/*
** Write the current contents of the in-memory linked-list to a PMA. Return
** SQLITE_OK if successful, or an SQLite error code otherwise.
**
** The format of a PMA is:
**
** * A varint. This varint contains the total number of bytes of content
** in the PMA (not including the varint itself).
**
** * One or more records packed end-to-end in order of ascending keys.
** Each record consists of a varint followed by a blob of data (the
** key). The varint is the number of bytes in the blob of data.
*/
static int vdbeSorterListToPMA(sqlite3 *db, VdbeCursor *pCsr){
int rc = SQLITE_OK; /* Return code */
VdbeSorter *pSorter = pCsr->pSorter;
if( pSorter->nInMemory==0 ){
assert( pSorter->pRecord==0 );
return rc;
}
rc = vdbeSorterSort(pCsr);
/* If the first temporary PMA file has not been opened, open it now. */
if( rc==SQLITE_OK && pSorter->pTemp1==0 ){
rc = vdbeSorterOpenTempFile(db, &pSorter->pTemp1);
assert( rc!=SQLITE_OK || pSorter->pTemp1 );
assert( pSorter->iWriteOff==0 );
assert( pSorter->nPMA==0 );
}
if( rc==SQLITE_OK ){
i64 iOff = pSorter->iWriteOff;
SorterRecord *p;
SorterRecord *pNext = 0;
static const char eightZeros[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };
pSorter->nPMA++;
rc = vdbeSorterWriteVarint(pSorter->pTemp1, pSorter->nInMemory, &iOff);
for(p=pSorter->pRecord; rc==SQLITE_OK && p; p=pNext){
pNext = p->pNext;
rc = vdbeSorterWriteVarint(pSorter->pTemp1, p->nVal, &iOff);
if( rc==SQLITE_OK ){
rc = sqlite3OsWrite(pSorter->pTemp1, p->pVal, p->nVal, iOff);
iOff += p->nVal;
}
sqlite3DbFree(db, p);
}
/* This assert verifies that unless an error has occurred, the size of
** the PMA on disk is the same as the expected size stored in
** pSorter->nInMemory. */
assert( rc!=SQLITE_OK || pSorter->nInMemory==(
iOff-pSorter->iWriteOff-sqlite3VarintLen(pSorter->nInMemory)
));
pSorter->iWriteOff = iOff;
if( rc==SQLITE_OK ){
/* Terminate each file with 8 extra bytes so that from any offset
** in the file we can always read 9 bytes without a SHORT_READ error */
rc = sqlite3OsWrite(pSorter->pTemp1, eightZeros, 8, iOff);
}
pSorter->pRecord = p;
}
return rc;
}
/*
** Write the current contents of the in-memory linked-list to a PMA. Return
** SQLITE_OK if successful, or an SQLite error code otherwise.
**
** The format of a PMA is:
**
** * A varint. This varint contains the total number of bytes of content
** in the PMA (not including the varint itself).
**
** * One or more records packed end-to-end in order of ascending keys.
** Each record consists of a varint followed by a blob of data (the
** key). The varint is the number of bytes in the blob of data.
*/
static int vdbeSorterListToPMA(sqlite3 *db, const VdbeCursor *pCsr){
int rc = SQLITE_OK; /* Return code */
VdbeSorter *pSorter = pCsr->pSorter;
FileWriter writer;
memset(&writer, 0, sizeof(FileWriter));
if( pSorter->nInMemory==0 ){
assert( pSorter->pRecord==0 );
return rc;
}
rc = vdbeSorterSort(pCsr);
/* If the first temporary PMA file has not been opened, open it now. */
if( rc==SQLITE_OK && pSorter->pTemp1==0 ){
rc = vdbeSorterOpenTempFile(db, &pSorter->pTemp1);
assert( rc!=SQLITE_OK || pSorter->pTemp1 );
assert( pSorter->iWriteOff==0 );
assert( pSorter->nPMA==0 );
}
if( rc==SQLITE_OK ){
SorterRecord *p;
SorterRecord *pNext = 0;
fileWriterInit(db, pSorter->pTemp1, &writer, pSorter->iWriteOff);
pSorter->nPMA++;
fileWriterWriteVarint(&writer, pSorter->nInMemory);
for(p=pSorter->pRecord; p; p=pNext){
pNext = p->pNext;
fileWriterWriteVarint(&writer, p->nVal);
fileWriterWrite(&writer, p->pVal, p->nVal);
sqlite3DbFree(db, p);
}
pSorter->pRecord = p;
rc = fileWriterFinish(db, &writer, &pSorter->iWriteOff);
}
return rc;
}
/*
** Once the sorter has been populated, this function is called to prepare
** for iterating through its contents in sorted order.
*/
int sqlite3VdbeSorterRewind(sqlite3 *db, VdbeCursor *pCsr, int *pbEof){
VdbeSorter *pSorter = pCsr->pSorter;
int rc; /* Return code */
sqlite3_file *pTemp2 = 0; /* Second temp file to use */
i64 iWrite2 = 0; /* Write offset for pTemp2 */
int nIter; /* Number of iterators used */
int nByte; /* Bytes of space required for aIter/aTree */
int N = 2; /* Power of 2 >= nIter */
assert( pSorter );
/* If no data has been written to disk, then do not do so now. Instead,
** sort the VdbeSorter.pRecord list. The vdbe layer will read data directly
** from the in-memory list. */
if( pSorter->nPMA==0 ){
*pbEof = !pSorter->pRecord;
assert( pSorter->aTree==0 );
return vdbeSorterSort(pCsr);
}
/* Write the current b-tree to a PMA. Close the b-tree cursor. */
rc = vdbeSorterListToPMA(db, pCsr);
if( rc!=SQLITE_OK ) return rc;
/* Allocate space for aIter[] and aTree[]. */
nIter = pSorter->nPMA;
if( nIter>SORTER_MAX_MERGE_COUNT ) nIter = SORTER_MAX_MERGE_COUNT;
assert( nIter>0 );
while( N<nIter ) N += N;
nByte = N * (sizeof(int) + sizeof(VdbeSorterIter));
pSorter->aIter = (VdbeSorterIter *)sqlite3DbMallocZero(db, nByte);
if( !pSorter->aIter ) return SQLITE_NOMEM;
pSorter->aTree = (int *)&pSorter->aIter[N];
pSorter->nTree = N;
do {
int iNew; /* Index of new, merged, PMA */
for(iNew=0;
rc==SQLITE_OK && iNew*SORTER_MAX_MERGE_COUNT<pSorter->nPMA;
iNew++
){
i64 nWrite; /* Number of bytes in new PMA */
/* If there are SORTER_MAX_MERGE_COUNT or less PMAs in file pTemp1,
** initialize an iterator for each of them and break out of the loop.
** These iterators will be incrementally merged as the VDBE layer calls
** sqlite3VdbeSorterNext().
**
** Otherwise, if pTemp1 contains more than SORTER_MAX_MERGE_COUNT PMAs,
** initialize interators for SORTER_MAX_MERGE_COUNT of them. These PMAs
** are merged into a single PMA that is written to file pTemp2.
*/
rc = vdbeSorterInitMerge(db, pCsr, &nWrite);
assert( rc!=SQLITE_OK || pSorter->aIter[ pSorter->aTree[1] ].pFile );
if( rc!=SQLITE_OK || pSorter->nPMA<=SORTER_MAX_MERGE_COUNT ){
break;
}
/* Open the second temp file, if it is not already open. */
if( pTemp2==0 ){
assert( iWrite2==0 );
rc = vdbeSorterOpenTempFile(db, &pTemp2);
}
if( rc==SQLITE_OK ){
rc = vdbeSorterWriteVarint(pTemp2, nWrite, &iWrite2);
}
if( rc==SQLITE_OK ){
int bEof = 0;
while( rc==SQLITE_OK && bEof==0 ){
int nToWrite;
VdbeSorterIter *pIter = &pSorter->aIter[ pSorter->aTree[1] ];
assert( pIter->pFile );
nToWrite = pIter->nKey + sqlite3VarintLen(pIter->nKey);
rc = sqlite3OsWrite(pTemp2, pIter->aAlloc, nToWrite, iWrite2);
iWrite2 += nToWrite;
if( rc==SQLITE_OK ){
rc = sqlite3VdbeSorterNext(db, pCsr, &bEof);
}
}
}
}
if( pSorter->nPMA<=SORTER_MAX_MERGE_COUNT ){
break;
}else{
sqlite3_file *pTmp = pSorter->pTemp1;
pSorter->nPMA = iNew;
pSorter->pTemp1 = pTemp2;
pTemp2 = pTmp;
pSorter->iWriteOff = iWrite2;
pSorter->iReadOff = 0;
iWrite2 = 0;
}
}while( rc==SQLITE_OK );
if( pTemp2 ){
sqlite3OsCloseFree(pTemp2);
}
*pbEof = (pSorter->aIter[pSorter->aTree[1]].pFile==0);
return rc;
}
