562 int rc; /* Return Code */ |
| 3563 sqlite3_stmt *pStmt = 0; /* Statement used to read %_segdir entry */ |
| 3564 int nByte; /* Bytes allocated at pCsr->apSegment[] */ |
| 3565 |
| 3566 /* Allocate space for the Fts3MultiSegReader.aCsr[] array */ |
| 3567 memset(pCsr, 0, sizeof(*pCsr)); |
| 3568 nByte = sizeof(Fts3SegReader *) * nSeg; |
| 3569 pCsr->apSegment = (Fts3SegReader **)sqlite3_malloc(nByte); |
| 3570 |
| 3571 if( pCsr->apSegment==0 ){ |
| 3572 rc = SQLITE_NOMEM; |
| 3573 }else{ |
| 3574 memset(pCsr->apSegment, 0, nByte); |
| 3575 rc = fts3SqlStmt(p, SQL_SELECT_LEVEL, &pStmt, 0); |
| 3576 } |
| 3577 if( rc==SQLITE_OK ){ |
| 3578 int i; |
| 3579 int rc2; |
| 3580 sqlite3_bind_int64(pStmt, 1, iAbsLevel); |
| 3581 assert( pCsr->nSegment==0 ); |
| 3582 for(i=0; rc==SQLITE_OK && sqlite3_step(pStmt)==SQLITE_ROW && i<nSeg; i++){ |
| 3583 rc = sqlite3Fts3SegReaderNew(i, 0, |
| 3584 sqlite3_column_int64(pStmt, 1), /* segdir.start_block */ |
| 3585 sqlite3_column_int64(pStmt, 2), /* segdir.leaves_end_block */ |
| 3586 sqlite3_column_int64(pStmt, 3), /* segdir.end_block */ |
| 3587 sqlite3_column_blob(pStmt, 4), /* segdir.root */ |
| 3588 sqlite3_column_bytes(pStmt, 4), /* segdir.root */ |
| 3589 &pCsr->apSegment[i] |
| 3590 ); |
| 3591 pCsr->nSegment++; |
| 3592 } |
| 3593 rc2 = sqlite3_reset(pStmt); |
| 3594 if( rc==SQLITE_OK ) rc = rc2; |
| 3595 } |
| 3596 |
| 3597 return rc; |
| 3598 } |
| 3599 |
| 3600 typedef struct IncrmergeWriter IncrmergeWriter; |
| 3601 typedef struct NodeWriter NodeWriter; |
| 3602 typedef struct Blob Blob; |
| 3603 typedef struct NodeReader NodeReader; |
| 3604 |
| 3605 /* |
| 3606 ** An instance of the following structure is used as a dynamic buffer |
| 3607 ** to build up nodes or other blobs of data in. |
| 3608 ** |
| 3609 ** The function blobGrowBuffer() is used to extend the allocation. |
| 3610 */ |
| 3611 struct Blob { |
| 3612 char *a; /* Pointer to allocation */ |
| 3613 int n; /* Number of valid bytes of data in a[] */ |
| 3614 int nAlloc; /* Allocated size of a[] (nAlloc>=n) */ |
| 3615 }; |
| 3616 |
| 3617 /* |
| 3618 ** This structure is used to build up buffers containing segment b-tree |
| 3619 ** nodes (blocks). |
| 3620 */ |
| 3621 struct NodeWriter { |
| 3622 sqlite3_int64 iBlock; /* Current block id */ |
| 3623 Blob key; /* Last key written to the current block */ |
| 3624 Blob block; /* Current block image */ |
| 3625 }; |
| 3626 |
| 3627 /* |
| 3628 ** An object of this type contains the state required to create or append |
| 3629 ** to an appendable b-tree segment. |
| 3630 */ |
| 3631 struct IncrmergeWriter { |
| 3632 int nLeafEst; /* Space allocated for leaf blocks */ |
| 3633 int nWork; /* Number of leaf pages flushed */ |
| 3634 sqlite3_int64 iAbsLevel; /* Absolute level of input segments */ |
| 3635 int iIdx; /* Index of *output* segment in iAbsLevel+1 */ |
| 3636 sqlite3_int64 iStart; /* Block number of first allocated block */ |
| 3637 sqlite3_int64 iEnd; /* Block number of last allocated block */ |
| 3638 sqlite3_int64 nLeafData; /* Bytes of leaf page data so far */ |
| 3639 u8 bNoLeafData; /* If true, store 0 for segment size */ |
| 3640 NodeWriter aNodeWriter[FTS_MAX_APPENDABLE_HEIGHT]; |
| 3641 }; |
| 3642 |
| 3643 /* |
| 3644 ** An object of the following type is used to read data from a single |
| 3645 ** FTS segment node. See the following functions: |
| 3646 ** |
| 3647 ** nodeReaderInit() |
| 3648 ** nodeReaderNext() |
| 3649 ** nodeReaderRelease() |
| 3650 */ |
| 3651 struct NodeReader { |
| 3652 const char *aNode; |
| 3653 int nNode; |
| 3654 int iOff; /* Current offset within aNode[] */ |
| 3655 |
| 3656 /* Output variables. Containing the current node entry. */ |
| 3657 sqlite3_int64 iChild; /* Pointer to child node */ |
| 3658 Blob term; /* Current term */ |
| 3659 const char *aDoclist; /* Pointer to doclist */ |
| 3660 int nDoclist; /* Size of doclist in bytes */ |
| 3661 }; |
| 3662 |
| 3663 /* |
| 3664 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op. |
| 3665 ** Otherwise, if the allocation at pBlob->a is not already at least nMin |
| 3666 ** bytes in size, extend (realloc) it to be so. |
| 3667 ** |
| 3668 ** If an OOM error occurs, set *pRc to SQLITE_NOMEM and leave pBlob->a |
| 3669 ** unmodified. Otherwise, if the allocation succeeds, update pBlob->nAlloc |
| 3670 ** to reflect the new size of the pBlob->a[] buffer. |
| 3671 */ |
| 3672 static void blobGrowBuffer(Blob *pBlob, int nMin, int *pRc){ |
| 3673 if( *pRc==SQLITE_OK && nMin>pBlob->nAlloc ){ |
| 3674 int nAlloc = nMin; |
| 3675 char *a = (char *)sqlite3_realloc(pBlob->a, nAlloc); |
| 3676 if( a ){ |
| 3677 pBlob->nAlloc = nAlloc; |
| 3678 pBlob->a = a; |
| 3679 }else{ |
| 3680 *pRc = SQLITE_NOMEM; |
| 3681 } |
| 3682 } |
| 3683 } |
| 3684 |
| 3685 /* |
| 3686 ** Attempt to advance the node-reader object passed as the first argument to |
| 3687 ** the next entry on the node. |
| 3688 ** |
| 3689 ** Return an error code if an error occurs (SQLITE_NOMEM is possible). |
| 3690 ** Otherwise return SQLITE_OK. If there is no next entry on the node |
| 3691 ** (e.g. because the current entry is the last) set NodeReader->aNode to |
| 3692 ** NULL to indicate EOF. Otherwise, populate the NodeReader structure output |
| 3693 ** variables for the new entry. |
| 3694 */ |
| 3695 static int nodeReaderNext(NodeReader *p){ |
| 3696 int bFirst = (p->term.n==0); /* True for first term on the node */ |
| 3697 int nPrefix = 0; /* Bytes to copy from previous term */ |
| 3698 int nSuffix = 0; /* Bytes to append to the prefix */ |
| 3699 int rc = SQLITE_OK; /* Return code */ |
| 3700 |
| 3701 assert( p->aNode ); |
| 3702 if( p->iChild && bFirst==0 ) p->iChild++; |
| 3703 if( p->iOff>=p->nNode ){ |
| 3704 /* EOF */ |
| 3705 p->aNode = 0; |
| 3706 }else{ |
| 3707 if( bFirst==0 ){ |
| 3708 p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &nPrefix); |
| 3709 } |
| 3710 p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &nSuffix); |
| 3711 |
| 3712 blobGrowBuffer(&p->term, nPrefix+nSuffix, &rc); |
| 3713 if( rc==SQLITE_OK ){ |
| 3714 memcpy(&p->term.a[nPrefix], &p->aNode[p->iOff], nSuffix); |
| 3715 p->term.n = nPrefix+nSuffix; |
| 3716 p->iOff += nSuffix; |
| 3717 if( p->iChild==0 ){ |
| 3718 p->iOff += fts3GetVarint32(&p->aNode[p->iOff], &p->nDoclist); |
| 3719 p->aDoclist = &p->aNode[p->iOff]; |
| 3720 p->iOff += p->nDoclist; |
| 3721 } |
| 3722 } |
| 3723 } |
| 3724 |
| 3725 assert( p->iOff<=p->nNode ); |
| 3726 |
| 3727 return rc; |
| 3728 } |
| 3729 |
| 3730 /* |
| 3731 ** Release all dynamic resources held by node-reader object *p. |
| 3732 */ |
| 3733 static void nodeReaderRelease(NodeReader *p){ |
| 3734 sqlite3_free(p->term.a); |
| 3735 } |
| 3736 |
| 3737 /* |
| 3738 ** Initialize a node-reader object to read the node in buffer aNode/nNode. |
| 3739 ** |
| 3740 ** If successful, SQLITE_OK is returned and the NodeReader object set to |
| 3741 ** point to the first entry on the node (if any). Otherwise, an SQLite |
| 3742 ** error code is returned. |
| 3743 */ |
| 3744 static int nodeReaderInit(NodeReader *p, const char *aNode, int nNode){ |
| 3745 memset(p, 0, sizeof(NodeReader)); |
| 3746 p->aNode = aNode; |
| 3747 p->nNode = nNode; |
| 3748 |
| 3749 /* Figure out if this is a leaf or an internal node. */ |
| 3750 if( p->aNode[0] ){ |
| 3751 /* An internal node. */ |
| 3752 p->iOff = 1 + sqlite3Fts3GetVarint(&p->aNode[1], &p->iChild); |
| 3753 }else{ |
| 3754 p->iOff = 1; |
| 3755 } |
| 3756 |
| 3757 return nodeReaderNext(p); |
| 3758 } |
| 3759 |
| 3760 /* |
| 3761 ** This function is called while writing an FTS segment each time a leaf o |
| 3762 ** node is finished and written to disk. The key (zTerm/nTerm) is guaranteed |
| 3763 ** to be greater than the largest key on the node just written, but smaller |
| 3764 ** than or equal to the first key that will be written to the next leaf |
| 3765 ** node. |
| 3766 ** |
| 3767 ** The block id of the leaf node just written to disk may be found in |
| 3768 ** (pWriter->aNodeWriter[0].iBlock) when this function is called. |
| 3769 */ |
| 3770 static int fts3IncrmergePush( |
| 3771 Fts3Table *p, /* Fts3 table handle */ |
| 3772 IncrmergeWriter *pWriter, /* Writer object */ |
| 3773 const char *zTerm, /* Term to write to internal node */ |
| 3774 int nTerm /* Bytes at zTerm */ |
| 3775 ){ |
| 3776 sqlite3_int64 iPtr = pWriter->aNodeWriter[0].iBlock; |
| 3777 int iLayer; |
| 3778 |
| 3779 assert( nTerm>0 ); |
| 3780 for(iLayer=1; ALWAYS(iLayer<FTS_MAX_APPENDABLE_HEIGHT); iLayer++){ |
| 3781 sqlite3_int64 iNextPtr = 0; |
| 3782 NodeWriter *pNode = &pWriter->aNodeWriter[iLayer]; |
| 3783 int rc = SQLITE_OK; |
| 3784 int nPrefix; |
| 3785 int nSuffix; |
| 3786 int nSpace; |
| 3787 |
| 3788 /* Figure out how much space the key will consume if it is written to |
| 3789 ** the current node of layer iLayer. Due to the prefix compression, |
| 3790 ** the space required changes depending on which node the key is to |
| 3791 ** be added to. */ |
| 3792 nPrefix = fts3PrefixCompress(pNode->key.a, pNode->key.n, zTerm, nTerm); |
| 3793 nSuffix = nTerm - nPrefix; |
| 3794 nSpace = sqlite3Fts3VarintLen(nPrefix); |
| 3795 nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; |
| 3796 |
| 3797 if( pNode->key.n==0 || (pNode->block.n + nSpace)<=p->nNodeSize ){ |
| 3798 /* If the current node of layer iLayer contains zero keys, or if adding |
| 3799 ** the key to it will not cause it to grow to larger than nNodeSize |
| 3800 ** bytes in size, write the key here. */ |
| 3801 |
| 3802 Blob *pBlk = &pNode->block; |
| 3803 if( pBlk->n==0 ){ |
| 3804 blobGrowBuffer(pBlk, p->nNodeSize, &rc); |
| 3805 if( rc==SQLITE_OK ){ |
| 3806 pBlk->a[0] = (char)iLayer; |
| 3807 pBlk->n = 1 + sqlite3Fts3PutVarint(&pBlk->a[1], iPtr); |
| 3808 } |
| 3809 } |
| 3810 blobGrowBuffer(pBlk, pBlk->n + nSpace, &rc); |
| 3811 blobGrowBuffer(&pNode->key, nTerm, &rc); |
| 3812 |
| 3813 if( rc==SQLITE_OK ){ |
| 3814 if( pNode->key.n ){ |
| 3815 pBlk->n += sqlite3Fts3PutVarint(&pBlk->a[pBlk->n], nPrefix); |
| 3816 } |
| 3817 pBlk->n += sqlite3Fts3PutVarint(&pBlk->a[pBlk->n], nSuffix); |
| 3818 memcpy(&pBlk->a[pBlk->n], &zTerm[nPrefix], nSuffix); |
| 3819 pBlk->n += nSuffix; |
| 3820 |
| 3821 memcpy(pNode->key.a, zTerm, nTerm); |
| 3822 pNode->key.n = nTerm; |
| 3823 } |
| 3824 }else{ |
| 3825 /* Otherwise, flush the current node of layer iLayer to disk. |
| 3826 ** Then allocate a new, empty sibling node. The key will be written |
| 3827 ** into the parent of this node. */ |
| 3828 rc = fts3WriteSegment(p, pNode->iBlock, pNode->block.a, pNode->block.n); |
| 3829 |
| 3830 assert( pNode->block.nAlloc>=p->nNodeSize ); |
| 3831 pNode->block.a[0] = (char)iLayer; |
| 3832 pNode->block.n = 1 + sqlite3Fts3PutVarint(&pNode->block.a[1], iPtr+1); |
| 3833 |
| 3834 iNextPtr = pNode->iBlock; |
| 3835 pNode->iBlock++; |
| 3836 pNode->key.n = 0; |
| 3837 } |
| 3838 |
| 3839 if( rc!=SQLITE_OK || iNextPtr==0 ) return rc; |
| 3840 iPtr = iNextPtr; |
| 3841 } |
| 3842 |
| 3843 assert( 0 ); |
| 3844 return 0; |
| 3845 } |
| 3846 |
| 3847 /* |
| 3848 ** Append a term and (optionally) doclist to the FTS segment node currently |
| 3849 ** stored in blob *pNode. The node need not contain any terms, but the |
| 3850 ** header must be written before this function is called. |
| 3851 ** |
| 3852 ** A node header is a single 0x00 byte for a leaf node, or a height varint |
| 3853 ** followed by the left-hand-child varint for an internal node. |
| 3854 ** |
| 3855 ** The term to be appended is passed via arguments zTerm/nTerm. For a |
| 3856 ** leaf node, the doclist is passed as aDoclist/nDoclist. For an internal |
| 3857 ** node, both aDoclist and nDoclist must be passed 0. |
| 3858 ** |
| 3859 ** If the size of the value in blob pPrev is zero, then this is the first |
| 3860 ** term written to the node. Otherwise, pPrev contains a copy of the |
| 3861 ** previous term. Before this function returns, it is updated to contain a |
| 3862 ** copy of zTerm/nTerm. |
| 3863 ** |
| 3864 ** It is assumed that the buffer associated with pNode is already large |
| 3865 ** enough to accommodate the new entry. The buffer associated with pPrev |
| 3866 ** is extended by this function if requrired. |
| 3867 ** |
| 3868 ** If an error (i.e. OOM condition) occurs, an SQLite error code is |
| 3869 ** returned. Otherwise, SQLITE_OK. |
| 3870 */ |
| 3871 static int fts3AppendToNode( |
| 3872 Blob *pNode, /* Current node image to append to */ |
| 3873 Blob *pPrev, /* Buffer containing previous term written */ |
| 3874 const char *zTerm, /* New term to write */ |
| 3875 int nTerm, /* Size of zTerm in bytes */ |
| 3876 const char *aDoclist, /* Doclist (or NULL) to write */ |
| 3877 int nDoclist /* Size of aDoclist in bytes */ |
| 3878 ){ |
| 3879 int rc = SQLITE_OK; /* Return code */ |
| 3880 int bFirst = (pPrev->n==0); /* True if this is the first term written */ |
| 3881 int nPrefix; /* Size of term prefix in bytes */ |
| 3882 int nSuffix; /* Size of term suffix in bytes */ |
| 3883 |
| 3884 /* Node must have already been started. There must be a doclist for a |
| 3885 ** leaf node, and there must not be a doclist for an internal node. */ |
| 3886 assert( pNode->n>0 ); |
| 3887 assert( (pNode->a[0]=='\0')==(aDoclist!=0) ); |
| 3888 |
| 3889 blobGrowBuffer(pPrev, nTerm, &rc); |
| 3890 if( rc!=SQLITE_OK ) return rc; |
| 3891 |
| 3892 nPrefix = fts3PrefixCompress(pPrev->a, pPrev->n, zTerm, nTerm); |
| 3893 nSuffix = nTerm - nPrefix; |
| 3894 memcpy(pPrev->a, zTerm, nTerm); |
| 3895 pPrev->n = nTerm; |
| 3896 |
| 3897 if( bFirst==0 ){ |
| 3898 pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nPrefix); |
| 3899 } |
| 3900 pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nSuffix); |
| 3901 memcpy(&pNode->a[pNode->n], &zTerm[nPrefix], nSuffix); |
| 3902 pNode->n += nSuffix; |
| 3903 |
| 3904 if( aDoclist ){ |
| 3905 pNode->n += sqlite3Fts3PutVarint(&pNode->a[pNode->n], nDoclist); |
| 3906 memcpy(&pNode->a[pNode->n], aDoclist, nDoclist); |
| 3907 pNode->n += nDoclist; |
| 3908 } |
| 3909 |
| 3910 assert( pNode->n<=pNode->nAlloc ); |
| 3911 |
| 3912 return SQLITE_OK; |
| 3913 } |
| 3914 |
| 3915 /* |
| 3916 ** Append the current term and doclist pointed to by cursor pCsr to the |
| 3917 ** appendable b-tree segment opened for writing by pWriter. |
| 3918 ** |
| 3919 ** Return SQLITE_OK if successful, or an SQLite error code otherwise. |
| 3920 */ |
| 3921 static int fts3IncrmergeAppend( |
| 3922 Fts3Table *p, /* Fts3 table handle */ |
| 3923 IncrmergeWriter *pWriter, /* Writer object */ |
| 3924 Fts3MultiSegReader *pCsr /* Cursor containing term and doclist */ |
| 3925 ){ |
| 3926 const char *zTerm = pCsr->zTerm; |
| 3927 int nTerm = pCsr->nTerm; |
| 3928 const char *aDoclist = pCsr->aDoclist; |
| 3929 int nDoclist = pCsr->nDoclist; |
| 3930 int rc = SQLITE_OK; /* Return code */ |
| 3931 int nSpace; /* Total space in bytes required on leaf */ |
| 3932 int nPrefix; /* Size of prefix shared with previous term */ |
| 3933 int nSuffix; /* Size of suffix (nTerm - nPrefix) */ |
| 3934 NodeWriter *pLeaf; /* Object used to write leaf nodes */ |
| 3935 |
| 3936 pLeaf = &pWriter->aNodeWriter[0]; |
| 3937 nPrefix = fts3PrefixCompress(pLeaf->key.a, pLeaf->key.n, zTerm, nTerm); |
| 3938 nSuffix = nTerm - nPrefix; |
| 3939 |
| 3940 nSpace = sqlite3Fts3VarintLen(nPrefix); |
| 3941 nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; |
| 3942 nSpace += sqlite3Fts3VarintLen(nDoclist) + nDoclist; |
| 3943 |
| 3944 /* If the current block is not empty, and if adding this term/doclist |
| 3945 ** to the current block would make it larger than Fts3Table.nNodeSize |
| 3946 ** bytes, write this block out to the database. */ |
| 3947 if( pLeaf->block.n>0 && (pLeaf->block.n + nSpace)>p->nNodeSize ){ |
| 3948 rc = fts3WriteSegment(p, pLeaf->iBlock, pLeaf->block.a, pLeaf->block.n); |
| 3949 pWriter->nWork++; |
| 3950 |
| 3951 /* Add the current term to the parent node. The term added to the |
| 3952 ** parent must: |
| 3953 ** |
| 3954 ** a) be greater than the largest term on the leaf node just written |
| 3955 ** to the database (still available in pLeaf->key), and |
| 3956 ** |
| 3957 ** b) be less than or equal to the term about to be added to the new |
| 3958 ** leaf node (zTerm/nTerm). |
| 3959 ** |
| 3960 ** In other words, it must be the prefix of zTerm 1 byte longer than |
| 3961 ** the common prefix (if any) of zTerm and pWriter->zTerm. |
| 3962 */ |
| 3963 if( rc==SQLITE_OK ){ |
| 3964 rc = fts3IncrmergePush(p, pWriter, zTerm, nPrefix+1); |
| 3965 } |
| 3966 |
| 3967 /* Advance to the next output block */ |
| 3968 pLeaf->iBlock++; |
| 3969 pLeaf->key.n = 0; |
| 3970 pLeaf->block.n = 0; |
| 3971 |
| 3972 nSuffix = nTerm; |
| 3973 nSpace = 1; |
| 3974 nSpace += sqlite3Fts3VarintLen(nSuffix) + nSuffix; |
| 3975 nSpace += sqlite3Fts3VarintLen(nDoclist) + nDoclist; |
| 3976 } |
| 3977 |
| 3978 pWriter->nLeafData += nSpace; |
| 3979 blobGrowBuffer(&pLeaf->block, pLeaf->block.n + nSpace, &rc); |
| 3980 if( rc==SQLITE_OK ){ |
| 3981 if( pLeaf->block.n==0 ){ |
| 3982 pLeaf->block.n = 1; |
| 3983 pLeaf->block.a[0] = '\0'; |
| 3984 } |
| 3985 rc = fts3AppendToNode( |
| 3986 &pLeaf->block, &pLeaf->key, zTerm, nTerm, aDoclist, nDoclist |
| 3987 ); |
| 3988 } |
| 3989 |
| 3990 return rc; |
| 3991 } |
| 3992 |
| 3993 /* |
| 3994 ** This function is called to release all dynamic resources held by the |
| 3995 ** merge-writer object pWriter, and if no error has occurred, to flush |
| 3996 ** all outstanding node buffers held by pWriter to disk. |
| 3997 ** |
| 3998 ** If *pRc is not SQLITE_OK when this function is called, then no attempt |
| 3999 ** is made to write any data to disk. Instead, this function serves only |
| 4000 ** to release outstanding resources. |
| 4001 ** |
| 4002 ** Otherwise, if *pRc is initially SQLITE_OK and an error occurs while |
| 4003 ** flushing buffers to disk, *pRc is set to an SQLite error code before |
| 4004 ** returning. |
| 4005 */ |
| 4006 static void fts3IncrmergeRelease( |
| 4007 Fts3Table *p, /* FTS3 table handle */ |
| 4008 IncrmergeWriter *pWriter, /* Merge-writer object */ |
| 4009 int *pRc /* IN/OUT: Error code */ |
| 4010 ){ |
| 4011 int i; /* Used to iterate through non-root layers */ |
| 4012 int iRoot; /* Index of root in pWriter->aNodeWriter */ |
| 4013 NodeWriter *pRoot; /* NodeWriter for root node */ |
| 4014 int rc = *pRc; /* Error code */ |
| 4015 |
| 4016 /* Set iRoot to the index in pWriter->aNodeWriter[] of the output segment |
| 4017 ** root node. If the segment fits entirely on a single leaf node, iRoot |
| 4018 ** will be set to 0. If the root node is the parent of the leaves, iRoot |
| 4019 ** will be 1. And so on. */ |
| 4020 for(iRoot=FTS_MAX_APPENDABLE_HEIGHT-1; iRoot>=0; iRoot--){ |
| 4021 NodeWriter *pNode = &pWriter->aNodeWriter[iRoot]; |
| 4022 if( pNode->block.n>0 ) break; |
| 4023 assert( *pRc || pNode->block.nAlloc==0 ); |
| 4024 assert( *pRc || pNode->key.nAlloc==0 ); |
| 4025 sqlite3_free(pNode->block.a); |
| 4026 sqlite3_free(pNode->key.a); |
| 4027 } |
| 4028 |
| 4029 /* Empty output segment. This is a no-op. */ |
| 4030 if( iRoot<0 ) return; |
| 4031 |
| 4032 /* The entire output segment fits on a single node. Normally, this means |
| 4033 ** the node would be stored as a blob in the "root" column of the %_segdir |
| 4034 ** table. However, this is not permitted in this case. The problem is that |
| 4035 ** space has already been reserved in the %_segments table, and so the |
| 4036 ** start_block and end_block fields of the %_segdir table must be populated. |
| 4037 ** And, by design or by accident, released versions of FTS cannot handle |
| 4038 ** segments that fit entirely on the root node with start_block!=0. |
| 4039 ** |
| 4040 ** Instead, create a synthetic root node that contains nothing but a |
| 4041 ** pointer to the single content node. So that the segment consists of a |
| 4042 ** single leaf and a single interior (root) node. |
| 4043 ** |
| 4044 ** Todo: Better might be to defer allocating space in the %_segments |
| 4045 ** table until we are sure it is needed. |
| 4046 */ |
| 4047 if( iRoot==0 ){ |
| 4048 Blob *pBlock = &pWriter->aNodeWriter[1].block; |
| 4049 blobGrowBuffer(pBlock, 1 + FTS3_VARINT_MAX, &rc); |
| 4050 if( rc==SQLITE_OK ){ |
| 4051 pBlock->a[0] = 0x01; |
| 4052 pBlock->n = 1 + sqlite3Fts3PutVarint( |
| 4053 &pBlock->a[1], pWriter->aNodeWriter[0].iBlock |
| 4054 ); |
| 4055 } |
| 4056 iRoot = 1; |
| 4057 } |
| 4058 pRoot = &pWriter->aNodeWriter[iRoot]; |
| 4059 |
| 4060 /* Flush all currently outstanding nodes to disk. */ |
| 4061 for(i=0; i<iRoot; i++){ |
| 4062 NodeWriter *pNode = &pWriter->aNodeWriter[i]; |
| 4063 if( pNode->block.n>0 && rc==SQLITE_OK ){ |
| 4064 rc = fts3WriteSegment(p, pNode->iBlock, pNode->block.a, pNode->block.n); |
| 4065 } |
| 4066 sqlite3_free(pNode->block.a); |
| 4067 sqlite3_free(pNode->key.a); |
| 4068 } |
| 4069 |
| 4070 /* Write the %_segdir record. */ |
| 4071 if( rc==SQLITE_OK ){ |
| 4072 rc = fts3WriteSegdir(p, |
| 4073 pWriter->iAbsLevel+1, /* level */ |
| 4074 pWriter->iIdx, /* idx */ |
| 4075 pWriter->iStart, /* start_block */ |
| 4076 pWriter->aNodeWriter[0].iBlock, /* leaves_end_block */ |
| 4077 pWriter->iEnd, /* end_block */ |
| 4078 (pWriter->bNoLeafData==0 ? pWriter->nLeafData : 0), /* end_block */ |
| 4079 pRoot->block.a, pRoot->block.n /* root */ |
| 4080 ); |
| 4081 } |
| 4082 sqlite3_free(pRoot->block.a); |
| 4083 sqlite3_free(pRoot->key.a); |
| 4084 |
| 4085 *pRc = rc; |
| 4086 } |
| 4087 |
| 4088 /* |
| 4089 ** Compare the term in buffer zLhs (size in bytes nLhs) with that in |
| 4090 ** zRhs (size in bytes nRhs) using memcmp. If one term is a prefix of |
| 4091 ** the other, it is considered to be smaller than the other. |
| 4092 ** |
| 4093 ** Return -ve if zLhs is smaller than zRhs, 0 if it is equal, or +ve |
| 4094 ** if it is greater. |
| 4095 */ |
| 4096 static int fts3TermCmp( |
| 4097 const char *zLhs, int nLhs, /* LHS of comparison */ |
| 4098 const char *zRhs, int nRhs /* RHS of comparison */ |
| 4099 ){ |
| 4100 int nCmp = MIN(nLhs, nRhs); |
| 4101 int res; |
| 4102 |
| 4103 res = memcmp(zLhs, zRhs, nCmp); |
| 4104 if( res==0 ) res = nLhs - nRhs; |
| 4105 |
| 4106 return res; |
| 4107 } |
| 4108 |
| 4109 |
| 4110 /* |
| 4111 ** Query to see if the entry in the %_segments table with blockid iEnd is |
| 4112 ** NULL. If no error occurs and the entry is NULL, set *pbRes 1 before |
| 4113 ** returning. Otherwise, set *pbRes to 0. |
| 4114 ** |
| 4115 ** Or, if an error occurs while querying the database, return an SQLite |
| 4116 ** error code. The final value of *pbRes is undefined in this case. |
| 4117 ** |
| 4118 ** This is used to test if a segment is an "appendable" segment. If it |
| 4119 ** is, then a NULL entry has been inserted into the %_segments table |
| 4120 ** with blockid %_segdir.end_block. |
| 4121 */ |
| 4122 static int fts3IsAppendable(Fts3Table *p, sqlite3_int64 iEnd, int *pbRes){ |
| 4123 int bRes = 0; /* Result to set *pbRes to */ |
| 4124 sqlite3_stmt *pCheck = 0; /* Statement to query database with */ |
| 4125 int rc; /* Return code */ |
| 4126 |
| 4127 rc = fts3SqlStmt(p, SQL_SEGMENT_IS_APPENDABLE, &pCheck, 0); |
| 4128 if( rc==SQLITE_OK ){ |
| 4129 sqlite3_bind_int64(pCheck, 1, iEnd); |
| 4130 if( SQLITE_ROW==sqlite3_step(pCheck) ) bRes = 1; |
| 4131 rc = sqlite3_reset(pCheck); |
| 4132 } |
| 4133 |
| 4134 *pbRes = bRes; |
| 4135 return rc; |
| 4136 } |
| 4137 |
| 4138 /* |
| 4139 ** This function is called when initializing an incremental-merge operation. |
| 4140 ** It checks if the existing segment with index value iIdx at absolute level |
| 4141 ** (iAbsLevel+1) can be appended to by the incremental merge. If it can, the |
| 4142 ** merge-writer object *pWriter is initialized to write to it. |
| 4143 ** |
| 4144 ** An existing segment can be appended to by an incremental merge if: |
| 4145 ** |
| 4146 ** * It was initially created as an appendable segment (with all required |
| 4147 ** space pre-allocated), and |
| 4148 ** |
| 4149 ** * The first key read from the input (arguments zKey and nKey) is |
| 4150 ** greater than the largest key currently stored in the potential |
| 4151 ** output segment. |
| 4152 */ |
| 4153 static int fts3IncrmergeLoad( |
| 4154 Fts3Table *p, /* Fts3 table handle */ |
| 4155 sqlite3_int64 iAbsLevel, /* Absolute level of input segments */ |
| 4156 int iIdx, /* Index of candidate output segment */ |
| 4157 const char *zKey, /* First key to write */ |
| 4158 int nKey, /* Number of bytes in nKey */ |
| 4159 IncrmergeWriter *pWriter /* Populate this object */ |
| 4160 ){ |
| 4161 int rc; /* Return code */ |
| 4162 sqlite3_stmt *pSelect = 0; /* SELECT to read %_segdir entry */ |
| 4163 |
| 4164 rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR, &pSelect, 0); |
| 4165 if( rc==SQLITE_OK ){ |
| 4166 sqlite3_int64 iStart = 0; /* Value of %_segdir.start_block */ |
| 4167 sqlite3_int64 iLeafEnd = 0; /* Value of %_segdir.leaves_end_block */ |
| 4168 sqlite3_int64 iEnd = 0; /* Value of %_segdir.end_block */ |
| 4169 const char *aRoot = 0; /* Pointer to %_segdir.root buffer */ |
| 4170 int nRoot = 0; /* Size of aRoot[] in bytes */ |
| 4171 int rc2; /* Return code from sqlite3_reset() */ |
| 4172 int bAppendable = 0; /* Set to true if segment is appendable */ |
| 4173 |
| 4174 /* Read the %_segdir entry for index iIdx absolute level (iAbsLevel+1) */ |
| 4175 sqlite3_bind_int64(pSelect, 1, iAbsLevel+1); |
| 4176 sqlite3_bind_int(pSelect, 2, iIdx); |
| 4177 if( sqlite3_step(pSelect)==SQLITE_ROW ){ |
| 4178 iStart = sqlite3_column_int64(pSelect, 1); |
| 4179 iLeafEnd = sqlite3_column_int64(pSelect, 2); |
| 4180 fts3ReadEndBlockField(pSelect, 3, &iEnd, &pWriter->nLeafData); |
| 4181 if( pWriter->nLeafData<0 ){ |
| 4182 pWriter->nLeafData = pWriter->nLeafData * -1; |
| 4183 } |
| 4184 pWriter->bNoLeafData = (pWriter->nLeafData==0); |
| 4185 nRoot = sqlite3_column_bytes(pSelect, 4); |
| 4186 aRoot = sqlite3_column_blob(pSelect, 4); |
| 4187 }else{ |
| 4188 return sqlite3_reset(pSelect); |
| 4189 } |
| 4190 |
| 4191 /* Check for the zero-length marker in the %_segments table */ |
| 4192 rc = fts3IsAppendable(p, iEnd, &bAppendable); |
| 4193 |
| 4194 /* Check that zKey/nKey is larger than the largest key the candidate */ |
| 4195 if( rc==SQLITE_OK && bAppendable ){ |
| 4196 char *aLeaf = 0; |
| 4197 int nLeaf = 0; |
| 4198 |
| 4199 rc = sqlite3Fts3ReadBlock(p, iLeafEnd, &aLeaf, &nLeaf, 0); |
| 4200 if( rc==SQLITE_OK ){ |
| 4201 NodeReader reader; |
| 4202 for(rc = nodeReaderInit(&reader, aLeaf, nLeaf); |
| 4203 rc==SQLITE_OK && reader.aNode; |
| 4204 rc = nodeReaderNext(&reader) |
| 4205 ){ |
| 4206 assert( reader.aNode ); |
| 4207 } |
| 4208 if( fts3TermCmp(zKey, nKey, reader.term.a, reader.term.n)<=0 ){ |
| 4209 bAppendable = 0; |
| 4210 } |
| 4211 nodeReaderRelease(&reader); |
| 4212 } |
| 4213 sqlite3_free(aLeaf); |
| 4214 } |
| 4215 |
| 4216 if( rc==SQLITE_OK && bAppendable ){ |
| 4217 /* It is possible to append to this segment. Set up the IncrmergeWriter |
| 4218 ** object to do so. */ |
| 4219 int i; |
| 4220 int nHeight = (int)aRoot[0]; |
| 4221 NodeWriter *pNode; |
| 4222 |
| 4223 pWriter->nLeafEst = (int)((iEnd - iStart) + 1)/FTS_MAX_APPENDABLE_HEIGHT; |
| 4224 pWriter->iStart = iStart; |
| 4225 pWriter->iEnd = iEnd; |
| 4226 pWriter->iAbsLevel = iAbsLevel; |
| 4227 pWriter->iIdx = iIdx; |
| 4228 |
| 4229 for(i=nHeight+1; i<FTS_MAX_APPENDABLE_HEIGHT; i++){ |
| 4230 pWriter->aNodeWriter[i].iBlock = pWriter->iStart + i*pWriter->nLeafEst; |
| 4231 } |
| 4232 |
| 4233 pNode = &pWriter->aNodeWriter[nHeight]; |
| 4234 pNode->iBlock = pWriter->iStart + pWriter->nLeafEst*nHeight; |
| 4235 blobGrowBuffer(&pNode->block, MAX(nRoot, p->nNodeSize), &rc); |
| 4236 if( rc==SQLITE_OK ){ |
| 4237 memcpy(pNode->block.a, aRoot, nRoot); |
| 4238 pNode->block.n = nRoot; |
| 4239 } |
| 4240 |
| 4241 for(i=nHeight; i>=0 && rc==SQLITE_OK; i--){ |
| 4242 NodeReader reader; |
| 4243 pNode = &pWriter->aNodeWriter[i]; |
| 4244 |
| 4245 rc = nodeReaderInit(&reader, pNode->block.a, pNode->block.n); |
| 4246 while( reader.aNode && rc==SQLITE_OK ) rc = nodeReaderNext(&reader); |
| 4247 blobGrowBuffer(&pNode->key, reader.term.n, &rc); |
| 4248 if( rc==SQLITE_OK ){ |
| 4249 memcpy(pNode->key.a, reader.term.a, reader.term.n); |
| 4250 pNode->key.n = reader.term.n; |
| 4251 if( i>0 ){ |
| 4252 char *aBlock = 0; |
| 4253 int nBlock = 0; |
| 4254 pNode = &pWriter->aNodeWriter[i-1]; |
| 4255 pNode->iBlock = reader.iChild; |
| 4256 rc = sqlite3Fts3ReadBlock(p, reader.iChild, &aBlock, &nBlock, 0); |
| 4257 blobGrowBuffer(&pNode->block, MAX(nBlock, p->nNodeSize), &rc); |
| 4258 if( rc==SQLITE_OK ){ |
| 4259 memcpy(pNode->block.a, aBlock, nBlock); |
| 4260 pNode->block.n = nBlock; |
| 4261 } |
| 4262 sqlite3_free(aBlock); |
| 4263 } |
| 4264 } |
| 4265 nodeReaderRelease(&reader); |
| 4266 } |
| 4267 } |
| 4268 |
| 4269 rc2 = sqlite3_reset(pSelect); |
| 4270 if( rc==SQLITE_OK ) rc = rc2; |
| 4271 } |
| 4272 |
| 4273 return rc; |
| 4274 } |
| 4275 |
| 4276 /* |
| 4277 ** Determine the largest segment index value that exists within absolute |
| 4278 ** level iAbsLevel+1. If no error occurs, set *piIdx to this value plus |
| 4279 ** one before returning SQLITE_OK. Or, if there are no segments at all |
| 4280 ** within level iAbsLevel, set *piIdx to zero. |
| 4281 ** |
| 4282 ** If an error occurs, return an SQLite error code. The final value of |
| 4283 ** *piIdx is undefined in this case. |
| 4284 */ |
| 4285 static int fts3IncrmergeOutputIdx( |
| 4286 Fts3Table *p, /* FTS Table handle */ |
| 4287 sqlite3_int64 iAbsLevel, /* Absolute index of input segments */ |
| 4288 int *piIdx /* OUT: Next free index at iAbsLevel+1 */ |
| 4289 ){ |
| 4290 int rc; |
| 4291 sqlite3_stmt *pOutputIdx = 0; /* SQL used to find output index */ |
| 4292 |
| 4293 rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pOutputIdx, 0); |
| 4294 if( rc==SQLITE_OK ){ |
| 4295 sqlite3_bind_int64(pOutputIdx, 1, iAbsLevel+1); |
| 4296 sqlite3_step(pOutputIdx); |
| 4297 *piIdx = sqlite3_column_int(pOutputIdx, 0); |
| 4298 rc = sqlite3_reset(pOutputIdx); |
| 4299 } |
| 4300 |
| 4301 return rc; |
| 4302 } |
| 4303 |
| 4304 /* |
| 4305 ** Allocate an appendable output segment on absolute level iAbsLevel+1 |
| 4306 ** with idx value iIdx. |
| 4307 ** |
| 4308 ** In the %_segdir table, a segment is defined by the values in three |
| 4309 ** columns: |
| 4310 ** |
| 4311 ** start_block |
| 4312 ** leaves_end_block |
| 4313 ** end_block |
| 4314 ** |
| 4315 ** When an appendable segment is allocated, it is estimated that the |
| 4316 ** maximum number of leaf blocks that may be required is the sum of the |
| 4317 ** number of leaf blocks consumed by the input segments, plus the number |
| 4318 ** of input segments, multiplied by two. This value is stored in stack |
| 4319 ** variable nLeafEst. |
| 4320 ** |
| 4321 ** A total of 16*nLeafEst blocks are allocated when an appendable segment |
| 4322 ** is created ((1 + end_block - start_block)==16*nLeafEst). The contiguous |
| 4323 ** array of leaf nodes starts at the first block allocated. The array |
| 4324 ** of interior nodes that are parents of the leaf nodes start at block |
| 4325 ** (start_block + (1 + end_block - start_block) / 16). And so on. |
| 4326 ** |
| 4327 ** In the actual code below, the value "16" is replaced with the |
| 4328 ** pre-processor macro FTS_MAX_APPENDABLE_HEIGHT. |
| 4329 */ |
| 4330 static int fts3IncrmergeWriter( |
| 4331 Fts3Table *p, /* Fts3 table handle */ |
| 4332 sqlite3_int64 iAbsLevel, /* Absolute level of input segments */ |
| 4333 int iIdx, /* Index of new output segment */ |
| 4334 Fts3MultiSegReader *pCsr, /* Cursor that data will be read from */ |
| 4335 IncrmergeWriter *pWriter /* Populate this object */ |
| 4336 ){ |
| 4337 int rc; /* Return Code */ |
| 4338 int i; /* Iterator variable */ |
| 4339 int nLeafEst = 0; /* Blocks allocated for leaf nodes */ |
| 4340 sqlite3_stmt *pLeafEst = 0; /* SQL used to determine nLeafEst */ |
| 4341 sqlite3_stmt *pFirstBlock = 0; /* SQL used to determine first block */ |
| 4342 |
| 4343 /* Calculate nLeafEst. */ |
| 4344 rc = fts3SqlStmt(p, SQL_MAX_LEAF_NODE_ESTIMATE, &pLeafEst, 0); |
| 4345 if( rc==SQLITE_OK ){ |
| 4346 sqlite3_bind_int64(pLeafEst, 1, iAbsLevel); |
| 4347 sqlite3_bind_int64(pLeafEst, 2, pCsr->nSegment); |
| 4348 if( SQLITE_ROW==sqlite3_step(pLeafEst) ){ |
| 4349 nLeafEst = sqlite3_column_int(pLeafEst, 0); |
| 4350 } |
| 4351 rc = sqlite3_reset(pLeafEst); |
| 4352 } |
| 4353 if( rc!=SQLITE_OK ) return rc; |
| 4354 |
| 4355 /* Calculate the first block to use in the output segment */ |
| 4356 rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pFirstBlock, 0); |
| 4357 if( rc==SQLITE_OK ){ |
| 4358 if( SQLITE_ROW==sqlite3_step(pFirstBlock) ){ |
| 4359 pWriter->iStart = sqlite3_column_int64(pFirstBlock, 0); |
| 4360 pWriter->iEnd = pWriter->iStart - 1; |
| 4361 pWriter->iEnd += nLeafEst * FTS_MAX_APPENDABLE_HEIGHT; |
| 4362 } |
| 4363 rc = sqlite3_reset(pFirstBlock); |
| 4364 } |
| 4365 if( rc!=SQLITE_OK ) return rc; |
| 4366 |
| 4367 /* Insert the marker in the %_segments table to make sure nobody tries |
| 4368 ** to steal the space just allocated. This is also used to identify |
| 4369 ** appendable segments. */ |
| 4370 rc = fts3WriteSegment(p, pWriter->iEnd, 0, 0); |
| 4371 if( rc!=SQLITE_OK ) return rc; |
| 4372 |
| 4373 pWriter->iAbsLevel = iAbsLevel; |
| 4374 pWriter->nLeafEst = nLeafEst; |
| 4375 pWriter->iIdx = iIdx; |
| 4376 |
| 4377 /* Set up the array of NodeWriter objects */ |
| 4378 for(i=0; i<FTS_MAX_APPENDABLE_HEIGHT; i++){ |
| 4379 pWriter->aNodeWriter[i].iBlock = pWriter->iStart + i*pWriter->nLeafEst; |
| 4380 } |
| 4381 return SQLITE_OK; |
| 4382 } |
| 4383 |
| 4384 /* |
| 4385 ** Remove an entry from the %_segdir table. This involves running the |
| 4386 ** following two statements: |
| 4387 ** |
| 4388 ** DELETE FROM %_segdir WHERE level = :iAbsLevel AND idx = :iIdx |
| 4389 ** UPDATE %_segdir SET idx = idx - 1 WHERE level = :iAbsLevel AND idx > :iIdx |
| 4390 ** |
| 4391 ** The DELETE statement removes the specific %_segdir level. The UPDATE |
| 4392 ** statement ensures that the remaining segments have contiguously allocated |
| 4393 ** idx values. |
| 4394 */ |
| 4395 static int fts3RemoveSegdirEntry( |
| 4396 Fts3Table *p, /* FTS3 table handle */ |
| 4397 sqlite3_int64 iAbsLevel, /* Absolute level to delete from */ |
| 4398 int iIdx /* Index of %_segdir entry to delete */ |
| 4399 ){ |
| 4400 int rc; /* Return code */ |
| 4401 sqlite3_stmt *pDelete = 0; /* DELETE statement */ |
| 4402 |
| 4403 rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_ENTRY, &pDelete, 0); |
| 4404 if( rc==SQLITE_OK ){ |
| 4405 sqlite3_bind_int64(pDelete, 1, iAbsLevel); |
| 4406 sqlite3_bind_int(pDelete, 2, iIdx); |
| 4407 sqlite3_step(pDelete); |
| 4408 rc = sqlite3_reset(pDelete); |
| 4409 } |
| 4410 |
| 4411 return rc; |
| 4412 } |
| 4413 |
| 4414 /* |
| 4415 ** One or more segments have just been removed from absolute level iAbsLevel. |
| 4416 ** Update the 'idx' values of the remaining segments in the level so that |
| 4417 ** the idx values are a contiguous sequence starting from 0. |
| 4418 */ |
| 4419 static int fts3RepackSegdirLevel( |
| 4420 Fts3Table *p, /* FTS3 table handle */ |
| 4421 sqlite3_int64 iAbsLevel /* Absolute level to repack */ |
| 4422 ){ |
| 4423 int rc; /* Return code */ |
| 4424 int *aIdx = 0; /* Array of remaining idx values */ |
| 4425 int nIdx = 0; /* Valid entries in aIdx[] */ |
| 4426 int nAlloc = 0; /* Allocated size of aIdx[] */ |
| 4427 int i; /* Iterator variable */ |
| 4428 sqlite3_stmt *pSelect = 0; /* Select statement to read idx values */ |
| 4429 sqlite3_stmt *pUpdate = 0; /* Update statement to modify idx values */ |
| 4430 |
| 4431 rc = fts3SqlStmt(p, SQL_SELECT_INDEXES, &pSelect, 0); |
| 4432 if( rc==SQLITE_OK ){ |
| 4433 int rc2; |
| 4434 sqlite3_bind_int64(pSelect, 1, iAbsLevel); |
| 4435 while( SQLITE_ROW==sqlite3_step(pSelect) ){ |
| 4436 if( nIdx>=nAlloc ){ |
| 4437 int *aNew; |
| 4438 nAlloc += 16; |
| 4439 aNew = sqlite3_realloc(aIdx, nAlloc*sizeof(int)); |
| 4440 if( !aNew ){ |
| 4441 rc = SQLITE_NOMEM; |
| 4442 break; |
| 4443 } |
| 4444 aIdx = aNew; |
| 4445 } |
| 4446 aIdx[nIdx++] = sqlite3_column_int(pSelect, 0); |
| 4447 } |
| 4448 rc2 = sqlite3_reset(pSelect); |
| 4449 if( rc==SQLITE_OK ) rc = rc2; |
| 4450 } |
| 4451 |
| 4452 if( rc==SQLITE_OK ){ |
| 4453 rc = fts3SqlStmt(p, SQL_SHIFT_SEGDIR_ENTRY, &pUpdate, 0); |
| 4454 } |
| 4455 if( rc==SQLITE_OK ){ |
| 4456 sqlite3_bind_int64(pUpdate, 2, iAbsLevel); |
| 4457 } |
| 4458 |
| 4459 assert( p->bIgnoreSavepoint==0 ); |
| 4460 p->bIgnoreSavepoint = 1; |
| 4461 for(i=0; rc==SQLITE_OK && i<nIdx; i++){ |
| 4462 if( aIdx[i]!=i ){ |
| 4463 sqlite3_bind_int(pUpdate, 3, aIdx[i]); |
| 4464 sqlite3_bind_int(pUpdate, 1, i); |
| 4465 sqlite3_step(pUpdate); |
| 4466 rc = sqlite3_reset(pUpdate); |
| 4467 } |
| 4468 } |
| 4469 p->bIgnoreSavepoint = 0; |
| 4470 |
| 4471 sqlite3_free(aIdx); |
| 4472 return rc; |
| 4473 } |
| 4474 |
| 4475 static void fts3StartNode(Blob *pNode, int iHeight, sqlite3_int64 iChild){ |
| 4476 pNode->a[0] = (char)iHeight; |
| 4477 if( iChild ){ |
| 4478 assert( pNode->nAlloc>=1+sqlite3Fts3VarintLen(iChild) ); |
| 4479 pNode->n = 1 + sqlite3Fts3PutVarint(&pNode->a[1], iChild); |
| 4480 }else{ |
| 4481 assert( pNode->nAlloc>=1 ); |
| 4482 pNode->n = 1; |
| 4483 } |
| 4484 } |
| 4485 |
| 4486 /* |
| 4487 ** The first two arguments are a pointer to and the size of a segment b-tree |
| 4488 ** node. The node may be a leaf or an internal node. |
| 4489 ** |
| 4490 ** This function creates a new node image in blob object *pNew by copying |
| 4491 ** all terms that are greater than or equal to zTerm/nTerm (for leaf nodes) |
| 4492 ** or greater than zTerm/nTerm (for internal nodes) from aNode/nNode. |
| 4493 */ |
| 4494 static int fts3TruncateNode( |
| 4495 const char *aNode, /* Current node image */ |
| 4496 int nNode, /* Size of aNode in bytes */ |
| 4497 Blob *pNew, /* OUT: Write new node image here */ |
| 4498 const char *zTerm, /* Omit all terms smaller than this */ |
| 4499 int nTerm, /* Size of zTerm in bytes */ |
| 4500 sqlite3_int64 *piBlock /* OUT: Block number in next layer down */ |
| 4501 ){ |
| 4502 NodeReader reader; /* Reader object */ |
| 4503 Blob prev = {0, 0, 0}; /* Previous term written to new node */ |
| 4504 int rc = SQLITE_OK; /* Return code */ |
| 4505 int bLeaf = aNode[0]=='\0'; /* True for a leaf node */ |
| 4506 |
| 4507 /* Allocate required output space */ |
| 4508 blobGrowBuffer(pNew, nNode, &rc); |
| 4509 if( rc!=SQLITE_OK ) return rc; |
| 4510 pNew->n = 0; |
| 4511 |
| 4512 /* Populate new node buffer */ |
| 4513 for(rc = nodeReaderInit(&reader, aNode, nNode); |
| 4514 rc==SQLITE_OK && reader.aNode; |
| 4515 rc = nodeReaderNext(&reader) |
| 4516 ){ |
| 4517 if( pNew->n==0 ){ |
| 4518 int res = fts3TermCmp(reader.term.a, reader.term.n, zTerm, nTerm); |
| 4519 if( res<0 || (bLeaf==0 && res==0) ) continue; |
| 4520 fts3StartNode(pNew, (int)aNode[0], reader.iChild); |
| 4521 *piBlock = reader.iChild; |
| 4522 } |
| 4523 rc = fts3AppendToNode( |
| 4524 pNew, &prev, reader.term.a, reader.term.n, |
| 4525 reader.aDoclist, reader.nDoclist |
| 4526 ); |
| 4527 if( rc!=SQLITE_OK ) break; |
| 4528 } |
| 4529 if( pNew->n==0 ){ |
| 4530 fts3StartNode(pNew, (int)aNode[0], reader.iChild); |
| 4531 *piBlock = reader.iChild; |
| 4532 } |
| 4533 assert( pNew->n<=pNew->nAlloc ); |
| 4534 |
| 4535 nodeReaderRelease(&reader); |
| 4536 sqlite3_free(prev.a); |
| 4537 return rc; |
| 4538 } |
| 4539 |
| 4540 /* |
| 4541 ** Remove all terms smaller than zTerm/nTerm from segment iIdx in absolute |
| 4542 ** level iAbsLevel. This may involve deleting entries from the %_segments |
| 4543 ** table, and modifying existing entries in both the %_segments and %_segdir |
| 4544 ** tables. |
| 4545 ** |
| 4546 ** SQLITE_OK is returned if the segment is updated successfully. Or an |
| 4547 ** SQLite error code otherwise. |
| 4548 */ |
| 4549 static int fts3TruncateSegment( |
| 4550 Fts3Table *p, /* FTS3 table handle */ |
| 4551 sqlite3_int64 iAbsLevel, /* Absolute level of segment to modify */ |
| 4552 int iIdx, /* Index within level of segment to modify */ |
| 4553 const char *zTerm, /* Remove terms smaller than this */ |
| 4554 int nTerm /* Number of bytes in buffer zTerm */ |
| 4555 ){ |
| 4556 int rc = SQLITE_OK; /* Return code */ |
| 4557 Blob root = {0,0,0}; /* New root page image */ |
| 4558 Blob block = {0,0,0}; /* Buffer used for any other block */ |
| 4559 sqlite3_int64 iBlock = 0; /* Block id */ |
| 4560 sqlite3_int64 iNewStart = 0; /* New value for iStartBlock */ |
| 4561 sqlite3_int64 iOldStart = 0; /* Old value for iStartBlock */ |
| 4562 sqlite3_stmt *pFetch = 0; /* Statement used to fetch segdir */ |
| 4563 |
| 4564 rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR, &pFetch, 0); |
| 4565 if( rc==SQLITE_OK ){ |
| 4566 int rc2; /* sqlite3_reset() return code */ |
| 4567 sqlite3_bind_int64(pFetch, 1, iAbsLevel); |
| 4568 sqlite3_bind_int(pFetch, 2, iIdx); |
| 4569 if( SQLITE_ROW==sqlite3_step(pFetch) ){ |
| 4570 const char *aRoot = sqlite3_column_blob(pFetch, 4); |
| 4571 int nRoot = sqlite3_column_bytes(pFetch, 4); |
| 4572 iOldStart = sqlite3_column_int64(pFetch, 1); |
| 4573 rc = fts3TruncateNode(aRoot, nRoot, &root, zTerm, nTerm, &iBlock); |
| 4574 } |
| 4575 rc2 = sqlite3_reset(pFetch); |
| 4576 if( rc==SQLITE_OK ) rc = rc2; |
| 4577 } |
| 4578 |
| 4579 while( rc==SQLITE_OK && iBlock ){ |
| 4580 char *aBlock = 0; |
| 4581 int nBlock = 0; |
| 4582 iNewStart = iBlock; |
| 4583 |
| 4584 rc = sqlite3Fts3ReadBlock(p, iBlock, &aBlock, &nBlock, 0); |
| 4585 if( rc==SQLITE_OK ){ |
| 4586 rc = fts3TruncateNode(aBlock, nBlock, &block, zTerm, nTerm, &iBlock); |
| 4587 } |
| 4588 if( rc==SQLITE_OK ){ |
| 4589 rc = fts3WriteSegment(p, iNewStart, block.a, block.n); |
| 4590 } |
| 4591 sqlite3_free(aBlock); |
| 4592 } |
| 4593 |
| 4594 /* Variable iNewStart now contains the first valid leaf node. */ |
| 4595 if( rc==SQLITE_OK && iNewStart ){ |
| 4596 sqlite3_stmt *pDel = 0; |
| 4597 rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDel, 0); |
| 4598 if( rc==SQLITE_OK ){ |
| 4599 sqlite3_bind_int64(pDel, 1, iOldStart); |
| 4600 sqlite3_bind_int64(pDel, 2, iNewStart-1); |
| 4601 sqlite3_step(pDel); |
| 4602 rc = sqlite3_reset(pDel); |
| 4603 } |
| 4604 } |
| 4605 |
| 4606 if( rc==SQLITE_OK ){ |
| 4607 sqlite3_stmt *pChomp = 0; |
| 4608 rc = fts3SqlStmt(p, SQL_CHOMP_SEGDIR, &pChomp, 0); |
| 4609 if( rc==SQLITE_OK ){ |
| 4610 sqlite3_bind_int64(pChomp, 1, iNewStart); |
| 4611 sqlite3_bind_blob(pChomp, 2, root.a, root.n, SQLITE_STATIC); |
| 4612 sqlite3_bind_int64(pChomp, 3, iAbsLevel); |
| 4613 sqlite3_bind_int(pChomp, 4, iIdx); |
| 4614 sqlite3_step(pChomp); |
| 4615 rc = sqlite3_reset(pChomp); |
| 4616 } |
| 4617 } |
| 4618 |
| 4619 sqlite3_free(root.a); |
| 4620 sqlite3_free(block.a); |
| 4621 return rc; |
| 4622 } |
| 4623 |
| 4624 /* |
| 4625 ** This function is called after an incrmental-merge operation has run to |
| 4626 ** merge (or partially merge) two or more segments from absolute level |
| 4627 ** iAbsLevel. |
| 4628 ** |
| 4629 ** Each input segment is either removed from the db completely (if all of |
| 4630 ** its data was copied to the output segment by the incrmerge operation) |
| 4631 ** or modified in place so that it no longer contains those entries that |
| 4632 ** have been duplicated in the output segment. |
| 4633 */ |
| 4634 static int fts3IncrmergeChomp( |
| 4635 Fts3Table *p, /* FTS table handle */ |
| 4636 sqlite3_int64 iAbsLevel, /* Absolute level containing segments */ |
| 4637 Fts3MultiSegReader *pCsr, /* Chomp all segments opened by this cursor */ |
| 4638 int *pnRem /* Number of segments not deleted */ |
| 4639 ){ |
| 4640 int i; |
| 4641 int nRem = 0; |
| 4642 int rc = SQLITE_OK; |
| 4643 |
| 4644 for(i=pCsr->nSegment-1; i>=0 && rc==SQLITE_OK; i--){ |
| 4645 Fts3SegReader *pSeg = 0; |
| 4646 int j; |
| 4647 |
| 4648 /* Find the Fts3SegReader object with Fts3SegReader.iIdx==i. It is hiding |
| 4649 ** somewhere in the pCsr->apSegment[] array. */ |
| 4650 for(j=0; ALWAYS(j<pCsr->nSegment); j++){ |
| 4651 pSeg = pCsr->apSegment[j]; |
| 4652 if( pSeg->iIdx==i ) break; |
| 4653 } |
| 4654 assert( j<pCsr->nSegment && pSeg->iIdx==i ); |
| 4655 |
| 4656 if( pSeg->aNode==0 ){ |
| 4657 /* Seg-reader is at EOF. Remove the entire input segment. */ |
| 4658 rc = fts3DeleteSegment(p, pSeg); |
| 4659 if( rc==SQLITE_OK ){ |
| 4660 rc = fts3RemoveSegdirEntry(p, iAbsLevel, pSeg->iIdx); |
| 4661 } |
| 4662 *pnRem = 0; |
| 4663 }else{ |
| 4664 /* The incremental merge did not copy all the data from this |
| 4665 ** segment to the upper level. The segment is modified in place |
| 4666 ** so that it contains no keys smaller than zTerm/nTerm. */ |
| 4667 const char *zTerm = pSeg->zTerm; |
| 4668 int nTerm = pSeg->nTerm; |
| 4669 rc = fts3TruncateSegment(p, iAbsLevel, pSeg->iIdx, zTerm, nTerm); |
| 4670 nRem++; |
| 4671 } |
| 4672 } |
| 4673 |
| 4674 if( rc==SQLITE_OK && nRem!=pCsr->nSegment ){ |
| 4675 rc = fts3RepackSegdirLevel(p, iAbsLevel); |
| 4676 } |
| 4677 |
| 4678 *pnRem = nRem; |
| 4679 return rc; |
| 4680 } |
| 4681 |
| 4682 /* |
| 4683 ** Store an incr-merge hint in the database. |
| 4684 */ |
| 4685 static int fts3IncrmergeHintStore(Fts3Table *p, Blob *pHint){ |
| 4686 sqlite3_stmt *pReplace = 0; |
| 4687 int rc; /* Return code */ |
| 4688 |
| 4689 rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pReplace, 0); |
| 4690 if( rc==SQLITE_OK ){ |
| 4691 sqlite3_bind_int(pReplace, 1, FTS_STAT_INCRMERGEHINT); |
| 4692 sqlite3_bind_blob(pReplace, 2, pHint->a, pHint->n, SQLITE_STATIC); |
| 4693 sqlite3_step(pReplace); |
| 4694 rc = sqlite3_reset(pReplace); |
| 4695 } |
| 4696 |
| 4697 return rc; |
| 4698 } |
| 4699 |
| 4700 /* |
| 4701 ** Load an incr-merge hint from the database. The incr-merge hint, if one |
| 4702 ** exists, is stored in the rowid==1 row of the %_stat table. |
| 4703 ** |
| 4704 ** If successful, populate blob *pHint with the value read from the %_stat |
| 4705 ** table and return SQLITE_OK. Otherwise, if an error occurs, return an |
| 4706 ** SQLite error code. |
| 4707 */ |
| 4708 static int fts3IncrmergeHintLoad(Fts3Table *p, Blob *pHint){ |
| 4709 sqlite3_stmt *pSelect = 0; |
| 4710 int rc; |
| 4711 |
| 4712 pHint->n = 0; |
| 4713 rc = fts3SqlStmt(p, SQL_SELECT_STAT, &pSelect, 0); |
| 4714 if( rc==SQLITE_OK ){ |
| 4715 int rc2; |
| 4716 sqlite3_bind_int(pSelect, 1, FTS_STAT_INCRMERGEHINT); |
| 4717 if( SQLITE_ROW==sqlite3_step(pSelect) ){ |
| 4718 const char *aHint = sqlite3_column_blob(pSelect, 0); |
| 4719 int nHint = sqlite3_column_bytes(pSelect, 0); |
| 4720 if( aHint ){ |
| 4721 blobGrowBuffer(pHint, nHint, &rc); |
| 4722 if( rc==SQLITE_OK ){ |
| 4723 memcpy(pHint->a, aHint, nHint); |
| 4724 pHint->n = nHint; |
| 4725 } |
| 4726 } |
| 4727 } |
| 4728 rc2 = sqlite3_reset(pSelect); |
| 4729 if( rc==SQLITE_OK ) rc = rc2; |
| 4730 } |
| 4731 |
| 4732 return rc; |
| 4733 } |
| 4734 |
| 4735 /* |
| 4736 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op. |
| 4737 ** Otherwise, append an entry to the hint stored in blob *pHint. Each entry |
| 4738 ** consists of two varints, the absolute level number of the input segments |
| 4739 ** and the number of input segments. |
| 4740 ** |
| 4741 ** If successful, leave *pRc set to SQLITE_OK and return. If an error occurs, |
| 4742 ** set *pRc to an SQLite error code before returning. |
| 4743 */ |
| 4744 static void fts3IncrmergeHintPush( |
| 4745 Blob *pHint, /* Hint blob to append to */ |
| 4746 i64 iAbsLevel, /* First varint to store in hint */ |
| 4747 int nInput, /* Second varint to store in hint */ |
| 4748 int *pRc /* IN/OUT: Error code */ |
| 4749 ){ |
| 4750 blobGrowBuffer(pHint, pHint->n + 2*FTS3_VARINT_MAX, pRc); |
| 4751 if( *pRc==SQLITE_OK ){ |
| 4752 pHint->n += sqlite3Fts3PutVarint(&pHint->a[pHint->n], iAbsLevel); |
| 4753 pHint->n += sqlite3Fts3PutVarint(&pHint->a[pHint->n], (i64)nInput); |
| 4754 } |
| 4755 } |
| 4756 |
| 4757 /* |
| 4758 ** Read the last entry (most recently pushed) from the hint blob *pHint |
| 4759 ** and then remove the entry. Write the two values read to *piAbsLevel and |
| 4760 ** *pnInput before returning. |
| 4761 ** |
| 4762 ** If no error occurs, return SQLITE_OK. If the hint blob in *pHint does |
| 4763 ** not contain at least two valid varints, return SQLITE_CORRUPT_VTAB. |
| 4764 */ |
| 4765 static int fts3IncrmergeHintPop(Blob *pHint, i64 *piAbsLevel, int *pnInput){ |
| 4766 const int nHint = pHint->n; |
| 4767 int i; |
| 4768 |
| 4769 i = pHint->n-2; |
| 4770 while( i>0 && (pHint->a[i-1] & 0x80) ) i--; |
| 4771 while( i>0 && (pHint->a[i-1] & 0x80) ) i--; |
| 4772 |
| 4773 pHint->n = i; |
| 4774 i += sqlite3Fts3GetVarint(&pHint->a[i], piAbsLevel); |
| 4775 i += fts3GetVarint32(&pHint->a[i], pnInput); |
| 4776 if( i!=nHint ) return SQLITE_CORRUPT_VTAB; |
| 4777 |
| 4778 return SQLITE_OK; |
| 4779 } |
| 4780 |
| 4781 |
| 4782 /* |
| 4783 ** Attempt an incremental merge that writes nMerge leaf blocks. |
| 4784 ** |
| 4785 ** Incremental merges happen nMin segments at a time. The segments |
| 4786 ** to be merged are the nMin oldest segments (the ones with the smallest |
| 4787 ** values for the _segdir.idx field) in the highest level that contains |
| 4788 ** at least nMin segments. Multiple merges might occur in an attempt to |
| 4789 ** write the quota of nMerge leaf blocks. |
| 4790 */ |
| 4791 int sqlite3Fts3Incrmerge(Fts3Table *p, int nMerge, int nMin){ |
| 4792 int rc; /* Return code */ |
| 4793 int nRem = nMerge; /* Number of leaf pages yet to be written */ |
| 4794 Fts3MultiSegReader *pCsr; /* Cursor used to read input data */ |
| 4795 Fts3SegFilter *pFilter; /* Filter used with cursor pCsr */ |
| 4796 IncrmergeWriter *pWriter; /* Writer object */ |
| 4797 int nSeg = 0; /* Number of input segments */ |
| 4798 sqlite3_int64 iAbsLevel = 0; /* Absolute level number to work on */ |
| 4799 Blob hint = {0, 0, 0}; /* Hint read from %_stat table */ |
| 4800 int bDirtyHint = 0; /* True if blob 'hint' has been modified */ |
| 4801 |
| 4802 /* Allocate space for the cursor, filter and writer objects */ |
| 4803 const int nAlloc = sizeof(*pCsr) + sizeof(*pFilter) + sizeof(*pWriter); |
| 4804 pWriter = (IncrmergeWriter *)sqlite3_malloc(nAlloc); |
| 4805 if( !pWriter ) return SQLITE_NOMEM; |
| 4806 pFilter = (Fts3SegFilter *)&pWriter[1]; |
| 4807 pCsr = (Fts3MultiSegReader *)&pFilter[1]; |
| 4808 |
| 4809 rc = fts3IncrmergeHintLoad(p, &hint); |
| 4810 while( rc==SQLITE_OK && nRem>0 ){ |
| 4811 const i64 nMod = FTS3_SEGDIR_MAXLEVEL * p->nIndex; |
| 4812 sqlite3_stmt *pFindLevel = 0; /* SQL used to determine iAbsLevel */ |
| 4813 int bUseHint = 0; /* True if attempting to append */ |
| 4814 int iIdx = 0; /* Largest idx in level (iAbsLevel+1) */ |
| 4815 |
| 4816 /* Search the %_segdir table for the absolute level with the smallest |
| 4817 ** relative level number that contains at least nMin segments, if any. |
| 4818 ** If one is found, set iAbsLevel to the absolute level number and |
| 4819 ** nSeg to nMin. If no level with at least nMin segments can be found, |
| 4820 ** set nSeg to -1. |
| 4821 */ |
| 4822 rc = fts3SqlStmt(p, SQL_FIND_MERGE_LEVEL, &pFindLevel, 0); |
| 4823 sqlite3_bind_int(pFindLevel, 1, nMin); |
| 4824 if( sqlite3_step(pFindLevel)==SQLITE_ROW ){ |
| 4825 iAbsLevel = sqlite3_column_int64(pFindLevel, 0); |
| 4826 nSeg = nMin; |
| 4827 }else{ |
| 4828 nSeg = -1; |
| 4829 } |
| 4830 rc = sqlite3_reset(pFindLevel); |
| 4831 |
| 4832 /* If the hint read from the %_stat table is not empty, check if the |
| 4833 ** last entry in it specifies a relative level smaller than or equal |
| 4834 ** to the level identified by the block above (if any). If so, this |
| 4835 ** iteration of the loop will work on merging at the hinted level. |
| 4836 */ |
| 4837 if( rc==SQLITE_OK && hint.n ){ |
| 4838 int nHint = hint.n; |
| 4839 sqlite3_int64 iHintAbsLevel = 0; /* Hint level */ |
| 4840 int nHintSeg = 0; /* Hint number of segments */ |
| 4841 |
| 4842 rc = fts3IncrmergeHintPop(&hint, &iHintAbsLevel, &nHintSeg); |
| 4843 if( nSeg<0 || (iAbsLevel % nMod) >= (iHintAbsLevel % nMod) ){ |
| 4844 iAbsLevel = iHintAbsLevel; |
| 4845 nSeg = nHintSeg; |
| 4846 bUseHint = 1; |
| 4847 bDirtyHint = 1; |
| 4848 }else{ |
| 4849 /* This undoes the effect of the HintPop() above - so that no entry |
| 4850 ** is removed from the hint blob. */ |
| 4851 hint.n = nHint; |
| 4852 } |
| 4853 } |
| 4854 |
| 4855 /* If nSeg is less that zero, then there is no level with at least |
| 4856 ** nMin segments and no hint in the %_stat table. No work to do. |
| 4857 ** Exit early in this case. */ |
| 4858 if( nSeg<0 ) break; |
| 4859 |
| 4860 /* Open a cursor to iterate through the contents of the oldest nSeg |
| 4861 ** indexes of absolute level iAbsLevel. If this cursor is opened using |
| 4862 ** the 'hint' parameters, it is possible that there are less than nSeg |
| 4863 ** segments available in level iAbsLevel. In this case, no work is |
| 4864 ** done on iAbsLevel - fall through to the next iteration of the loop |
| 4865 ** to start work on some other level. */ |
| 4866 memset(pWriter, 0, nAlloc); |
| 4867 pFilter->flags = FTS3_SEGMENT_REQUIRE_POS; |
| 4868 |
| 4869 if( rc==SQLITE_OK ){ |
| 4870 rc = fts3IncrmergeOutputIdx(p, iAbsLevel, &iIdx); |
| 4871 assert( bUseHint==1 || bUseHint==0 ); |
| 4872 if( iIdx==0 || (bUseHint && iIdx==1) ){ |
| 4873 int bIgnore = 0; |
| 4874 rc = fts3SegmentIsMaxLevel(p, iAbsLevel+1, &bIgnore); |
| 4875 if( bIgnore ){ |
| 4876 pFilter->flags |= FTS3_SEGMENT_IGNORE_EMPTY; |
| 4877 } |
| 4878 } |
| 4879 } |
| 4880 |
| 4881 if( rc==SQLITE_OK ){ |
| 4882 rc = fts3IncrmergeCsr(p, iAbsLevel, nSeg, pCsr); |
| 4883 } |
| 4884 if( SQLITE_OK==rc && pCsr->nSegment==nSeg |
| 4885 && SQLITE_OK==(rc = sqlite3Fts3SegReaderStart(p, pCsr, pFilter)) |
| 4886 && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pCsr)) |
| 4887 ){ |
| 4888 if( bUseHint && iIdx>0 ){ |
| 4889 const char *zKey = pCsr->zTerm; |
| 4890 int nKey = pCsr->nTerm; |
| 4891 rc = fts3IncrmergeLoad(p, iAbsLevel, iIdx-1, zKey, nKey, pWriter); |
| 4892 }else{ |
| 4893 rc = fts3IncrmergeWriter(p, iAbsLevel, iIdx, pCsr, pWriter); |
| 4894 } |
| 4895 |
| 4896 if( rc==SQLITE_OK && pWriter->nLeafEst ){ |
| 4897 fts3LogMerge(nSeg, iAbsLevel); |
| 4898 do { |
| 4899 rc = fts3IncrmergeAppend(p, pWriter, pCsr); |
| 4900 if( rc==SQLITE_OK ) rc = sqlite3Fts3SegReaderStep(p, pCsr); |
| 4901 if( pWriter->nWork>=nRem && rc==SQLITE_ROW ) rc = SQLITE_OK; |
| 4902 }while( rc==SQLITE_ROW ); |
| 4903 |
| 4904 /* Update or delete the input segments */ |
| 4905 if( rc==SQLITE_OK ){ |
| 4906 nRem -= (1 + pWriter->nWork); |
| 4907 rc = fts3IncrmergeChomp(p, iAbsLevel, pCsr, &nSeg); |
| 4908 if( nSeg!=0 ){ |
| 4909 bDirtyHint = 1; |
| 4910 fts3IncrmergeHintPush(&hint, iAbsLevel, nSeg, &rc); |
| 4911 } |
| 4912 } |
| 4913 } |
| 4914 |
| 4915 if( nSeg!=0 ){ |
| 4916 pWriter->nLeafData = pWriter->nLeafData * -1; |
| 4917 } |
| 4918 fts3IncrmergeRelease(p, pWriter, &rc); |
| 4919 if( nSeg==0 && pWriter->bNoLeafData==0 ){ |
| 4920 fts3PromoteSegments(p, iAbsLevel+1, pWriter->nLeafData); |
| 4921 } |
| 4922 } |
| 4923 |
| 4924 sqlite3Fts3SegReaderFinish(pCsr); |
| 4925 } |
| 4926 |
| 4927 /* Write the hint values into the %_stat table for the next incr-merger */ |
| 4928 if( bDirtyHint && rc==SQLITE_OK ){ |
| 4929 rc = fts3IncrmergeHintStore(p, &hint); |
| 4930 } |
| 4931 |
| 4932 sqlite3_free(pWriter); |
| 4933 sqlite3_free(hint.a); |
| 4934 return rc; |
| 4935 } |
| 4936 |
| 4937 /* |
| 4938 ** Convert the text beginning at *pz into an integer and return |
| 4939 ** its value. Advance *pz to point to the first character past |
| 4940 ** the integer. |
| 4941 */ |
| 4942 static int fts3Getint(const char **pz){ |
| 4943 const char *z = *pz; |
| 4944 int i = 0; |
| 4945 while( (*z)>='0' && (*z)<='9' ) i = 10*i + *(z++) - '0'; |
| 4946 *pz = z; |
| 4947 return i; |
| 4948 } |
| 4949 |
| 4950 /* |
| 4951 ** Process statements of the form: |
| 4952 ** |
| 4953 ** INSERT INTO table(table) VALUES('merge=A,B'); |
| 4954 ** |
| 4955 ** A and B are integers that decode to be the number of leaf pages |
| 4956 ** written for the merge, and the minimum number of segments on a level |
| 4957 ** before it will be selected for a merge, respectively. |
| 4958 */ |
| 4959 static int fts3DoIncrmerge( |
| 4960 Fts3Table *p, /* FTS3 table handle */ |
| 4961 const char *zParam /* Nul-terminated string containing "A,B" */ |
| 4962 ){ |
| 4963 int rc; |
| 4964 int nMin = (FTS3_MERGE_COUNT / 2); |
| 4965 int nMerge = 0; |
| 4966 const char *z = zParam; |
| 4967 |
| 4968 /* Read the first integer value */ |
| 4969 nMerge = fts3Getint(&z); |
| 4970 |
| 4971 /* If the first integer value is followed by a ',', read the second |
| 4972 ** integer value. */ |
| 4973 if( z[0]==',' && z[1]!='\0' ){ |
| 4974 z++; |
| 4975 nMin = fts3Getint(&z); |
| 4976 } |
| 4977 |
| 4978 if( z[0]!='\0' || nMin<2 ){ |
| 4979 rc = SQLITE_ERROR; |
| 4980 }else{ |
| 4981 rc = SQLITE_OK; |
| 4982 if( !p->bHasStat ){ |
| 4983 assert( p->bFts4==0 ); |
| 4984 sqlite3Fts3CreateStatTable(&rc, p); |
| 4985 } |
| 4986 if( rc==SQLITE_OK ){ |
| 4987 rc = sqlite3Fts3Incrmerge(p, nMerge, nMin); |
| 4988 } |
| 4989 sqlite3Fts3SegmentsClose(p); |
| 4990 } |
| 4991 return rc; |
| 4992 } |
| 4993 |
| 4994 /* |
| 4995 ** Process statements of the form: |
| 4996 ** |
| 4997 ** INSERT INTO table(table) VALUES('automerge=X'); |
| 4998 ** |
| 4999 ** where X is an integer. X==0 means to turn automerge off. X!=0 means |
| 5000 ** turn it on. The setting is persistent. |
| 5001 */ |
| 5002 static int fts3DoAutoincrmerge( |
| 5003 Fts3Table *p, /* FTS3 table handle */ |
| 5004 const char *zParam /* Nul-terminated string containing boolean */ |
| 5005 ){ |
| 5006 int rc = SQLITE_OK; |
| 5007 sqlite3_stmt *pStmt = 0; |
| 5008 p->nAutoincrmerge = fts3Getint(&zParam); |
| 5009 if( p->nAutoincrmerge==1 || p->nAutoincrmerge>FTS3_MERGE_COUNT ){ |
| 5010 p->nAutoincrmerge = 8; |
| 5011 } |
| 5012 if( !p->bHasStat ){ |
| 5013 assert( p->bFts4==0 ); |
| 5014 sqlite3Fts3CreateStatTable(&rc, p); |
| 5015 if( rc ) return rc; |
| 5016 } |
| 5017 rc = fts3SqlStmt(p, SQL_REPLACE_STAT, &pStmt, 0); |
| 5018 if( rc ) return rc; |
| 5019 sqlite3_bind_int(pStmt, 1, FTS_STAT_AUTOINCRMERGE); |
| 5020 sqlite3_bind_int(pStmt, 2, p->nAutoincrmerge); |
| 5021 sqlite3_step(pStmt); |
| 5022 rc = sqlite3_reset(pStmt); |
| 5023 return rc; |
| 5024 } |
| 5025 |
| 5026 /* |
| 5027 ** Return a 64-bit checksum for the FTS index entry specified by the |
| 5028 ** arguments to this function. |
| 5029 */ |
| 5030 static u64 fts3ChecksumEntry( |
| 5031 const char *zTerm, /* Pointer to buffer containing term */ |
| 5032 int nTerm, /* Size of zTerm in bytes */ |
| 5033 int iLangid, /* Language id for current row */ |
| 5034 int iIndex, /* Index (0..Fts3Table.nIndex-1) */ |
| 5035 i64 iDocid, /* Docid for current row. */ |
| 5036 int iCol, /* Column number */ |
| 5037 int iPos /* Position */ |
| 5038 ){ |
| 5039 int i; |
| 5040 u64 ret = (u64)iDocid; |
| 5041 |
| 5042 ret += (ret<<3) + iLangid; |
| 5043 ret += (ret<<3) + iIndex; |
| 5044 ret += (ret<<3) + iCol; |
| 5045 ret += (ret<<3) + iPos; |
| 5046 for(i=0; i<nTerm; i++) ret += (ret<<3) + zTerm[i]; |
| 5047 |
| 5048 return ret; |
| 5049 } |
| 5050 |
| 5051 /* |
| 5052 ** Return a checksum of all entries in the FTS index that correspond to |
| 5053 ** language id iLangid. The checksum is calculated by XORing the checksums |
| 5054 ** of each individual entry (see fts3ChecksumEntry()) together. |
| 5055 ** |
| 5056 ** If successful, the checksum value is returned and *pRc set to SQLITE_OK. |
| 5057 ** Otherwise, if an error occurs, *pRc is set to an SQLite error code. The |
| 5058 ** return value is undefined in this case. |
| 5059 */ |
| 5060 static u64 fts3ChecksumIndex( |
| 5061 Fts3Table *p, /* FTS3 table handle */ |
| 5062 int iLangid, /* Language id to return cksum for */ |
| 5063 int iIndex, /* Index to cksum (0..p->nIndex-1) */ |
| 5064 int *pRc /* OUT: Return code */ |
| 5065 ){ |
| 5066 Fts3SegFilter filter; |
| 5067 Fts3MultiSegReader csr; |
| 5068 int rc; |
| 5069 u64 cksum = 0; |
| 5070 |
| 5071 assert( *pRc==SQLITE_OK ); |
| 5072 |
| 5073 memset(&filter, 0, sizeof(filter)); |
| 5074 memset(&csr, 0, sizeof(csr)); |
| 5075 filter.flags = FTS3_SEGMENT_REQUIRE_POS|FTS3_SEGMENT_IGNORE_EMPTY; |
| 5076 filter.flags |= FTS3_SEGMENT_SCAN; |
| 5077 |
| 5078 rc = sqlite3Fts3SegReaderCursor( |
| 5079 p, iLangid, iIndex, FTS3_SEGCURSOR_ALL, 0, 0, 0, 1,&csr |
| 5080 ); |
| 5081 if( rc==SQLITE_OK ){ |
| 5082 rc = sqlite3Fts3SegReaderStart(p, &csr, &filter); |
| 5083 } |
| 5084 |
| 5085 if( rc==SQLITE_OK ){ |
| 5086 while( SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, &csr)) ){ |
| 5087 char *pCsr = csr.aDoclist; |
| 5088 char *pEnd = &pCsr[csr.nDoclist]; |
| 5089 |
| 5090 i64 iDocid = 0; |
| 5091 i64 iCol = 0; |
| 5092 i64 iPos = 0; |
| 5093 |
| 5094 pCsr += sqlite3Fts3GetVarint(pCsr, &iDocid); |
| 5095 while( pCsr<pEnd ){ |
| 5096 i64 iVal = 0; |
| 5097 pCsr += sqlite3Fts3GetVarint(pCsr, &iVal); |
| 5098 if( pCsr<pEnd ){ |
| 5099 if( iVal==0 || iVal==1 ){ |
| 5100 iCol = 0; |
| 5101 iPos = 0; |
| 5102 if( iVal ){ |
| 5103 pCsr += sqlite3Fts3GetVarint(pCsr, &iCol); |
| 5104 }else{ |
| 5105 pCsr += sqlite3Fts3GetVarint(pCsr, &iVal); |
| 5106 iDocid += iVal; |
| 5107 } |
| 5108 }else{ |
| 5109 iPos += (iVal - 2); |
| 5110 cksum = cksum ^ fts3ChecksumEntry( |
| 5111 csr.zTerm, csr.nTerm, iLangid, iIndex, iDocid, |
| 5112 (int)iCol, (int)iPos |
| 5113 ); |
| 5114 } |
| 5115 } |
| 5116 } |
| 5117 } |
| 5118 } |
| 5119 sqlite3Fts3SegReaderFinish(&csr); |
| 5120 |
| 5121 *pRc = rc; |
| 5122 return cksum; |
| 5123 } |
| 5124 |
| 5125 /* |
| 5126 ** Check if the contents of the FTS index match the current contents of the |
| 5127 ** content table. If no error occurs and the contents do match, set *pbOk |
| 5128 ** to true and return SQLITE_OK. Or if the contents do not match, set *pbOk |
| 5129 ** to false before returning. |
| 5130 ** |
| 5131 ** If an error occurs (e.g. an OOM or IO error), return an SQLite error |
| 5132 ** code. The final value of *pbOk is undefined in this case. |
| 5133 */ |
| 5134 static int fts3IntegrityCheck(Fts3Table *p, int *pbOk){ |
| 5135 int rc = SQLITE_OK; /* Return code */ |
| 5136 u64 cksum1 = 0; /* Checksum based on FTS index contents */ |
| 5137 u64 cksum2 = 0; /* Checksum based on %_content contents */ |
| 5138 sqlite3_stmt *pAllLangid = 0; /* Statement to return all language-ids */ |
| 5139 |
| 5140 /* This block calculates the checksum according to the FTS index. */ |
| 5141 rc = fts3SqlStmt(p, SQL_SELECT_ALL_LANGID, &pAllLangid, 0); |
| 5142 if( rc==SQLITE_OK ){ |
| 5143 int rc2; |
| 5144 sqlite3_bind_int(pAllLangid, 1, p->nIndex); |
| 5145 while( rc==SQLITE_OK && sqlite3_step(pAllLangid)==SQLITE_ROW ){ |
| 5146 int iLangid = sqlite3_column_int(pAllLangid, 0); |
| 5147 int i; |
| 5148 for(i=0; i<p->nIndex; i++){ |
| 5149 cksum1 = cksum1 ^ fts3ChecksumIndex(p, iLangid, i, &rc); |
| 5150 } |
| 5151 } |
| 5152 rc2 = sqlite3_reset(pAllLangid); |
| 5153 if( rc==SQLITE_OK ) rc = rc2; |
| 5154 } |
| 5155 |
| 5156 /* This block calculates the checksum according to the %_content table */ |
| 5157 rc = fts3SqlStmt(p, SQL_SELECT_ALL_LANGID, &pAllLangid, 0); |
| 5158 if( rc==SQLITE_OK ){ |
| 5159 sqlite3_tokenizer_module const *pModule = p->pTokenizer->pModule; |
| 5160 sqlite3_stmt *pStmt = 0; |
| 5161 char *zSql; |
| 5162 |
| 5163 zSql = sqlite3_mprintf("SELECT %s" , p->zReadExprlist); |
| 5164 if( !zSql ){ |
| 5165 rc = SQLITE_NOMEM; |
| 5166 }else{ |
| 5167 rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); |
| 5168 sqlite3_free(zSql); |
| 5169 } |
| 5170 |
| 5171 while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ |
| 5172 i64 iDocid = sqlite3_column_int64(pStmt, 0); |
| 5173 int iLang = langidFromSelect(p, pStmt); |
| 5174 int iCol; |
| 5175 |
| 5176 for(iCol=0; rc==SQLITE_OK && iCol<p->nColumn; iCol++){ |
| 5177 if( p->abNotindexed[iCol]==0 ){ |
| 5178 const char *zText = (const char *)sqlite3_column_text(pStmt, iCol+1); |
| 5179 int nText = sqlite3_column_bytes(pStmt, iCol+1); |
| 5180 sqlite3_tokenizer_cursor *pT = 0; |
| 5181 |
| 5182 rc = sqlite3Fts3OpenTokenizer(p->pTokenizer, iLang, zText, nText,&pT); |
| 5183 while( rc==SQLITE_OK ){ |
| 5184 char const *zToken; /* Buffer containing token */ |
| 5185 int nToken = 0; /* Number of bytes in token */ |
| 5186 int iDum1 = 0, iDum2 = 0; /* Dummy variables */ |
| 5187 int iPos = 0; /* Position of token in zText */ |
| 5188 |
| 5189 rc = pModule->xNext(pT, &zToken, &nToken, &iDum1, &iDum2, &iPos); |
| 5190 if( rc==SQLITE_OK ){ |
| 5191 int i; |
| 5192 cksum2 = cksum2 ^ fts3ChecksumEntry( |
| 5193 zToken, nToken, iLang, 0, iDocid, iCol, iPos |
| 5194 ); |
| 5195 for(i=1; i<p->nIndex; i++){ |
| 5196 if( p->aIndex[i].nPrefix<=nToken ){ |
| 5197 cksum2 = cksum2 ^ fts3ChecksumEntry( |
| 5198 zToken, p->aIndex[i].nPrefix, iLang, i, iDocid, iCol, iPos |
| 5199 ); |
| 5200 } |
| 5201 } |
| 5202 } |
| 5203 } |
| 5204 if( pT ) pModule->xClose(pT); |
| 5205 if( rc==SQLITE_DONE ) rc = SQLITE_OK; |
| 5206 } |
| 5207 } |
| 5208 } |
| 5209 |
| 5210 sqlite3_finalize(pStmt); |
| 5211 } |
| 5212 |
| 5213 *pbOk = (cksum1==cksum2); |
| 5214 return rc; |
| 5215 } |
| 5216 |
| 5217 /* |
| 5218 ** Run the integrity-check. If no error occurs and the current contents of |
| 5219 ** the FTS index are correct, return SQLITE_OK. Or, if the contents of the |
| 5220 ** FTS index are incorrect, return SQLITE_CORRUPT_VTAB. |
| 5221 ** |
| 5222 ** Or, if an error (e.g. an OOM or IO error) occurs, return an SQLite |
| 5223 ** error code. |
| 5224 ** |
| 5225 ** The integrity-check works as follows. For each token and indexed token |
| 5226 ** prefix in the document set, a 64-bit checksum is calculated (by code |
| 5227 ** in fts3ChecksumEntry()) based on the following: |
| 5228 ** |
| 5229 ** + The index number (0 for the main index, 1 for the first prefix |
| 5230 ** index etc.), |
| 5231 ** + The token (or token prefix) text itself, |
| 5232 ** + The language-id of the row it appears in, |
| 5233 ** + The docid of the row it appears in, |
| 5234 ** + The column it appears in, and |
| 5235 ** + The tokens position within that column. |
| 5236 ** |
| 5237 ** The checksums for all entries in the index are XORed together to create |
| 5238 ** a single checksum for the entire index. |
| 5239 ** |
| 5240 ** The integrity-check code calculates the same checksum in two ways: |
| 5241 ** |
| 5242 ** 1. By scanning the contents of the FTS index, and |
| 5243 ** 2. By scanning and tokenizing the content table. |
| 5244 ** |
| 5245 ** If the two checksums are identical, the integrity-check is deemed to have |
| 5246 ** passed. |
| 5247 */ |
| 5248 static int fts3DoIntegrityCheck( |
| 5249 Fts3Table *p /* FTS3 table handle */ |
| 5250 ){ |
| 5251 int rc; |
| 5252 int bOk = 0; |
| 5253 rc = fts3IntegrityCheck(p, &bOk); |
| 5254 if( rc==SQLITE_OK && bOk==0 ) rc = SQLITE_CORRUPT_VTAB; |
| 5255 return rc; |
| 5256 } |
| 5257 |
| 5258 /* |
| 5259 ** Handle a 'special' INSERT of the form: |
| 5260 ** |
| 5261 ** "INSERT INTO tbl(tbl) VALUES(<expr>)" |
| 5262 ** |
| 5263 ** Argument pVal contains the result of <expr>. Currently the only |
| 5264 ** meaningful value to insert is the text 'optimize'. |
| 5265 */ |
| 5266 static int fts3SpecialInsert(Fts3Table *p, sqlite3_value *pVal){ |
| 5267 int rc; /* Return Code */ |
| 5268 const char *zVal = (const char *)sqlite3_value_text(pVal); |
| 5269 int nVal = sqlite3_value_bytes(pVal); |
| 5270 |
| 5271 if( !zVal ){ |
| 5272 return SQLITE_NOMEM; |
| 5273 }else if( nVal==8 && 0==sqlite3_strnicmp(zVal, "optimize", 8) ){ |
| 5274 rc = fts3DoOptimize(p, 0); |
| 5275 }else if( nVal==7 && 0==sqlite3_strnicmp(zVal, "rebuild", 7) ){ |
| 5276 rc = fts3DoRebuild(p); |
| 5277 }else if( nVal==15 && 0==sqlite3_strnicmp(zVal, "integrity-check", 15) ){ |
| 5278 rc = fts3DoIntegrityCheck(p); |
| 5279 }else if( nVal>6 && 0==sqlite3_strnicmp(zVal, "merge=", 6) ){ |
| 5280 rc = fts3DoIncrmerge(p, &zVal[6]); |
| 5281 }else if( nVal>10 && 0==sqlite3_strnicmp(zVal, "automerge=", 10) ){ |
| 5282 rc = fts3DoAutoincrmerge(p, &zVal[10]); |
| 5283 #ifdef SQLITE_TEST |
| 5284 }else if( nVal>9 && 0==sqlite3_strnicmp(zVal, "nodesize=", 9) ){ |
| 5285 p->nNodeSize = atoi(&zVal[9]); |
| 5286 rc = SQLITE_OK; |
| 5287 }else if( nVal>11 && 0==sqlite3_strnicmp(zVal, "maxpending=", 9) ){ |
| 5288 p->nMaxPendingData = atoi(&zVal[11]); |
| 5289 rc = SQLITE_OK; |
| 5290 }else if( nVal>21 && 0==sqlite3_strnicmp(zVal, "test-no-incr-doclist=", 21) ){ |
| 5291 p->bNoIncrDoclist = atoi(&zVal[21]); |
| 5292 rc = SQLITE_OK; |
| 5293 #endif |
| 5294 }else{ |
| 5295 rc = SQLITE_ERROR; |
| 5296 } |
| 5297 |
| 5298 return rc; |
| 5299 } |
| 5300 |
| 5301 #ifndef SQLITE_DISABLE_FTS4_DEFERRED |
| 5302 /* |
| 5303 ** Delete all cached deferred doclists. Deferred doclists are cached |
| 5304 ** (allocated) by the sqlite3Fts3CacheDeferredDoclists() function. |
| 5305 */ |
| 5306 void sqlite3Fts3FreeDeferredDoclists(Fts3Cursor *pCsr){ |
| 5307 Fts3DeferredToken *pDef; |
| 5308 for(pDef=pCsr->pDeferred; pDef; pDef=pDef->pNext){ |
| 5309 fts3PendingListDelete(pDef->pList); |
| 5310 pDef->pList = 0; |
| 5311 } |
| 5312 } |
| 5313 |
| 5314 /* |
| 5315 ** Free all entries in the pCsr->pDeffered list. Entries are added to |
| 5316 ** this list using sqlite3Fts3DeferToken(). |
| 5317 */ |
| 5318 void sqlite3Fts3FreeDeferredTokens(Fts3Cursor *pCsr){ |
| 5319 Fts3DeferredToken *pDef; |
| 5320 Fts3DeferredToken *pNext; |
| 5321 for(pDef=pCsr->pDeferred; pDef; pDef=pNext){ |
| 5322 pNext = pDef->pNext; |
| 5323 fts3PendingListDelete(pDef->pList); |
| 5324 sqlite3_free(pDef); |
| 5325 } |
| 5326 pCsr->pDeferred = 0; |
| 5327 } |
| 5328 |
| 5329 /* |
| 5330 ** Generate deferred-doclists for all tokens in the pCsr->pDeferred list |
| 5331 ** based on the row that pCsr currently points to. |
| 5332 ** |
| 5333 ** A deferred-doclist is like any other doclist with position information |
| 5334 ** included, except that it only contains entries for a single row of the |
| 5335 ** table, not for all rows. |
| 5336 */ |
| 5337 int sqlite3Fts3CacheDeferredDoclists(Fts3Cursor *pCsr){ |
| 5338 int rc = SQLITE_OK; /* Return code */ |
| 5339 if( pCsr->pDeferred ){ |
| 5340 int i; /* Used to iterate through table columns */ |
| 5341 sqlite3_int64 iDocid; /* Docid of the row pCsr points to */ |
| 5342 Fts3DeferredToken *pDef; /* Used to iterate through deferred tokens */ |
| 5343 |
| 5344 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; |
| 5345 sqlite3_tokenizer *pT = p->pTokenizer; |
| 5346 sqlite3_tokenizer_module const *pModule = pT->pModule; |
| 5347 |
| 5348 assert( pCsr->isRequireSeek==0 ); |
| 5349 iDocid = sqlite3_column_int64(pCsr->pStmt, 0); |
| 5350 |
| 5351 for(i=0; i<p->nColumn && rc==SQLITE_OK; i++){ |
| 5352 if( p->abNotindexed[i]==0 ){ |
| 5353 const char *zText = (const char *)sqlite3_column_text(pCsr->pStmt, i+1); |
| 5354 sqlite3_tokenizer_cursor *pTC = 0; |
| 5355 |
| 5356 rc = sqlite3Fts3OpenTokenizer(pT, pCsr->iLangid, zText, -1, &pTC); |
| 5357 while( rc==SQLITE_OK ){ |
| 5358 char const *zToken; /* Buffer containing token */ |
| 5359 int nToken = 0; /* Number of bytes in token */ |
| 5360 int iDum1 = 0, iDum2 = 0; /* Dummy variables */ |
| 5361 int iPos = 0; /* Position of token in zText */ |
| 5362 |
| 5363 rc = pModule->xNext(pTC, &zToken, &nToken, &iDum1, &iDum2, &iPos); |
| 5364 for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ |
| 5365 Fts3PhraseToken *pPT = pDef->pToken; |
| 5366 if( (pDef->iCol>=p->nColumn || pDef->iCol==i) |
| 5367 && (pPT->bFirst==0 || iPos==0) |
| 5368 && (pPT->n==nToken || (pPT->isPrefix && pPT->n<nToken)) |
| 5369 && (0==memcmp(zToken, pPT->z, pPT->n)) |
| 5370 ){ |
| 5371 fts3PendingListAppend(&pDef->pList, iDocid, i, iPos, &rc); |
| 5372 } |
| 5373 } |
| 5374 } |
| 5375 if( pTC ) pModule->xClose(pTC); |
| 5376 if( rc==SQLITE_DONE ) rc = SQLITE_OK; |
| 5377 } |
| 5378 } |
| 5379 |
| 5380 for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ |
| 5381 if( pDef->pList ){ |
| 5382 rc = fts3PendingListAppendVarint(&pDef->pList, 0); |
| 5383 } |
| 5384 } |
| 5385 } |
| 5386 |
| 5387 return rc; |
| 5388 } |
| 5389 |
| 5390 int sqlite3Fts3DeferredTokenList( |
| 5391 Fts3DeferredToken *p, |
| 5392 char **ppData, |
| 5393 int *pnData |
| 5394 ){ |
| 5395 char *pRet; |
| 5396 int nSkip; |
| 5397 sqlite3_int64 dummy; |
| 5398 |
| 5399 *ppData = 0; |
| 5400 *pnData = 0; |
| 5401 |
| 5402 if( p->pList==0 ){ |
| 5403 return SQLITE_OK; |
| 5404 } |
| 5405 |
| 5406 pRet = (char *)sqlite3_malloc(p->pList->nData); |
| 5407 if( !pRet ) return SQLITE_NOMEM; |
| 5408 |
| 5409 nSkip = sqlite3Fts3GetVarint(p->pList->aData, &dummy); |
| 5410 *pnData = p->pList->nData - nSkip; |
| 5411 *ppData = pRet; |
| 5412 |
| 5413 memcpy(pRet, &p->pList->aData[nSkip], *pnData); |
| 5414 return SQLITE_OK; |
| 5415 } |
| 5416 |
| 5417 /* |
| 5418 ** Add an entry for token pToken to the pCsr->pDeferred list. |
| 5419 */ |
| 5420 int sqlite3Fts3DeferToken( |
| 5421 Fts3Cursor *pCsr, /* Fts3 table cursor */ |
| 5422 Fts3PhraseToken *pToken, /* Token to defer */ |
| 5423 int iCol /* Column that token must appear in (or -1) */ |
| 5424 ){ |
| 5425 Fts3DeferredToken *pDeferred; |
| 5426 pDeferred = sqlite3_malloc(sizeof(*pDeferred)); |
| 5427 if( !pDeferred ){ |
| 5428 return SQLITE_NOMEM; |
| 5429 } |
| 5430 memset(pDeferred, 0, sizeof(*pDeferred)); |
| 5431 pDeferred->pToken = pToken; |
| 5432 pDeferred->pNext = pCsr->pDeferred; |
| 5433 pDeferred->iCol = iCol; |
| 5434 pCsr->pDeferred = pDeferred; |
| 5435 |
| 5436 assert( pToken->pDeferred==0 ); |
| 5437 pToken->pDeferred = pDeferred; |
| 5438 |
| 5439 return SQLITE_OK; |
| 5440 } |
| 5441 #endif |
| 5442 |
| 5443 /* |
| 5444 ** SQLite value pRowid contains the rowid of a row that may or may not be |
| 5445 ** present in the FTS3 table. If it is, delete it and adjust the contents |
| 5446 ** of subsiduary data structures accordingly. |
| 5447 */ |
| 5448 static int fts3DeleteByRowid( |
| 5449 Fts3Table *p, |
| 5450 sqlite3_value *pRowid, |
| 5451 int *pnChng, /* IN/OUT: Decrement if row is deleted */ |
| 5452 u32 *aSzDel |
| 5453 ){ |
| 5454 int rc = SQLITE_OK; /* Return code */ |
| 5455 int bFound = 0; /* True if *pRowid really is in the table */ |
| 5456 |
| 5457 fts3DeleteTerms(&rc, p, pRowid, aSzDel, &bFound); |
| 5458 if( bFound && rc==SQLITE_OK ){ |
| 5459 int isEmpty = 0; /* Deleting *pRowid leaves the table empty */ |
| 5460 rc = fts3IsEmpty(p, pRowid, &isEmpty); |
| 5461 if( rc==SQLITE_OK ){ |
| 5462 if( isEmpty ){ |
| 5463 /* Deleting this row means the whole table is empty. In this case |
| 5464 ** delete the contents of all three tables and throw away any |
| 5465 ** data in the pendingTerms hash table. */ |
| 5466 rc = fts3DeleteAll(p, 1); |
| 5467 *pnChng = 0; |
| 5468 memset(aSzDel, 0, sizeof(u32) * (p->nColumn+1) * 2); |
| 5469 }else{ |
| 5470 *pnChng = *pnChng - 1; |
| 5471 if( p->zContentTbl==0 ){ |
| 5472 fts3SqlExec(&rc, p, SQL_DELETE_CONTENT, &pRowid); |
| 5473 } |
| 5474 if( p->bHasDocsize ){ |
| 5475 fts3SqlExec(&rc, p, SQL_DELETE_DOCSIZE, &pRowid); |
| 5476 } |
| 5477 } |
| 5478 } |
| 5479 } |
| 5480 |
| 5481 return rc; |
| 5482 } |
| 5483 |
| 5484 /* |
| 5485 ** This function does the work for the xUpdate method of FTS3 virtual |
| 5486 ** tables. The schema of the virtual table being: |
| 5487 ** |
| 5488 ** CREATE TABLE <table name>( |
| 5489 ** <user columns>, |
| 5490 ** <table name> HIDDEN, |
| 5491 ** docid HIDDEN, |
| 5492 ** <langid> HIDDEN |
| 5493 ** ); |
| 5494 ** |
| 5495 ** |
| 5496 */ |
| 5497 int sqlite3Fts3UpdateMethod( |
| 5498 sqlite3_vtab *pVtab, /* FTS3 vtab object */ |
| 5499 int nArg, /* Size of argument array */ |
| 5500 sqlite3_value **apVal, /* Array of arguments */ |
| 5501 sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ |
| 5502 ){ |
| 5503 Fts3Table *p = (Fts3Table *)pVtab; |
| 5504 int rc = SQLITE_OK; /* Return Code */ |
| 5505 int isRemove = 0; /* True for an UPDATE or DELETE */ |
| 5506 u32 *aSzIns = 0; /* Sizes of inserted documents */ |
| 5507 u32 *aSzDel = 0; /* Sizes of deleted documents */ |
| 5508 int nChng = 0; /* Net change in number of documents */ |
| 5509 int bInsertDone = 0; |
| 5510 |
| 5511 /* At this point it must be known if the %_stat table exists or not. |
| 5512 ** So bHasStat may not be 2. */ |
| 5513 assert( p->bHasStat==0 || p->bHasStat==1 ); |
| 5514 |
| 5515 assert( p->pSegments==0 ); |
| 5516 assert( |
| 5517 nArg==1 /* DELETE operations */ |
| 5518 || nArg==(2 + p->nColumn + 3) /* INSERT or UPDATE operations */ |
| 5519 ); |
| 5520 |
| 5521 /* Check for a "special" INSERT operation. One of the form: |
| 5522 ** |
| 5523 ** INSERT INTO xyz(xyz) VALUES('command'); |
| 5524 */ |
| 5525 if( nArg>1 |
| 5526 && sqlite3_value_type(apVal[0])==SQLITE_NULL |
| 5527 && sqlite3_value_type(apVal[p->nColumn+2])!=SQLITE_NULL |
| 5528 ){ |
| 5529 rc = fts3SpecialInsert(p, apVal[p->nColumn+2]); |
| 5530 goto update_out; |
| 5531 } |
| 5532 |
| 5533 if( nArg>1 && sqlite3_value_int(apVal[2 + p->nColumn + 2])<0 ){ |
| 5534 rc = SQLITE_CONSTRAINT; |
| 5535 goto update_out; |
| 5536 } |
| 5537 |
| 5538 /* Allocate space to hold the change in document sizes */ |
| 5539 aSzDel = sqlite3_malloc( sizeof(aSzDel[0])*(p->nColumn+1)*2 ); |
| 5540 if( aSzDel==0 ){ |
| 5541 rc = SQLITE_NOMEM; |
| 5542 goto update_out; |
| 5543 } |
| 5544 aSzIns = &aSzDel[p->nColumn+1]; |
| 5545 memset(aSzDel, 0, sizeof(aSzDel[0])*(p->nColumn+1)*2); |
| 5546 |
| 5547 rc = fts3Writelock(p); |
| 5548 if( rc!=SQLITE_OK ) goto update_out; |
| 5549 |
| 5550 /* If this is an INSERT operation, or an UPDATE that modifies the rowid |
| 5551 ** value, then this operation requires constraint handling. |
| 5552 ** |
| 5553 ** If the on-conflict mode is REPLACE, this means that the existing row |
| 5554 ** should be deleted from the database before inserting the new row. Or, |
| 5555 ** if the on-conflict mode is other than REPLACE, then this method must |
| 5556 ** detect the conflict and return SQLITE_CONSTRAINT before beginning to |
| 5557 ** modify the database file. |
| 5558 */ |
| 5559 if( nArg>1 && p->zContentTbl==0 ){ |
| 5560 /* Find the value object that holds the new rowid value. */ |
| 5561 sqlite3_value *pNewRowid = apVal[3+p->nColumn]; |
| 5562 if( sqlite3_value_type(pNewRowid)==SQLITE_NULL ){ |
| 5563 pNewRowid = apVal[1]; |
| 5564 } |
| 5565 |
| 5566 if( sqlite3_value_type(pNewRowid)!=SQLITE_NULL && ( |
| 5567 sqlite3_value_type(apVal[0])==SQLITE_NULL |
| 5568 || sqlite3_value_int64(apVal[0])!=sqlite3_value_int64(pNewRowid) |
| 5569 )){ |
| 5570 /* The new rowid is not NULL (in this case the rowid will be |
| 5571 ** automatically assigned and there is no chance of a conflict), and |
| 5572 ** the statement is either an INSERT or an UPDATE that modifies the |
| 5573 ** rowid column. So if the conflict mode is REPLACE, then delete any |
| 5574 ** existing row with rowid=pNewRowid. |
| 5575 ** |
| 5576 ** Or, if the conflict mode is not REPLACE, insert the new record into |
| 5577 ** the %_content table. If we hit the duplicate rowid constraint (or any |
| 5578 ** other error) while doing so, return immediately. |
| 5579 ** |
| 5580 ** This branch may also run if pNewRowid contains a value that cannot |
| 5581 ** be losslessly converted to an integer. In this case, the eventual |
| 5582 ** call to fts3InsertData() (either just below or further on in this |
| 5583 ** function) will return SQLITE_MISMATCH. If fts3DeleteByRowid is |
| 5584 ** invoked, it will delete zero rows (since no row will have |
| 5585 ** docid=$pNewRowid if $pNewRowid is not an integer value). |
| 5586 */ |
| 5587 if( sqlite3_vtab_on_conflict(p->db)==SQLITE_REPLACE ){ |
| 5588 rc = fts3DeleteByRowid(p, pNewRowid, &nChng, aSzDel); |
| 5589 }else{ |
| 5590 rc = fts3InsertData(p, apVal, pRowid); |
| 5591 bInsertDone = 1; |
| 5592 } |
| 5593 } |
| 5594 } |
| 5595 if( rc!=SQLITE_OK ){ |
| 5596 goto update_out; |
| 5597 } |
| 5598 |
| 5599 /* If this is a DELETE or UPDATE operation, remove the old record. */ |
| 5600 if( sqlite3_value_type(apVal[0])!=SQLITE_NULL ){ |
| 5601 assert( sqlite3_value_type(apVal[0])==SQLITE_INTEGER ); |
| 5602 rc = fts3DeleteByRowid(p, apVal[0], &nChng, aSzDel); |
| 5603 isRemove = 1; |
| 5604 } |
| 5605 |
| 5606 /* If this is an INSERT or UPDATE operation, insert the new record. */ |
| 5607 if( nArg>1 && rc==SQLITE_OK ){ |
| 5608 int iLangid = sqlite3_value_int(apVal[2 + p->nColumn + 2]); |
| 5609 if( bInsertDone==0 ){ |
| 5610 rc = fts3InsertData(p, apVal, pRowid); |
| 5611 if( rc==SQLITE_CONSTRAINT && p->zContentTbl==0 ){ |
| 5612 rc = FTS_CORRUPT_VTAB; |
| 5613 } |
| 5614 } |
| 5615 if( rc==SQLITE_OK && (!isRemove || *pRowid!=p->iPrevDocid ) ){ |
| 5616 rc = fts3PendingTermsDocid(p, iLangid, *pRowid); |
| 5617 } |
| 5618 if( rc==SQLITE_OK ){ |
| 5619 assert( p->iPrevDocid==*pRowid ); |
| 5620 rc = fts3InsertTerms(p, iLangid, apVal, aSzIns); |
| 5621 } |
| 5622 if( p->bHasDocsize ){ |
| 5623 fts3InsertDocsize(&rc, p, aSzIns); |
| 5624 } |
| 5625 nChng++; |
| 5626 } |
| 5627 |
| 5628 if( p->bFts4 ){ |
| 5629 fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nChng); |
| 5630 } |
| 5631 |
| 5632 update_out: |
| 5633 sqlite3_free(aSzDel); |
| 5634 sqlite3Fts3SegmentsClose(p); |
| 5635 return rc; |
| 5636 } |
| 5637 |
| 5638 /* |
| 5639 ** Flush any data in the pending-terms hash table to disk. If successful, |
| 5640 ** merge all segments in the database (including the new segment, if |
| 5641 ** there was any data to flush) into a single segment. |
| 5642 */ |
| 5643 int sqlite3Fts3Optimize(Fts3Table *p){ |
| 5644 int rc; |
| 5645 rc = sqlite3_exec(p->db, "SAVEPOINT fts3", 0, 0, 0); |
| 5646 if( rc==SQLITE_OK ){ |
| 5647 rc = fts3DoOptimize(p, 1); |
| 5648 if( rc==SQLITE_OK || rc==SQLITE_DONE ){ |
| 5649 int rc2 = sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); |
| 5650 if( rc2!=SQLITE_OK ) rc = rc2; |
| 5651 }else{ |
| 5652 sqlite3_exec(p->db, "ROLLBACK TO fts3", 0, 0, 0); |
| 5653 sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); |
| 5654 } |
| 5655 } |
| 5656 sqlite3Fts3SegmentsClose(p); |
| 5657 return rc; |
| 5658 } |
| 5659 |
| 5660 #endif |
OLD | NEW |