OLD | NEW |
| (Empty) |
1 /* | |
2 ** 2005-07-08 | |
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 ** This file contains code associated with the ANALYZE command. | |
13 ** | |
14 ** The ANALYZE command gather statistics about the content of tables | |
15 ** and indices. These statistics are made available to the query planner | |
16 ** to help it make better decisions about how to perform queries. | |
17 ** | |
18 ** The following system tables are or have been supported: | |
19 ** | |
20 ** CREATE TABLE sqlite_stat1(tbl, idx, stat); | |
21 ** CREATE TABLE sqlite_stat2(tbl, idx, sampleno, sample); | |
22 ** CREATE TABLE sqlite_stat3(tbl, idx, nEq, nLt, nDLt, sample); | |
23 ** CREATE TABLE sqlite_stat4(tbl, idx, nEq, nLt, nDLt, sample); | |
24 ** | |
25 ** Additional tables might be added in future releases of SQLite. | |
26 ** The sqlite_stat2 table is not created or used unless the SQLite version | |
27 ** is between 3.6.18 and 3.7.8, inclusive, and unless SQLite is compiled | |
28 ** with SQLITE_ENABLE_STAT2. The sqlite_stat2 table is deprecated. | |
29 ** The sqlite_stat2 table is superseded by sqlite_stat3, which is only | |
30 ** created and used by SQLite versions 3.7.9 and later and with | |
31 ** SQLITE_ENABLE_STAT3 defined. The functionality of sqlite_stat3 | |
32 ** is a superset of sqlite_stat2. The sqlite_stat4 is an enhanced | |
33 ** version of sqlite_stat3 and is only available when compiled with | |
34 ** SQLITE_ENABLE_STAT4 and in SQLite versions 3.8.1 and later. It is | |
35 ** not possible to enable both STAT3 and STAT4 at the same time. If they | |
36 ** are both enabled, then STAT4 takes precedence. | |
37 ** | |
38 ** For most applications, sqlite_stat1 provides all the statistics required | |
39 ** for the query planner to make good choices. | |
40 ** | |
41 ** Format of sqlite_stat1: | |
42 ** | |
43 ** There is normally one row per index, with the index identified by the | |
44 ** name in the idx column. The tbl column is the name of the table to | |
45 ** which the index belongs. In each such row, the stat column will be | |
46 ** a string consisting of a list of integers. The first integer in this | |
47 ** list is the number of rows in the index. (This is the same as the | |
48 ** number of rows in the table, except for partial indices.) The second | |
49 ** integer is the average number of rows in the index that have the same | |
50 ** value in the first column of the index. The third integer is the average | |
51 ** number of rows in the index that have the same value for the first two | |
52 ** columns. The N-th integer (for N>1) is the average number of rows in | |
53 ** the index which have the same value for the first N-1 columns. For | |
54 ** a K-column index, there will be K+1 integers in the stat column. If | |
55 ** the index is unique, then the last integer will be 1. | |
56 ** | |
57 ** The list of integers in the stat column can optionally be followed | |
58 ** by the keyword "unordered". The "unordered" keyword, if it is present, | |
59 ** must be separated from the last integer by a single space. If the | |
60 ** "unordered" keyword is present, then the query planner assumes that | |
61 ** the index is unordered and will not use the index for a range query. | |
62 ** | |
63 ** If the sqlite_stat1.idx column is NULL, then the sqlite_stat1.stat | |
64 ** column contains a single integer which is the (estimated) number of | |
65 ** rows in the table identified by sqlite_stat1.tbl. | |
66 ** | |
67 ** Format of sqlite_stat2: | |
68 ** | |
69 ** The sqlite_stat2 is only created and is only used if SQLite is compiled | |
70 ** with SQLITE_ENABLE_STAT2 and if the SQLite version number is between | |
71 ** 3.6.18 and 3.7.8. The "stat2" table contains additional information | |
72 ** about the distribution of keys within an index. The index is identified by | |
73 ** the "idx" column and the "tbl" column is the name of the table to which | |
74 ** the index belongs. There are usually 10 rows in the sqlite_stat2 | |
75 ** table for each index. | |
76 ** | |
77 ** The sqlite_stat2 entries for an index that have sampleno between 0 and 9 | |
78 ** inclusive are samples of the left-most key value in the index taken at | |
79 ** evenly spaced points along the index. Let the number of samples be S | |
80 ** (10 in the standard build) and let C be the number of rows in the index. | |
81 ** Then the sampled rows are given by: | |
82 ** | |
83 ** rownumber = (i*C*2 + C)/(S*2) | |
84 ** | |
85 ** For i between 0 and S-1. Conceptually, the index space is divided into | |
86 ** S uniform buckets and the samples are the middle row from each bucket. | |
87 ** | |
88 ** The format for sqlite_stat2 is recorded here for legacy reference. This | |
89 ** version of SQLite does not support sqlite_stat2. It neither reads nor | |
90 ** writes the sqlite_stat2 table. This version of SQLite only supports | |
91 ** sqlite_stat3. | |
92 ** | |
93 ** Format for sqlite_stat3: | |
94 ** | |
95 ** The sqlite_stat3 format is a subset of sqlite_stat4. Hence, the | |
96 ** sqlite_stat4 format will be described first. Further information | |
97 ** about sqlite_stat3 follows the sqlite_stat4 description. | |
98 ** | |
99 ** Format for sqlite_stat4: | |
100 ** | |
101 ** As with sqlite_stat2, the sqlite_stat4 table contains histogram data | |
102 ** to aid the query planner in choosing good indices based on the values | |
103 ** that indexed columns are compared against in the WHERE clauses of | |
104 ** queries. | |
105 ** | |
106 ** The sqlite_stat4 table contains multiple entries for each index. | |
107 ** The idx column names the index and the tbl column is the table of the | |
108 ** index. If the idx and tbl columns are the same, then the sample is | |
109 ** of the INTEGER PRIMARY KEY. The sample column is a blob which is the | |
110 ** binary encoding of a key from the index. The nEq column is a | |
111 ** list of integers. The first integer is the approximate number | |
112 ** of entries in the index whose left-most column exactly matches | |
113 ** the left-most column of the sample. The second integer in nEq | |
114 ** is the approximate number of entries in the index where the | |
115 ** first two columns match the first two columns of the sample. | |
116 ** And so forth. nLt is another list of integers that show the approximate | |
117 ** number of entries that are strictly less than the sample. The first | |
118 ** integer in nLt contains the number of entries in the index where the | |
119 ** left-most column is less than the left-most column of the sample. | |
120 ** The K-th integer in the nLt entry is the number of index entries | |
121 ** where the first K columns are less than the first K columns of the | |
122 ** sample. The nDLt column is like nLt except that it contains the | |
123 ** number of distinct entries in the index that are less than the | |
124 ** sample. | |
125 ** | |
126 ** There can be an arbitrary number of sqlite_stat4 entries per index. | |
127 ** The ANALYZE command will typically generate sqlite_stat4 tables | |
128 ** that contain between 10 and 40 samples which are distributed across | |
129 ** the key space, though not uniformly, and which include samples with | |
130 ** large nEq values. | |
131 ** | |
132 ** Format for sqlite_stat3 redux: | |
133 ** | |
134 ** The sqlite_stat3 table is like sqlite_stat4 except that it only | |
135 ** looks at the left-most column of the index. The sqlite_stat3.sample | |
136 ** column contains the actual value of the left-most column instead | |
137 ** of a blob encoding of the complete index key as is found in | |
138 ** sqlite_stat4.sample. The nEq, nLt, and nDLt entries of sqlite_stat3 | |
139 ** all contain just a single integer which is the same as the first | |
140 ** integer in the equivalent columns in sqlite_stat4. | |
141 */ | |
142 #ifndef SQLITE_OMIT_ANALYZE | |
143 #include "sqliteInt.h" | |
144 | |
145 #if defined(SQLITE_ENABLE_STAT4) | |
146 # define IsStat4 1 | |
147 # define IsStat3 0 | |
148 #elif defined(SQLITE_ENABLE_STAT3) | |
149 # define IsStat4 0 | |
150 # define IsStat3 1 | |
151 #else | |
152 # define IsStat4 0 | |
153 # define IsStat3 0 | |
154 # undef SQLITE_STAT4_SAMPLES | |
155 # define SQLITE_STAT4_SAMPLES 1 | |
156 #endif | |
157 #define IsStat34 (IsStat3+IsStat4) /* 1 for STAT3 or STAT4. 0 otherwise */ | |
158 | |
159 /* | |
160 ** This routine generates code that opens the sqlite_statN tables. | |
161 ** The sqlite_stat1 table is always relevant. sqlite_stat2 is now | |
162 ** obsolete. sqlite_stat3 and sqlite_stat4 are only opened when | |
163 ** appropriate compile-time options are provided. | |
164 ** | |
165 ** If the sqlite_statN tables do not previously exist, it is created. | |
166 ** | |
167 ** Argument zWhere may be a pointer to a buffer containing a table name, | |
168 ** or it may be a NULL pointer. If it is not NULL, then all entries in | |
169 ** the sqlite_statN tables associated with the named table are deleted. | |
170 ** If zWhere==0, then code is generated to delete all stat table entries. | |
171 */ | |
172 static void openStatTable( | |
173 Parse *pParse, /* Parsing context */ | |
174 int iDb, /* The database we are looking in */ | |
175 int iStatCur, /* Open the sqlite_stat1 table on this cursor */ | |
176 const char *zWhere, /* Delete entries for this table or index */ | |
177 const char *zWhereType /* Either "tbl" or "idx" */ | |
178 ){ | |
179 static const struct { | |
180 const char *zName; | |
181 const char *zCols; | |
182 } aTable[] = { | |
183 { "sqlite_stat1", "tbl,idx,stat" }, | |
184 #if defined(SQLITE_ENABLE_STAT4) | |
185 { "sqlite_stat4", "tbl,idx,neq,nlt,ndlt,sample" }, | |
186 { "sqlite_stat3", 0 }, | |
187 #elif defined(SQLITE_ENABLE_STAT3) | |
188 { "sqlite_stat3", "tbl,idx,neq,nlt,ndlt,sample" }, | |
189 { "sqlite_stat4", 0 }, | |
190 #else | |
191 { "sqlite_stat3", 0 }, | |
192 { "sqlite_stat4", 0 }, | |
193 #endif | |
194 }; | |
195 int i; | |
196 sqlite3 *db = pParse->db; | |
197 Db *pDb; | |
198 Vdbe *v = sqlite3GetVdbe(pParse); | |
199 int aRoot[ArraySize(aTable)]; | |
200 u8 aCreateTbl[ArraySize(aTable)]; | |
201 | |
202 if( v==0 ) return; | |
203 assert( sqlite3BtreeHoldsAllMutexes(db) ); | |
204 assert( sqlite3VdbeDb(v)==db ); | |
205 pDb = &db->aDb[iDb]; | |
206 | |
207 /* Create new statistic tables if they do not exist, or clear them | |
208 ** if they do already exist. | |
209 */ | |
210 for(i=0; i<ArraySize(aTable); i++){ | |
211 const char *zTab = aTable[i].zName; | |
212 Table *pStat; | |
213 if( (pStat = sqlite3FindTable(db, zTab, pDb->zName))==0 ){ | |
214 if( aTable[i].zCols ){ | |
215 /* The sqlite_statN table does not exist. Create it. Note that a | |
216 ** side-effect of the CREATE TABLE statement is to leave the rootpage | |
217 ** of the new table in register pParse->regRoot. This is important | |
218 ** because the OpenWrite opcode below will be needing it. */ | |
219 sqlite3NestedParse(pParse, | |
220 "CREATE TABLE %Q.%s(%s)", pDb->zName, zTab, aTable[i].zCols | |
221 ); | |
222 aRoot[i] = pParse->regRoot; | |
223 aCreateTbl[i] = OPFLAG_P2ISREG; | |
224 } | |
225 }else{ | |
226 /* The table already exists. If zWhere is not NULL, delete all entries | |
227 ** associated with the table zWhere. If zWhere is NULL, delete the | |
228 ** entire contents of the table. */ | |
229 aRoot[i] = pStat->tnum; | |
230 aCreateTbl[i] = 0; | |
231 sqlite3TableLock(pParse, iDb, aRoot[i], 1, zTab); | |
232 if( zWhere ){ | |
233 sqlite3NestedParse(pParse, | |
234 "DELETE FROM %Q.%s WHERE %s=%Q", | |
235 pDb->zName, zTab, zWhereType, zWhere | |
236 ); | |
237 }else{ | |
238 /* The sqlite_stat[134] table already exists. Delete all rows. */ | |
239 sqlite3VdbeAddOp2(v, OP_Clear, aRoot[i], iDb); | |
240 } | |
241 } | |
242 } | |
243 | |
244 /* Open the sqlite_stat[134] tables for writing. */ | |
245 for(i=0; aTable[i].zCols; i++){ | |
246 assert( i<ArraySize(aTable) ); | |
247 sqlite3VdbeAddOp4Int(v, OP_OpenWrite, iStatCur+i, aRoot[i], iDb, 3); | |
248 sqlite3VdbeChangeP5(v, aCreateTbl[i]); | |
249 VdbeComment((v, aTable[i].zName)); | |
250 } | |
251 } | |
252 | |
253 /* | |
254 ** Recommended number of samples for sqlite_stat4 | |
255 */ | |
256 #ifndef SQLITE_STAT4_SAMPLES | |
257 # define SQLITE_STAT4_SAMPLES 24 | |
258 #endif | |
259 | |
260 /* | |
261 ** Three SQL functions - stat_init(), stat_push(), and stat_get() - | |
262 ** share an instance of the following structure to hold their state | |
263 ** information. | |
264 */ | |
265 typedef struct Stat4Accum Stat4Accum; | |
266 typedef struct Stat4Sample Stat4Sample; | |
267 struct Stat4Sample { | |
268 tRowcnt *anEq; /* sqlite_stat4.nEq */ | |
269 tRowcnt *anDLt; /* sqlite_stat4.nDLt */ | |
270 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
271 tRowcnt *anLt; /* sqlite_stat4.nLt */ | |
272 union { | |
273 i64 iRowid; /* Rowid in main table of the key */ | |
274 u8 *aRowid; /* Key for WITHOUT ROWID tables */ | |
275 } u; | |
276 u32 nRowid; /* Sizeof aRowid[] */ | |
277 u8 isPSample; /* True if a periodic sample */ | |
278 int iCol; /* If !isPSample, the reason for inclusion */ | |
279 u32 iHash; /* Tiebreaker hash */ | |
280 #endif | |
281 }; | |
282 struct Stat4Accum { | |
283 tRowcnt nRow; /* Number of rows in the entire table */ | |
284 tRowcnt nPSample; /* How often to do a periodic sample */ | |
285 int nCol; /* Number of columns in index + pk/rowid */ | |
286 int nKeyCol; /* Number of index columns w/o the pk/rowid */ | |
287 int mxSample; /* Maximum number of samples to accumulate */ | |
288 Stat4Sample current; /* Current row as a Stat4Sample */ | |
289 u32 iPrn; /* Pseudo-random number used for sampling */ | |
290 Stat4Sample *aBest; /* Array of nCol best samples */ | |
291 int iMin; /* Index in a[] of entry with minimum score */ | |
292 int nSample; /* Current number of samples */ | |
293 int iGet; /* Index of current sample accessed by stat_get() */ | |
294 Stat4Sample *a; /* Array of mxSample Stat4Sample objects */ | |
295 sqlite3 *db; /* Database connection, for malloc() */ | |
296 }; | |
297 | |
298 /* Reclaim memory used by a Stat4Sample | |
299 */ | |
300 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
301 static void sampleClear(sqlite3 *db, Stat4Sample *p){ | |
302 assert( db!=0 ); | |
303 if( p->nRowid ){ | |
304 sqlite3DbFree(db, p->u.aRowid); | |
305 p->nRowid = 0; | |
306 } | |
307 } | |
308 #endif | |
309 | |
310 /* Initialize the BLOB value of a ROWID | |
311 */ | |
312 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
313 static void sampleSetRowid(sqlite3 *db, Stat4Sample *p, int n, const u8 *pData){ | |
314 assert( db!=0 ); | |
315 if( p->nRowid ) sqlite3DbFree(db, p->u.aRowid); | |
316 p->u.aRowid = sqlite3DbMallocRaw(db, n); | |
317 if( p->u.aRowid ){ | |
318 p->nRowid = n; | |
319 memcpy(p->u.aRowid, pData, n); | |
320 }else{ | |
321 p->nRowid = 0; | |
322 } | |
323 } | |
324 #endif | |
325 | |
326 /* Initialize the INTEGER value of a ROWID. | |
327 */ | |
328 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
329 static void sampleSetRowidInt64(sqlite3 *db, Stat4Sample *p, i64 iRowid){ | |
330 assert( db!=0 ); | |
331 if( p->nRowid ) sqlite3DbFree(db, p->u.aRowid); | |
332 p->nRowid = 0; | |
333 p->u.iRowid = iRowid; | |
334 } | |
335 #endif | |
336 | |
337 | |
338 /* | |
339 ** Copy the contents of object (*pFrom) into (*pTo). | |
340 */ | |
341 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
342 static void sampleCopy(Stat4Accum *p, Stat4Sample *pTo, Stat4Sample *pFrom){ | |
343 pTo->isPSample = pFrom->isPSample; | |
344 pTo->iCol = pFrom->iCol; | |
345 pTo->iHash = pFrom->iHash; | |
346 memcpy(pTo->anEq, pFrom->anEq, sizeof(tRowcnt)*p->nCol); | |
347 memcpy(pTo->anLt, pFrom->anLt, sizeof(tRowcnt)*p->nCol); | |
348 memcpy(pTo->anDLt, pFrom->anDLt, sizeof(tRowcnt)*p->nCol); | |
349 if( pFrom->nRowid ){ | |
350 sampleSetRowid(p->db, pTo, pFrom->nRowid, pFrom->u.aRowid); | |
351 }else{ | |
352 sampleSetRowidInt64(p->db, pTo, pFrom->u.iRowid); | |
353 } | |
354 } | |
355 #endif | |
356 | |
357 /* | |
358 ** Reclaim all memory of a Stat4Accum structure. | |
359 */ | |
360 static void stat4Destructor(void *pOld){ | |
361 Stat4Accum *p = (Stat4Accum*)pOld; | |
362 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
363 int i; | |
364 for(i=0; i<p->nCol; i++) sampleClear(p->db, p->aBest+i); | |
365 for(i=0; i<p->mxSample; i++) sampleClear(p->db, p->a+i); | |
366 sampleClear(p->db, &p->current); | |
367 #endif | |
368 sqlite3DbFree(p->db, p); | |
369 } | |
370 | |
371 /* | |
372 ** Implementation of the stat_init(N,K,C) SQL function. The three parameters | |
373 ** are: | |
374 ** N: The number of columns in the index including the rowid/pk (note 1) | |
375 ** K: The number of columns in the index excluding the rowid/pk. | |
376 ** C: The number of rows in the index (note 2) | |
377 ** | |
378 ** Note 1: In the special case of the covering index that implements a | |
379 ** WITHOUT ROWID table, N is the number of PRIMARY KEY columns, not the | |
380 ** total number of columns in the table. | |
381 ** | |
382 ** Note 2: C is only used for STAT3 and STAT4. | |
383 ** | |
384 ** For indexes on ordinary rowid tables, N==K+1. But for indexes on | |
385 ** WITHOUT ROWID tables, N=K+P where P is the number of columns in the | |
386 ** PRIMARY KEY of the table. The covering index that implements the | |
387 ** original WITHOUT ROWID table as N==K as a special case. | |
388 ** | |
389 ** This routine allocates the Stat4Accum object in heap memory. The return | |
390 ** value is a pointer to the Stat4Accum object. The datatype of the | |
391 ** return value is BLOB, but it is really just a pointer to the Stat4Accum | |
392 ** object. | |
393 */ | |
394 static void statInit( | |
395 sqlite3_context *context, | |
396 int argc, | |
397 sqlite3_value **argv | |
398 ){ | |
399 Stat4Accum *p; | |
400 int nCol; /* Number of columns in index being sampled */ | |
401 int nKeyCol; /* Number of key columns */ | |
402 int nColUp; /* nCol rounded up for alignment */ | |
403 int n; /* Bytes of space to allocate */ | |
404 sqlite3 *db; /* Database connection */ | |
405 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
406 int mxSample = SQLITE_STAT4_SAMPLES; | |
407 #endif | |
408 | |
409 /* Decode the three function arguments */ | |
410 UNUSED_PARAMETER(argc); | |
411 nCol = sqlite3_value_int(argv[0]); | |
412 assert( nCol>0 ); | |
413 nColUp = sizeof(tRowcnt)<8 ? (nCol+1)&~1 : nCol; | |
414 nKeyCol = sqlite3_value_int(argv[1]); | |
415 assert( nKeyCol<=nCol ); | |
416 assert( nKeyCol>0 ); | |
417 | |
418 /* Allocate the space required for the Stat4Accum object */ | |
419 n = sizeof(*p) | |
420 + sizeof(tRowcnt)*nColUp /* Stat4Accum.anEq */ | |
421 + sizeof(tRowcnt)*nColUp /* Stat4Accum.anDLt */ | |
422 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
423 + sizeof(tRowcnt)*nColUp /* Stat4Accum.anLt */ | |
424 + sizeof(Stat4Sample)*(nCol+mxSample) /* Stat4Accum.aBest[], a[] */ | |
425 + sizeof(tRowcnt)*3*nColUp*(nCol+mxSample) | |
426 #endif | |
427 ; | |
428 db = sqlite3_context_db_handle(context); | |
429 p = sqlite3DbMallocZero(db, n); | |
430 if( p==0 ){ | |
431 sqlite3_result_error_nomem(context); | |
432 return; | |
433 } | |
434 | |
435 p->db = db; | |
436 p->nRow = 0; | |
437 p->nCol = nCol; | |
438 p->nKeyCol = nKeyCol; | |
439 p->current.anDLt = (tRowcnt*)&p[1]; | |
440 p->current.anEq = &p->current.anDLt[nColUp]; | |
441 | |
442 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
443 { | |
444 u8 *pSpace; /* Allocated space not yet assigned */ | |
445 int i; /* Used to iterate through p->aSample[] */ | |
446 | |
447 p->iGet = -1; | |
448 p->mxSample = mxSample; | |
449 p->nPSample = (tRowcnt)(sqlite3_value_int64(argv[2])/(mxSample/3+1) + 1); | |
450 p->current.anLt = &p->current.anEq[nColUp]; | |
451 p->iPrn = nCol*0x689e962d ^ sqlite3_value_int(argv[2])*0xd0944565; | |
452 | |
453 /* Set up the Stat4Accum.a[] and aBest[] arrays */ | |
454 p->a = (struct Stat4Sample*)&p->current.anLt[nColUp]; | |
455 p->aBest = &p->a[mxSample]; | |
456 pSpace = (u8*)(&p->a[mxSample+nCol]); | |
457 for(i=0; i<(mxSample+nCol); i++){ | |
458 p->a[i].anEq = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); | |
459 p->a[i].anLt = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); | |
460 p->a[i].anDLt = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp); | |
461 } | |
462 assert( (pSpace - (u8*)p)==n ); | |
463 | |
464 for(i=0; i<nCol; i++){ | |
465 p->aBest[i].iCol = i; | |
466 } | |
467 } | |
468 #endif | |
469 | |
470 /* Return a pointer to the allocated object to the caller. Note that | |
471 ** only the pointer (the 2nd parameter) matters. The size of the object | |
472 ** (given by the 3rd parameter) is never used and can be any positive | |
473 ** value. */ | |
474 sqlite3_result_blob(context, p, sizeof(*p), stat4Destructor); | |
475 } | |
476 static const FuncDef statInitFuncdef = { | |
477 2+IsStat34, /* nArg */ | |
478 SQLITE_UTF8, /* funcFlags */ | |
479 0, /* pUserData */ | |
480 0, /* pNext */ | |
481 statInit, /* xFunc */ | |
482 0, /* xStep */ | |
483 0, /* xFinalize */ | |
484 "stat_init", /* zName */ | |
485 0, /* pHash */ | |
486 0 /* pDestructor */ | |
487 }; | |
488 | |
489 #ifdef SQLITE_ENABLE_STAT4 | |
490 /* | |
491 ** pNew and pOld are both candidate non-periodic samples selected for | |
492 ** the same column (pNew->iCol==pOld->iCol). Ignoring this column and | |
493 ** considering only any trailing columns and the sample hash value, this | |
494 ** function returns true if sample pNew is to be preferred over pOld. | |
495 ** In other words, if we assume that the cardinalities of the selected | |
496 ** column for pNew and pOld are equal, is pNew to be preferred over pOld. | |
497 ** | |
498 ** This function assumes that for each argument sample, the contents of | |
499 ** the anEq[] array from pSample->anEq[pSample->iCol+1] onwards are valid. | |
500 */ | |
501 static int sampleIsBetterPost( | |
502 Stat4Accum *pAccum, | |
503 Stat4Sample *pNew, | |
504 Stat4Sample *pOld | |
505 ){ | |
506 int nCol = pAccum->nCol; | |
507 int i; | |
508 assert( pNew->iCol==pOld->iCol ); | |
509 for(i=pNew->iCol+1; i<nCol; i++){ | |
510 if( pNew->anEq[i]>pOld->anEq[i] ) return 1; | |
511 if( pNew->anEq[i]<pOld->anEq[i] ) return 0; | |
512 } | |
513 if( pNew->iHash>pOld->iHash ) return 1; | |
514 return 0; | |
515 } | |
516 #endif | |
517 | |
518 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
519 /* | |
520 ** Return true if pNew is to be preferred over pOld. | |
521 ** | |
522 ** This function assumes that for each argument sample, the contents of | |
523 ** the anEq[] array from pSample->anEq[pSample->iCol] onwards are valid. | |
524 */ | |
525 static int sampleIsBetter( | |
526 Stat4Accum *pAccum, | |
527 Stat4Sample *pNew, | |
528 Stat4Sample *pOld | |
529 ){ | |
530 tRowcnt nEqNew = pNew->anEq[pNew->iCol]; | |
531 tRowcnt nEqOld = pOld->anEq[pOld->iCol]; | |
532 | |
533 assert( pOld->isPSample==0 && pNew->isPSample==0 ); | |
534 assert( IsStat4 || (pNew->iCol==0 && pOld->iCol==0) ); | |
535 | |
536 if( (nEqNew>nEqOld) ) return 1; | |
537 #ifdef SQLITE_ENABLE_STAT4 | |
538 if( nEqNew==nEqOld ){ | |
539 if( pNew->iCol<pOld->iCol ) return 1; | |
540 return (pNew->iCol==pOld->iCol && sampleIsBetterPost(pAccum, pNew, pOld)); | |
541 } | |
542 return 0; | |
543 #else | |
544 return (nEqNew==nEqOld && pNew->iHash>pOld->iHash); | |
545 #endif | |
546 } | |
547 | |
548 /* | |
549 ** Copy the contents of sample *pNew into the p->a[] array. If necessary, | |
550 ** remove the least desirable sample from p->a[] to make room. | |
551 */ | |
552 static void sampleInsert(Stat4Accum *p, Stat4Sample *pNew, int nEqZero){ | |
553 Stat4Sample *pSample = 0; | |
554 int i; | |
555 | |
556 assert( IsStat4 || nEqZero==0 ); | |
557 | |
558 #ifdef SQLITE_ENABLE_STAT4 | |
559 if( pNew->isPSample==0 ){ | |
560 Stat4Sample *pUpgrade = 0; | |
561 assert( pNew->anEq[pNew->iCol]>0 ); | |
562 | |
563 /* This sample is being added because the prefix that ends in column | |
564 ** iCol occurs many times in the table. However, if we have already | |
565 ** added a sample that shares this prefix, there is no need to add | |
566 ** this one. Instead, upgrade the priority of the highest priority | |
567 ** existing sample that shares this prefix. */ | |
568 for(i=p->nSample-1; i>=0; i--){ | |
569 Stat4Sample *pOld = &p->a[i]; | |
570 if( pOld->anEq[pNew->iCol]==0 ){ | |
571 if( pOld->isPSample ) return; | |
572 assert( pOld->iCol>pNew->iCol ); | |
573 assert( sampleIsBetter(p, pNew, pOld) ); | |
574 if( pUpgrade==0 || sampleIsBetter(p, pOld, pUpgrade) ){ | |
575 pUpgrade = pOld; | |
576 } | |
577 } | |
578 } | |
579 if( pUpgrade ){ | |
580 pUpgrade->iCol = pNew->iCol; | |
581 pUpgrade->anEq[pUpgrade->iCol] = pNew->anEq[pUpgrade->iCol]; | |
582 goto find_new_min; | |
583 } | |
584 } | |
585 #endif | |
586 | |
587 /* If necessary, remove sample iMin to make room for the new sample. */ | |
588 if( p->nSample>=p->mxSample ){ | |
589 Stat4Sample *pMin = &p->a[p->iMin]; | |
590 tRowcnt *anEq = pMin->anEq; | |
591 tRowcnt *anLt = pMin->anLt; | |
592 tRowcnt *anDLt = pMin->anDLt; | |
593 sampleClear(p->db, pMin); | |
594 memmove(pMin, &pMin[1], sizeof(p->a[0])*(p->nSample-p->iMin-1)); | |
595 pSample = &p->a[p->nSample-1]; | |
596 pSample->nRowid = 0; | |
597 pSample->anEq = anEq; | |
598 pSample->anDLt = anDLt; | |
599 pSample->anLt = anLt; | |
600 p->nSample = p->mxSample-1; | |
601 } | |
602 | |
603 /* The "rows less-than" for the rowid column must be greater than that | |
604 ** for the last sample in the p->a[] array. Otherwise, the samples would | |
605 ** be out of order. */ | |
606 #ifdef SQLITE_ENABLE_STAT4 | |
607 assert( p->nSample==0 | |
608 || pNew->anLt[p->nCol-1] > p->a[p->nSample-1].anLt[p->nCol-1] ); | |
609 #endif | |
610 | |
611 /* Insert the new sample */ | |
612 pSample = &p->a[p->nSample]; | |
613 sampleCopy(p, pSample, pNew); | |
614 p->nSample++; | |
615 | |
616 /* Zero the first nEqZero entries in the anEq[] array. */ | |
617 memset(pSample->anEq, 0, sizeof(tRowcnt)*nEqZero); | |
618 | |
619 #ifdef SQLITE_ENABLE_STAT4 | |
620 find_new_min: | |
621 #endif | |
622 if( p->nSample>=p->mxSample ){ | |
623 int iMin = -1; | |
624 for(i=0; i<p->mxSample; i++){ | |
625 if( p->a[i].isPSample ) continue; | |
626 if( iMin<0 || sampleIsBetter(p, &p->a[iMin], &p->a[i]) ){ | |
627 iMin = i; | |
628 } | |
629 } | |
630 assert( iMin>=0 ); | |
631 p->iMin = iMin; | |
632 } | |
633 } | |
634 #endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */ | |
635 | |
636 /* | |
637 ** Field iChng of the index being scanned has changed. So at this point | |
638 ** p->current contains a sample that reflects the previous row of the | |
639 ** index. The value of anEq[iChng] and subsequent anEq[] elements are | |
640 ** correct at this point. | |
641 */ | |
642 static void samplePushPrevious(Stat4Accum *p, int iChng){ | |
643 #ifdef SQLITE_ENABLE_STAT4 | |
644 int i; | |
645 | |
646 /* Check if any samples from the aBest[] array should be pushed | |
647 ** into IndexSample.a[] at this point. */ | |
648 for(i=(p->nCol-2); i>=iChng; i--){ | |
649 Stat4Sample *pBest = &p->aBest[i]; | |
650 pBest->anEq[i] = p->current.anEq[i]; | |
651 if( p->nSample<p->mxSample || sampleIsBetter(p, pBest, &p->a[p->iMin]) ){ | |
652 sampleInsert(p, pBest, i); | |
653 } | |
654 } | |
655 | |
656 /* Update the anEq[] fields of any samples already collected. */ | |
657 for(i=p->nSample-1; i>=0; i--){ | |
658 int j; | |
659 for(j=iChng; j<p->nCol; j++){ | |
660 if( p->a[i].anEq[j]==0 ) p->a[i].anEq[j] = p->current.anEq[j]; | |
661 } | |
662 } | |
663 #endif | |
664 | |
665 #if defined(SQLITE_ENABLE_STAT3) && !defined(SQLITE_ENABLE_STAT4) | |
666 if( iChng==0 ){ | |
667 tRowcnt nLt = p->current.anLt[0]; | |
668 tRowcnt nEq = p->current.anEq[0]; | |
669 | |
670 /* Check if this is to be a periodic sample. If so, add it. */ | |
671 if( (nLt/p->nPSample)!=(nLt+nEq)/p->nPSample ){ | |
672 p->current.isPSample = 1; | |
673 sampleInsert(p, &p->current, 0); | |
674 p->current.isPSample = 0; | |
675 }else | |
676 | |
677 /* Or if it is a non-periodic sample. Add it in this case too. */ | |
678 if( p->nSample<p->mxSample | |
679 || sampleIsBetter(p, &p->current, &p->a[p->iMin]) | |
680 ){ | |
681 sampleInsert(p, &p->current, 0); | |
682 } | |
683 } | |
684 #endif | |
685 | |
686 #ifndef SQLITE_ENABLE_STAT3_OR_STAT4 | |
687 UNUSED_PARAMETER( p ); | |
688 UNUSED_PARAMETER( iChng ); | |
689 #endif | |
690 } | |
691 | |
692 /* | |
693 ** Implementation of the stat_push SQL function: stat_push(P,C,R) | |
694 ** Arguments: | |
695 ** | |
696 ** P Pointer to the Stat4Accum object created by stat_init() | |
697 ** C Index of left-most column to differ from previous row | |
698 ** R Rowid for the current row. Might be a key record for | |
699 ** WITHOUT ROWID tables. | |
700 ** | |
701 ** This SQL function always returns NULL. It's purpose it to accumulate | |
702 ** statistical data and/or samples in the Stat4Accum object about the | |
703 ** index being analyzed. The stat_get() SQL function will later be used to | |
704 ** extract relevant information for constructing the sqlite_statN tables. | |
705 ** | |
706 ** The R parameter is only used for STAT3 and STAT4 | |
707 */ | |
708 static void statPush( | |
709 sqlite3_context *context, | |
710 int argc, | |
711 sqlite3_value **argv | |
712 ){ | |
713 int i; | |
714 | |
715 /* The three function arguments */ | |
716 Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]); | |
717 int iChng = sqlite3_value_int(argv[1]); | |
718 | |
719 UNUSED_PARAMETER( argc ); | |
720 UNUSED_PARAMETER( context ); | |
721 assert( p->nCol>0 ); | |
722 assert( iChng<p->nCol ); | |
723 | |
724 if( p->nRow==0 ){ | |
725 /* This is the first call to this function. Do initialization. */ | |
726 for(i=0; i<p->nCol; i++) p->current.anEq[i] = 1; | |
727 }else{ | |
728 /* Second and subsequent calls get processed here */ | |
729 samplePushPrevious(p, iChng); | |
730 | |
731 /* Update anDLt[], anLt[] and anEq[] to reflect the values that apply | |
732 ** to the current row of the index. */ | |
733 for(i=0; i<iChng; i++){ | |
734 p->current.anEq[i]++; | |
735 } | |
736 for(i=iChng; i<p->nCol; i++){ | |
737 p->current.anDLt[i]++; | |
738 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
739 p->current.anLt[i] += p->current.anEq[i]; | |
740 #endif | |
741 p->current.anEq[i] = 1; | |
742 } | |
743 } | |
744 p->nRow++; | |
745 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
746 if( sqlite3_value_type(argv[2])==SQLITE_INTEGER ){ | |
747 sampleSetRowidInt64(p->db, &p->current, sqlite3_value_int64(argv[2])); | |
748 }else{ | |
749 sampleSetRowid(p->db, &p->current, sqlite3_value_bytes(argv[2]), | |
750 sqlite3_value_blob(argv[2])); | |
751 } | |
752 p->current.iHash = p->iPrn = p->iPrn*1103515245 + 12345; | |
753 #endif | |
754 | |
755 #ifdef SQLITE_ENABLE_STAT4 | |
756 { | |
757 tRowcnt nLt = p->current.anLt[p->nCol-1]; | |
758 | |
759 /* Check if this is to be a periodic sample. If so, add it. */ | |
760 if( (nLt/p->nPSample)!=(nLt+1)/p->nPSample ){ | |
761 p->current.isPSample = 1; | |
762 p->current.iCol = 0; | |
763 sampleInsert(p, &p->current, p->nCol-1); | |
764 p->current.isPSample = 0; | |
765 } | |
766 | |
767 /* Update the aBest[] array. */ | |
768 for(i=0; i<(p->nCol-1); i++){ | |
769 p->current.iCol = i; | |
770 if( i>=iChng || sampleIsBetterPost(p, &p->current, &p->aBest[i]) ){ | |
771 sampleCopy(p, &p->aBest[i], &p->current); | |
772 } | |
773 } | |
774 } | |
775 #endif | |
776 } | |
777 static const FuncDef statPushFuncdef = { | |
778 2+IsStat34, /* nArg */ | |
779 SQLITE_UTF8, /* funcFlags */ | |
780 0, /* pUserData */ | |
781 0, /* pNext */ | |
782 statPush, /* xFunc */ | |
783 0, /* xStep */ | |
784 0, /* xFinalize */ | |
785 "stat_push", /* zName */ | |
786 0, /* pHash */ | |
787 0 /* pDestructor */ | |
788 }; | |
789 | |
790 #define STAT_GET_STAT1 0 /* "stat" column of stat1 table */ | |
791 #define STAT_GET_ROWID 1 /* "rowid" column of stat[34] entry */ | |
792 #define STAT_GET_NEQ 2 /* "neq" column of stat[34] entry */ | |
793 #define STAT_GET_NLT 3 /* "nlt" column of stat[34] entry */ | |
794 #define STAT_GET_NDLT 4 /* "ndlt" column of stat[34] entry */ | |
795 | |
796 /* | |
797 ** Implementation of the stat_get(P,J) SQL function. This routine is | |
798 ** used to query statistical information that has been gathered into | |
799 ** the Stat4Accum object by prior calls to stat_push(). The P parameter | |
800 ** has type BLOB but it is really just a pointer to the Stat4Accum object. | |
801 ** The content to returned is determined by the parameter J | |
802 ** which is one of the STAT_GET_xxxx values defined above. | |
803 ** | |
804 ** If neither STAT3 nor STAT4 are enabled, then J is always | |
805 ** STAT_GET_STAT1 and is hence omitted and this routine becomes | |
806 ** a one-parameter function, stat_get(P), that always returns the | |
807 ** stat1 table entry information. | |
808 */ | |
809 static void statGet( | |
810 sqlite3_context *context, | |
811 int argc, | |
812 sqlite3_value **argv | |
813 ){ | |
814 Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]); | |
815 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
816 /* STAT3 and STAT4 have a parameter on this routine. */ | |
817 int eCall = sqlite3_value_int(argv[1]); | |
818 assert( argc==2 ); | |
819 assert( eCall==STAT_GET_STAT1 || eCall==STAT_GET_NEQ | |
820 || eCall==STAT_GET_ROWID || eCall==STAT_GET_NLT | |
821 || eCall==STAT_GET_NDLT | |
822 ); | |
823 if( eCall==STAT_GET_STAT1 ) | |
824 #else | |
825 assert( argc==1 ); | |
826 #endif | |
827 { | |
828 /* Return the value to store in the "stat" column of the sqlite_stat1 | |
829 ** table for this index. | |
830 ** | |
831 ** The value is a string composed of a list of integers describing | |
832 ** the index. The first integer in the list is the total number of | |
833 ** entries in the index. There is one additional integer in the list | |
834 ** for each indexed column. This additional integer is an estimate of | |
835 ** the number of rows matched by a stabbing query on the index using | |
836 ** a key with the corresponding number of fields. In other words, | |
837 ** if the index is on columns (a,b) and the sqlite_stat1 value is | |
838 ** "100 10 2", then SQLite estimates that: | |
839 ** | |
840 ** * the index contains 100 rows, | |
841 ** * "WHERE a=?" matches 10 rows, and | |
842 ** * "WHERE a=? AND b=?" matches 2 rows. | |
843 ** | |
844 ** If D is the count of distinct values and K is the total number of | |
845 ** rows, then each estimate is computed as: | |
846 ** | |
847 ** I = (K+D-1)/D | |
848 */ | |
849 char *z; | |
850 int i; | |
851 | |
852 char *zRet = sqlite3MallocZero( (p->nKeyCol+1)*25 ); | |
853 if( zRet==0 ){ | |
854 sqlite3_result_error_nomem(context); | |
855 return; | |
856 } | |
857 | |
858 sqlite3_snprintf(24, zRet, "%llu", (u64)p->nRow); | |
859 z = zRet + sqlite3Strlen30(zRet); | |
860 for(i=0; i<p->nKeyCol; i++){ | |
861 u64 nDistinct = p->current.anDLt[i] + 1; | |
862 u64 iVal = (p->nRow + nDistinct - 1) / nDistinct; | |
863 sqlite3_snprintf(24, z, " %llu", iVal); | |
864 z += sqlite3Strlen30(z); | |
865 assert( p->current.anEq[i] ); | |
866 } | |
867 assert( z[0]=='\0' && z>zRet ); | |
868 | |
869 sqlite3_result_text(context, zRet, -1, sqlite3_free); | |
870 } | |
871 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
872 else if( eCall==STAT_GET_ROWID ){ | |
873 if( p->iGet<0 ){ | |
874 samplePushPrevious(p, 0); | |
875 p->iGet = 0; | |
876 } | |
877 if( p->iGet<p->nSample ){ | |
878 Stat4Sample *pS = p->a + p->iGet; | |
879 if( pS->nRowid==0 ){ | |
880 sqlite3_result_int64(context, pS->u.iRowid); | |
881 }else{ | |
882 sqlite3_result_blob(context, pS->u.aRowid, pS->nRowid, | |
883 SQLITE_TRANSIENT); | |
884 } | |
885 } | |
886 }else{ | |
887 tRowcnt *aCnt = 0; | |
888 | |
889 assert( p->iGet<p->nSample ); | |
890 switch( eCall ){ | |
891 case STAT_GET_NEQ: aCnt = p->a[p->iGet].anEq; break; | |
892 case STAT_GET_NLT: aCnt = p->a[p->iGet].anLt; break; | |
893 default: { | |
894 aCnt = p->a[p->iGet].anDLt; | |
895 p->iGet++; | |
896 break; | |
897 } | |
898 } | |
899 | |
900 if( IsStat3 ){ | |
901 sqlite3_result_int64(context, (i64)aCnt[0]); | |
902 }else{ | |
903 char *zRet = sqlite3MallocZero(p->nCol * 25); | |
904 if( zRet==0 ){ | |
905 sqlite3_result_error_nomem(context); | |
906 }else{ | |
907 int i; | |
908 char *z = zRet; | |
909 for(i=0; i<p->nCol; i++){ | |
910 sqlite3_snprintf(24, z, "%llu ", (u64)aCnt[i]); | |
911 z += sqlite3Strlen30(z); | |
912 } | |
913 assert( z[0]=='\0' && z>zRet ); | |
914 z[-1] = '\0'; | |
915 sqlite3_result_text(context, zRet, -1, sqlite3_free); | |
916 } | |
917 } | |
918 } | |
919 #endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */ | |
920 #ifndef SQLITE_DEBUG | |
921 UNUSED_PARAMETER( argc ); | |
922 #endif | |
923 } | |
924 static const FuncDef statGetFuncdef = { | |
925 1+IsStat34, /* nArg */ | |
926 SQLITE_UTF8, /* funcFlags */ | |
927 0, /* pUserData */ | |
928 0, /* pNext */ | |
929 statGet, /* xFunc */ | |
930 0, /* xStep */ | |
931 0, /* xFinalize */ | |
932 "stat_get", /* zName */ | |
933 0, /* pHash */ | |
934 0 /* pDestructor */ | |
935 }; | |
936 | |
937 static void callStatGet(Vdbe *v, int regStat4, int iParam, int regOut){ | |
938 assert( regOut!=regStat4 && regOut!=regStat4+1 ); | |
939 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
940 sqlite3VdbeAddOp2(v, OP_Integer, iParam, regStat4+1); | |
941 #elif SQLITE_DEBUG | |
942 assert( iParam==STAT_GET_STAT1 ); | |
943 #else | |
944 UNUSED_PARAMETER( iParam ); | |
945 #endif | |
946 sqlite3VdbeAddOp3(v, OP_Function, 0, regStat4, regOut); | |
947 sqlite3VdbeChangeP4(v, -1, (char*)&statGetFuncdef, P4_FUNCDEF); | |
948 sqlite3VdbeChangeP5(v, 1 + IsStat34); | |
949 } | |
950 | |
951 /* | |
952 ** Generate code to do an analysis of all indices associated with | |
953 ** a single table. | |
954 */ | |
955 static void analyzeOneTable( | |
956 Parse *pParse, /* Parser context */ | |
957 Table *pTab, /* Table whose indices are to be analyzed */ | |
958 Index *pOnlyIdx, /* If not NULL, only analyze this one index */ | |
959 int iStatCur, /* Index of VdbeCursor that writes the sqlite_stat1 table */ | |
960 int iMem, /* Available memory locations begin here */ | |
961 int iTab /* Next available cursor */ | |
962 ){ | |
963 sqlite3 *db = pParse->db; /* Database handle */ | |
964 Index *pIdx; /* An index to being analyzed */ | |
965 int iIdxCur; /* Cursor open on index being analyzed */ | |
966 int iTabCur; /* Table cursor */ | |
967 Vdbe *v; /* The virtual machine being built up */ | |
968 int i; /* Loop counter */ | |
969 int jZeroRows = -1; /* Jump from here if number of rows is zero */ | |
970 int iDb; /* Index of database containing pTab */ | |
971 u8 needTableCnt = 1; /* True to count the table */ | |
972 int regNewRowid = iMem++; /* Rowid for the inserted record */ | |
973 int regStat4 = iMem++; /* Register to hold Stat4Accum object */ | |
974 int regChng = iMem++; /* Index of changed index field */ | |
975 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
976 int regRowid = iMem++; /* Rowid argument passed to stat_push() */ | |
977 #endif | |
978 int regTemp = iMem++; /* Temporary use register */ | |
979 int regTabname = iMem++; /* Register containing table name */ | |
980 int regIdxname = iMem++; /* Register containing index name */ | |
981 int regStat1 = iMem++; /* Value for the stat column of sqlite_stat1 */ | |
982 int regPrev = iMem; /* MUST BE LAST (see below) */ | |
983 | |
984 pParse->nMem = MAX(pParse->nMem, iMem); | |
985 v = sqlite3GetVdbe(pParse); | |
986 if( v==0 || NEVER(pTab==0) ){ | |
987 return; | |
988 } | |
989 if( pTab->tnum==0 ){ | |
990 /* Do not gather statistics on views or virtual tables */ | |
991 return; | |
992 } | |
993 if( sqlite3_strnicmp(pTab->zName, "sqlite_", 7)==0 ){ | |
994 /* Do not gather statistics on system tables */ | |
995 return; | |
996 } | |
997 assert( sqlite3BtreeHoldsAllMutexes(db) ); | |
998 iDb = sqlite3SchemaToIndex(db, pTab->pSchema); | |
999 assert( iDb>=0 ); | |
1000 assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); | |
1001 #ifndef SQLITE_OMIT_AUTHORIZATION | |
1002 if( sqlite3AuthCheck(pParse, SQLITE_ANALYZE, pTab->zName, 0, | |
1003 db->aDb[iDb].zName ) ){ | |
1004 return; | |
1005 } | |
1006 #endif | |
1007 | |
1008 /* Establish a read-lock on the table at the shared-cache level. | |
1009 ** Open a read-only cursor on the table. Also allocate a cursor number | |
1010 ** to use for scanning indexes (iIdxCur). No index cursor is opened at | |
1011 ** this time though. */ | |
1012 sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); | |
1013 iTabCur = iTab++; | |
1014 iIdxCur = iTab++; | |
1015 pParse->nTab = MAX(pParse->nTab, iTab); | |
1016 sqlite3OpenTable(pParse, iTabCur, iDb, pTab, OP_OpenRead); | |
1017 sqlite3VdbeAddOp4(v, OP_String8, 0, regTabname, 0, pTab->zName, 0); | |
1018 | |
1019 for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){ | |
1020 int nCol; /* Number of columns in pIdx. "N" */ | |
1021 int addrRewind; /* Address of "OP_Rewind iIdxCur" */ | |
1022 int addrNextRow; /* Address of "next_row:" */ | |
1023 const char *zIdxName; /* Name of the index */ | |
1024 int nColTest; /* Number of columns to test for changes */ | |
1025 | |
1026 if( pOnlyIdx && pOnlyIdx!=pIdx ) continue; | |
1027 if( pIdx->pPartIdxWhere==0 ) needTableCnt = 0; | |
1028 if( !HasRowid(pTab) && IsPrimaryKeyIndex(pIdx) ){ | |
1029 nCol = pIdx->nKeyCol; | |
1030 zIdxName = pTab->zName; | |
1031 nColTest = nCol - 1; | |
1032 }else{ | |
1033 nCol = pIdx->nColumn; | |
1034 zIdxName = pIdx->zName; | |
1035 nColTest = pIdx->uniqNotNull ? pIdx->nKeyCol-1 : nCol-1; | |
1036 } | |
1037 | |
1038 /* Populate the register containing the index name. */ | |
1039 sqlite3VdbeAddOp4(v, OP_String8, 0, regIdxname, 0, zIdxName, 0); | |
1040 VdbeComment((v, "Analysis for %s.%s", pTab->zName, zIdxName)); | |
1041 | |
1042 /* | |
1043 ** Pseudo-code for loop that calls stat_push(): | |
1044 ** | |
1045 ** Rewind csr | |
1046 ** if eof(csr) goto end_of_scan; | |
1047 ** regChng = 0 | |
1048 ** goto chng_addr_0; | |
1049 ** | |
1050 ** next_row: | |
1051 ** regChng = 0 | |
1052 ** if( idx(0) != regPrev(0) ) goto chng_addr_0 | |
1053 ** regChng = 1 | |
1054 ** if( idx(1) != regPrev(1) ) goto chng_addr_1 | |
1055 ** ... | |
1056 ** regChng = N | |
1057 ** goto chng_addr_N | |
1058 ** | |
1059 ** chng_addr_0: | |
1060 ** regPrev(0) = idx(0) | |
1061 ** chng_addr_1: | |
1062 ** regPrev(1) = idx(1) | |
1063 ** ... | |
1064 ** | |
1065 ** endDistinctTest: | |
1066 ** regRowid = idx(rowid) | |
1067 ** stat_push(P, regChng, regRowid) | |
1068 ** Next csr | |
1069 ** if !eof(csr) goto next_row; | |
1070 ** | |
1071 ** end_of_scan: | |
1072 */ | |
1073 | |
1074 /* Make sure there are enough memory cells allocated to accommodate | |
1075 ** the regPrev array and a trailing rowid (the rowid slot is required | |
1076 ** when building a record to insert into the sample column of | |
1077 ** the sqlite_stat4 table. */ | |
1078 pParse->nMem = MAX(pParse->nMem, regPrev+nColTest); | |
1079 | |
1080 /* Open a read-only cursor on the index being analyzed. */ | |
1081 assert( iDb==sqlite3SchemaToIndex(db, pIdx->pSchema) ); | |
1082 sqlite3VdbeAddOp3(v, OP_OpenRead, iIdxCur, pIdx->tnum, iDb); | |
1083 sqlite3VdbeSetP4KeyInfo(pParse, pIdx); | |
1084 VdbeComment((v, "%s", pIdx->zName)); | |
1085 | |
1086 /* Invoke the stat_init() function. The arguments are: | |
1087 ** | |
1088 ** (1) the number of columns in the index including the rowid | |
1089 ** (or for a WITHOUT ROWID table, the number of PK columns), | |
1090 ** (2) the number of columns in the key without the rowid/pk | |
1091 ** (3) the number of rows in the index, | |
1092 ** | |
1093 ** | |
1094 ** The third argument is only used for STAT3 and STAT4 | |
1095 */ | |
1096 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1097 sqlite3VdbeAddOp2(v, OP_Count, iIdxCur, regStat4+3); | |
1098 #endif | |
1099 sqlite3VdbeAddOp2(v, OP_Integer, nCol, regStat4+1); | |
1100 sqlite3VdbeAddOp2(v, OP_Integer, pIdx->nKeyCol, regStat4+2); | |
1101 sqlite3VdbeAddOp3(v, OP_Function, 0, regStat4+1, regStat4); | |
1102 sqlite3VdbeChangeP4(v, -1, (char*)&statInitFuncdef, P4_FUNCDEF); | |
1103 sqlite3VdbeChangeP5(v, 2+IsStat34); | |
1104 | |
1105 /* Implementation of the following: | |
1106 ** | |
1107 ** Rewind csr | |
1108 ** if eof(csr) goto end_of_scan; | |
1109 ** regChng = 0 | |
1110 ** goto next_push_0; | |
1111 ** | |
1112 */ | |
1113 addrRewind = sqlite3VdbeAddOp1(v, OP_Rewind, iIdxCur); | |
1114 VdbeCoverage(v); | |
1115 sqlite3VdbeAddOp2(v, OP_Integer, 0, regChng); | |
1116 addrNextRow = sqlite3VdbeCurrentAddr(v); | |
1117 | |
1118 if( nColTest>0 ){ | |
1119 int endDistinctTest = sqlite3VdbeMakeLabel(v); | |
1120 int *aGotoChng; /* Array of jump instruction addresses */ | |
1121 aGotoChng = sqlite3DbMallocRaw(db, sizeof(int)*nColTest); | |
1122 if( aGotoChng==0 ) continue; | |
1123 | |
1124 /* | |
1125 ** next_row: | |
1126 ** regChng = 0 | |
1127 ** if( idx(0) != regPrev(0) ) goto chng_addr_0 | |
1128 ** regChng = 1 | |
1129 ** if( idx(1) != regPrev(1) ) goto chng_addr_1 | |
1130 ** ... | |
1131 ** regChng = N | |
1132 ** goto endDistinctTest | |
1133 */ | |
1134 sqlite3VdbeAddOp0(v, OP_Goto); | |
1135 addrNextRow = sqlite3VdbeCurrentAddr(v); | |
1136 if( nColTest==1 && pIdx->nKeyCol==1 && IsUniqueIndex(pIdx) ){ | |
1137 /* For a single-column UNIQUE index, once we have found a non-NULL | |
1138 ** row, we know that all the rest will be distinct, so skip | |
1139 ** subsequent distinctness tests. */ | |
1140 sqlite3VdbeAddOp2(v, OP_NotNull, regPrev, endDistinctTest); | |
1141 VdbeCoverage(v); | |
1142 } | |
1143 for(i=0; i<nColTest; i++){ | |
1144 char *pColl = (char*)sqlite3LocateCollSeq(pParse, pIdx->azColl[i]); | |
1145 sqlite3VdbeAddOp2(v, OP_Integer, i, regChng); | |
1146 sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, i, regTemp); | |
1147 aGotoChng[i] = | |
1148 sqlite3VdbeAddOp4(v, OP_Ne, regTemp, 0, regPrev+i, pColl, P4_COLLSEQ); | |
1149 sqlite3VdbeChangeP5(v, SQLITE_NULLEQ); | |
1150 VdbeCoverage(v); | |
1151 } | |
1152 sqlite3VdbeAddOp2(v, OP_Integer, nColTest, regChng); | |
1153 sqlite3VdbeAddOp2(v, OP_Goto, 0, endDistinctTest); | |
1154 | |
1155 | |
1156 /* | |
1157 ** chng_addr_0: | |
1158 ** regPrev(0) = idx(0) | |
1159 ** chng_addr_1: | |
1160 ** regPrev(1) = idx(1) | |
1161 ** ... | |
1162 */ | |
1163 sqlite3VdbeJumpHere(v, addrNextRow-1); | |
1164 for(i=0; i<nColTest; i++){ | |
1165 sqlite3VdbeJumpHere(v, aGotoChng[i]); | |
1166 sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, i, regPrev+i); | |
1167 } | |
1168 sqlite3VdbeResolveLabel(v, endDistinctTest); | |
1169 sqlite3DbFree(db, aGotoChng); | |
1170 } | |
1171 | |
1172 /* | |
1173 ** chng_addr_N: | |
1174 ** regRowid = idx(rowid) // STAT34 only | |
1175 ** stat_push(P, regChng, regRowid) // 3rd parameter STAT34 only | |
1176 ** Next csr | |
1177 ** if !eof(csr) goto next_row; | |
1178 */ | |
1179 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1180 assert( regRowid==(regStat4+2) ); | |
1181 if( HasRowid(pTab) ){ | |
1182 sqlite3VdbeAddOp2(v, OP_IdxRowid, iIdxCur, regRowid); | |
1183 }else{ | |
1184 Index *pPk = sqlite3PrimaryKeyIndex(pIdx->pTable); | |
1185 int j, k, regKey; | |
1186 regKey = sqlite3GetTempRange(pParse, pPk->nKeyCol); | |
1187 for(j=0; j<pPk->nKeyCol; j++){ | |
1188 k = sqlite3ColumnOfIndex(pIdx, pPk->aiColumn[j]); | |
1189 sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, k, regKey+j); | |
1190 VdbeComment((v, "%s", pTab->aCol[pPk->aiColumn[j]].zName)); | |
1191 } | |
1192 sqlite3VdbeAddOp3(v, OP_MakeRecord, regKey, pPk->nKeyCol, regRowid); | |
1193 sqlite3ReleaseTempRange(pParse, regKey, pPk->nKeyCol); | |
1194 } | |
1195 #endif | |
1196 assert( regChng==(regStat4+1) ); | |
1197 sqlite3VdbeAddOp3(v, OP_Function, 1, regStat4, regTemp); | |
1198 sqlite3VdbeChangeP4(v, -1, (char*)&statPushFuncdef, P4_FUNCDEF); | |
1199 sqlite3VdbeChangeP5(v, 2+IsStat34); | |
1200 sqlite3VdbeAddOp2(v, OP_Next, iIdxCur, addrNextRow); VdbeCoverage(v); | |
1201 | |
1202 /* Add the entry to the stat1 table. */ | |
1203 callStatGet(v, regStat4, STAT_GET_STAT1, regStat1); | |
1204 assert( "BBB"[0]==SQLITE_AFF_TEXT ); | |
1205 sqlite3VdbeAddOp4(v, OP_MakeRecord, regTabname, 3, regTemp, "BBB", 0); | |
1206 sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur, regNewRowid); | |
1207 sqlite3VdbeAddOp3(v, OP_Insert, iStatCur, regTemp, regNewRowid); | |
1208 sqlite3VdbeChangeP5(v, OPFLAG_APPEND); | |
1209 | |
1210 /* Add the entries to the stat3 or stat4 table. */ | |
1211 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1212 { | |
1213 int regEq = regStat1; | |
1214 int regLt = regStat1+1; | |
1215 int regDLt = regStat1+2; | |
1216 int regSample = regStat1+3; | |
1217 int regCol = regStat1+4; | |
1218 int regSampleRowid = regCol + nCol; | |
1219 int addrNext; | |
1220 int addrIsNull; | |
1221 u8 seekOp = HasRowid(pTab) ? OP_NotExists : OP_NotFound; | |
1222 | |
1223 pParse->nMem = MAX(pParse->nMem, regCol+nCol); | |
1224 | |
1225 addrNext = sqlite3VdbeCurrentAddr(v); | |
1226 callStatGet(v, regStat4, STAT_GET_ROWID, regSampleRowid); | |
1227 addrIsNull = sqlite3VdbeAddOp1(v, OP_IsNull, regSampleRowid); | |
1228 VdbeCoverage(v); | |
1229 callStatGet(v, regStat4, STAT_GET_NEQ, regEq); | |
1230 callStatGet(v, regStat4, STAT_GET_NLT, regLt); | |
1231 callStatGet(v, regStat4, STAT_GET_NDLT, regDLt); | |
1232 sqlite3VdbeAddOp4Int(v, seekOp, iTabCur, addrNext, regSampleRowid, 0); | |
1233 /* We know that the regSampleRowid row exists because it was read by | |
1234 ** the previous loop. Thus the not-found jump of seekOp will never | |
1235 ** be taken */ | |
1236 VdbeCoverageNeverTaken(v); | |
1237 #ifdef SQLITE_ENABLE_STAT3 | |
1238 sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, | |
1239 pIdx->aiColumn[0], regSample); | |
1240 #else | |
1241 for(i=0; i<nCol; i++){ | |
1242 i16 iCol = pIdx->aiColumn[i]; | |
1243 sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, iCol, regCol+i); | |
1244 } | |
1245 sqlite3VdbeAddOp3(v, OP_MakeRecord, regCol, nCol, regSample); | |
1246 #endif | |
1247 sqlite3VdbeAddOp3(v, OP_MakeRecord, regTabname, 6, regTemp); | |
1248 sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur+1, regNewRowid); | |
1249 sqlite3VdbeAddOp3(v, OP_Insert, iStatCur+1, regTemp, regNewRowid); | |
1250 sqlite3VdbeAddOp2(v, OP_Goto, 1, addrNext); /* P1==1 for end-of-loop */ | |
1251 sqlite3VdbeJumpHere(v, addrIsNull); | |
1252 } | |
1253 #endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */ | |
1254 | |
1255 /* End of analysis */ | |
1256 sqlite3VdbeJumpHere(v, addrRewind); | |
1257 } | |
1258 | |
1259 | |
1260 /* Create a single sqlite_stat1 entry containing NULL as the index | |
1261 ** name and the row count as the content. | |
1262 */ | |
1263 if( pOnlyIdx==0 && needTableCnt ){ | |
1264 VdbeComment((v, "%s", pTab->zName)); | |
1265 sqlite3VdbeAddOp2(v, OP_Count, iTabCur, regStat1); | |
1266 jZeroRows = sqlite3VdbeAddOp1(v, OP_IfNot, regStat1); VdbeCoverage(v); | |
1267 sqlite3VdbeAddOp2(v, OP_Null, 0, regIdxname); | |
1268 assert( "BBB"[0]==SQLITE_AFF_TEXT ); | |
1269 sqlite3VdbeAddOp4(v, OP_MakeRecord, regTabname, 3, regTemp, "BBB", 0); | |
1270 sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur, regNewRowid); | |
1271 sqlite3VdbeAddOp3(v, OP_Insert, iStatCur, regTemp, regNewRowid); | |
1272 sqlite3VdbeChangeP5(v, OPFLAG_APPEND); | |
1273 sqlite3VdbeJumpHere(v, jZeroRows); | |
1274 } | |
1275 } | |
1276 | |
1277 | |
1278 /* | |
1279 ** Generate code that will cause the most recent index analysis to | |
1280 ** be loaded into internal hash tables where is can be used. | |
1281 */ | |
1282 static void loadAnalysis(Parse *pParse, int iDb){ | |
1283 Vdbe *v = sqlite3GetVdbe(pParse); | |
1284 if( v ){ | |
1285 sqlite3VdbeAddOp1(v, OP_LoadAnalysis, iDb); | |
1286 } | |
1287 } | |
1288 | |
1289 /* | |
1290 ** Generate code that will do an analysis of an entire database | |
1291 */ | |
1292 static void analyzeDatabase(Parse *pParse, int iDb){ | |
1293 sqlite3 *db = pParse->db; | |
1294 Schema *pSchema = db->aDb[iDb].pSchema; /* Schema of database iDb */ | |
1295 HashElem *k; | |
1296 int iStatCur; | |
1297 int iMem; | |
1298 int iTab; | |
1299 | |
1300 sqlite3BeginWriteOperation(pParse, 0, iDb); | |
1301 iStatCur = pParse->nTab; | |
1302 pParse->nTab += 3; | |
1303 openStatTable(pParse, iDb, iStatCur, 0, 0); | |
1304 iMem = pParse->nMem+1; | |
1305 iTab = pParse->nTab; | |
1306 assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); | |
1307 for(k=sqliteHashFirst(&pSchema->tblHash); k; k=sqliteHashNext(k)){ | |
1308 Table *pTab = (Table*)sqliteHashData(k); | |
1309 analyzeOneTable(pParse, pTab, 0, iStatCur, iMem, iTab); | |
1310 } | |
1311 loadAnalysis(pParse, iDb); | |
1312 } | |
1313 | |
1314 /* | |
1315 ** Generate code that will do an analysis of a single table in | |
1316 ** a database. If pOnlyIdx is not NULL then it is a single index | |
1317 ** in pTab that should be analyzed. | |
1318 */ | |
1319 static void analyzeTable(Parse *pParse, Table *pTab, Index *pOnlyIdx){ | |
1320 int iDb; | |
1321 int iStatCur; | |
1322 | |
1323 assert( pTab!=0 ); | |
1324 assert( sqlite3BtreeHoldsAllMutexes(pParse->db) ); | |
1325 iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); | |
1326 sqlite3BeginWriteOperation(pParse, 0, iDb); | |
1327 iStatCur = pParse->nTab; | |
1328 pParse->nTab += 3; | |
1329 if( pOnlyIdx ){ | |
1330 openStatTable(pParse, iDb, iStatCur, pOnlyIdx->zName, "idx"); | |
1331 }else{ | |
1332 openStatTable(pParse, iDb, iStatCur, pTab->zName, "tbl"); | |
1333 } | |
1334 analyzeOneTable(pParse, pTab, pOnlyIdx, iStatCur,pParse->nMem+1,pParse->nTab); | |
1335 loadAnalysis(pParse, iDb); | |
1336 } | |
1337 | |
1338 /* | |
1339 ** Generate code for the ANALYZE command. The parser calls this routine | |
1340 ** when it recognizes an ANALYZE command. | |
1341 ** | |
1342 ** ANALYZE -- 1 | |
1343 ** ANALYZE <database> -- 2 | |
1344 ** ANALYZE ?<database>.?<tablename> -- 3 | |
1345 ** | |
1346 ** Form 1 causes all indices in all attached databases to be analyzed. | |
1347 ** Form 2 analyzes all indices the single database named. | |
1348 ** Form 3 analyzes all indices associated with the named table. | |
1349 */ | |
1350 void sqlite3Analyze(Parse *pParse, Token *pName1, Token *pName2){ | |
1351 sqlite3 *db = pParse->db; | |
1352 int iDb; | |
1353 int i; | |
1354 char *z, *zDb; | |
1355 Table *pTab; | |
1356 Index *pIdx; | |
1357 Token *pTableName; | |
1358 Vdbe *v; | |
1359 | |
1360 /* Read the database schema. If an error occurs, leave an error message | |
1361 ** and code in pParse and return NULL. */ | |
1362 assert( sqlite3BtreeHoldsAllMutexes(pParse->db) ); | |
1363 if( SQLITE_OK!=sqlite3ReadSchema(pParse) ){ | |
1364 return; | |
1365 } | |
1366 | |
1367 assert( pName2!=0 || pName1==0 ); | |
1368 if( pName1==0 ){ | |
1369 /* Form 1: Analyze everything */ | |
1370 for(i=0; i<db->nDb; i++){ | |
1371 if( i==1 ) continue; /* Do not analyze the TEMP database */ | |
1372 analyzeDatabase(pParse, i); | |
1373 } | |
1374 }else if( pName2->n==0 ){ | |
1375 /* Form 2: Analyze the database or table named */ | |
1376 iDb = sqlite3FindDb(db, pName1); | |
1377 if( iDb>=0 ){ | |
1378 analyzeDatabase(pParse, iDb); | |
1379 }else{ | |
1380 z = sqlite3NameFromToken(db, pName1); | |
1381 if( z ){ | |
1382 if( (pIdx = sqlite3FindIndex(db, z, 0))!=0 ){ | |
1383 analyzeTable(pParse, pIdx->pTable, pIdx); | |
1384 }else if( (pTab = sqlite3LocateTable(pParse, 0, z, 0))!=0 ){ | |
1385 analyzeTable(pParse, pTab, 0); | |
1386 } | |
1387 sqlite3DbFree(db, z); | |
1388 } | |
1389 } | |
1390 }else{ | |
1391 /* Form 3: Analyze the fully qualified table name */ | |
1392 iDb = sqlite3TwoPartName(pParse, pName1, pName2, &pTableName); | |
1393 if( iDb>=0 ){ | |
1394 zDb = db->aDb[iDb].zName; | |
1395 z = sqlite3NameFromToken(db, pTableName); | |
1396 if( z ){ | |
1397 if( (pIdx = sqlite3FindIndex(db, z, zDb))!=0 ){ | |
1398 analyzeTable(pParse, pIdx->pTable, pIdx); | |
1399 }else if( (pTab = sqlite3LocateTable(pParse, 0, z, zDb))!=0 ){ | |
1400 analyzeTable(pParse, pTab, 0); | |
1401 } | |
1402 sqlite3DbFree(db, z); | |
1403 } | |
1404 } | |
1405 } | |
1406 v = sqlite3GetVdbe(pParse); | |
1407 if( v ) sqlite3VdbeAddOp0(v, OP_Expire); | |
1408 } | |
1409 | |
1410 /* | |
1411 ** Used to pass information from the analyzer reader through to the | |
1412 ** callback routine. | |
1413 */ | |
1414 typedef struct analysisInfo analysisInfo; | |
1415 struct analysisInfo { | |
1416 sqlite3 *db; | |
1417 const char *zDatabase; | |
1418 }; | |
1419 | |
1420 /* | |
1421 ** The first argument points to a nul-terminated string containing a | |
1422 ** list of space separated integers. Read the first nOut of these into | |
1423 ** the array aOut[]. | |
1424 */ | |
1425 static void decodeIntArray( | |
1426 char *zIntArray, /* String containing int array to decode */ | |
1427 int nOut, /* Number of slots in aOut[] */ | |
1428 tRowcnt *aOut, /* Store integers here */ | |
1429 LogEst *aLog, /* Or, if aOut==0, here */ | |
1430 Index *pIndex /* Handle extra flags for this index, if not NULL */ | |
1431 ){ | |
1432 char *z = zIntArray; | |
1433 int c; | |
1434 int i; | |
1435 tRowcnt v; | |
1436 | |
1437 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1438 if( z==0 ) z = ""; | |
1439 #else | |
1440 assert( z!=0 ); | |
1441 #endif | |
1442 for(i=0; *z && i<nOut; i++){ | |
1443 v = 0; | |
1444 while( (c=z[0])>='0' && c<='9' ){ | |
1445 v = v*10 + c - '0'; | |
1446 z++; | |
1447 } | |
1448 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1449 if( aOut ) aOut[i] = v; | |
1450 if( aLog ) aLog[i] = sqlite3LogEst(v); | |
1451 #else | |
1452 assert( aOut==0 ); | |
1453 UNUSED_PARAMETER(aOut); | |
1454 assert( aLog!=0 ); | |
1455 aLog[i] = sqlite3LogEst(v); | |
1456 #endif | |
1457 if( *z==' ' ) z++; | |
1458 } | |
1459 #ifndef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1460 assert( pIndex!=0 ); | |
1461 #else | |
1462 if( pIndex ) | |
1463 #endif | |
1464 while( z[0] ){ | |
1465 if( sqlite3_strglob("unordered*", z)==0 ){ | |
1466 pIndex->bUnordered = 1; | |
1467 }else if( sqlite3_strglob("sz=[0-9]*", z)==0 ){ | |
1468 pIndex->szIdxRow = sqlite3LogEst(sqlite3Atoi(z+3)); | |
1469 } | |
1470 #ifdef SQLITE_ENABLE_COSTMULT | |
1471 else if( sqlite3_strglob("costmult=[0-9]*",z)==0 ){ | |
1472 pIndex->pTable->costMult = sqlite3LogEst(sqlite3Atoi(z+9)); | |
1473 } | |
1474 #endif | |
1475 while( z[0]!=0 && z[0]!=' ' ) z++; | |
1476 while( z[0]==' ' ) z++; | |
1477 } | |
1478 } | |
1479 | |
1480 /* | |
1481 ** This callback is invoked once for each index when reading the | |
1482 ** sqlite_stat1 table. | |
1483 ** | |
1484 ** argv[0] = name of the table | |
1485 ** argv[1] = name of the index (might be NULL) | |
1486 ** argv[2] = results of analysis - on integer for each column | |
1487 ** | |
1488 ** Entries for which argv[1]==NULL simply record the number of rows in | |
1489 ** the table. | |
1490 */ | |
1491 static int analysisLoader(void *pData, int argc, char **argv, char **NotUsed){ | |
1492 analysisInfo *pInfo = (analysisInfo*)pData; | |
1493 Index *pIndex; | |
1494 Table *pTable; | |
1495 const char *z; | |
1496 | |
1497 assert( argc==3 ); | |
1498 UNUSED_PARAMETER2(NotUsed, argc); | |
1499 | |
1500 if( argv==0 || argv[0]==0 || argv[2]==0 ){ | |
1501 return 0; | |
1502 } | |
1503 pTable = sqlite3FindTable(pInfo->db, argv[0], pInfo->zDatabase); | |
1504 if( pTable==0 ){ | |
1505 return 0; | |
1506 } | |
1507 if( argv[1]==0 ){ | |
1508 pIndex = 0; | |
1509 }else if( sqlite3_stricmp(argv[0],argv[1])==0 ){ | |
1510 pIndex = sqlite3PrimaryKeyIndex(pTable); | |
1511 }else{ | |
1512 pIndex = sqlite3FindIndex(pInfo->db, argv[1], pInfo->zDatabase); | |
1513 } | |
1514 z = argv[2]; | |
1515 | |
1516 if( pIndex ){ | |
1517 int nCol = pIndex->nKeyCol+1; | |
1518 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1519 tRowcnt * const aiRowEst = pIndex->aiRowEst = (tRowcnt*)sqlite3MallocZero( | |
1520 sizeof(tRowcnt) * nCol | |
1521 ); | |
1522 if( aiRowEst==0 ) pInfo->db->mallocFailed = 1; | |
1523 #else | |
1524 tRowcnt * const aiRowEst = 0; | |
1525 #endif | |
1526 pIndex->bUnordered = 0; | |
1527 decodeIntArray((char*)z, nCol, aiRowEst, pIndex->aiRowLogEst, pIndex); | |
1528 if( pIndex->pPartIdxWhere==0 ) pTable->nRowLogEst = pIndex->aiRowLogEst[0]; | |
1529 }else{ | |
1530 Index fakeIdx; | |
1531 fakeIdx.szIdxRow = pTable->szTabRow; | |
1532 #ifdef SQLITE_ENABLE_COSTMULT | |
1533 fakeIdx.pTable = pTable; | |
1534 #endif | |
1535 decodeIntArray((char*)z, 1, 0, &pTable->nRowLogEst, &fakeIdx); | |
1536 pTable->szTabRow = fakeIdx.szIdxRow; | |
1537 } | |
1538 | |
1539 return 0; | |
1540 } | |
1541 | |
1542 /* | |
1543 ** If the Index.aSample variable is not NULL, delete the aSample[] array | |
1544 ** and its contents. | |
1545 */ | |
1546 void sqlite3DeleteIndexSamples(sqlite3 *db, Index *pIdx){ | |
1547 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1548 if( pIdx->aSample ){ | |
1549 int j; | |
1550 for(j=0; j<pIdx->nSample; j++){ | |
1551 IndexSample *p = &pIdx->aSample[j]; | |
1552 sqlite3DbFree(db, p->p); | |
1553 } | |
1554 sqlite3DbFree(db, pIdx->aSample); | |
1555 } | |
1556 if( db && db->pnBytesFreed==0 ){ | |
1557 pIdx->nSample = 0; | |
1558 pIdx->aSample = 0; | |
1559 } | |
1560 #else | |
1561 UNUSED_PARAMETER(db); | |
1562 UNUSED_PARAMETER(pIdx); | |
1563 #endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */ | |
1564 } | |
1565 | |
1566 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1567 /* | |
1568 ** Populate the pIdx->aAvgEq[] array based on the samples currently | |
1569 ** stored in pIdx->aSample[]. | |
1570 */ | |
1571 static void initAvgEq(Index *pIdx){ | |
1572 if( pIdx ){ | |
1573 IndexSample *aSample = pIdx->aSample; | |
1574 IndexSample *pFinal = &aSample[pIdx->nSample-1]; | |
1575 int iCol; | |
1576 int nCol = 1; | |
1577 if( pIdx->nSampleCol>1 ){ | |
1578 /* If this is stat4 data, then calculate aAvgEq[] values for all | |
1579 ** sample columns except the last. The last is always set to 1, as | |
1580 ** once the trailing PK fields are considered all index keys are | |
1581 ** unique. */ | |
1582 nCol = pIdx->nSampleCol-1; | |
1583 pIdx->aAvgEq[nCol] = 1; | |
1584 } | |
1585 for(iCol=0; iCol<nCol; iCol++){ | |
1586 int nSample = pIdx->nSample; | |
1587 int i; /* Used to iterate through samples */ | |
1588 tRowcnt sumEq = 0; /* Sum of the nEq values */ | |
1589 tRowcnt avgEq = 0; | |
1590 tRowcnt nRow; /* Number of rows in index */ | |
1591 i64 nSum100 = 0; /* Number of terms contributing to sumEq */ | |
1592 i64 nDist100; /* Number of distinct values in index */ | |
1593 | |
1594 if( !pIdx->aiRowEst || iCol>=pIdx->nKeyCol || pIdx->aiRowEst[iCol+1]==0 ){ | |
1595 nRow = pFinal->anLt[iCol]; | |
1596 nDist100 = (i64)100 * pFinal->anDLt[iCol]; | |
1597 nSample--; | |
1598 }else{ | |
1599 nRow = pIdx->aiRowEst[0]; | |
1600 nDist100 = ((i64)100 * pIdx->aiRowEst[0]) / pIdx->aiRowEst[iCol+1]; | |
1601 } | |
1602 | |
1603 /* Set nSum to the number of distinct (iCol+1) field prefixes that | |
1604 ** occur in the stat4 table for this index. Set sumEq to the sum of | |
1605 ** the nEq values for column iCol for the same set (adding the value | |
1606 ** only once where there exist duplicate prefixes). */ | |
1607 for(i=0; i<nSample; i++){ | |
1608 if( i==(pIdx->nSample-1) | |
1609 || aSample[i].anDLt[iCol]!=aSample[i+1].anDLt[iCol] | |
1610 ){ | |
1611 sumEq += aSample[i].anEq[iCol]; | |
1612 nSum100 += 100; | |
1613 } | |
1614 } | |
1615 | |
1616 if( nDist100>nSum100 ){ | |
1617 avgEq = ((i64)100 * (nRow - sumEq))/(nDist100 - nSum100); | |
1618 } | |
1619 if( avgEq==0 ) avgEq = 1; | |
1620 pIdx->aAvgEq[iCol] = avgEq; | |
1621 } | |
1622 } | |
1623 } | |
1624 | |
1625 /* | |
1626 ** Look up an index by name. Or, if the name of a WITHOUT ROWID table | |
1627 ** is supplied instead, find the PRIMARY KEY index for that table. | |
1628 */ | |
1629 static Index *findIndexOrPrimaryKey( | |
1630 sqlite3 *db, | |
1631 const char *zName, | |
1632 const char *zDb | |
1633 ){ | |
1634 Index *pIdx = sqlite3FindIndex(db, zName, zDb); | |
1635 if( pIdx==0 ){ | |
1636 Table *pTab = sqlite3FindTable(db, zName, zDb); | |
1637 if( pTab && !HasRowid(pTab) ) pIdx = sqlite3PrimaryKeyIndex(pTab); | |
1638 } | |
1639 return pIdx; | |
1640 } | |
1641 | |
1642 /* | |
1643 ** Load the content from either the sqlite_stat4 or sqlite_stat3 table | |
1644 ** into the relevant Index.aSample[] arrays. | |
1645 ** | |
1646 ** Arguments zSql1 and zSql2 must point to SQL statements that return | |
1647 ** data equivalent to the following (statements are different for stat3, | |
1648 ** see the caller of this function for details): | |
1649 ** | |
1650 ** zSql1: SELECT idx,count(*) FROM %Q.sqlite_stat4 GROUP BY idx | |
1651 ** zSql2: SELECT idx,neq,nlt,ndlt,sample FROM %Q.sqlite_stat4 | |
1652 ** | |
1653 ** where %Q is replaced with the database name before the SQL is executed. | |
1654 */ | |
1655 static int loadStatTbl( | |
1656 sqlite3 *db, /* Database handle */ | |
1657 int bStat3, /* Assume single column records only */ | |
1658 const char *zSql1, /* SQL statement 1 (see above) */ | |
1659 const char *zSql2, /* SQL statement 2 (see above) */ | |
1660 const char *zDb /* Database name (e.g. "main") */ | |
1661 ){ | |
1662 int rc; /* Result codes from subroutines */ | |
1663 sqlite3_stmt *pStmt = 0; /* An SQL statement being run */ | |
1664 char *zSql; /* Text of the SQL statement */ | |
1665 Index *pPrevIdx = 0; /* Previous index in the loop */ | |
1666 IndexSample *pSample; /* A slot in pIdx->aSample[] */ | |
1667 | |
1668 assert( db->lookaside.bEnabled==0 ); | |
1669 zSql = sqlite3MPrintf(db, zSql1, zDb); | |
1670 if( !zSql ){ | |
1671 return SQLITE_NOMEM; | |
1672 } | |
1673 rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); | |
1674 sqlite3DbFree(db, zSql); | |
1675 if( rc ) return rc; | |
1676 | |
1677 while( sqlite3_step(pStmt)==SQLITE_ROW ){ | |
1678 int nIdxCol = 1; /* Number of columns in stat4 records */ | |
1679 | |
1680 char *zIndex; /* Index name */ | |
1681 Index *pIdx; /* Pointer to the index object */ | |
1682 int nSample; /* Number of samples */ | |
1683 int nByte; /* Bytes of space required */ | |
1684 int i; /* Bytes of space required */ | |
1685 tRowcnt *pSpace; | |
1686 | |
1687 zIndex = (char *)sqlite3_column_text(pStmt, 0); | |
1688 if( zIndex==0 ) continue; | |
1689 nSample = sqlite3_column_int(pStmt, 1); | |
1690 pIdx = findIndexOrPrimaryKey(db, zIndex, zDb); | |
1691 assert( pIdx==0 || bStat3 || pIdx->nSample==0 ); | |
1692 /* Index.nSample is non-zero at this point if data has already been | |
1693 ** loaded from the stat4 table. In this case ignore stat3 data. */ | |
1694 if( pIdx==0 || pIdx->nSample ) continue; | |
1695 if( bStat3==0 ){ | |
1696 assert( !HasRowid(pIdx->pTable) || pIdx->nColumn==pIdx->nKeyCol+1 ); | |
1697 if( !HasRowid(pIdx->pTable) && IsPrimaryKeyIndex(pIdx) ){ | |
1698 nIdxCol = pIdx->nKeyCol; | |
1699 }else{ | |
1700 nIdxCol = pIdx->nColumn; | |
1701 } | |
1702 } | |
1703 pIdx->nSampleCol = nIdxCol; | |
1704 nByte = sizeof(IndexSample) * nSample; | |
1705 nByte += sizeof(tRowcnt) * nIdxCol * 3 * nSample; | |
1706 nByte += nIdxCol * sizeof(tRowcnt); /* Space for Index.aAvgEq[] */ | |
1707 | |
1708 pIdx->aSample = sqlite3DbMallocZero(db, nByte); | |
1709 if( pIdx->aSample==0 ){ | |
1710 sqlite3_finalize(pStmt); | |
1711 return SQLITE_NOMEM; | |
1712 } | |
1713 pSpace = (tRowcnt*)&pIdx->aSample[nSample]; | |
1714 pIdx->aAvgEq = pSpace; pSpace += nIdxCol; | |
1715 for(i=0; i<nSample; i++){ | |
1716 pIdx->aSample[i].anEq = pSpace; pSpace += nIdxCol; | |
1717 pIdx->aSample[i].anLt = pSpace; pSpace += nIdxCol; | |
1718 pIdx->aSample[i].anDLt = pSpace; pSpace += nIdxCol; | |
1719 } | |
1720 assert( ((u8*)pSpace)-nByte==(u8*)(pIdx->aSample) ); | |
1721 } | |
1722 rc = sqlite3_finalize(pStmt); | |
1723 if( rc ) return rc; | |
1724 | |
1725 zSql = sqlite3MPrintf(db, zSql2, zDb); | |
1726 if( !zSql ){ | |
1727 return SQLITE_NOMEM; | |
1728 } | |
1729 rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); | |
1730 sqlite3DbFree(db, zSql); | |
1731 if( rc ) return rc; | |
1732 | |
1733 while( sqlite3_step(pStmt)==SQLITE_ROW ){ | |
1734 char *zIndex; /* Index name */ | |
1735 Index *pIdx; /* Pointer to the index object */ | |
1736 int nCol = 1; /* Number of columns in index */ | |
1737 | |
1738 zIndex = (char *)sqlite3_column_text(pStmt, 0); | |
1739 if( zIndex==0 ) continue; | |
1740 pIdx = findIndexOrPrimaryKey(db, zIndex, zDb); | |
1741 if( pIdx==0 ) continue; | |
1742 /* This next condition is true if data has already been loaded from | |
1743 ** the sqlite_stat4 table. In this case ignore stat3 data. */ | |
1744 nCol = pIdx->nSampleCol; | |
1745 if( bStat3 && nCol>1 ) continue; | |
1746 if( pIdx!=pPrevIdx ){ | |
1747 initAvgEq(pPrevIdx); | |
1748 pPrevIdx = pIdx; | |
1749 } | |
1750 pSample = &pIdx->aSample[pIdx->nSample]; | |
1751 decodeIntArray((char*)sqlite3_column_text(pStmt,1),nCol,pSample->anEq,0,0); | |
1752 decodeIntArray((char*)sqlite3_column_text(pStmt,2),nCol,pSample->anLt,0,0); | |
1753 decodeIntArray((char*)sqlite3_column_text(pStmt,3),nCol,pSample->anDLt,0,0); | |
1754 | |
1755 /* Take a copy of the sample. Add two 0x00 bytes the end of the buffer. | |
1756 ** This is in case the sample record is corrupted. In that case, the | |
1757 ** sqlite3VdbeRecordCompare() may read up to two varints past the | |
1758 ** end of the allocated buffer before it realizes it is dealing with | |
1759 ** a corrupt record. Adding the two 0x00 bytes prevents this from causing | |
1760 ** a buffer overread. */ | |
1761 pSample->n = sqlite3_column_bytes(pStmt, 4); | |
1762 pSample->p = sqlite3DbMallocZero(db, pSample->n + 2); | |
1763 if( pSample->p==0 ){ | |
1764 sqlite3_finalize(pStmt); | |
1765 return SQLITE_NOMEM; | |
1766 } | |
1767 memcpy(pSample->p, sqlite3_column_blob(pStmt, 4), pSample->n); | |
1768 pIdx->nSample++; | |
1769 } | |
1770 rc = sqlite3_finalize(pStmt); | |
1771 if( rc==SQLITE_OK ) initAvgEq(pPrevIdx); | |
1772 return rc; | |
1773 } | |
1774 | |
1775 /* | |
1776 ** Load content from the sqlite_stat4 and sqlite_stat3 tables into | |
1777 ** the Index.aSample[] arrays of all indices. | |
1778 */ | |
1779 static int loadStat4(sqlite3 *db, const char *zDb){ | |
1780 int rc = SQLITE_OK; /* Result codes from subroutines */ | |
1781 | |
1782 assert( db->lookaside.bEnabled==0 ); | |
1783 if( sqlite3FindTable(db, "sqlite_stat4", zDb) ){ | |
1784 rc = loadStatTbl(db, 0, | |
1785 "SELECT idx,count(*) FROM %Q.sqlite_stat4 GROUP BY idx", | |
1786 "SELECT idx,neq,nlt,ndlt,sample FROM %Q.sqlite_stat4", | |
1787 zDb | |
1788 ); | |
1789 } | |
1790 | |
1791 if( rc==SQLITE_OK && sqlite3FindTable(db, "sqlite_stat3", zDb) ){ | |
1792 rc = loadStatTbl(db, 1, | |
1793 "SELECT idx,count(*) FROM %Q.sqlite_stat3 GROUP BY idx", | |
1794 "SELECT idx,neq,nlt,ndlt,sqlite_record(sample) FROM %Q.sqlite_stat3", | |
1795 zDb | |
1796 ); | |
1797 } | |
1798 | |
1799 return rc; | |
1800 } | |
1801 #endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */ | |
1802 | |
1803 /* | |
1804 ** Load the content of the sqlite_stat1 and sqlite_stat3/4 tables. The | |
1805 ** contents of sqlite_stat1 are used to populate the Index.aiRowEst[] | |
1806 ** arrays. The contents of sqlite_stat3/4 are used to populate the | |
1807 ** Index.aSample[] arrays. | |
1808 ** | |
1809 ** If the sqlite_stat1 table is not present in the database, SQLITE_ERROR | |
1810 ** is returned. In this case, even if SQLITE_ENABLE_STAT3/4 was defined | |
1811 ** during compilation and the sqlite_stat3/4 table is present, no data is | |
1812 ** read from it. | |
1813 ** | |
1814 ** If SQLITE_ENABLE_STAT3/4 was defined during compilation and the | |
1815 ** sqlite_stat4 table is not present in the database, SQLITE_ERROR is | |
1816 ** returned. However, in this case, data is read from the sqlite_stat1 | |
1817 ** table (if it is present) before returning. | |
1818 ** | |
1819 ** If an OOM error occurs, this function always sets db->mallocFailed. | |
1820 ** This means if the caller does not care about other errors, the return | |
1821 ** code may be ignored. | |
1822 */ | |
1823 int sqlite3AnalysisLoad(sqlite3 *db, int iDb){ | |
1824 analysisInfo sInfo; | |
1825 HashElem *i; | |
1826 char *zSql; | |
1827 int rc; | |
1828 | |
1829 assert( iDb>=0 && iDb<db->nDb ); | |
1830 assert( db->aDb[iDb].pBt!=0 ); | |
1831 | |
1832 /* Clear any prior statistics */ | |
1833 assert( sqlite3SchemaMutexHeld(db, iDb, 0) ); | |
1834 for(i=sqliteHashFirst(&db->aDb[iDb].pSchema->idxHash);i;i=sqliteHashNext(i)){ | |
1835 Index *pIdx = sqliteHashData(i); | |
1836 sqlite3DefaultRowEst(pIdx); | |
1837 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1838 sqlite3DeleteIndexSamples(db, pIdx); | |
1839 pIdx->aSample = 0; | |
1840 #endif | |
1841 } | |
1842 | |
1843 /* Check to make sure the sqlite_stat1 table exists */ | |
1844 sInfo.db = db; | |
1845 sInfo.zDatabase = db->aDb[iDb].zName; | |
1846 if( sqlite3FindTable(db, "sqlite_stat1", sInfo.zDatabase)==0 ){ | |
1847 return SQLITE_ERROR; | |
1848 } | |
1849 | |
1850 /* Load new statistics out of the sqlite_stat1 table */ | |
1851 zSql = sqlite3MPrintf(db, | |
1852 "SELECT tbl,idx,stat FROM %Q.sqlite_stat1", sInfo.zDatabase); | |
1853 if( zSql==0 ){ | |
1854 rc = SQLITE_NOMEM; | |
1855 }else{ | |
1856 rc = sqlite3_exec(db, zSql, analysisLoader, &sInfo, 0); | |
1857 sqlite3DbFree(db, zSql); | |
1858 } | |
1859 | |
1860 | |
1861 /* Load the statistics from the sqlite_stat4 table. */ | |
1862 #ifdef SQLITE_ENABLE_STAT3_OR_STAT4 | |
1863 if( rc==SQLITE_OK ){ | |
1864 int lookasideEnabled = db->lookaside.bEnabled; | |
1865 db->lookaside.bEnabled = 0; | |
1866 rc = loadStat4(db, sInfo.zDatabase); | |
1867 db->lookaside.bEnabled = lookasideEnabled; | |
1868 } | |
1869 for(i=sqliteHashFirst(&db->aDb[iDb].pSchema->idxHash);i;i=sqliteHashNext(i)){ | |
1870 Index *pIdx = sqliteHashData(i); | |
1871 sqlite3_free(pIdx->aiRowEst); | |
1872 pIdx->aiRowEst = 0; | |
1873 } | |
1874 #endif | |
1875 | |
1876 if( rc==SQLITE_NOMEM ){ | |
1877 db->mallocFailed = 1; | |
1878 } | |
1879 return rc; | |
1880 } | |
1881 | |
1882 | |
1883 #endif /* SQLITE_OMIT_ANALYZE */ | |
OLD | NEW |