Fix a problem causing the recovery extension to use excessive memory and CPU time...
[sqlite.git] / src / rowset.c
blob5956cb2ad82afa0caef631f9efc9ec6e677d317b
1 /*
2 ** 2008 December 3
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.
11 *************************************************************************
13 ** This module implements an object we call a "RowSet".
15 ** The RowSet object is a collection of rowids. Rowids
16 ** are inserted into the RowSet in an arbitrary order. Inserts
17 ** can be intermixed with tests to see if a given rowid has been
18 ** previously inserted into the RowSet.
20 ** After all inserts are finished, it is possible to extract the
21 ** elements of the RowSet in sorted order. Once this extraction
22 ** process has started, no new elements may be inserted.
24 ** Hence, the primitive operations for a RowSet are:
26 ** CREATE
27 ** INSERT
28 ** TEST
29 ** SMALLEST
30 ** DESTROY
32 ** The CREATE and DESTROY primitives are the constructor and destructor,
33 ** obviously. The INSERT primitive adds a new element to the RowSet.
34 ** TEST checks to see if an element is already in the RowSet. SMALLEST
35 ** extracts the least value from the RowSet.
37 ** The INSERT primitive might allocate additional memory. Memory is
38 ** allocated in chunks so most INSERTs do no allocation. There is an
39 ** upper bound on the size of allocated memory. No memory is freed
40 ** until DESTROY.
42 ** The TEST primitive includes a "batch" number. The TEST primitive
43 ** will only see elements that were inserted before the last change
44 ** in the batch number. In other words, if an INSERT occurs between
45 ** two TESTs where the TESTs have the same batch number, then the
46 ** value added by the INSERT will not be visible to the second TEST.
47 ** The initial batch number is zero, so if the very first TEST contains
48 ** a non-zero batch number, it will see all prior INSERTs.
50 ** No INSERTs may occurs after a SMALLEST. An assertion will fail if
51 ** that is attempted.
53 ** The cost of an INSERT is roughly constant. (Sometimes new memory
54 ** has to be allocated on an INSERT.) The cost of a TEST with a new
55 ** batch number is O(NlogN) where N is the number of elements in the RowSet.
56 ** The cost of a TEST using the same batch number is O(logN). The cost
57 ** of the first SMALLEST is O(NlogN). Second and subsequent SMALLEST
58 ** primitives are constant time. The cost of DESTROY is O(N).
60 ** TEST and SMALLEST may not be used by the same RowSet. This used to
61 ** be possible, but the feature was not used, so it was removed in order
62 ** to simplify the code.
64 #include "sqliteInt.h"
68 ** Target size for allocation chunks.
70 #define ROWSET_ALLOCATION_SIZE 1024
73 ** The number of rowset entries per allocation chunk.
75 #define ROWSET_ENTRY_PER_CHUNK \
76 ((ROWSET_ALLOCATION_SIZE-8)/sizeof(struct RowSetEntry))
79 ** Each entry in a RowSet is an instance of the following object.
81 ** This same object is reused to store a linked list of trees of RowSetEntry
82 ** objects. In that alternative use, pRight points to the next entry
83 ** in the list, pLeft points to the tree, and v is unused. The
84 ** RowSet.pForest value points to the head of this forest list.
86 struct RowSetEntry {
87 i64 v; /* ROWID value for this entry */
88 struct RowSetEntry *pRight; /* Right subtree (larger entries) or list */
89 struct RowSetEntry *pLeft; /* Left subtree (smaller entries) */
93 ** RowSetEntry objects are allocated in large chunks (instances of the
94 ** following structure) to reduce memory allocation overhead. The
95 ** chunks are kept on a linked list so that they can be deallocated
96 ** when the RowSet is destroyed.
98 struct RowSetChunk {
99 struct RowSetChunk *pNextChunk; /* Next chunk on list of them all */
100 struct RowSetEntry aEntry[ROWSET_ENTRY_PER_CHUNK]; /* Allocated entries */
104 ** A RowSet in an instance of the following structure.
106 ** A typedef of this structure if found in sqliteInt.h.
108 struct RowSet {
109 struct RowSetChunk *pChunk; /* List of all chunk allocations */
110 sqlite3 *db; /* The database connection */
111 struct RowSetEntry *pEntry; /* List of entries using pRight */
112 struct RowSetEntry *pLast; /* Last entry on the pEntry list */
113 struct RowSetEntry *pFresh; /* Source of new entry objects */
114 struct RowSetEntry *pForest; /* List of binary trees of entries */
115 u16 nFresh; /* Number of objects on pFresh */
116 u16 rsFlags; /* Various flags */
117 int iBatch; /* Current insert batch */
121 ** Allowed values for RowSet.rsFlags
123 #define ROWSET_SORTED 0x01 /* True if RowSet.pEntry is sorted */
124 #define ROWSET_NEXT 0x02 /* True if sqlite3RowSetNext() has been called */
127 ** Allocate a RowSet object. Return NULL if a memory allocation
128 ** error occurs.
130 RowSet *sqlite3RowSetInit(sqlite3 *db){
131 RowSet *p = sqlite3DbMallocRawNN(db, sizeof(*p));
132 if( p ){
133 int N = sqlite3DbMallocSize(db, p);
134 p->pChunk = 0;
135 p->db = db;
136 p->pEntry = 0;
137 p->pLast = 0;
138 p->pForest = 0;
139 p->pFresh = (struct RowSetEntry*)(ROUND8(sizeof(*p)) + (char*)p);
140 p->nFresh = (u16)((N - ROUND8(sizeof(*p)))/sizeof(struct RowSetEntry));
141 p->rsFlags = ROWSET_SORTED;
142 p->iBatch = 0;
144 return p;
148 ** Deallocate all chunks from a RowSet. This frees all memory that
149 ** the RowSet has allocated over its lifetime. This routine is
150 ** the destructor for the RowSet.
152 void sqlite3RowSetClear(void *pArg){
153 RowSet *p = (RowSet*)pArg;
154 struct RowSetChunk *pChunk, *pNextChunk;
155 for(pChunk=p->pChunk; pChunk; pChunk = pNextChunk){
156 pNextChunk = pChunk->pNextChunk;
157 sqlite3DbFree(p->db, pChunk);
159 p->pChunk = 0;
160 p->nFresh = 0;
161 p->pEntry = 0;
162 p->pLast = 0;
163 p->pForest = 0;
164 p->rsFlags = ROWSET_SORTED;
168 ** Deallocate all chunks from a RowSet. This frees all memory that
169 ** the RowSet has allocated over its lifetime. This routine is
170 ** the destructor for the RowSet.
172 void sqlite3RowSetDelete(void *pArg){
173 sqlite3RowSetClear(pArg);
174 sqlite3DbFree(((RowSet*)pArg)->db, pArg);
178 ** Allocate a new RowSetEntry object that is associated with the
179 ** given RowSet. Return a pointer to the new and completely uninitialized
180 ** object.
182 ** In an OOM situation, the RowSet.db->mallocFailed flag is set and this
183 ** routine returns NULL.
185 static struct RowSetEntry *rowSetEntryAlloc(RowSet *p){
186 assert( p!=0 );
187 if( p->nFresh==0 ){ /*OPTIMIZATION-IF-FALSE*/
188 /* We could allocate a fresh RowSetEntry each time one is needed, but it
189 ** is more efficient to pull a preallocated entry from the pool */
190 struct RowSetChunk *pNew;
191 pNew = sqlite3DbMallocRawNN(p->db, sizeof(*pNew));
192 if( pNew==0 ){
193 return 0;
195 pNew->pNextChunk = p->pChunk;
196 p->pChunk = pNew;
197 p->pFresh = pNew->aEntry;
198 p->nFresh = ROWSET_ENTRY_PER_CHUNK;
200 p->nFresh--;
201 return p->pFresh++;
205 ** Insert a new value into a RowSet.
207 ** The mallocFailed flag of the database connection is set if a
208 ** memory allocation fails.
210 void sqlite3RowSetInsert(RowSet *p, i64 rowid){
211 struct RowSetEntry *pEntry; /* The new entry */
212 struct RowSetEntry *pLast; /* The last prior entry */
214 /* This routine is never called after sqlite3RowSetNext() */
215 assert( p!=0 && (p->rsFlags & ROWSET_NEXT)==0 );
217 pEntry = rowSetEntryAlloc(p);
218 if( pEntry==0 ) return;
219 pEntry->v = rowid;
220 pEntry->pRight = 0;
221 pLast = p->pLast;
222 if( pLast ){
223 if( rowid<=pLast->v ){ /*OPTIMIZATION-IF-FALSE*/
224 /* Avoid unnecessary sorts by preserving the ROWSET_SORTED flags
225 ** where possible */
226 p->rsFlags &= ~ROWSET_SORTED;
228 pLast->pRight = pEntry;
229 }else{
230 p->pEntry = pEntry;
232 p->pLast = pEntry;
236 ** Merge two lists of RowSetEntry objects. Remove duplicates.
238 ** The input lists are connected via pRight pointers and are
239 ** assumed to each already be in sorted order.
241 static struct RowSetEntry *rowSetEntryMerge(
242 struct RowSetEntry *pA, /* First sorted list to be merged */
243 struct RowSetEntry *pB /* Second sorted list to be merged */
245 struct RowSetEntry head;
246 struct RowSetEntry *pTail;
248 pTail = &head;
249 assert( pA!=0 && pB!=0 );
250 for(;;){
251 assert( pA->pRight==0 || pA->v<=pA->pRight->v );
252 assert( pB->pRight==0 || pB->v<=pB->pRight->v );
253 if( pA->v<=pB->v ){
254 if( pA->v<pB->v ) pTail = pTail->pRight = pA;
255 pA = pA->pRight;
256 if( pA==0 ){
257 pTail->pRight = pB;
258 break;
260 }else{
261 pTail = pTail->pRight = pB;
262 pB = pB->pRight;
263 if( pB==0 ){
264 pTail->pRight = pA;
265 break;
269 return head.pRight;
273 ** Sort all elements on the list of RowSetEntry objects into order of
274 ** increasing v.
276 static struct RowSetEntry *rowSetEntrySort(struct RowSetEntry *pIn){
277 unsigned int i;
278 struct RowSetEntry *pNext, *aBucket[40];
280 memset(aBucket, 0, sizeof(aBucket));
281 while( pIn ){
282 pNext = pIn->pRight;
283 pIn->pRight = 0;
284 for(i=0; aBucket[i]; i++){
285 pIn = rowSetEntryMerge(aBucket[i], pIn);
286 aBucket[i] = 0;
288 aBucket[i] = pIn;
289 pIn = pNext;
291 pIn = aBucket[0];
292 for(i=1; i<sizeof(aBucket)/sizeof(aBucket[0]); i++){
293 if( aBucket[i]==0 ) continue;
294 pIn = pIn ? rowSetEntryMerge(pIn, aBucket[i]) : aBucket[i];
296 return pIn;
301 ** The input, pIn, is a binary tree (or subtree) of RowSetEntry objects.
302 ** Convert this tree into a linked list connected by the pRight pointers
303 ** and return pointers to the first and last elements of the new list.
305 static void rowSetTreeToList(
306 struct RowSetEntry *pIn, /* Root of the input tree */
307 struct RowSetEntry **ppFirst, /* Write head of the output list here */
308 struct RowSetEntry **ppLast /* Write tail of the output list here */
310 assert( pIn!=0 );
311 if( pIn->pLeft ){
312 struct RowSetEntry *p;
313 rowSetTreeToList(pIn->pLeft, ppFirst, &p);
314 p->pRight = pIn;
315 }else{
316 *ppFirst = pIn;
318 if( pIn->pRight ){
319 rowSetTreeToList(pIn->pRight, &pIn->pRight, ppLast);
320 }else{
321 *ppLast = pIn;
323 assert( (*ppLast)->pRight==0 );
328 ** Convert a sorted list of elements (connected by pRight) into a binary
329 ** tree with depth of iDepth. A depth of 1 means the tree contains a single
330 ** node taken from the head of *ppList. A depth of 2 means a tree with
331 ** three nodes. And so forth.
333 ** Use as many entries from the input list as required and update the
334 ** *ppList to point to the unused elements of the list. If the input
335 ** list contains too few elements, then construct an incomplete tree
336 ** and leave *ppList set to NULL.
338 ** Return a pointer to the root of the constructed binary tree.
340 static struct RowSetEntry *rowSetNDeepTree(
341 struct RowSetEntry **ppList,
342 int iDepth
344 struct RowSetEntry *p; /* Root of the new tree */
345 struct RowSetEntry *pLeft; /* Left subtree */
346 if( *ppList==0 ){ /*OPTIMIZATION-IF-TRUE*/
347 /* Prevent unnecessary deep recursion when we run out of entries */
348 return 0;
350 if( iDepth>1 ){ /*OPTIMIZATION-IF-TRUE*/
351 /* This branch causes a *balanced* tree to be generated. A valid tree
352 ** is still generated without this branch, but the tree is wildly
353 ** unbalanced and inefficient. */
354 pLeft = rowSetNDeepTree(ppList, iDepth-1);
355 p = *ppList;
356 if( p==0 ){ /*OPTIMIZATION-IF-FALSE*/
357 /* It is safe to always return here, but the resulting tree
358 ** would be unbalanced */
359 return pLeft;
361 p->pLeft = pLeft;
362 *ppList = p->pRight;
363 p->pRight = rowSetNDeepTree(ppList, iDepth-1);
364 }else{
365 p = *ppList;
366 *ppList = p->pRight;
367 p->pLeft = p->pRight = 0;
369 return p;
373 ** Convert a sorted list of elements into a binary tree. Make the tree
374 ** as deep as it needs to be in order to contain the entire list.
376 static struct RowSetEntry *rowSetListToTree(struct RowSetEntry *pList){
377 int iDepth; /* Depth of the tree so far */
378 struct RowSetEntry *p; /* Current tree root */
379 struct RowSetEntry *pLeft; /* Left subtree */
381 assert( pList!=0 );
382 p = pList;
383 pList = p->pRight;
384 p->pLeft = p->pRight = 0;
385 for(iDepth=1; pList; iDepth++){
386 pLeft = p;
387 p = pList;
388 pList = p->pRight;
389 p->pLeft = pLeft;
390 p->pRight = rowSetNDeepTree(&pList, iDepth);
392 return p;
396 ** Extract the smallest element from the RowSet.
397 ** Write the element into *pRowid. Return 1 on success. Return
398 ** 0 if the RowSet is already empty.
400 ** After this routine has been called, the sqlite3RowSetInsert()
401 ** routine may not be called again.
403 ** This routine may not be called after sqlite3RowSetTest() has
404 ** been used. Older versions of RowSet allowed that, but as the
405 ** capability was not used by the code generator, it was removed
406 ** for code economy.
408 int sqlite3RowSetNext(RowSet *p, i64 *pRowid){
409 assert( p!=0 );
410 assert( p->pForest==0 ); /* Cannot be used with sqlite3RowSetText() */
412 /* Merge the forest into a single sorted list on first call */
413 if( (p->rsFlags & ROWSET_NEXT)==0 ){ /*OPTIMIZATION-IF-FALSE*/
414 if( (p->rsFlags & ROWSET_SORTED)==0 ){ /*OPTIMIZATION-IF-FALSE*/
415 p->pEntry = rowSetEntrySort(p->pEntry);
417 p->rsFlags |= ROWSET_SORTED|ROWSET_NEXT;
420 /* Return the next entry on the list */
421 if( p->pEntry ){
422 *pRowid = p->pEntry->v;
423 p->pEntry = p->pEntry->pRight;
424 if( p->pEntry==0 ){ /*OPTIMIZATION-IF-TRUE*/
425 /* Free memory immediately, rather than waiting on sqlite3_finalize() */
426 sqlite3RowSetClear(p);
428 return 1;
429 }else{
430 return 0;
435 ** Check to see if element iRowid was inserted into the rowset as
436 ** part of any insert batch prior to iBatch. Return 1 or 0.
438 ** If this is the first test of a new batch and if there exist entries
439 ** on pRowSet->pEntry, then sort those entries into the forest at
440 ** pRowSet->pForest so that they can be tested.
442 int sqlite3RowSetTest(RowSet *pRowSet, int iBatch, sqlite3_int64 iRowid){
443 struct RowSetEntry *p, *pTree;
445 /* This routine is never called after sqlite3RowSetNext() */
446 assert( pRowSet!=0 && (pRowSet->rsFlags & ROWSET_NEXT)==0 );
448 /* Sort entries into the forest on the first test of a new batch.
449 ** To save unnecessary work, only do this when the batch number changes.
451 if( iBatch!=pRowSet->iBatch ){ /*OPTIMIZATION-IF-FALSE*/
452 p = pRowSet->pEntry;
453 if( p ){
454 struct RowSetEntry **ppPrevTree = &pRowSet->pForest;
455 if( (pRowSet->rsFlags & ROWSET_SORTED)==0 ){ /*OPTIMIZATION-IF-FALSE*/
456 /* Only sort the current set of entries if they need it */
457 p = rowSetEntrySort(p);
459 for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){
460 ppPrevTree = &pTree->pRight;
461 if( pTree->pLeft==0 ){
462 pTree->pLeft = rowSetListToTree(p);
463 break;
464 }else{
465 struct RowSetEntry *pAux, *pTail;
466 rowSetTreeToList(pTree->pLeft, &pAux, &pTail);
467 pTree->pLeft = 0;
468 p = rowSetEntryMerge(pAux, p);
471 if( pTree==0 ){
472 *ppPrevTree = pTree = rowSetEntryAlloc(pRowSet);
473 if( pTree ){
474 pTree->v = 0;
475 pTree->pRight = 0;
476 pTree->pLeft = rowSetListToTree(p);
479 pRowSet->pEntry = 0;
480 pRowSet->pLast = 0;
481 pRowSet->rsFlags |= ROWSET_SORTED;
483 pRowSet->iBatch = iBatch;
486 /* Test to see if the iRowid value appears anywhere in the forest.
487 ** Return 1 if it does and 0 if not.
489 for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){
490 p = pTree->pLeft;
491 while( p ){
492 if( p->v<iRowid ){
493 p = p->pRight;
494 }else if( p->v>iRowid ){
495 p = p->pLeft;
496 }else{
497 return 1;
501 return 0;