/*
** 2008 December 3
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
*************************************************************************
**
** This module implements an object we call a "RowSet".
**
** The RowSet object is a collection of rowids. Rowids
** are inserted into the RowSet in an arbitrary order. Inserts
** can be intermixed with tests to see if a given rowid has been
** previously inserted into the RowSet.
**
** After all inserts are finished, it is possible to extract the
** elements of the RowSet in sorted order. Once this extraction
** process has started, no new elements may be inserted.
**
** Hence, the primitive operations for a RowSet are:
**
** CREATE
** INSERT
** TEST
** SMALLEST
** DESTROY
**
** The CREATE and DESTROY primitives are the constructor and destructor,
** obviously. The INSERT primitive adds a new element to the RowSet.
** TEST checks to see if an element is already in the RowSet. SMALLEST
** extracts the least value from the RowSet.
**
** The INSERT primitive might allocate additional memory. Memory is
** allocated in chunks so most INSERTs do no allocation. There is an
** upper bound on the size of allocated memory. No memory is freed
** until DESTROY.
**
** The TEST primitive includes a "batch" number. The TEST primitive
** will only see elements that were inserted before the last change
** in the batch number. In other words, if an INSERT occurs between
** two TESTs where the TESTs have the same batch number, then the
** value added by the INSERT will not be visible to the second TEST.
** The initial batch number is zero, so if the very first TEST contains
** a non-zero batch number, it will see all prior INSERTs.
**
** No INSERTs may occurs after a SMALLEST. An assertion will fail if
** that is attempted.
**
** The cost of an INSERT is roughly constant. (Sometimes new memory
** has to be allocated on an INSERT.) The cost of a TEST with a new
** batch number is O(NlogN) where N is the number of elements in the RowSet.
** The cost of a TEST using the same batch number is O(logN). The cost
** of the first SMALLEST is O(NlogN). Second and subsequent SMALLEST
** primitives are constant time. The cost of DESTROY is O(N).
**
** TEST and SMALLEST may not be used by the same RowSet. This used to
** be possible, but the feature was not used, so it was removed in order
** to simplify the code.
*/
#include "sqliteInt.h"
/*
** Target size for allocation chunks.
*/
#define ROWSET_ALLOCATION_SIZE 1024
/*
** The number of rowset entries per allocation chunk.
*/
#define ROWSET_ENTRY_PER_CHUNK \
((ROWSET_ALLOCATION_SIZE-8)/sizeof(struct RowSetEntry))
/*
** Each entry in a RowSet is an instance of the following object.
**
** This same object is reused to store a linked list of trees of RowSetEntry
** objects. In that alternative use, pRight points to the next entry
** in the list, pLeft points to the tree, and v is unused. The
** RowSet.pForest value points to the head of this forest list.
*/
struct RowSetEntry {
i64 v; /* ROWID value for this entry */
struct RowSetEntry *pRight; /* Right subtree (larger entries) or list */
struct RowSetEntry *pLeft; /* Left subtree (smaller entries) */
};
/*
** RowSetEntry objects are allocated in large chunks (instances of the
** following structure) to reduce memory allocation overhead. The
** chunks are kept on a linked list so that they can be deallocated
** when the RowSet is destroyed.
*/
struct RowSetChunk {
struct RowSetChunk *pNextChunk; /* Next chunk on list of them all */
struct RowSetEntry aEntry[ROWSET_ENTRY_PER_CHUNK]; /* Allocated entries */
};
/*
** A RowSet in an instance of the following structure.
**
** A typedef of this structure if found in sqliteInt.h.
*/
struct RowSet {
struct RowSetChunk *pChunk; /* List of all chunk allocations */
sqlite3 *db; /* The database connection */
struct RowSetEntry *pEntry; /* List of entries using pRight */
struct RowSetEntry *pLast; /* Last entry on the pEntry list */
struct RowSetEntry *pFresh; /* Source of new entry objects */
struct RowSetEntry *pForest; /* List of binary trees of entries */
u16 nFresh; /* Number of objects on pFresh */
u16 rsFlags; /* Various flags */
int iBatch; /* Current insert batch */
};
/*
** Allowed values for RowSet.rsFlags
*/
#define ROWSET_SORTED 0x01 /* True if RowSet.pEntry is sorted */
#define ROWSET_NEXT 0x02 /* True if sqlite3RowSetNext() has been called */
/*
** Allocate a RowSet object. Return NULL if a memory allocation
** error occurs.
*/
RowSet *sqlite3RowSetInit(sqlite3 *db){
RowSet *p = sqlite3DbMallocRawNN(db, sizeof(*p));
if( p ){
int N = sqlite3DbMallocSize(db, p);
p->pChunk = 0;
p->db = db;
p->pEntry = 0;
p->pLast = 0;
p->pForest = 0;
p->pFresh = (struct RowSetEntry*)(ROUND8(sizeof(*p)) + (char*)p);
p->nFresh = (u16)((N - ROUND8(sizeof(*p)))/sizeof(struct RowSetEntry));
p->rsFlags = ROWSET_SORTED;
p->iBatch = 0;
}
return p;
}
/*
** Deallocate all chunks from a RowSet. This frees all memory that
** the RowSet has allocated over its lifetime. This routine is
** the destructor for the RowSet.
*/
void sqlite3RowSetClear(void *pArg){
RowSet *p = (RowSet*)pArg;
struct RowSetChunk *pChunk, *pNextChunk;
for(pChunk=p->pChunk; pChunk; pChunk = pNextChunk){
pNextChunk = pChunk->pNextChunk;
sqlite3DbFree(p->db, pChunk);
}
p->pChunk = 0;
p->nFresh = 0;
p->pEntry = 0;
p->pLast = 0;
p->pForest = 0;
p->rsFlags = ROWSET_SORTED;
}
/*
** Deallocate all chunks from a RowSet. This frees all memory that
** the RowSet has allocated over its lifetime. This routine is
** the destructor for the RowSet.
*/
void sqlite3RowSetDelete(void *pArg){
sqlite3RowSetClear(pArg);
sqlite3DbFree(((RowSet*)pArg)->db, pArg);
}
/*
** Allocate a new RowSetEntry object that is associated with the
** given RowSet. Return a pointer to the new and completely uninitialized
** object.
**
** In an OOM situation, the RowSet.db->mallocFailed flag is set and this
** routine returns NULL.
*/
static struct RowSetEntry *rowSetEntryAlloc(RowSet *p){
assert( p!=0 );
if( p->nFresh==0 ){ /*OPTIMIZATION-IF-FALSE*/
/* We could allocate a fresh RowSetEntry each time one is needed, but it
** is more efficient to pull a preallocated entry from the pool */
struct RowSetChunk *pNew;
pNew = sqlite3DbMallocRawNN(p->db, sizeof(*pNew));
if( pNew==0 ){
return 0;
}
pNew->pNextChunk = p->pChunk;
p->pChunk = pNew;
p->pFresh = pNew->aEntry;
p->nFresh = ROWSET_ENTRY_PER_CHUNK;
}
p->nFresh--;
return p->pFresh++;
}
/*
** Insert a new value into a RowSet.
**
** The mallocFailed flag of the database connection is set if a
** memory allocation fails.
*/
void sqlite3RowSetInsert(RowSet *p, i64 rowid){
struct RowSetEntry *pEntry; /* The new entry */
struct RowSetEntry *pLast; /* The last prior entry */
/* This routine is never called after sqlite3RowSetNext() */
assert( p!=0 && (p->rsFlags & ROWSET_NEXT)==0 );
pEntry = rowSetEntryAlloc(p);
if( pEntry==0 ) return;
pEntry->v = rowid;
pEntry->pRight = 0;
pLast = p->pLast;
if( pLast ){
if( rowid<=pLast->v ){ /*OPTIMIZATION-IF-FALSE*/
/* Avoid unnecessary sorts by preserving the ROWSET_SORTED flags
** where possible */
p->rsFlags &= ~ROWSET_SORTED;
}
pLast->pRight = pEntry;
}else{
p->pEntry = pEntry;
}
p->pLast = pEntry;
}
/*
** Merge two lists of RowSetEntry objects. Remove duplicates.
**
** The input lists are connected via pRight pointers and are
** assumed to each already be in sorted order.
*/
static struct RowSetEntry *rowSetEntryMerge(
struct RowSetEntry *pA, /* First sorted list to be merged */
struct RowSetEntry *pB /* Second sorted list to be merged */
){
struct RowSetEntry head;
struct RowSetEntry *pTail;
pTail = &head;
assert( pA!=0 && pB!=0 );
for(;;){
assert( pA->pRight==0 || pA->v<=pA->pRight->v );
assert( pB->pRight==0 || pB->v<=pB->pRight->v );
if( pA->v<=pB->v ){
if( pA->v