SQLite4  Check-in [6666862302]

#include "lsmtest.h"


int do_work(int nArg, char **azArg){
  struct Option {
    const char *zName;
  } aOpt [] = {
    { "-nmerge" },
    { "-nkb" },
    { 0 }
  };

  lsm_db *pDb;
  int rc;
  int i;
  const char *zDb;
  int nMerge = 1;
  int nKB = (1<<30);

  if( nArg==0 ) goto usage;
  zDb = azArg[nArg-1];
  for(i=0; i<(nArg-1); i++){
    int iSel;
    rc = testArgSelect(aOpt, "option", azArg[i], &iSel);
    if( rc ) return rc;
    switch( iSel ){
      case 0:
        i++;
        if( i==(nArg-1) ) goto usage;
        nMerge = atoi(azArg[i]);
        break;
      case 1:
        i++;
        if( i==(nArg-1) ) goto usage;
        nKB = atoi(azArg[i]);
        break;
    }
  }

  rc = lsm_new(0, &pDb);
  if( rc!=LSM_OK ){
    testPrintError("lsm_open(): rc=%d\n", rc);
  }else{
    rc = lsm_open(pDb, zDb);
    if( rc!=LSM_OK ){
      testPrintError("lsm_open(): rc=%d\n", rc);
    }else{
      int n = -1;
      lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &n);
      n = n*2;
      lsm_config(pDb, LSM_CONFIG_AUTOCHECKPOINT, &n);

      rc = lsm_work(pDb, nMerge, nKB, 0);
      if( rc!=LSM_OK ){
        testPrintError("lsm_work(): rc=%d\n", rc);
      }
    }
  }
  if( rc==LSM_OK ){
    rc = lsm_checkpoint(pDb, 0);

"Options are:\n"
"  -repeat  $repeat                 (default value 10)\n"
"  -write   $write                  (default value 10000)\n"
"  -pause   $pause                  (default value 0)\n"
"  -fetch   $fetch                  (default value 0)\n"
"  -keysize $keysize                (default value 12)\n"
"  -valsize $valsize                (default value 100)\n"
"  -system  $system                 (default value \"lsm\")\n"
"\n"
);
}

int do_speed_test2(int nArg, char **azArg){
  struct Option {
    const char *zOpt;

  }
  
  printf("#");
  for(i=0; i<ArraySize(aOpt); i++){
    if( aOpt[i].zOpt ){
      if( aOpt[i].eVal>=0 ){
        printf(" %s=%d", &aOpt[i].zOpt[1], aParam[aOpt[i].eVal]);
      }else if( aOpt[i].eVal==-1 ){
        printf(" %s=\"%s\"", &aOpt[i].zOpt[1], zSystem);
      }
    }
  }
  printf("\n");

  defn.nMinKey = defn.nMaxKey = aParam[ST_KEYSIZE];

  if( strcmp(azArg[0], "-")==0 ){
    pInput = stdin;
  }else{
    pClose = pInput = fopen(azArg[0], "r");
  }
  zDb = azArg[1];
  pEnv = tdb_lsm_env();
  rc = pEnv->xOpen(pEnv, zDb, 0, &pOut);
  if( rc!=LSM_OK ) return rc;

  while( feof(pInput)==0 ){
    char zLine[80];
    fgets(zLine, sizeof(zLine)-1, pInput);
    zLine[sizeof(zLine)-1] = '\0';

#ifdef __linux__
#include <sys/time.h>
#include <sys/resource.h>

static void lsmtest_rusage_report(void){
  int res;
  struct rusage r;
  memset(&r, 0, sizeof(r));

  res = getrusage(RUSAGE_SELF, &r);
  assert( res==0 );

  printf("# getrusage: { ru_maxrss %d ru_oublock %d ru_inblock %d }\n", 
      (int)r.ru_maxrss, (int)r.ru_oublock, (int)r.ru_inblock
  );

#include <sys/time.h>

typedef struct LsmDb LsmDb;
typedef struct LsmWorker LsmWorker;
typedef struct LsmFile LsmFile;

#define LSMTEST_DFLT_MT_MAX_CKPT (8*1024)
#define LSMTEST_DFLT_MT_MIN_CKPT (2*1024)

#ifdef LSM_MUTEX_PTHREADS
#include <pthread.h>

#define LSMTEST_THREAD_CKPT      1
#define LSMTEST_THREAD_WORKER    2
#define LSMTEST_THREAD_WORKER_AC 3

  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xFullpath(pRealEnv, zFile, zOut, pnOut);
}

static int testEnvOpen(
  lsm_env *pEnv,                  /* Environment for current LsmDb */
  const char *zFile,              /* Name of file to open */
  int flags,
  lsm_file **ppFile               /* OUT: New file handle object */
){
  lsm_env *pRealEnv = tdb_lsm_env();
  LsmDb *pDb = (LsmDb *)pEnv->pVfsCtx;
  int rc;                         /* Return Code */
  LsmFile *pRet;                  /* The new file handle */
  int nFile;                      /* Length of string zFile in bytes */

  nFile = strlen(zFile);
  pRet = (LsmFile *)testMalloc(sizeof(LsmFile));
  pRet->pDb = pDb;
  pRet->bLog = (nFile > 4 && 0==memcmp("-log", &zFile[nFile-4], 4));

  rc = pRealEnv->xOpen(pRealEnv, zFile, flags, &pRet->pReal);
  if( rc!=LSM_OK ){
    testFree(pRet);
    pRet = 0;
  }

  *ppFile = (lsm_file *)pRet;
  return rc;

  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xLock(p->pReal, iLock, eType);
}

static int testEnvTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xTestLock(p->pReal, iLock, nLock, eType);
}

static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
}

  char *zFile = pDb->zName;
  char *zFree = 0;

  for(iFile=0; iFile<2; iFile++){
    lsm_file *pFile = 0;
    int i;

    pEnv->xOpen(pEnv, zFile, 0, &pFile);
    for(i=0; i<pDb->aFile[iFile].nSector; i++){
      u8 *aOld = pDb->aFile[iFile].aSector[i].aOld;
      if( aOld ){
        int iOpt = testPrngValue(iSeed++) % 3;
        switch( iOpt ){
          case 0:
            break;

  testFree((char *)pDb->pBuf);
  testFree((char *)pDb);
  return rc;
}

static int waitOnCheckpointer(LsmDb *pDb, lsm_db *db){
  int nSleep = 0;
  int nKB;
  int rc;

  do {
    nKB = 0;
    rc = lsm_info(db, LSM_INFO_CHECKPOINT_SIZE, &nKB);
    if( rc!=LSM_OK || nKB<pDb->nMtMaxCkpt ) break;
    usleep(5000);
    nSleep += 5;
  }while( 1 );

#if 0
    if( nSleep ) printf("# waitOnCheckpointer(): nSleep=%d\n", nSleep);
#endif

  return rc;
}

static int waitOnWorker(LsmDb *pDb){
  int rc;
  int nLimit = -1;
  int nSleep = 0;

  rc = lsm_config(pDb->db, LSM_CONFIG_AUTOFLUSH, &nLimit);
  do {
    int nOld, nNew, rc;
    rc = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
    if( rc!=LSM_OK ) return rc;
    if( nOld==0 || nNew<(nLimit/2) ) break;
    usleep(5000);
    nSleep += 5;
  }while( 1 );

#if 0
  if( nSleep ) printf("# waitOnWorker(): nSleep=%d\n", nSleep);
#endif

  } aParam[] = {
    { "autoflush",        0, LSM_CONFIG_AUTOFLUSH },
    { "page_size",        0, LSM_CONFIG_PAGE_SIZE },
    { "block_size",       0, LSM_CONFIG_BLOCK_SIZE },
    { "safety",           0, LSM_CONFIG_SAFETY },
    { "autowork",         0, LSM_CONFIG_AUTOWORK },
    { "autocheckpoint",   0, LSM_CONFIG_AUTOCHECKPOINT },

    { "mmap",             0, LSM_CONFIG_MMAP },
    { "use_log",          0, LSM_CONFIG_USE_LOG },
    { "automerge",        0, LSM_CONFIG_AUTOMERGE },
    { "max_freelist",     0, LSM_CONFIG_MAX_FREELIST },
    { "multi_proc",       0, LSM_CONFIG_MULTIPLE_PROCESSES },
    { "worker_automerge", 1, LSM_CONFIG_AUTOMERGE },
    { "test_no_recovery", 0, TEST_NO_RECOVERY },

      if( rc!=0 ) return rc;
      eParam = aParam[i].eParam;

      z++;
      zStart = z;
      while( *z>='0' && *z<='9' ) z++;
      if( *z=='k' || *z=='K' ){
        iMul = 1;
        z++;
      }else if( *z=='M' || *z=='M' ){
        iMul = 1024;
        z++;
      }
      nParam = z-zStart;
      if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
      memcpy(zParam, zStart, nParam);
      zParam[nParam] = '\0';
      iVal = atoi(zParam) * iMul;

          case TEST_NO_RECOVERY:
            if( pLsm ) pLsm->bNoRecovery = iVal;
            break;
          case TEST_MT_MODE:
            if( pLsm ) nThread = iVal;
            break;
          case TEST_MT_MIN_CKPT:
            if( pLsm && iVal>0 ) pLsm->nMtMinCkpt = iVal*1024;
            break;
          case TEST_MT_MAX_CKPT:
            if( pLsm && iVal>0 ) pLsm->nMtMaxCkpt = iVal*1024;
            break;
#ifdef HAVE_ZLIB
          case TEST_COMPRESSION:
            testConfigureCompression(db);
            break;
#endif
        }

#endif
  pDb->env.xMap = testEnvMap;
  pDb->env.xUnmap = testEnvUnmap;
  pDb->env.xFileid = testEnvFileid;
  pDb->env.xClose = testEnvClose;
  pDb->env.xUnlink = testEnvUnlink;
  pDb->env.xLock = testEnvLock;
  pDb->env.xTestLock = testEnvTestLock;
  pDb->env.xShmBarrier = testEnvShmBarrier;
  pDb->env.xShmMap = testEnvShmMap;
  pDb->env.xShmUnmap = testEnvShmUnmap;
  pDb->env.xSleep = testEnvSleep;

  rc = lsm_new(&pDb->env, &pDb->db);
  if( rc==LSM_OK ){

}

int test_lsm_small_open(
  const char *zFile, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = "page_size=256 block_size=64";
  return testLsmOpen(zCfg, zFile, bClear, ppDb);
}

int test_lsm_lomem_open(
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
    /* "max_freelist=4 autocheckpoint=32" */
  const char *zCfg = 
    "page_size=256 block_size=64 autoflush=16 "
    "autocheckpoint=32"
    "mmap=0 "
  ;
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

int test_lsm_zip_open(
  const char *zFilename, 
  int bClear, 
  TestDb **ppDb
){
  const char *zCfg = 
    "page_size=256 block_size=64 autoflush=16 "
    "autocheckpoint=32 compression=1 mmap=0 "
  ;
  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
}

lsm_db *tdb_lsm(TestDb *pDb){
  if( pDb->pMethods->xClose==test_lsm_close ){
    return ((LsmDb *)pDb)->db;

    int rc = LSM_OK;
    int nCkpt = -1;

    /* Do some work. If an error occurs, exit. */

    pthread_mutex_unlock(&p->worker_mutex);
    if( p->eType==LSMTEST_THREAD_CKPT ){
      int nKB = 0;
      rc = lsm_info(pWorker, LSM_INFO_CHECKPOINT_SIZE, &nKB);
      if( rc==LSM_OK && nKB>=p->pDb->nMtMinCkpt ){
        rc = lsm_checkpoint(pWorker, 0);
      }
    }else{
      int nWrite;
      do {

        if( p->eType==LSMTEST_THREAD_WORKER ){

#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#include <sys/time.h>


/*
** Global variables used within this module.
*/
static struct TestutilGlobal {
  char **argv;
  int argc;
} g = {0, 0};

static struct TestutilRnd {
  unsigned int aRand1[2048];          /* Bits 0..10 */
  unsigned int aRand2[2048];          /* Bits 11..21 */
  unsigned int aRand3[1024];          /* Bits 22..31 */
} r;

/*************************************************************************
** The following block is a copy of the implementation of SQLite function
** sqlite3_randomness. This version has two important differences:
**
**   1. It always uses the same seed. So the sequence of random data output

    0xC4, 0xEC, 0x80, 0xD0, 0x98, 0xA7, 0x76, 0xCC, 
    0x9C, 0x2F, 0x7B, 0xFF, 0x8E, 0x0E, 0xBB, 0x90, 
    0xAE, 0x13, 0x06, 0xF5, 0x1C, 0x4E, 0x52, 0xF7
  }
};

/* Generate and return single random byte */
static unsigned char randomByte(void){
  unsigned char t;
  sqlite3Prng.i++;
  t = sqlite3Prng.s[sqlite3Prng.i];
  sqlite3Prng.j += t;
  sqlite3Prng.s[sqlite3Prng.i] = sqlite3Prng.s[sqlite3Prng.j];
  sqlite3Prng.s[sqlite3Prng.j] = t;
  t += sqlite3Prng.s[sqlite3Prng.i];

}
/*
** End of code copied from SQLite.
*************************************************************************/


int testPrngInit(void){

  randomBlob(sizeof(r.aRand1), (unsigned char *)r.aRand1);
  randomBlob(sizeof(r.aRand2), (unsigned char *)r.aRand2);
  randomBlob(sizeof(r.aRand3), (unsigned char *)r.aRand3);
  return 0;
}

unsigned int testPrngValue(unsigned int iVal){
  return
    r.aRand1[iVal & 0x000007FF] ^
    r.aRand2[(iVal>>11) & 0x000007FF] ^
    r.aRand3[(iVal>>22) & 0x000003FF]
  ;
}

void testPrngArray(unsigned int iVal, unsigned int *aOut, int nOut){
  int i;
  for(i=0; i<nOut; i++){
    aOut[i] = testPrngValue(iVal+i);
  }
}

void testPrngString(unsigned int iVal, char *aOut, int nOut){
  int i;
  for(i=0; i<(nOut-1); i++){
    aOut[i] = 'a' + (testPrngValue(iVal+i) % 26);
  }
  aOut[i] = '\0';
}

  struct Entry { const char *zName; };
  struct Entry *pEntry;
  const char *zPrev = 0;

  testPrintError("unrecognized %s \"%s\": must be ", zType, zArg);
  for(pEntry=(struct Entry *)aData; 
      pEntry->zName; 
      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
  ){
    if( zPrev ){ testPrintError("%s, ", zPrev); }
    zPrev = pEntry->zName;
  }
  testPrintError("or %s\n", zPrev);
}

  int i = 0;
  int iOut = -1;
  int nOut = 0;

  for(pEntry=(struct Entry *)aData; 
      pEntry->zName; 
      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
  ){
    int nName = strlen(pEntry->zName);
    if( nArg<=nName && memcmp(pEntry->zName, zArg, nArg)==0 ){
      iOut = i;
      if( nName==nArg ){
        nOut = 1;
        break;

/*
** 2013 March 1
**
** 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 file contains C code for a program that links against SQLite
** versions 3 and 4. It contains a few simple performance test routines
** that can be run against either database system.
*/

#include "sqlite4.h"
#include "sqlite3.h"
#include "lsm.h"

#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include <unistd.h>

#define SQLITE3_DB_FILE "test.db3"
#define SQLITE4_DB_FILE "test.db4"

#include "lsmtest_util.c"

/*
** Unlink database zDb and its supporting files (wal, shm, journal, and log).
** This function works with both lsm and sqlite3 databases.
*/
static int unlink_db(const char *zDb){
  int i;
  const char *azExt[] = { "", "-shm", "-wal", "-journal", "-log", 0 };

  for(i=0; azExt[i]; i++){
    char *zFile = sqlite4_mprintf(0, "%s%s", zDb, azExt[i]);
    unlink(zFile);
    sqlite4_free(0, zFile);
  }

  return 0;
}

static char *create_schema_sql(int nIdx){
  char *zSchema;
  int i;

  zSchema = sqlite4_mprintf(0, "CREATE TABLE t1(k PRIMARY KEY,");
  for(i=0; i<nIdx; i++){
    zSchema = sqlite4_mprintf(0, "%z c%d BLOB,", zSchema, i);
  }
  zSchema = sqlite4_mprintf(0, "%z v BLOB);", zSchema);

  for(i=0; i<nIdx; i++){
    zSchema = sqlite4_mprintf(
      0, "%z\nCREATE INDEX i%d ON t1 (c%d);", zSchema, i, i
    );
  }

  return zSchema;
}

static char *create_insert_sql(int nIdx){
  char *zInsert;
  int i;

  zInsert = sqlite4_mprintf(0, "INSERT INTO t1 VALUES(rblob(:1, 8, 20),");
  for(i=0; i<nIdx; i++){
    zInsert = sqlite4_mprintf(0, "%z rblob((:1<<%d)+:1, 8, 20),", zInsert, i);
  }
  zInsert = sqlite4_mprintf(0, "%z rblob((:1<<%d)+:1, 100, 150));", zInsert, i);

  return zInsert;
}

static char *create_select_sql(int iIdx){
  char *zSql;
  if( iIdx==0 ){
    zSql = sqlite4_mprintf(0, "SELECT * FROM t1 WHERE k = rblob(:1, 8, 20)");
  }else{
    int iCol = iIdx-1;
    zSql = sqlite4_mprintf(0, 
        "SELECT * FROM t1 WHERE c%d = rblob((:1<<%d)+:1, 8, 20)", iCol, iCol
    );
  }
  return zSql;
}

static int do_explode(const char *zLine, int rc, int iLine){
  if( rc ){
    fprintf(stderr, "ERROR: \"%s\" at line %d failed. rc=%d\n", 
        zLine, iLine, rc
    );
    exit(-1);
  }
  return 0;
}
#define EXPLODE(rc) do_explode(#rc, rc, __LINE__)


/*************************************************************************
** Implementations of the rblob(nMin, nMax) function. One for src4 and
** one for sqlite3.
*/

/* src4 implementation */
static void rblobFunc4(sqlite4_context *ctx, int nArg, sqlite4_value **apArg){
  unsigned char aBlob[1000];

  int iSeed = sqlite4_value_int(apArg[0]);
  int nMin = sqlite4_value_int(apArg[1]);
  int nMax = sqlite4_value_int(apArg[2]);
  int nByte;

  nByte = testPrngValue(iSeed + 1000000) & 0x7FFFFFFF;
  nByte = (nByte % (nMax+1-nMin)) + nMin;
  assert( nByte>=nMin && nByte<=nMax );
  if( nByte>sizeof(aBlob) ) nByte = sizeof(aBlob);
  testPrngArray(iSeed, (unsigned int *)aBlob, (nByte+3)/4);

  sqlite4_result_blob(ctx, aBlob, nByte, SQLITE4_TRANSIENT, 0);
}
static void install_rblob_function4(sqlite4 *db){
  testPrngInit();
  sqlite4_create_function(db, "rblob", 3, SQLITE4_UTF8, 0, rblobFunc4, 0, 0);
}

/* sqlite3 implementation */
static void rblobFunc3(sqlite3_context *ctx, int nArg, sqlite3_value **apArg){
  unsigned char aBlob[1000];

  int iSeed = sqlite3_value_int(apArg[0]);
  int nMin = sqlite3_value_int(apArg[1]);
  int nMax = sqlite3_value_int(apArg[2]);
  int nByte;

  nByte = testPrngValue(iSeed + 1000000) & 0x7FFFFFFF;
  nByte = (nByte % (nMax+1-nMin)) + nMin;
  assert( nByte>=nMin && nByte<=nMax );
  if( nByte>sizeof(aBlob) ) nByte = sizeof(aBlob);
  testPrngArray(iSeed, (unsigned int *)aBlob, (nByte+3)/4);

  sqlite3_result_blob(ctx, aBlob, nByte, SQLITE_TRANSIENT);
}
static void install_rblob_function3(sqlite3 *db){
  testPrngInit();
  sqlite3_create_function(db, "rblob", 3, SQLITE_UTF8, 0, rblobFunc3, 0, 0);
}
/*
** End of rblob() implementations.
*************************************************************************/

/*************************************************************************
** Integer query functions for sqlite3 and src4.
*/
static int integer_query4(sqlite4 *db, const char *zSql){
  int iRet;
  sqlite4_stmt *pStmt;

  EXPLODE( sqlite4_prepare(db, zSql, -1, &pStmt, 0) );
  EXPLODE( SQLITE_ROW!=sqlite4_step(pStmt) );
  iRet = sqlite4_column_int(pStmt, 0);
  EXPLODE( sqlite4_finalize(pStmt) );

  return iRet;
}
static int integer_query3(sqlite3 *db, const char *zSql){
  int iRet;
  sqlite3_stmt *pStmt;

  EXPLODE( sqlite3_prepare(db, zSql, -1, &pStmt, 0) );
  EXPLODE( SQLITE_ROW!=sqlite3_step(pStmt) );
  iRet = sqlite3_column_int(pStmt, 0);
  EXPLODE( sqlite3_finalize(pStmt) );

  return iRet;
}
/*
** End of integer query implementations.
*************************************************************************/

static int do_insert1_test4(
  int nRow,                       /* Number of rows to insert in total */
  int nRowPerTrans,               /* Number of rows per transaction */
  int nIdx,                       /* Number of aux indexes (aside from PK) */
  int iSync                       /* PRAGMA synchronous value (0, 1 or 2) */
){
  char *zCreateTbl;               /* Create table statement */
  char *zInsert;                  /* INSERT statement */
  sqlite4_stmt *pInsert;          /* Compiled INSERT statement */
  sqlite4 *db = 0;                /* Database handle */
  int i;                          /* Counter to count nRow rows */
  int nMs;                        /* Test time in ms */

  lsm_db *pLsm;

  unlink_db(SQLITE4_DB_FILE);
  EXPLODE(  sqlite4_open(0, SQLITE4_DB_FILE, &db)  );
  sqlite4_kvstore_control(db, "main", SQLITE4_KVCTRL_LSM_HANDLE, &pLsm);
  i = iSync;
  lsm_config(pLsm, LSM_CONFIG_SAFETY, &i);
  assert( i==iSync );

  install_rblob_function4(db);

  zCreateTbl = create_schema_sql(nIdx);
  zInsert = create_insert_sql(nIdx);

  /* Create the db schema and prepare the INSERT statement */
  EXPLODE(  sqlite4_exec(db, zCreateTbl, 0, 0, 0)  );
  EXPLODE(  sqlite4_prepare(db, zInsert, -1, &pInsert, 0)  );

  /* Run the test */
  testTimeInit();
  for(i=0; i<nRow; i++){
    if( (i % nRowPerTrans)==0 ){
      if( i!=0 ) EXPLODE(  sqlite4_exec(db, "COMMIT", 0, 0, 0)  );
      EXPLODE(  sqlite4_exec(db, "BEGIN", 0, 0, 0)  );
    }
    sqlite4_bind_int(pInsert, 1, i);
    sqlite4_step(pInsert);
    EXPLODE(  sqlite4_reset(pInsert)  );
  }
  EXPLODE(  sqlite4_exec(db, "COMMIT", 0, 0, 0)  );

  /* Free all the stuff allocated above */
  sqlite4_finalize(pInsert);
  sqlite4_free(0, zCreateTbl);
  sqlite4_free(0, zInsert);
  sqlite4_close(db);
  nMs = testTimeGet();

  /* Print out the time taken by the test */
  printf("%.3f seconds\n", (double)nMs / 1000.0);
  return 0;
}
static int do_insert1_test3(
  int nRow,                       /* Number of rows to insert in total */
  int nRowPerTrans,               /* Number of rows per transaction */
  int nIdx,                       /* Number of aux indexes (aside from PK) */
  int iSync                       /* PRAGMA synchronous value (0, 1 or 2) */
){
  char *zCreateTbl;               /* Create table statement */
  char *zInsert;                  /* INSERT statement */
  char *zSync;                    /* "PRAGMA synchronous=" statement */
  sqlite3_stmt *pInsert;          /* Compiled INSERT statement */
  sqlite3 *db = 0;                /* Database handle */
  int i;                          /* Counter to count nRow rows */
  int nMs;                        /* Test time in ms */

  unlink_db(SQLITE3_DB_FILE);
  EXPLODE( sqlite3_open(SQLITE3_DB_FILE, &db) );
  EXPLODE( sqlite3_exec(db, "PRAGMA journal_mode=WAL", 0, 0, 0) );
  zSync = sqlite4_mprintf(0, "PRAGMA synchronous=%d", iSync);
  EXPLODE( sqlite3_exec(db, zSync, 0, 0, 0) );
  sqlite4_free(0, zSync);

  install_rblob_function3(db);

  zCreateTbl = create_schema_sql(nIdx);
  zInsert = create_insert_sql(nIdx);

  /* Create the db schema and prepare the INSERT statement */
  EXPLODE(  sqlite3_exec(db, zCreateTbl, 0, 0, 0)  );
  EXPLODE(  sqlite3_prepare(db, zInsert, -1, &pInsert, 0)  );

  /* Run the test */
  testTimeInit();
  for(i=0; i<nRow; i++){
    if( (i % nRowPerTrans)==0 ){
      if( i!=0 ) EXPLODE(  sqlite3_exec(db, "COMMIT", 0, 0, 0)  );
      EXPLODE(  sqlite3_exec(db, "BEGIN", 0, 0, 0)  );
    }
    sqlite3_bind_int(pInsert, 1, i);
    sqlite3_step(pInsert);
    EXPLODE(  sqlite3_reset(pInsert)  );
  }
  EXPLODE(  sqlite3_exec(db, "COMMIT", 0, 0, 0)  );

  /* Finalize the statement and close the db. */
  sqlite3_finalize(pInsert);
  sqlite3_close(db);
  nMs = testTimeGet();

  /* Free the stuff allocated above */
  sqlite4_free(0, zCreateTbl);
  sqlite4_free(0, zInsert);

  /* Print out the time taken by the test */
  printf("%.3f seconds\n", (double)nMs / 1000.0);
  return 0;
}

static int do_insert1(int argc, char **argv){
  struct Insert1Arg {
    const char *zArg;
    int nMin;
    int nMax;
  } aArg[] = { 
    {"-db",           3,    4}, 
    {"-rows",         1,    10000000}, 
    {"-rowspertrans", 1,    10000000}, 
    {"-indexes",      0,    20}, 
    {"-sync",         0,    2}, 
    {0,0,0}
  };
  int i;

  int iDb = 4;                    /* SQLite 3 or 4 */
  int nRow = 50000;               /* Total rows: 50000 */
  int nRowPerTrans = 10;          /* Total rows each transaction: 50000 */
  int nIdx = 3;                   /* Number of auxilliary indexes */
  int iSync = 1;                  /* PRAGMA synchronous setting */

  for(i=0; i<argc; i++){
    int iSel;
    int iVal;
    int rc;

    rc = testArgSelectX(aArg, "argument", sizeof(aArg[0]), argv[i], &iSel);
    if( rc!=0 ) return -1;
    if( i==argc-1 ){
      fprintf(stderr, "option %s requires an argument\n", aArg[iSel].zArg);
      return -1;
    }
    iVal = atoi(argv[++i]);
    if( iVal<aArg[iSel].nMin || iVal>aArg[iSel].nMax ){
      fprintf(stderr, "option %s out of range (%d..%d)\n", 
          aArg[iSel].zArg, aArg[iSel].nMin, aArg[iSel].nMax 
      );
      return -1;
    }

    switch( iSel ){
      case 0: iDb = iVal;          break;
      case 1: nRow = iVal;         break;
      case 2: nRowPerTrans = iVal; break;
      case 3: nIdx = iVal;         break;
      case 4: iSync = iVal;        break;
    }
  }

  printf("insert1: db=%d rows=%d rowspertrans=%d indexes=%d sync=%d ... ", 
      iDb, nRow, nRowPerTrans, nIdx, iSync
  );
  fflush(stdout);
  if( iDb==3 ){
    do_insert1_test3(nRow, nRowPerTrans, nIdx, iSync);
  }else{
    do_insert1_test4(nRow, nRowPerTrans, nIdx, iSync);
  }

  return 0;
}

static int do_select1_test4(
  int nRow,                       /* Number of rows to read in total */
  int nRowPerTrans,               /* Number of rows per transaction */
  int iIdx
){
  int nMs = 0;
  sqlite4_stmt *pSelect = 0;
  char *zSelect;
  sqlite4 *db;
  int i;
  int nTblRow;

  EXPLODE( sqlite4_open(0, SQLITE4_DB_FILE, &db) );
  install_rblob_function4(db);

  nTblRow = integer_query4(db, "SELECT count(*) FROM t1");

  /* Create the db schema and prepare the INSERT statement */
  zSelect = create_select_sql(iIdx);
  EXPLODE(  sqlite4_prepare(db, zSelect, -1, &pSelect, 0)  );

  testTimeInit();
  for(i=0; i<nRow; i++){
    if( (i % nRowPerTrans)==0 ){
      if( i!=0 ) EXPLODE(  sqlite4_exec(db, "COMMIT", 0, 0, 0)  );
      EXPLODE(  sqlite4_exec(db, "BEGIN", 0, 0, 0)  );
    }
    sqlite4_bind_int(pSelect, 1, (i*211)%nTblRow);
    EXPLODE(  SQLITE_ROW!=sqlite4_step(pSelect)  );
    EXPLODE(  sqlite4_reset(pSelect)  );
  }
  EXPLODE(  sqlite4_exec(db, "COMMIT", 0, 0, 0)  );
  nMs = testTimeGet();

  sqlite4_finalize(pSelect);
  sqlite4_close(db);
  sqlite4_free(0, zSelect);

  printf("%.3f seconds\n", (double)nMs / 1000.0);
  return 0;
}
static int do_select1_test3(
  int nRow,                       /* Number of rows to read in total */
  int nRowPerTrans,               /* Number of rows per transaction */
  int iIdx
){
  int nMs = 0;
  sqlite3_stmt *pSelect = 0;
  char *zSelect;
  sqlite3 *db;
  int i;
  int nTblRow;

  EXPLODE( sqlite3_open(SQLITE3_DB_FILE, &db) );
  install_rblob_function3(db);

  nTblRow = integer_query3(db, "SELECT count(*) FROM t1");

  /* Create the db schema and prepare the INSERT statement */
  zSelect = create_select_sql(iIdx);
  EXPLODE(  sqlite3_prepare(db, zSelect, -1, &pSelect, 0)  );

  testTimeInit();
  for(i=0; i<nRow; i++){
    if( (i % nRowPerTrans)==0 ){
      if( i!=0 ) EXPLODE(  sqlite3_exec(db, "COMMIT", 0, 0, 0)  );
      EXPLODE(  sqlite3_exec(db, "BEGIN", 0, 0, 0)  );
    }
    sqlite3_bind_int(pSelect, 1, (i*211)%nTblRow);
    EXPLODE(  SQLITE_ROW!=sqlite3_step(pSelect)  );
    EXPLODE(  sqlite3_reset(pSelect)  );
  }
  EXPLODE(  sqlite3_exec(db, "COMMIT", 0, 0, 0)  );
  nMs = testTimeGet();

  sqlite3_finalize(pSelect);
  sqlite3_close(db);
  sqlite4_free(0, zSelect);

  printf("%.3f seconds\n", (double)nMs / 1000.0);
  return 0;
}

static int do_select1(int argc, char **argv){
  struct Insert1Arg {
    const char *zArg;
    int nMin;
    int nMax;
  } aArg[] = {
    {"-db",           3,    4}, 
    {"-rows",         1,    10000000}, 
    {"-rowspertrans", 1,    10000000}, 
    {"-index",        0,    21}, 
    {0,0,0}
  };
  int i;

  int iDb = 4;                    /* SQLite 3 or 4 */
  int nRow = 50000;               /* Total rows: 50000 */
  int nRowPerTrans = 10;          /* Total rows each transaction: 50000 */
  int iIdx = 0;

  for(i=0; i<argc; i++){
    int iSel;
    int iVal;
    int rc;

    rc = testArgSelectX(aArg, "argument", sizeof(aArg[0]), argv[i], &iSel);
    if( rc!=0 ) return -1;
    if( i==argc-1 ){
      fprintf(stderr, "option %s requires an argument\n", aArg[iSel].zArg);
      return -1;
    }
    iVal = atoi(argv[++i]);
    if( iVal<aArg[iSel].nMin || iVal>aArg[iSel].nMax ){
      fprintf(stderr, "option %s out of range (%d..%d)\n", 
          aArg[iSel].zArg, aArg[iSel].nMin, aArg[iSel].nMax 
      );
      return -1;
    }

    switch( iSel ){
      case 0: iDb = iVal;          break;
      case 1: nRow = iVal;         break;
      case 2: nRowPerTrans = iVal; break;
      case 3: iIdx = iVal;         break;
    }
  }

  printf("select1: db=%d rows=%d rowspertrans=%d index=%d ... ", 
      iDb, nRow, nRowPerTrans, iIdx
  );
  fflush(stdout);
  if( iDb==3 ){
    do_select1_test3(nRow, nRowPerTrans, iIdx);
  }else{
    do_select1_test4(nRow, nRowPerTrans, iIdx);
  }

  return 0;
}

int main(int argc, char **argv){
  struct SqltestArg {
    const char *zPrg;
    int (*xPrg)(int, char **);
  } aArg[] = { 
    {"select", do_select1},
    {"insert", do_insert1},
    {0, 0}
  };
  int iSel;
  int rc;

  if( argc<2 ){
    fprintf(stderr, "Usage: %s sub-program...\n", argv[0]);
    return -1;
  }

  rc = testArgSelectX(aArg, "sub-program", sizeof(aArg[0]), argv[1], &iSel);
  if( rc!=0 ) return -1;

  aArg[iSel].xPrg(argc-2, argv+2);
  return 0;
}

# Once the macros above are defined, the rest of this make script will
# build the SQLite library and testing tools.
################################################################################

# FIXME:  Required options for now.
#
OPTS += -DLSM_MUTEX_NONE
#OPTS += -DSQLITE4_DEBUG=1 -DLSM_DEBUG=1
OPTS += -DHAVE_GMTIME_R
OPTS += -DHAVE_LOCALTIME_R
OPTS += -DHAVE_MALLOC_USABLE_SIZE
OPTS += -DHAVE_USLEEP
#OPTS += -DSQLITE4_MEMDEBUG=1
#OPTS += -DSQLITE4_NO_SYNC=1 -DLSM_NO_SYNC=1
OPTS += -DSQLITE4_OMIT_ANALYZE
OPTS += -DSQLITE4_OMIT_AUTOMATIC_INDEX
OPTS += -DSQLITE4_OMIT_BTREECOUNT
OPTS += -DSQLITE4_OMIT_VIRTUALTABLE=1
OPTS += -DSQLITE4_OMIT_XFER_OPT
OPTS += -DSQLITE4_THREADSAFE=0

# This is how we compile
#
TCCX =  $(TCC) $(OPTS) -I. -I$(TOP)/src -I$(TOP) 
TCCX += -I$(TOP)/ext/rtree -I$(TOP)/ext/icu -I$(TOP)/ext/fts3
TCCX += -I$(TOP)/ext/async

TCPPX = g++ -Wall -g -I. -I$(TOP)/src $(OPTS)


LIBOBJ+= vdbe.o parse.o \
         alter.o analyze.o attach.o auth.o \
         build.o \
         callback.o complete.o ctime.o date.o delete.o env.o expr.o \
         fault.o fkey.o fts5.o fts5func.o \
         func.o global.o hash.o \
         icu.o insert.o kv.o kvlsm.o kvmem.o legacy.o \
         lsm_ckpt.o lsm_file.o lsm_log.o lsm_main.o lsm_mem.o lsm_mutex.o \
         lsm_shared.o lsm_str.o lsm_sorted.o lsm_tree.o \
         lsm_unix.o lsm_varint.o \
         main.o malloc.o math.o mem.o mem0.o mem1.o mem2.o mem3.o mem5.o \
         mutex.o mutex_noop.o mutex_unix.o mutex_w32.o \
         opcodes.o os.o \
         pragma.o prepare.o printf.o \
         random.o resolve.o rowset.o rtree.o select.o status.o \
         tokenize.o trigger.o \
         update.o util.o varint.o \
         vdbeapi.o vdbeaux.o vdbecodec.o vdbecursor.o \

  $(TOP)/src/auth.c \
  $(TOP)/src/build.c \
  $(TOP)/src/callback.c \
  $(TOP)/src/complete.c \
  $(TOP)/src/ctime.c \
  $(TOP)/src/date.c \
  $(TOP)/src/delete.c \
  $(TOP)/src/env.c \
  $(TOP)/src/expr.c \
  $(TOP)/src/fault.c \
  $(TOP)/src/fkey.c \
  $(TOP)/src/fts5.c \
  $(TOP)/src/fts5func.c \
  $(TOP)/src/func.c \
  $(TOP)/src/global.c \

  $(TOP)/src/lsm_sorted.c \
  $(TOP)/src/lsm_tree.c \
  $(TOP)/src/lsm_unix.c \
  $(TOP)/src/lsm_varint.c \
  $(TOP)/src/main.c \
  $(TOP)/src/malloc.c \
  $(TOP)/src/math.c \
  $(TOP)/src/mem.c \
  $(TOP)/src/mem0.c \
  $(TOP)/src/mem1.c \
  $(TOP)/src/mem2.c \
  $(TOP)/src/mem3.c \
  $(TOP)/src/mem5.c \
  $(TOP)/src/mutex.c \
  $(TOP)/src/mutex.h \

TESTFIXTURE_PREREQ  = $(TESTSRC) $(TESTSRC2) 
TESTFIXTURE_PREREQ += $(TOP)/src/tclsqlite.c
TESTFIXTURE_PREREQ += libsqlite4.a

testfixture$(EXE): $(TESTFIXTURE_PREREQ)
	$(TCCX) $(TCL_FLAGS) -DTCLSH=1 $(TESTFIXTURE_FLAGS)                  \
		$(TESTSRC) $(TESTSRC2) $(TOP)/src/tclsqlite.c                \
		-o testfixture$(EXE) $(LIBTCL) libsqlite4.a $(THREADLIB)

amalgamation-testfixture$(EXE): sqlite4.c $(TESTSRC) $(TOP)/src/tclsqlite.c
	$(TCCX) $(TCL_FLAGS) -DTCLSH=1 $(TESTFIXTURE_FLAGS)                  \
		$(TESTSRC) $(TOP)/src/tclsqlite.c sqlite4.c                  \
		-o testfixture$(EXE) $(LIBTCL) $(THREADLIB)

fts3-testfixture$(EXE): sqlite4.c fts3amal.c $(TESTSRC) $(TOP)/src/tclsqlite.c

# 
threadtest3$(EXE): sqlite4.o $(TOP)/test/threadtest3.c $(TOP)/test/tt3_checkpoint.c
	$(TCCX) -O2 sqlite4.o $(TOP)/test/threadtest3.c \
		-o threadtest3$(EXE) $(THREADLIB)

threadtest: threadtest3$(EXE)
	./threadtest3$(EXE)

SQLSRC = $(TOP)/lsm-test/sqltest.c $(TOP)/lsm-test/lsmtest_util.c 
sqltest$(EXE): $(SQLSRC) libsqlite4.a
	$(TCCX) $(TOP)/lsm-test/sqltest.c \
        -o sqltest$(EXE) -lsqlite3 libsqlite4.a $(THREADLIB)

TEST_EXTENSION = $(SHPREFIX)testloadext.$(SO)
$(TEST_EXTENSION): $(TOP)/test/test_loadext.c
	$(MKSHLIB) $(TOP)/test/test_loadext.c -o $(TEST_EXTENSION)

extensiontest: testfixture$(EXE) $(TEST_EXTENSION)
	./testfixture$(EXE) $(TOP)/test/loadext.test

        len = sqlite4GetToken(zCsr, &token);
      } while( token==TK_SPACE );
      assert( len>0 );
    } while( token!=TK_LP && token!=TK_USING );

    zRet = sqlite4MPrintf(db, "%.*s\"%w\"%s", ((u8*)tname.z) - zSql, zSql, 
       zTableName, tname.z+tname.n);
    sqlite4_result_text(context, zRet, -1, SQLITE4_TRANSIENT, 0);
    sqlite4DbFree(db, zRet);
  }
}

/*
** This C function implements an SQL user function that is used by SQL code
** generated by the ALTER TABLE ... RENAME command to modify the definition

        zInput = &z[n];
      }
      sqlite4DbFree(db, zParent);
    }
  }

  zResult = sqlite4MPrintf(db, "%s%s", (zOutput?zOutput:""), zInput), 
  sqlite4_result_text(context, zResult, -1, SQLITE4_TRANSIENT, 0);
  sqlite4DbFree(db, zOutput);
  sqlite4DbFree(db, zResult);
}
#endif

#ifndef SQLITE4_OMIT_TRIGGER
/* This function is used by SQL generated to implement the

    } while( dist!=2 || (token!=TK_WHEN && token!=TK_FOR && token!=TK_BEGIN) );

    /* Variable tname now contains the token that is the old table-name
    ** in the CREATE TRIGGER statement.
    */
    zRet = sqlite4MPrintf(db, "%.*s\"%w\"%s", ((u8*)tname.z) - zSql, zSql, 
       zTableName, tname.z+tname.n);
    sqlite4_result_text(context, zRet, -1, SQLITE4_TRANSIENT, 0);
    sqlite4DbFree(db, zRet);
  }
}
#endif   /* !SQLITE4_OMIT_TRIGGER */

/*
** Register built-in functions used to help implement ALTER TABLE

  if( pDb->pKV->iTransLevel ){
    sqlite4_snprintf(zErr,sizeof(zErr), "database %s is locked", zName);
    goto detach_error;
  }

  sqlite4KVStoreClose(pDb->pKV);
  pDb->pKV = 0;
  sqlite4SchemaClear(db->pEnv, pDb->pSchema);
  sqlite4DbFree(db, pDb->pSchema);
  pDb->pSchema = 0;
  sqlite4ResetInternalSchema(db, -1);
  return;

detach_error:
  sqlite4_result_error(context, zErr, -1);
}

    db->xCollNeeded(db->pCollNeededArg, db, enc, zExternal);
    sqlite4DbFree(db, zExternal);
  }
#ifndef SQLITE4_OMIT_UTF16
  if( db->xCollNeeded16 ){
    char const *zExternal;
    sqlite4_value *pTmp = sqlite4ValueNew(db);
    sqlite4ValueSetStr(pTmp, -1, zName, SQLITE4_UTF8, SQLITE4_STATIC, 0);
    zExternal = sqlite4ValueText(pTmp, SQLITE4_UTF16NATIVE);
    if( zExternal ){
      db->xCollNeeded16(db->pCollNeededArg, db, (int)ENC(db), zExternal);
    }
    sqlite4ValueFree(pTmp);
  }
#endif

  int rc = SQLITE4_NOMEM;

#ifndef SQLITE4_OMIT_AUTOINIT
  rc = sqlite4_initialize(0);
  if( rc ) return rc;
#endif
  pVal = sqlite4ValueNew(0);
  sqlite4ValueSetStr(pVal, -1, zSql, SQLITE4_UTF16NATIVE, SQLITE4_STATIC, 0);
  zSql8 = sqlite4ValueText(pVal, SQLITE4_UTF8);
  if( zSql8 ){
    rc = sqlite4_complete(zSql8);
  }else{
    rc = SQLITE4_NOMEM;
  }
  sqlite4ValueFree(pVal);

){
  DateTime x;
  if( isDate(context, argc, argv, &x)==0 ){
    char zBuf[100];
    computeYMD_HMS(&x);
    sqlite4_snprintf(zBuf,sizeof(zBuf), "%04d-%02d-%02d %02d:%02d:%02d",
                     x.Y, x.M, x.D, x.h, x.m, (int)(x.s));
    sqlite4_result_text(context, zBuf, -1, SQLITE4_TRANSIENT, 0);
  }
}

/*
**    time( TIMESTRING, MOD, MOD, ...)
**
** Return HH:MM:SS
*/
static void timeFunc(
  sqlite4_context *context,
  int argc,
  sqlite4_value **argv
){
  DateTime x;
  if( isDate(context, argc, argv, &x)==0 ){
    char zBuf[100];
    computeHMS(&x);
    sqlite4_snprintf(zBuf,sizeof(zBuf), "%02d:%02d:%02d", x.h, x.m, (int)x.s);
    sqlite4_result_text(context, zBuf, -1, SQLITE4_TRANSIENT, 0);
  }
}

/*
**    date( TIMESTRING, MOD, MOD, ...)
**
** Return YYYY-MM-DD
*/
static void dateFunc(
  sqlite4_context *context,
  int argc,
  sqlite4_value **argv
){
  DateTime x;
  if( isDate(context, argc, argv, &x)==0 ){
    char zBuf[100];
    computeYMD(&x);
    sqlite4_snprintf(zBuf,sizeof(zBuf), "%04d-%02d-%02d", x.Y, x.M, x.D);
    sqlite4_result_text(context, zBuf, -1, SQLITE4_TRANSIENT, 0);
  }
}

/*
**    strftime( FORMAT, TIMESTRING, MOD, MOD, ...)
**
** Return a string described by FORMAT.  Conversions as follows:

        }
        default:   z[j++] = '%'; break;
      }
    }
  }
  z[j] = 0;
  sqlite4_result_text(context, z, -1,
                      z==zBuf ? SQLITE4_TRANSIENT : SQLITE4_DYNAMIC, 0);
}

/*
** current_time()
**
** This function returns the same value as time('now').
*/

  sqlite4_mutex_enter(sqlite4MutexAlloc(SQLITE4_MUTEX_STATIC_MASTER));
  pTm = gmtime(&t);
  if( pTm ) memcpy(&sNow, pTm, sizeof(sNow));
  sqlite4_mutex_leave(sqlite4MutexAlloc(SQLITE4_MUTEX_STATIC_MASTER));
#endif
  if( pTm ){
    strftime(zBuf, 20, zFormat, &sNow);
    sqlite4_result_text(context, zBuf, -1, SQLITE4_TRANSIENT, 0);
  }
}
#endif

/*
** This function registered all of the above C functions as SQL
** functions.  This should be the only routine in this file with

  }
  sqlite4SelectDestInit(&dest, SRT_EphemTab, iCur);
  sqlite4Select(pParse, pDup, &dest);
  sqlite4SelectDelete(db, pDup);
}
#endif /* !defined(SQLITE4_OMIT_VIEW) && !defined(SQLITE4_OMIT_TRIGGER) */

#if defined(SQLITE4_ENABLE_UPDATE_DELETE_LIMIT) \
 && !defined(SQLITE4_OMIT_SUBQUERY)
/*
** Generate an expression tree to implement the WHERE, ORDER BY,
** and LIMIT/OFFSET portion of DELETE and UPDATE statements.
**
**     DELETE FROM table_wxyz WHERE a<5 ORDER BY a LIMIT 1;
**                            \__________________________/
**                               pLimitWhere (pInClause)

limit_where_cleanup_2:
  sqlite4ExprDelete(pParse->db, pWhere);
  sqlite4ExprListDelete(pParse->db, pOrderBy);
  sqlite4ExprDelete(pParse->db, pLimit);
  sqlite4ExprDelete(pParse->db, pOffset);
  return 0;
}
#endif /* defined(SQLITE4_ENABLE_UPDATE_DELETE_LIMIT) */
       /* && !defined(SQLITE4_OMIT_SUBQUERY) */

/*
** Generate code for a DELETE FROM statement.
**
**     DELETE FROM table_wxyz WHERE a<5 AND b NOT NULL;
**                 \________/       \________________/
**                  pTabList              pWhere

  ** delete.  */
  if( sqlite4IsReadOnly(pParse, pTab, pTrigger!=0) ) goto delete_from_cleanup;
  assert( !IsView(pTab) || pTrigger );
  assert( !IsView(pTab) || pTab->pIndex==0 );

  /* Invoke the authorization callback */
  rcauth = sqlite4AuthCheck(pParse, SQLITE4_DELETE, pTab->zName, 0, zDb);
  assert( rcauth==SQLITE4_OK || rcauth==SQLITE4_DENY
         || rcauth==SQLITE4_IGNORE );
  if( rcauth==SQLITE4_DENY ){
    goto delete_from_cleanup;
  }

  /* Assign a cursor number to the table or view this statement is 
  ** deleting from. If pTab is actually a view, this will be used as the
  ** ephemeral table cursor. 

/*
** 2013 January 7
**
** 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 file contains code used to help implement the sqlite4_env object.
*/
#include "sqliteInt.h"


/*
** Default factory objects
*/
static KVFactory memFactory = {
   0,
   "temp",
   sqlite4KVStoreOpenMem,
   1
};
KVFactory sqlite4BuiltinFactory = {
   &memFactory,
   "main",
   sqlite4KVStoreOpenLsm,
   1
};

/*
** The following singleton contains the global configuration for
** the SQLite library.
*/
struct sqlite4_env sqlite4DefaultEnv = {
   sizeof(sqlite4_env),       /* nByte */
   1,                         /* iVersion */
   SQLITE4_DEFAULT_MEMSTATUS, /* bMemstat */
   1,                         /* bCoreMutex */
   SQLITE4_THREADSAFE==1,     /* bFullMutex */
   0x7ffffffe,                /* mxStrlen */
   128,                       /* szLookaside */
   500,                       /* nLookaside */
   &sqlite4MMSystem,          /* pMM */
   {0,0,0,0,0,0,0,0,0},       /* m */
   {0,0,0,0,0,0,0,0,0,0},     /* mutex */
   (void*)0,                  /* pHeap */
   0,                         /* nHeap */
   0, 0,                      /* mnHeap, mxHeap */
   0,                         /* mxParserStack */
   &sqlite4BuiltinFactory,    /* pFactory */
   sqlite4OsRandomness,       /* xRandomness */
   sqlite4OsCurrentTime,      /* xCurrentTime */
   /* All the rest should always be initialized to zero */
   0,                         /* isInit */
   0,                         /* pFactoryMutex */
   0,                         /* pPrngMutex */
   0, 0,                      /* prngX, prngY */
   0,                         /* xLog */
   0,                         /* pLogArg */
   0,                         /* bLocaltimeFault */
   0,                         /* pMemMutex */
   {0,0,0,0},                 /* nowValue[] */
   {0,0,0,0},                 /* mxValue[] */
   {0,}                       /* hashGlobalFunc */
};

/*
** Return the default environment
*/
sqlite4_env *sqlite4_env_default(void){ return &sqlite4DefaultEnv; }

/*
** Initialize SQLite.  
**
** This routine must be called to initialize the run-time environment
** As long as you do not compile with SQLITE4_OMIT_AUTOINIT
** this routine will be called automatically by key routines such as
** sqlite4_open().  
**
** This routine is a no-op except on its very first call for a given
** sqlite4_env object, or for the first call after a call to sqlite4_shutdown.
**
** This routine is not threadsafe.  It should be called from a single
** thread to initialized the library in a multi-threaded system.  Other
** threads should avoid using the sqlite4_env object until after it has
** completely initialized.
*/
int sqlite4_initialize(sqlite4_env *pEnv){
  MUTEX_LOGIC( sqlite4_mutex *pMaster; )       /* The main static mutex */
  int rc;                                      /* Result code */

  if( pEnv==0 ) pEnv = &sqlite4DefaultEnv;

  /* If SQLite is already completely initialized, then this call
  ** to sqlite4_initialize() should be a no-op.  But the initialization
  ** must be complete.  So isInit must not be set until the very end
  ** of this routine.
  */
  if( pEnv->isInit ) return SQLITE4_OK;

  /* Initialize the mutex subsystem
  */
  rc = sqlite4MutexInit(pEnv);
  if( rc ){
    sqlite4MallocEnd(pEnv);
    return rc;
  }

  /* Initialize the memory allocation subsystem
  */
  rc = sqlite4MallocInit(pEnv);
  if( rc ) return rc;

  /* Create required mutexes
  */
  if( pEnv->bCoreMutex ){
    pEnv->pMemMutex = sqlite4MutexAlloc(pEnv, SQLITE4_MUTEX_FAST);
    pEnv->pPrngMutex = sqlite4MutexAlloc(pEnv, SQLITE4_MUTEX_FAST);
    pEnv->pFactoryMutex = sqlite4MutexAlloc(pEnv, SQLITE4_MUTEX_FAST);
    if( pEnv->pMemMutex==0
     || pEnv->pPrngMutex==0
     || pEnv->pFactoryMutex==0
    ){
      rc = SQLITE4_NOMEM;
    }
  }else{
    pEnv->pMemMutex = 0;
    pEnv->pPrngMutex = 0;
  }
  pEnv->isInit = 1;

  sqlite4OsInit(pEnv);

  /* Register global functions */
  if( rc==SQLITE4_OK ){
    sqlite4RegisterGlobalFunctions(pEnv);
  }

  /* The following is just a sanity check to make sure SQLite has
  ** been compiled correctly.  It is important to run this code, but
  ** we don't want to run it too often and soak up CPU cycles for no
  ** reason.  So we run it once during initialization.
  */
#ifndef NDEBUG
#ifndef SQLITE4_OMIT_FLOATING_POINT
  /* This section of code's only "output" is via assert() statements. */
  if ( rc==SQLITE4_OK ){
    u64 x = (((u64)1)<<63)-1;
    double y;
    assert(sizeof(x)==8);
    assert(sizeof(x)==sizeof(y));
    memcpy(&y, &x, 8);
    assert( sqlite4IsNaN(y) );
  }
#endif
#endif

  return rc;
}

/*
** Undo the effects of sqlite4_initialize().  Must not be called while
** there are outstanding database connections or memory allocations or
** while any part of SQLite is otherwise in use in any thread.  This
** routine is not threadsafe.  But it is safe to invoke this routine
** on when SQLite is already shut down.  If SQLite is already shut down
** when this routine is invoked, then this routine is a harmless no-op.
*/
int sqlite4_shutdown(sqlite4_env *pEnv){
  if( pEnv==0 ) pEnv = &sqlite4DefaultEnv;
  if( pEnv->isInit ){
    KVFactory *pMkr;
    sqlite4_mutex_free(pEnv->pFactoryMutex);
    sqlite4_mutex_free(pEnv->pPrngMutex);
    sqlite4_mutex_free(pEnv->pMemMutex);
    pEnv->pMemMutex = 0;
    while( (pMkr = pEnv->pFactory)!=0 && pMkr->isPerm==0 ){
      KVFactory *pNext = pMkr->pNext;
      sqlite4_free(pEnv, pMkr);
      pMkr = pNext;
    }
    sqlite4MutexEnd(pEnv);
    sqlite4MallocEnd(pEnv);
    pEnv->isInit = 0;
  }
  return SQLITE4_OK;
}

/*
** Return the size of an sqlite4_env object
*/
int sqlite4_env_size(void){ return sizeof(sqlite4_env); }

/*
** This API allows applications to modify the configuration described by
** an sqlite4_env object.
*/
int sqlite4_env_config(sqlite4_env *pEnv, int op, ...){
  va_list ap;
  int rc = SQLITE4_OK;

  if( pEnv==0 ) pEnv = sqlite4_env_default();

  va_start(ap, op);
  switch( op ){
    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_INIT, template);
    **
    ** Turn bulk memory into a new sqlite4_env object.  The template is
    ** a prior sqlite4_env that is used as a template in initializing the
    ** new sqlite4_env object.  The size of the bulk memory must be at
    ** least as many bytes as returned from sqlite4_env_size().
    */
    case SQLITE4_ENVCONFIG_INIT: {
      /* Disable all mutexing */
      sqlite4_env *pTemplate = va_arg(ap, sqlite4_env*);
      int n = pTemplate->nByte;
      if( n>sizeof(sqlite4_env) ) n = sizeof(sqlite4_env);
      memcpy(pEnv, pTemplate, n);
      pEnv->pFactory = &sqlite4BuiltinFactory;
      pEnv->isInit = 0;
      break;
    }

    /* Mutex configuration options are only available in a threadsafe
    ** compile. 
    */
#if defined(SQLITE4_THREADSAFE) && SQLITE4_THREADSAFE>0
    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_SINGLETHREAD);
    **
    ** Configure this environment for a single-threaded application.
    */
    case SQLITE4_ENVCONFIG_SINGLETHREAD: {
      /* Disable all mutexing */
      if( pEnv->isInit ){ rc = SQLITE4_MISUSE; break; }
      pEnv->bCoreMutex = 0;
      pEnv->bFullMutex = 0;
      break;
    }

    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_MULTITHREAD);
    **
    ** Configure this environment for a multi-threaded application where
    ** the same database connection is never used by more than a single
    ** thread at a time.
    */
    case SQLITE4_ENVCONFIG_MULTITHREAD: {
      /* Disable mutexing of database connections */
      /* Enable mutexing of core data structures */
      if( pEnv->isInit ){ rc = SQLITE4_MISUSE; break; }
      pEnv->bCoreMutex = 1;
      pEnv->bFullMutex = 0;
      break;
    }

    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_SERIALIZED);
    **
    ** Configure this environment for an unrestricted multi-threaded
    ** application where any thread can do whatever it wants with any
    ** database connection at any time.
    */
    case SQLITE4_ENVCONFIG_SERIALIZED: {
      /* Enable all mutexing */
      if( pEnv->isInit ){ rc = SQLITE4_MISUSE; break; }
      pEnv->bCoreMutex = 1;
      pEnv->bFullMutex = 1;
      break;
    }

    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_MUTEXT, sqlite4_mutex_methods*)
    **
    ** Configure this environment to use the mutex routines specified by the
    ** argument.
    */
    case SQLITE4_ENVCONFIG_MUTEX: {
      /* Specify an alternative mutex implementation */
      if( pEnv->isInit ){ rc = SQLITE4_MISUSE; break; }
      pEnv->mutex = *va_arg(ap, sqlite4_mutex_methods*);
      break;
    }

    /*
    ** sqlite4_env_config(p, SQLITE4_ENVCONFIG_GETMUTEX, sqlite4_mutex_methods*)
    **
    ** Copy the mutex routines in use by this environment into the structure
    ** given in the argument.
    */
    case SQLITE4_ENVCONFIG_GETMUTEX: {
      /* Retrieve the current mutex implementation */
      *va_arg(ap, sqlite4_mutex_methods*) = pEnv->mutex;
      break;
    }
#endif


    /*
    ** sqlite4_env_config(p, SQLITE4_ENVCONFIG_MALLOC, sqlite4_mem_methods*)
    **
    ** Set the memory allocation routines to be used by this environment.
    */
    case SQLITE4_ENVCONFIG_MALLOC: {
      /* Specify an alternative malloc implementation */
      if( pEnv->isInit ) return SQLITE4_MISUSE;
      pEnv->m = *va_arg(ap, sqlite4_mem_methods*);
      break;
    }

    /*
    ** sqlite4_env_config(p, SQLITE4_ENVCONFIG_GETMALLOC, sqlite4_mem_methods*)
    **
    ** Copy the memory allocation routines in use by this environment
    ** into the structure given in the argument.
    */
    case SQLITE4_ENVCONFIG_GETMALLOC: {
      /* Retrieve the current malloc() implementation */
      if( pEnv->m.xMalloc==0 ) sqlite4MemSetDefault(pEnv);
      *va_arg(ap, sqlite4_mem_methods*) = pEnv->m;
      break;
    }

    /* sqlite4_env_config(p, SQLITE4_ENVCONFIG_MEMSTAT, int onoff);
    **
    ** Enable or disable collection of memory usage statistics according to
    ** the onoff parameter.  
    */
    case SQLITE4_ENVCONFIG_MEMSTATUS: {
      /* Enable or disable the malloc status collection */
      pEnv->bMemstat = va_arg(ap, int);
      break;
    }

    /*
    ** sqlite4_env_config(p, SQLITE4_ENVCONFIG_LOOKASIDE, size, count);
    **
    ** Set the default lookaside memory settings for all subsequent
    ** database connections constructed in this environment.  The size
    ** parameter is the size of each lookaside memory buffer and the
    ** count parameter is the number of lookaside buffers.  Set both
    ** to zero to disable lookaside memory.
    */
    case SQLITE4_ENVCONFIG_LOOKASIDE: {
      pEnv->szLookaside = va_arg(ap, int);
      pEnv->nLookaside = va_arg(ap, int);
      break;
    }
    
    /*
    ** sqlite4_env_config(p, SQLITE4_ENVCONFIG_LOG, xOutput, pArg);
    **
    ** Set the log function that is called in response to sqlite4_log()
    ** calls.
    */
    case SQLITE4_ENVCONFIG_LOG: {
      /* MSVC is picky about pulling func ptrs from va lists.
      ** http://support.microsoft.com/kb/47961
      ** pEnv->xLog = va_arg(ap, void(*)(void*,int,const char*));
      */
      typedef void(*LOGFUNC_t)(void*,int,const char*);
      pEnv->xLog = va_arg(ap, LOGFUNC_t);
      pEnv->pLogArg = va_arg(ap, void*);
      break;
    }

    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_KVSTORE_PUSH, zName,xFactory);
    **
    ** Push a new KVStore factory onto the factory stack.  The new factory
    ** takes priority over prior factories.
    */
    case SQLITE4_ENVCONFIG_KVSTORE_PUSH: {
      const char *zName = va_arg(ap, const char*);
      int nName = sqlite4Strlen30(zName);
      KVFactory *pMkr = sqlite4_malloc(pEnv, sizeof(*pMkr)+nName+1);
      char *z;
      if( pMkr==0 ) return SQLITE4_NOMEM;
      z = (char*)&pMkr[1];
      memcpy(z, zName, nName+1);
      memset(pMkr, 0, sizeof(*pMkr));
      pMkr->zName = z;
      pMkr->xFactory = va_arg(ap, sqlite4_kvfactory);
      sqlite4_mutex_enter(pEnv->pFactoryMutex);
      pMkr->pNext = pEnv->pFactory;
      pEnv->pFactory = pMkr;
      sqlite4_mutex_leave(pEnv->pFactoryMutex);
      break;
    }

    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_KVSTORE_POP, zName, &pxFact);
    **
    ** Remove a KVStore factory from the stack.
    */
    /*
    ** sqlite4_env_config(pEnv, SQLITE4_ENVCONFIG_KVSTORE_GET, zName, &pxFact);
    **
    ** Get the current factory pointer with the given name but leave the
    ** factory on the stack.
    */
    case SQLITE4_ENVCONFIG_KVSTORE_POP:
    case SQLITE4_ENVCONFIG_KVSTORE_GET: {
      typedef int (**PxFact)(sqlite4_env*,KVStore**,const char*,unsigned);
      const char *zName = va_arg(ap, const char*);
      KVFactory *pMkr, **ppPrev;
      PxFact pxFact;

      pxFact = va_arg(ap,PxFact);
      *pxFact = 0;
      sqlite4_mutex_enter(pEnv->pFactoryMutex);
      ppPrev = &pEnv->pFactory;
      pMkr = *ppPrev;
      while( pMkr && strcmp(zName, pMkr->zName)!=0 ){
        ppPrev = &pMkr->pNext;
        pMkr = *ppPrev;
      }
      if( pMkr ){
        *pxFact = pMkr->xFactory;
        if( op==SQLITE4_ENVCONFIG_KVSTORE_POP && pMkr->isPerm==0 ){
          *ppPrev = pMkr->pNext;
          sqlite4_free(pEnv, pMkr);
        }
      }
      sqlite4_mutex_leave(pEnv->pFactoryMutex);
      break;
    }


    default: {
      rc = SQLITE4_ERROR;
      break;
    }
  }
  va_end(ap);
  return rc;
}

** for this node and for the pToken argument is a single allocation
** obtained from sqlite4DbMalloc().  The calling function
** is responsible for making sure the node eventually gets freed.
**
** If dequote is true, then the token (if it exists) is dequoted.
** If dequote is false, no dequoting is performance.  The deQuote
** parameter is ignored if pToken is NULL or if the token does not
** appear to be quoted.

**
** Special case:  If op==TK_INTEGER and pToken points to a string that
** can be translated into a 32-bit integer, then the token is not
** stored in u.zToken.  Instead, the integer values is written
** into u.iValue and the EP_IntValue flag is set.  No extra storage
** is allocated to hold the integer text and the dequote flag is ignored.
*/

      }else{
        int c;
        pNew->u.zToken = (char*)&pNew[1];
        assert( pToken->z!=0 || pToken->n==0 );
        if( pToken->n ) memcpy(pNew->u.zToken, pToken->z, pToken->n);
        pNew->u.zToken[pToken->n] = 0;
        if( dequote && nExtra>=3 
             && ((c = pToken->z[0])=='\'' || c=='"' || c=='[') ){
          sqlite4Dequote(pNew->u.zToken);

        }
      }
    }
#if SQLITE4_MAX_EXPR_DEPTH>0
    pNew->nHeight = 1;
#endif  
  }

    if( zErr==0 ){
      zErr = sqlite4MPrintf(db, "error parsing expression: %d", rc);
    }
    goto fts5_parse_expr_out;
  }

  fts5PrintExpr(db, azCol, pExpr, &zRet);
  sqlite4_result_text(pCtx, zRet, -1, SQLITE4_TRANSIENT, 0);
  fts5ExpressionFree(db, pExpr);
  sqlite4_free(sqlite4_db_env(db), zRet);

 fts5_parse_expr_out:
  if( p ) pTok->xDestroy(p);
  sqlite4DbFree(db, azCol);
  sqlite4_finalize(pStmt);

  int rc = sqlite4_create_function(
      db, "fts5_parse_expr", 3, SQLITE4_UTF8, 0, fts5_parse_expr, 0, 0
  );
  if( rc!=SQLITE4_OK ) return rc;
#endif
  return sqlite4InitFts5Func(db);
}

struct Fts5RankCtx {
  sqlite4 *db;
  double *aAvgdl;                 /* Average document size of each field */
  int nPhrase;                    /* Number of phrases in query */
  double *aIdf;                   /* IDF weights for each phrase in query */
};

static void fts5RankFreeCtx(void *pNotUsed, void *pCtx){
  if( pCtx ){
    Fts5RankCtx *p = (Fts5RankCtx *)pCtx;
    sqlite4DbFree(p->db, p);
  }
}

#define BM25_EXPLAIN  0x01

  if( p==0 ){
    int nPhrase;                  /* Number of phrases in query expression */
    int nByte;                    /* Number of bytes of data to allocate */

    sqlite4_mi_phrase_count(pCtx, &nPhrase);
    nByte = sizeof(Fts5RankCtx) + (nPhrase+nField) * sizeof(double);
    p = (Fts5RankCtx *)sqlite4DbMallocZero(db, nByte);
    sqlite4_set_auxdata(pCtx, 0, (void *)p, fts5RankFreeCtx, 0);
    p = sqlite4_get_auxdata(pCtx, 0);

    if( !p ){
      rc = SQLITE4_NOMEM;
    }else{
      int N;                      /* Total number of docs in collection */
      int ni;                     /* Number of docs with phrase i */

  }

  if( rc==SQLITE4_OK ){
    if( bExplain ){
      zExplain = sqlite4MAppendf(
          db, zExplain, "%s</table><b>overall rank=%.2f</b>", zExplain, rank
      );
      sqlite4_result_text(pCtx, zExplain, -1, SQLITE4_TRANSIENT, 0);
    }else{
      sqlite4_result_double(pCtx, rank);
    }
  }else{
    sqlite4_result_error_code(pCtx, rc);
  }
  sqlite4DbFree(db, zExplain);

  iOff += nShift;
  mask = mask >> nShift;

  pSnip->iOff = iOff;
  pSnip->hlmask = mask;
}

/*
** Parameter aSnip points to an array of nSnip Snippet objects, where nSnip
** is less than or equal to 4. This function sorts the array in place in
** ascending order of Snippet.iCol and Snippet.iOff. 
*/
static void fts5SnippetSort(Snippet *aSnip, int nSnip){
  Snippet aTmp[4];
  int i;

  assert( nSnip<=4 && nSnip>=1 );

  for(i=0; i<nSnip; i++){
    int iBest = -1;
    int iTry;
    for(iTry=0; iTry<nSnip; iTry++){
      Snippet *pTry = &aSnip[iTry];
      if( pTry->iCol>=0 && (iBest<0 
         || pTry->iCol<aSnip[iBest].iCol
         || (pTry->iCol==aSnip[iBest].iCol && pTry->iOff<aSnip[iBest].iOff)
      )){
        iBest = iTry;
      }
    }

    assert( iBest>=0 );
    memcpy(&aTmp[i], &aSnip[iBest], sizeof(Snippet));
    aSnip[iBest].iCol = -1;
  }

  memcpy(aSnip, aTmp, sizeof(Snippet)*nSnip);
}


static void fts5Snippet(sqlite4_context *pCtx, int nArg, sqlite4_value **apArg){
  Snippet aSnip[4];
  int nSnip;
  int iCol = -1;
  int nToken = -15;
  int rc;

    memset(aSnip, 0, sizeof(aSnip));
    for(i=0; rc==SQLITE4_OK && i<nSnip; i++){
      rc = fts5BestSnippet(pCtx, iCol, &mask, nTok, &aSnip[i]);
    }
    if( mask==0 || nSnip==4 ){
      SnippetText text = {0, 0, 0};
      fts5SnippetSort(aSnip, nSnip);
      for(i=0; rc==SQLITE4_OK && i<nSnip; i++){
        int nSz;
        rc = sqlite4_mi_size(pCtx, aSnip[i].iCol, -1, &nSz);
        if( rc==SQLITE4_OK ){
          fts5SnippetImprove(pCtx, nTok, nSz, &aSnip[i]);
          rc = fts5SnippetText(
              pCtx, &aSnip[i], &text, nTok, zStart, zEnd, zEllipses
          );
        }
      }
      sqlite4_result_text(pCtx, text.zOut, text.nOut, SQLITE4_TRANSIENT, 0);
      sqlite4DbFree(sqlite4_context_db_handle(pCtx), text.zOut);
      break;
    }
  }

  if( rc!=SQLITE4_OK ){
    sqlite4_result_error_code(pCtx, rc);

    void *p = SQLITE4_INT_TO_PTR(aRank[i].mask);
    const char *z = aRank[i].zName;
    rc = sqlite4_create_mi_function(db, z, -1, SQLITE4_UTF8, p, fts5Rank, 0);
  }

  return rc;
}

  switch( sqlite4_value_type(argv[0]) ){
    case SQLITE4_INTEGER: z = "integer"; break;
    case SQLITE4_TEXT:    z = "text";    break;
    case SQLITE4_FLOAT:   z = "real";    break;
    case SQLITE4_BLOB:    z = "blob";    break;
    default:             z = "null";    break;
  }
  sqlite4_result_text(context, z, -1, SQLITE4_STATIC, 0);
}


/*
** Implementation of the length() function
*/
static void lengthFunc(

    while( *z && p1 ){
      SQLITE4_SKIP_UTF8(z);
      p1--;
    }
    for(z2=z; *z2 && p2; p2--){
      SQLITE4_SKIP_UTF8(z2);
    }
    sqlite4_result_text(context, (char*)z, (int)(z2-z), SQLITE4_TRANSIENT, 0);
  }else{
    if( p1+p2>len ){
      p2 = len-p1;
      if( p2<0 ) p2 = 0;
    }
    sqlite4_result_blob(context, (char*)&z[p1], (int)p2, SQLITE4_TRANSIENT, 0);
  }
}

/*
** Implementation of the round() function
*/
#ifndef SQLITE4_OMIT_FLOATING_POINT

  assert( z2==(char*)sqlite4_value_text(argv[0]) );
  if( z2 ){
    z1 = contextMalloc(context, ((i64)n)+1);
    if( z1 ){
      for(i=0; i<n; i++){
        z1[i] = (char)sqlite4Toupper(z2[i]);
      }
      sqlite4_result_text(context, z1, n, SQLITE4_DYNAMIC, 0);
    }
  }
}
static void lowerFunc(sqlite4_context *context, int argc, sqlite4_value **argv){
  char *z1;
  const char *z2;
  int i, n;
  UNUSED_PARAMETER(argc);
  z2 = (char*)sqlite4_value_text(argv[0]);
  n = sqlite4_value_bytes(argv[0]);
  /* Verify that the call to _bytes() does not invalidate the _text() pointer */
  assert( z2==(char*)sqlite4_value_text(argv[0]) );
  if( z2 ){
    z1 = contextMalloc(context, ((i64)n)+1);
    if( z1 ){
      for(i=0; i<n; i++){
        z1[i] = sqlite4Tolower(z2[i]);
      }
      sqlite4_result_text(context, z1, n, SQLITE4_DYNAMIC, 0);
    }
  }
}


#if 0  /* This function is never used. */
/*

  n = sqlite4_value_int(argv[0]);
  if( n<1 ){
    n = 1;
  }
  p = contextMalloc(context, n);
  if( p ){
    sqlite4_randomness(sqlite4_context_env(context), n, p);
    sqlite4_result_blob(context, (char*)p, n, SQLITE4_DYNAMIC, 0);
  }

















}

/*
** Implementation of the changes() SQL function.
**
** IMP: R-62073-11209 The changes() SQL function is a wrapper
** around the sqlite4_changes() C/C++ function and hence follows the same

  sqlite4_context *context,
  int NotUsed,
  sqlite4_value **NotUsed2
){
  UNUSED_PARAMETER2(NotUsed, NotUsed2);
  /* IMP: R-48699-48617 This function is an SQL wrapper around the
  ** sqlite4_libversion() C-interface. */
  sqlite4_result_text(context, sqlite4_libversion(), -1, SQLITE4_STATIC, 0);
}

/*
** Implementation of the sqlite_source_id() function. The result is a string
** that identifies the particular version of the source code used to build
** SQLite.
*/
static void sourceidFunc(
  sqlite4_context *context,
  int NotUsed,
  sqlite4_value **NotUsed2
){
  UNUSED_PARAMETER2(NotUsed, NotUsed2);
  /* IMP: R-24470-31136 This function is an SQL wrapper around the
  ** sqlite4_sourceid() C interface. */
  sqlite4_result_text(context, sqlite4_sourceid(), -1, SQLITE4_STATIC, 0);
}

/*
** Implementation of the sqlite_log() function.  This is a wrapper around
** sqlite4_log().  The return value is NULL.  The function exists purely for
** its side-effects.
*/

  int n;
  assert( argc==1 );
  UNUSED_PARAMETER(argc);
  /* IMP: R-04922-24076 The sqlite_compileoption_get() SQL function
  ** is a wrapper around the sqlite4_compileoption_get() C/C++ function.
  */
  n = sqlite4_value_int(argv[0]);
  sqlite4_result_text(context, sqlite4_compileoption_get(n), -1,
                      SQLITE4_STATIC, 0);
}
#endif /* SQLITE4_OMIT_COMPILEOPTION_DIAGS */

/* Array for converting from half-bytes (nybbles) into ASCII hex
** digits. */
static const char hexdigits[] = {
  '0', '1', '2', '3', '4', '5', '6', '7',

          zText[(i*2)+2] = hexdigits[(zBlob[i]>>4)&0x0F];
          zText[(i*2)+3] = hexdigits[(zBlob[i])&0x0F];
        }
        zText[(nBlob*2)+2] = '\'';
        zText[(nBlob*2)+3] = '\0';
        zText[0] = 'x';
        zText[1] = '\'';
        sqlite4_result_text(context, zText, -1, SQLITE4_TRANSIENT, 0);
        sqlite4_free(sqlite4_context_env(context), zText);
      }
      break;
    }
    case SQLITE4_TEXT: {
      int i,j;
      u64 n;

          z[j++] = zArg[i];
          if( zArg[i]=='\'' ){
            z[j++] = '\'';
          }
        }
        z[j++] = '\'';
        z[j] = 0;
        sqlite4_result_text(context, z, j, SQLITE4_DYNAMIC, 0);
      }
      break;
    }
    default: {
      assert( sqlite4_value_type(argv[0])==SQLITE4_NULL );
      sqlite4_result_text(context, "NULL", 4, SQLITE4_STATIC, 0);
      break;
    }
  }
}

/*
** The hex() function.  Interpret the argument as a blob.  Return

  if( zHex ){
    for(i=0; i<n; i++, pBlob++){
      unsigned char c = *pBlob;
      *(z++) = hexdigits[(c>>4)&0xf];
      *(z++) = hexdigits[c&0xf];
    }
    *z = 0;
    sqlite4_result_text(context, zHex, n*2, SQLITE4_DYNAMIC, 0);






















  }
}

/*
** The replace() function.  Three arguments are all strings: call
** them A, B, and C. The result is also a string which is derived
** from A by replacing every occurance of B with C.  The match

    }
  }
  assert( j+nStr-i+1==nOut );
  memcpy(&zOut[j], &zStr[i], nStr-i);
  j += nStr - i;
  assert( j<=nOut );
  zOut[j] = 0;
  sqlite4_result_text(context, (char*)zOut, j, SQLITE4_DYNAMIC, 0);
}

/*
** Implementation of the TRIM(), LTRIM(), and RTRIM() functions.
** The userdata is 0x1 for left trim, 0x2 for right trim, 0x3 for both.
*/
static void trimFunc(

        nIn -= len;
      }
    }
    if( zCharSet ){
      sqlite4_free(sqlite4_context_env(context), azChar);
    }
  }
  sqlite4_result_text(context, (char*)zIn, nIn, SQLITE4_TRANSIENT, 0);
}


/* IMP: R-25361-16150 This function is omitted from SQLite by default. It
** is only available if the SQLITE4_SOUNDEX compile-time option is used
** when SQLite is built.
*/

        prevcode = 0;
      }
    }
    while( j<4 ){
      zResult[j++] = '0';
    }
    zResult[j] = 0;
    sqlite4_result_text(context, zResult, 4, SQLITE4_TRANSIENT, 0);
  }else{
    /* IMP: R-64894-50321 The string "?000" is returned if the argument
    ** is NULL or contains no ASCII alphabetic characters. */
    sqlite4_result_text(context, "?000", 4, SQLITE4_STATIC, 0);
  }
}
#endif /* SQLITE4_SOUNDEX */

#if 0 /*ndef SQLITE4_OMIT_LOAD_EXTENSION*/
/*
** A function that loads a shared-library extension then returns NULL.

*/
static void countStep(sqlite4_context *context, int argc, sqlite4_value **argv){
  CountCtx *p;
  p = sqlite4_aggregate_context(context, sizeof(*p));
  if( (argc==0 || SQLITE4_NULL!=sqlite4_value_type(argv[0])) && p ){
    p->n++;
  }









}   
static void countFinalize(sqlite4_context *context){
  CountCtx *p;
  p = sqlite4_aggregate_context(context, 0);
  sqlite4_result_int64(context, p ? p->n : 0);
}

  if( pAccum ){
    if( pAccum->tooBig ){
      sqlite4_result_error_toobig(context);
    }else if( pAccum->mallocFailed ){
      sqlite4_result_error_nomem(context);
    }else{    
      sqlite4_result_text(context, sqlite4StrAccumFinish(pAccum), -1, 
                          SQLITE4_DYNAMIC, 0);
    }
  }
}

/*
** This routine does per-connection function registration.  Most
** of the built-in functions above are part of the global function set.

    FUNCTION(sqlite_source_id,   0, 0, 0, sourceidFunc     ),
    FUNCTION(sqlite_log,         2, 0, 0, errlogFunc       ),
#ifndef SQLITE4_OMIT_COMPILEOPTION_DIAGS
    FUNCTION(sqlite_compileoption_used,1, 0, 0, compileoptionusedFunc  ),
    FUNCTION(sqlite_compileoption_get, 1, 0, 0, compileoptiongetFunc  ),
#endif /* SQLITE4_OMIT_COMPILEOPTION_DIAGS */
    FUNCTION(quote,              1, 0, 0, quoteFunc        ),

    FUNCTION(changes,            0, 0, 0, changes          ),
    FUNCTION(total_changes,      0, 0, 0, total_changes    ),
    FUNCTION(replace,            3, 0, 0, replaceFunc      ),

  #ifdef SQLITE4_SOUNDEX
    FUNCTION(soundex,            1, 0, 0, soundexFunc      ),
  #endif
  #if 0 /*ndef SQLITE4_OMIT_LOAD_EXTENSION*/
    FUNCTION(load_extension,     1, 0, 0, loadExt          ),
    FUNCTION(load_extension,     2, 0, 0, loadExt          ),
  #endif

  0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,  /* e0..e7    ........ */
  0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,  /* e8..ef    ........ */
  0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,  /* f0..f7    ........ */
  0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40   /* f8..ff    ........ */
};
#endif



























































/*
** Constant tokens for values 0 and 1.
*/
const Token sqlite4IntTokens[] = {
   { "0", 1 },
   { "1", 1 }

    if( pIdx!=pPk ){
      for(i=0; i<pPk->nColumn; i++){
        int idx = pPk->aiColumn[i];
        sqlite4VdbeAddOp2(v, OP_SCopy, regContent+idx, regTmp+i+pIdx->nColumn);
      }
    }
    sqlite4VdbeAddOp3(v, OP_MakeIdxKey, iIdx, regTmp, regKey);



    VdbeComment((v, "key for %s", pIdx->zName));

    /* If Index.onError==OE_None, then pIdx is not a UNIQUE or PRIMARY KEY 
    ** index. In this case there is no need to test the index for uniqueness
    ** - all that is required is to populate the regKey register. Jump 
    ** to the next iteration of the loop if this is the case.  */
    onError = pIdx->onError;

    addr1 = sqlite4VdbeAddOp2(v, OP_NewRowid, iDest, regRowid);
  }else{
    addr1 = sqlite4VdbeAddOp2(v, OP_Rowid, iSrc, regRowid);
    assert( (pDest->tabFlags & TF_Autoincrement)==0 );
  }
  sqlite4VdbeAddOp2(v, OP_RowData, iSrc, regData);
  sqlite4VdbeAddOp3(v, OP_Insert, iDest, regData, regRowid);
  sqlite4VdbeChangeP5(v, OPFLAG_NCHANGE|OPFLAG_APPEND);
  sqlite4VdbeChangeP4(v, -1, pDest->zName, 0);
  sqlite4VdbeAddOp2(v, OP_Next, iSrc, addr1);
  for(pDestIdx=pDest->pIndex; pDestIdx; pDestIdx=pDestIdx->pNext){
    for(pSrcIdx=pSrc->pIndex; ALWAYS(pSrcIdx); pSrcIdx=pSrcIdx->pNext){
      if( xferCompatibleIndex(pDestIdx, pSrcIdx) ) break;
    }
    assert( pSrcIdx );

    memset(pNew, 0, sizeof(KVLsm));
    pNew->base.pStoreVfunc = &kvlsmMethods;
    pNew->base.pEnv = pEnv;
    rc = lsm_new(0, &pNew->pDb);
    if( rc==SQLITE4_OK ){
      int i;
      int bMmap = 1;
      lsm_config(pNew->pDb, LSM_CONFIG_MMAP, &bMmap);
      for(i=0; i<ArraySize(aConfig); i++){
        const char *zVal = sqlite4_uri_parameter(zName, aConfig[i].zParam);
        if( zVal ){
          int nVal = sqlite4Atoi(zVal);
          lsm_config(pNew->pDb, aConfig[i].eParam, &nVal);
        }

extern "C" {
#endif

/*
** Opaque handle types.
*/
typedef struct lsm_compress lsm_compress;   /* Compression library functions */
typedef struct lsm_compress_factory lsm_compress_factory;
typedef struct lsm_cursor lsm_cursor;       /* Database cursor handle */
typedef struct lsm_db lsm_db;               /* Database connection handle */
typedef struct lsm_env lsm_env;             /* Runtime environment */
typedef struct lsm_file lsm_file;           /* OS file handle */
typedef struct lsm_mutex lsm_mutex;         /* Mutex handle */

/* 64-bit integer type used for file offsets. */
typedef long long int lsm_i64;              /* 64-bit signed integer type */

/* Candidate values for the 3rd argument to lsm_env.xLock() */
#define LSM_LOCK_UNLOCK 0
#define LSM_LOCK_SHARED 1
#define LSM_LOCK_EXCL   2

/* Flags for lsm_env.xOpen() */
#define LSM_OPEN_READONLY 0x0001

/*
** CAPI: Database Runtime Environment
**
** Run-time environment used by LSM
**
** xMap(pFile, iOff, nByte, ppOut, pnOut):
**   Memory map nByte bytes of file pFile starting at file offset iOff. If

*/
struct lsm_env {
  int nByte;                 /* Size of this structure in bytes */
  int iVersion;              /* Version number of this structure (1) */
  /****** file i/o ***********************************************/
  void *pVfsCtx;
  int (*xFullpath)(lsm_env*, const char *, char *, int *);
  int (*xOpen)(lsm_env*, const char *, int flags, lsm_file **);
  int (*xRead)(lsm_file *, lsm_i64, void *, int);
  int (*xWrite)(lsm_file *, lsm_i64, void *, int);
  int (*xTruncate)(lsm_file *, lsm_i64);
  int (*xSync)(lsm_file *);
  int (*xSectorSize)(lsm_file *);
#if 0
  int (*xRemap)(lsm_file *, lsm_i64, void **, lsm_i64*);
#endif
  int (*xMap)(lsm_file *, lsm_i64 iOff, lsm_i64 nByte, void **, lsm_i64 *);
  int (*xUnmap)(lsm_file *, void *, lsm_i64);
  int (*xFileid)(lsm_file *, void *pBuf, int *pnBuf);
  int (*xClose)(lsm_file *);
  int (*xUnlink)(lsm_env*, const char *);
  int (*xLock)(lsm_file*, int, int);
  int (*xTestLock)(lsm_file*, int, int, int);
  int (*xShmMap)(lsm_file*, int, int, void **);
  void (*xShmBarrier)(void);
  int (*xShmUnmap)(lsm_file*, int);
  /****** memory allocation ****************************************/
  void *pMemCtx;
  void *(*xMalloc)(lsm_env*, int);            /* malloc(3) function */
  void *(*xRealloc)(lsm_env*, void *, int);   /* realloc(3) function */

/*
** CAPI: LSM Error Codes
*/
#define LSM_OK         0
#define LSM_ERROR      1
#define LSM_BUSY       5
#define LSM_NOMEM      7
#define LSM_READONLY   8
#define LSM_IOERR     10
#define LSM_CORRUPT   11
#define LSM_FULL      13
#define LSM_CANTOPEN  14
#define LSM_PROTOCOL  15
#define LSM_MISUSE    21

#define LSM_MISMATCH  50


#define LSM_IOERR_NOENT (LSM_IOERR | (1<<8))

/* 
** CAPI: Creating and Destroying Database Connection Handles
**
** Open and close a database connection handle.
*/
int lsm_new(lsm_env*, lsm_db **ppDb);
int lsm_close(lsm_db *pDb);

/* 
** CAPI: Connecting to a Database
*/
int lsm_open(lsm_db *pDb, const char *zFilename);

/*
** CAPI: Obtaining pointers to database environments
**
** Return a pointer to the environment used by the database connection 
** passed as the first argument. Assuming the argument is valid, this 
** function always returns a valid environment pointer - it cannot fail.
*/
lsm_env *lsm_get_env(lsm_db *pDb);

*/
int lsm_config(lsm_db *, int, ...);

/*
** The following values may be passed as the second argument to lsm_config().
**
** LSM_CONFIG_AUTOFLUSH:
**   A read/write integer parameter. 
**
**   This value determines the amount of data allowed to accumulate in a
**   live in-memory tree before it is marked as old. After committing a
**   transaction, a connection checks if the size of the live in-memory tree,
**   including data structure overhead, is greater than the value of this
**   option in KB. If it is, and there is not already an old in-memory tree,
**   the live in-memory tree is marked as old.
**
**   The maximum allowable value is 1048576 (1GB). There is no minimum 
**   value. If this parameter is set to zero, then an attempt is made to
**   mark the live in-memory tree as old after each transaction is committed.
**
**   The default value is 1024 (1MB).
**
** LSM_CONFIG_PAGE_SIZE:
**   A read/write integer parameter. This parameter may only be set before
**   lsm_open() has been called.
**
** LSM_CONFIG_BLOCK_SIZE:
**   A read/write integer parameter. 
**
**   This parameter may only be set before lsm_open() has been called. It
**   must be set to a power of two between 64 and 65536, inclusive (block 
**   sizes between 64KB and 64MB).
**
**   If the connection creates a new database, the block size of the new
**   database is set to the value of this option in KB. After lsm_open()
**   has been called, querying this parameter returns the actual block
**   size of the opened database.
**
**   The default value is 1024 (1MB blocks).
**
** LSM_CONFIG_SAFETY:
**   A read/write integer parameter. Valid values are 0, 1 (the default) 
**   and 2. This parameter determines how robust the database is in the
**   face of a system crash (e.g. a power failure or operating system 
**   crash). As follows:
**

**                 contains all successfully committed transactions.
**
** LSM_CONFIG_AUTOWORK:
**   A read/write integer parameter.
**
** LSM_CONFIG_AUTOCHECKPOINT:
**   A read/write integer parameter.
**
**   If this option is set to non-zero value N, then a checkpoint is
**   automatically attempted after each N KB of data have been written to 
**   the database file.
**
**   The amount of uncheckpointed data already written to the database file
**   is a global parameter. After performing database work (writing to the
**   database file), the process checks if the total amount of uncheckpointed 
**   data exceeds the value of this paramter. If so, a checkpoint is performed.
**   This means that this option may cause the connection to perform a 
**   checkpoint even if the current connection has itself written very little
**   data into the database file.
**
**   The default value is 2048 (checkpoint every 2MB).
**
** LSM_CONFIG_MMAP:
**   A read/write integer parameter. True to use mmap() to access the 
**   database file. False otherwise.
**
** LSM_CONFIG_USE_LOG:
**   A read/write boolean parameter. True (the default) to use the log

**
**   This option may only be used before lsm_open() is called. Invoking it
**   after lsm_open() has been called results in an LSM_MISUSE error.
**
** LSM_CONFIG_GET_COMPRESSION:
**   Query the compression methods used to compress and decompress database
**   content.
**
** LSM_CONFIG_SET_COMPRESSION_FACTORY:
**   Configure a factory method to be invoked in case of an LSM_MISMATCH
**   error.
**
** LSM_CONFIG_READONLY:
**   A read/write boolean parameter. This parameter may only be set before
**   lsm_open() is called.
*/
#define LSM_CONFIG_AUTOFLUSH           1
#define LSM_CONFIG_PAGE_SIZE           2
#define LSM_CONFIG_SAFETY              3
#define LSM_CONFIG_BLOCK_SIZE          4
#define LSM_CONFIG_AUTOWORK            5

#define LSM_CONFIG_MMAP                7
#define LSM_CONFIG_USE_LOG             8
#define LSM_CONFIG_AUTOMERGE           9
#define LSM_CONFIG_MAX_FREELIST       10
#define LSM_CONFIG_MULTIPLE_PROCESSES 11
#define LSM_CONFIG_AUTOCHECKPOINT     12
#define LSM_CONFIG_SET_COMPRESSION    13
#define LSM_CONFIG_GET_COMPRESSION    14
#define LSM_CONFIG_SET_COMPRESSION_FACTORY 15
#define LSM_CONFIG_READONLY                16

#define LSM_SAFETY_OFF    0
#define LSM_SAFETY_NORMAL 1
#define LSM_SAFETY_FULL   2

#define LSM_MMAP_OFF     0
#define LSM_MMAP_FULL    1
#define LSM_MMAP_LIMITED 2

/*
** CAPI: Compression and/or Encryption Hooks
*/
struct lsm_compress {
  void *pCtx;
  unsigned int iId;
  int (*xBound)(void *, int nSrc);
  int (*xCompress)(void *, char *, int *, const char *, int);
  int (*xUncompress)(void *, char *, int *, const char *, int);
  void (*xFree)(void *pCtx);
};

struct lsm_compress_factory {
  void *pCtx;
  int (*xFactory)(void *, lsm_db *, unsigned int);
  void (*xFree)(void *pCtx);
};

#define LSM_COMPRESSION_EMPTY 0
#define LSM_COMPRESSION_NONE  1

/*
** CAPI: Allocating and Freeing Memory
**
** Invoke the memory allocation functions that belong to environment
** pEnv. Or the system defaults if no memory allocation functions have 
** been registered.

**
**   The Tcl structure returned is a list containing one element for each
**   free block in the database. The element itself consists of two 
**   integers - the block number and the id of the snapshot that freed it.
**
** LSM_INFO_CHECKPOINT_SIZE:
**   The third argument should be of type (int *). The location pointed to
**   by this argument is populated with the number of KB written to the
**   database file since the most recent checkpoint.
**
** LSM_INFO_TREE_SIZE:
**   If this value is passed as the second argument to an lsm_info() call, it
**   should be followed by two arguments of type (int *) (for a total of four
**   arguments).
**
**   At any time, there are either one or two tree structures held in shared
**   memory that new database clients will access (there may also be additional
**   tree structures being used by older clients - this API does not provide
**   information on them). One tree structure - the current tree - is used to
**   accumulate new data written to the database. The other tree structure -
**   the old tree - is a read-only tree holding older data and may be flushed 
**   to disk at any time.
** 
**   Assuming no error occurs, the location pointed to by the first of the two
**   (int *) arguments is set to the size of the old in-memory tree in KB.
**   The second is set to the size of the current, or live in-memory tree.
**
** LSM_INFO_COMPRESSION_ID:
**   This value should be followed by a single argument of type 
**   (unsigned int *). If successful, the location pointed to is populated 
**   with the database compression id before returning.
*/
#define LSM_INFO_NWRITE           1
#define LSM_INFO_NREAD            2
#define LSM_INFO_DB_STRUCTURE     3
#define LSM_INFO_LOG_STRUCTURE    4
#define LSM_INFO_ARRAY_STRUCTURE  5
#define LSM_INFO_PAGE_ASCII_DUMP  6
#define LSM_INFO_PAGE_HEX_DUMP    7
#define LSM_INFO_FREELIST         8
#define LSM_INFO_ARRAY_PAGES      9
#define LSM_INFO_CHECKPOINT_SIZE 10
#define LSM_INFO_TREE_SIZE       11

#define LSM_INFO_FREELIST_SIZE   12
#define LSM_INFO_COMPRESSION_ID  13


/* 
** CAPI: Opening and Closing Write Transactions
**
** These functions are used to open and close transactions and nested 
** sub-transactions.

);

/*
** CAPI: Explicit Database Work and Checkpointing
**
** This function is called by a thread to work on the database structure.
*/
int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite);

int lsm_flush(lsm_db *pDb);

/*
** Attempt to checkpoint the current database snapshot. Return an LSM
** error code if an error occurs or LSM_OK otherwise.
**
** If the current snapshot has already been checkpointed, calling this 
** function is a no-op. In this case if pnKB is not NULL, *pnKB is
** set to 0. Or, if the current snapshot is successfully checkpointed
** by this function and pbKB is not NULL, *pnKB is set to the number
** of bytes written to the database file since the previous checkpoint
** (the same measure as returned by the LSM_INFO_CHECKPOINT_SIZE query).
*/
int lsm_checkpoint(lsm_db *pDb, int *pnKB);

/*
** CAPI: Opening and Closing Database Cursors
**
** Open and close a database cursor.
*/
int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr);

/*
** Default values for various data structure parameters. These may be
** overridden by calls to lsm_config().
*/
#define LSM_DFLT_PAGE_SIZE          (4 * 1024)
#define LSM_DFLT_BLOCK_SIZE         (1 * 1024 * 1024)
#define LSM_DFLT_AUTOFLUSH          (1 * 1024 * 1024)
#define LSM_DFLT_AUTOCHECKPOINT     (i64)(2 * 1024 * 1024)
#define LSM_DFLT_AUTOWORK           1
#define LSM_DFLT_LOG_SIZE           (128*1024)
#define LSM_DFLT_AUTOMERGE          4
#define LSM_DFLT_SAFETY             LSM_SAFETY_NORMAL
#define LSM_DFLT_MMAP               (LSM_IS_64_BIT ? LSM_MMAP_FULL : 0)
#define LSM_DFLT_MULTIPLE_PROCESSES 1
#define LSM_DFLT_USE_LOG            1

#ifdef LSM_DEBUG
int lsmErrorBkpt(int);
#else
# define lsmErrorBkpt(x) (x)
#endif

#define LSM_PROTOCOL_BKPT lsmErrorBkpt(LSM_PROTOCOL)
#define LSM_IOERR_BKPT   lsmErrorBkpt(LSM_IOERR)
#define LSM_NOMEM_BKPT   lsmErrorBkpt(LSM_NOMEM)
#define LSM_CORRUPT_BKPT lsmErrorBkpt(LSM_CORRUPT)
#define LSM_MISUSE_BKPT  lsmErrorBkpt(LSM_MISUSE)

#define unused_parameter(x) (void)(x)
#define array_size(x) (sizeof(x)/sizeof(x[0]))


/* The size of each shared-memory chunk */
#define LSM_SHM_CHUNK_SIZE (32*1024)

/* The number of bytes reserved at the start of each shm chunk for MM. */
#define LSM_SHM_CHUNK_HDR  (sizeof(ShmChunk))

/* The number of available read locks. */
#define LSM_LOCK_NREADER   6

/* The number of available read-write client locks. */
#define LSM_LOCK_NRWCLIENT   16

/* Lock definitions. 
*/
#define LSM_LOCK_DMS1         1   /* Serialize connect/disconnect ops */
#define LSM_LOCK_DMS2         2   /* Read-write connections */
#define LSM_LOCK_DMS3         3   /* Read-only connections */
#define LSM_LOCK_WRITER       4
#define LSM_LOCK_WORKER       5
#define LSM_LOCK_CHECKPOINTER 6
#define LSM_LOCK_ROTRANS      7
#define LSM_LOCK_READER(i)    ((i) + LSM_LOCK_ROTRANS + 1)
#define LSM_LOCK_RWCLIENT(i)  ((i) + LSM_LOCK_READER(LSM_LOCK_NREADER))

/*
** Hard limit on the number of free-list entries that may be stored in 
** a checkpoint (the remainder are stored as a system record in the LSM).
** See also LSM_CONFIG_MAX_FREELIST.
*/
#define LSM_MAX_FREELIST_ENTRIES 24

*/
#define LSM_START_DELETE 0x01     /* Start of open-ended delete range */
#define LSM_END_DELETE   0x02     /* End of open-ended delete range */
#define LSM_POINT_DELETE 0x04     /* Delete this key */
#define LSM_INSERT       0x08     /* Insert this key and value */
#define LSM_SEPARATOR    0x10     /* True if entry is separator key only */
#define LSM_SYSTEMKEY    0x20     /* True if entry is a system key (FREELIST) */

#define LSM_CONTIGUOUS   0x40     /* Used in lsm_tree.c */

/*
** A string that can grow by appending.
*/
struct LsmString {
  lsm_env *pEnv;              /* Run-time environment */
  int n;                      /* Size of string.  -1 indicates error */

/*
** Database handle structure.
**
** mLock:
**   A bitmask representing the locks currently held by the connection.
**   An LSM database supports N distinct locks, where N is some number less
**   than or equal to 32. Locks are numbered starting from 1 (see the 
**   definitions for LSM_LOCK_WRITER and co.).
**
**   The least significant 32-bits in mLock represent EXCLUSIVE locks. The
**   most significant are SHARED locks. So, if a connection holds a SHARED
**   lock on lock region iLock, then the following is true:
**
**       (mLock & ((iLock+32-1) << 1))
**
**   Or for an EXCLUSIVE lock:
**
**       (mLock & ((iLock-1) << 1))
** 
** pCsr:
**   Points to the head of a linked list that contains all currently open
**   cursors. Once this list becomes empty, the user has no outstanding
**   cursors and the database handle can be successfully closed.
**
** pCsrCache:
**   This list contains cursor objects that have been closed using
**   lsm_csr_close(). Each time a cursor is closed, it is shifted from 
**   the pCsr list to this list. When a new cursor is opened, this list
**   is inspected to see if there exists a cursor object that can be
**   reused. This is an optimization only.
*/
struct lsm_db {

  /* Database handle configuration */
  lsm_env *pEnv;                            /* runtime environment */
  int (*xCmp)(void *, int, void *, int);    /* Compare function */

  /* Values configured by calls to lsm_config */
  int eSafety;                    /* LSM_SAFETY_OFF, NORMAL or FULL */
  int bAutowork;                  /* Configured by LSM_CONFIG_AUTOWORK */
  int nTreeLimit;                 /* Configured by LSM_CONFIG_AUTOFLUSH */
  int nMerge;                     /* Configured by LSM_CONFIG_AUTOMERGE */

  int bUseLog;                    /* Configured by LSM_CONFIG_USE_LOG */
  int nDfltPgsz;                  /* Configured by LSM_CONFIG_PAGE_SIZE */
  int nDfltBlksz;                 /* Configured by LSM_CONFIG_BLOCK_SIZE */
  int nMaxFreelist;               /* Configured by LSM_CONFIG_MAX_FREELIST */
  int eMmap;                      /* Configured by LSM_CONFIG_MMAP */
  i64 nAutockpt;                  /* Configured by LSM_CONFIG_AUTOCHECKPOINT */
  int bMultiProc;                 /* Configured by L_C_MULTIPLE_PROCESSES */
  int bReadonly;                  /* Configured by LSM_CONFIG_READONLY */
  lsm_compress compress;          /* Compression callbacks */
  lsm_compress_factory factory;   /* Compression callback factory */

  /* Sub-system handles */
  FileSystem *pFS;                /* On-disk portion of database */
  Database *pDatabase;            /* Database shared data */

  int iRwclient;                  /* Read-write client lock held (-1 == none) */

  /* Client transaction context */
  Snapshot *pClient;              /* Client snapshot */
  int iReader;                    /* Read lock held (-1 == unlocked) */
  int bRoTrans;                   /* True if a read-only db trans is open */
  MultiCursor *pCsr;              /* List of all open cursors */
  LogWriter *pLogWriter;          /* Context for writing to the log file */
  int nTransOpen;                 /* Number of opened write transactions */
  int nTransAlloc;                /* Allocated size of aTrans[] array */
  TransMark *aTrans;              /* Array of marks for transaction rollback */
  IntArray rollback;              /* List of tree-nodes to roll back */
  int bDiscardOld;                /* True if lsmTreeDiscardOld() was called */

  MultiCursor *pCsrCache;         /* List of all closed cursors */

  /* Worker context */
  Snapshot *pWorker;              /* Worker snapshot (or NULL) */
  Freelist *pFreelist;            /* See sortedNewToplevel() */
  int bUseFreelist;               /* True to use pFreelist */
  int bIncrMerge;                 /* True if currently doing a merge */

  int bInFactory;                 /* True if within factory.xFactory() */

  /* Debugging message callback */
  void (*xLog)(void *, int, const char *);
  void *pLogCtx;

  /* Work done notification callback */
  void (*xWork)(lsm_db *, void *);
  void *pWorkCtx;

  u64 mLock;                      /* Mask of current locks. See lsmShmLock(). */
  lsm_db *pNext;                  /* Next connection to same database */

  int nShm;                       /* Size of apShm[] array */
  void **apShm;                   /* Shared memory chunks */
  ShmHeader *pShmhdr;             /* Live shared-memory header */
  TreeHeader treehdr;             /* Local copy of tree-header */
  u32 aSnapshot[LSM_META_PAGE_SIZE / sizeof(u32)];

/*
** A snapshot of a database. A snapshot contains all the information required
** to read or write a database file on disk. See the description of struct
** Database below for futher details.
*/
struct Snapshot {
  Database *pDatabase;            /* Database this snapshot belongs to */
  u32 iCmpId;                     /* Id of compression scheme */
  Level *pLevel;                  /* Pointer to level 0 of snapshot (or NULL) */
  i64 iId;                        /* Snapshot id */
  i64 iLogOff;                    /* Log file offset */
  Redirect redirect;              /* Block redirection array */

  /* Used by worker snapshots only */
  int nBlock;                     /* Number of blocks in database file */

int lsmCheckpointSaveWorker(lsm_db *pDb, int);
int lsmDatabaseFull(lsm_db *pDb);
int lsmCheckpointSynced(lsm_db *pDb, i64 *piId, i64 *piLog, u32 *pnWrite);

int lsmCheckpointSize(lsm_db *db, int *pnByte);

int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId);

/* 
** Functions from file "lsm_tree.c".
*/
int lsmTreeNew(lsm_env *, int (*)(void *, int, void *, int), Tree **ppTree);
void lsmTreeRelease(lsm_env *, Tree *);
int lsmTreeInit(lsm_db *);
int lsmTreeRepair(lsm_db *);

void lsmTreeMakeOld(lsm_db *pDb);
void lsmTreeDiscardOld(lsm_db *pDb);
int lsmTreeHasOld(lsm_db *pDb);

int lsmTreeSize(lsm_db *);
int lsmTreeEndTransaction(lsm_db *pDb, int bCommit);
int lsmTreeLoadHeader(lsm_db *pDb, int *);
int lsmTreeLoadHeaderOk(lsm_db *, int);

int lsmTreeInsert(lsm_db *pDb, void *pKey, int nKey, void *pVal, int nVal);
int lsmTreeDelete(lsm_db *db, void *pKey1, int nKey1, void *pKey2, int nKey2);
void lsmTreeRollback(lsm_db *pDb, TreeMark *pMark);
void lsmTreeMark(lsm_db *pDb, TreeMark *pMark);

int lsmTreeCursorNew(lsm_db *pDb, int, TreeCursor **);
void lsmTreeCursorDestroy(TreeCursor *);

int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes);

int lsmMutexHeld(lsm_env *, lsm_mutex *);
int lsmMutexNotHeld(lsm_env *, lsm_mutex *);
#endif

/**************************************************************************
** Start of functions from "lsm_file.c".
*/
int lsmFsOpen(lsm_db *, const char *, int);
int lsmFsOpenLog(lsm_db *, int *);
void lsmFsCloseLog(lsm_db *);
void lsmFsClose(FileSystem *);

int lsmFsConfigure(lsm_db *db);

int lsmFsBlockSize(FileSystem *);
void lsmFsSetBlockSize(FileSystem *, int);

int lsmFsPageSize(FileSystem *);
void lsmFsSetPageSize(FileSystem *, int);

void lsmFsFlushWaiting(FileSystem *, int *);

/* Used by lsm_info(ARRAY_STRUCTURE) and lsm_config(MMAP) */
int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, Pgno iFirst, char **pzOut);
int lsmInfoArrayPages(lsm_db *pDb, Pgno iFirst, char **pzOut);
int lsmConfigMmap(lsm_db *pDb, int *piParam);

int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);
int lsmEnvTestLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int nLock, int);

int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
void lsmEnvShmBarrier(lsm_env *);
void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);

void lsmEnvSleep(lsm_env *, int);

int lsmSortedLoadFreelist(lsm_db *pDb, void **, int *);

void *lsmSortedSplitKey(Level *pLevel, int *pnByte);

void lsmSortedSaveTreeCursors(lsm_db *);

int lsmMCursorNew(lsm_db *, MultiCursor **);
void lsmMCursorClose(MultiCursor *, int);
int lsmMCursorSeek(MultiCursor *, int, void *, int , int);
int lsmMCursorFirst(MultiCursor *);
int lsmMCursorPrev(MultiCursor *);
int lsmMCursorLast(MultiCursor *);
int lsmMCursorValid(MultiCursor *);
int lsmMCursorNext(MultiCursor *);
int lsmMCursorKey(MultiCursor *, void **, int *);
int lsmMCursorValue(MultiCursor *, void **, int *);
int lsmMCursorType(MultiCursor *, int *);
lsm_db *lsmMCursorDb(MultiCursor *);
void lsmMCursorFreeCache(lsm_db *);

int lsmSaveCursors(lsm_db *pDb);
int lsmRestoreCursors(lsm_db *pDb);

void lsmSortedDumpStructure(lsm_db *pDb, Snapshot *, int, int, const char *);
void lsmFsDumpBlocklists(lsm_db *);

int lsmDbDatabaseConnect(lsm_db*, const char *);
void lsmDbDatabaseRelease(lsm_db *);

int lsmBeginReadTrans(lsm_db *);
int lsmBeginWriteTrans(lsm_db *);
int lsmBeginFlush(lsm_db *);

int lsmDetectRoTrans(lsm_db *db, int *);

int lsmBeginWork(lsm_db *);
void lsmFinishWork(lsm_db *, int, int *);

int lsmFinishRecovery(lsm_db *);
void lsmFinishReadTrans(lsm_db *);
int lsmFinishWriteTrans(lsm_db *, int);
int lsmFinishFlush(lsm_db *, int);

/* Candidate values for the 3rd argument to lsmShmLock() */
#define LSM_LOCK_UNLOCK 0
#define LSM_LOCK_SHARED 1
#define LSM_LOCK_EXCL   2

int lsmShmCacheChunks(lsm_db *db, int nChunk);
int lsmShmLock(lsm_db *db, int iLock, int eOp, int bBlock);
int lsmShmTestLock(lsm_db *db, int iLock, int nLock, int eOp);
void lsmShmBarrier(lsm_db *db);

#ifdef LSM_DEBUG
void lsmShmHasLock(lsm_db *db, int iLock, int eOp);
#else
# define lsmShmHasLock(x,y,z)
#endif

int lsmReadlock(lsm_db *, i64 iLsm, u32 iShmMin, u32 iShmMax);


int lsmLsmInUse(lsm_db *db, i64 iLsmId, int *pbInUse);
int lsmTreeInUse(lsm_db *db, u32 iLsmId, int *pbInUse);
int lsmFreelistAppend(lsm_env *pEnv, Freelist *p, int iBlk, i64 iId);

int lsmDbMultiProc(lsm_db *);
void lsmDbDeferredClose(lsm_db *, lsm_file *, LsmFile *);
LsmFile *lsmDbRecycleFd(lsm_db *);

int lsmWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);

int lsmCheckCompressionId(lsm_db *, u32);


/**************************************************************************
** functions in lsm_str.c
*/
void lsmStringInit(LsmString*, lsm_env *pEnv);
int lsmStringExtend(LsmString*, int);

**
**   Checkpoint header (see the CKPT_HDR_XXX #defines):
**
**     1. The checkpoint id MSW.
**     2. The checkpoint id LSW.
**     3. The number of integer values in the entire checkpoint, including 
**        the two checksum values.
**     4. The compression scheme id.
**     5. The total number of blocks in the database.
**     6. The block size.
**     7. The number of levels.
**     8. The nominal database page size.
**     9. The number of pages (in total) written to the database file.
**
**   Log pointer:
**
**     1. The log offset MSW.
**     2. The log offset LSW.
**     3. Log checksum 0.
**     4. Log checksum 1.
**
**     Note that the "log offset" is not the literal byte offset. Instead,
**     it is the byte offset multiplied by 2, with least significant bit
**     toggled each time the log pointer value is changed. This is to make
**     sure that this field changes each time the log pointer is updated,
**     even if the log file itself is disabled. See lsmTreeMakeOld().
**
**     See ckptExportLog() and ckptImportLog().
**
**   Append points:
**
**     8 integers (4 * 64-bit page numbers). See ckptExportAppendlist().
**

 + (((x)&0x00FF0000)>>8)  + (((x)&0xFF000000)>>24) \
)

static const int one = 1;
#define LSM_LITTLE_ENDIAN (*(u8 *)(&one))

/* Sizes, in integers, of various parts of the checkpoint. */
#define CKPT_HDR_SIZE         9
#define CKPT_LOGPTR_SIZE      4
#define CKPT_APPENDLIST_SIZE  (LSM_APPLIST_SZ * 2)

/* A #define to describe each integer in the checkpoint header. */
#define CKPT_HDR_ID_MSW   0
#define CKPT_HDR_ID_LSW   1
#define CKPT_HDR_NCKPT    2
#define CKPT_HDR_CMPID    3
#define CKPT_HDR_NBLOCK   4
#define CKPT_HDR_BLKSZ    5
#define CKPT_HDR_NLEVEL   6
#define CKPT_HDR_PGSZ     7
#define CKPT_HDR_NWRITE   8

#define CKPT_HDR_LO_MSW     9
#define CKPT_HDR_LO_LSW    10
#define CKPT_HDR_LO_CKSUM1 11
#define CKPT_HDR_LO_CKSUM2 12

typedef struct CkptBuffer CkptBuffer;

/*
** Dynamic buffer used to accumulate data for a checkpoint.
*/
struct CkptBuffer {

      ckptSetValue(&ckpt, iOut++, (p->iId >> 32) & 0xFFFFFFFF, &rc);
      ckptSetValue(&ckpt, iOut++, p->iId & 0xFFFFFFFF, &rc);
    }
  }

  /* Write the checkpoint header */
  assert( iId>=0 );
  assert( pSnap->iCmpId==pDb->compress.iId
       || pSnap->iCmpId==LSM_COMPRESSION_EMPTY 
  );
  ckptSetValue(&ckpt, CKPT_HDR_ID_MSW, (u32)(iId>>32), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_ID_LSW, (u32)(iId&0xFFFFFFFF), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NCKPT, iOut+2, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_CMPID, pDb->compress.iId, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NBLOCK, pSnap->nBlock, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_BLKSZ, lsmFsBlockSize(pFS), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NLEVEL, nLevel, &rc);
  ckptSetValue(&ckpt, CKPT_HDR_PGSZ, lsmFsPageSize(pFS), &rc);
  ckptSetValue(&ckpt, CKPT_HDR_NWRITE, pSnap->nWrite, &rc);

  if( bCksum ){

** is called when no valid snapshot can be found in the database header.
*/
static void ckptLoadEmpty(lsm_db *pDb){
  u32 aCkpt[] = {
    0,                  /* CKPT_HDR_ID_MSW */
    10,                 /* CKPT_HDR_ID_LSW */
    0,                  /* CKPT_HDR_NCKPT */
    LSM_COMPRESSION_EMPTY,   /* CKPT_HDR_CMPID */
    0,                  /* CKPT_HDR_NBLOCK */
    0,                  /* CKPT_HDR_BLKSZ */
    0,                  /* CKPT_HDR_NLEVEL */
    0,                  /* CKPT_HDR_PGSZ */
    0,                  /* CKPT_HDR_OVFL */
    0,                  /* CKPT_HDR_NWRITE */
    0, 0, 1234, 5678,   /* The log pointer and initial checksum */

        if( piRead ) *piRead = 2;
        return LSM_OK;
      }
    }

    lsmShmBarrier(pDb);
  }
  return LSM_PROTOCOL_BKPT;
}

int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId){
  int rc;

  assert( db->pClient==0 && db->pWorker==0 );
  rc = lsmCheckpointLoad(db, 0);
  if( rc==LSM_OK ){
    *piCmpId = db->aSnapshot[CKPT_HDR_CMPID];
  }

  return rc;
}

int lsmCheckpointLoadOk(lsm_db *pDb, int iSnap){
  u32 *aShm;
  assert( iSnap==1 || iSnap==2 );
  aShm = (iSnap==1) ? pDb->pShmhdr->aSnap1 : pDb->pShmhdr->aSnap2;
  return (lsmCheckpointId(pDb->aSnapshot, 0)==lsmCheckpointId(aShm, 0) );

  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

  /* Must be holding the WORKER lock to do this. Or DMS2. */
  assert( 
      lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) 
   || lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) 
  );

  /* Check that the two snapshots match. If not, repair them. */
  nInt1 = pShm->aSnap1[CKPT_HDR_NCKPT];
  nInt2 = pShm->aSnap2[CKPT_HDR_NCKPT];
  if( nInt1!=nInt2 || memcmp(pShm->aSnap1, pShm->aSnap2, nInt2*sizeof(u32)) ){
    if( ckptChecksumOk(pShm->aSnap1) ){
      memcpy(pShm->aSnap2, pShm->aSnap1, sizeof(u32)*nInt1);
    }else if( ckptChecksumOk(pShm->aSnap2) ){
      memcpy(pShm->aSnap1, pShm->aSnap2, sizeof(u32)*nInt2);
    }else{
      return LSM_PROTOCOL_BKPT;
    }
  }

  rc = lsmCheckpointDeserialize(pDb, 1, pShm->aSnap1, &pDb->pWorker);
  if( pDb->pWorker ) pDb->pWorker->pDatabase = pDb->pDatabase;

  if( rc==LSM_OK ){
    rc = lsmCheckCompressionId(pDb, pDb->pWorker->iCmpId);
  }

#if 0
  assert( rc!=LSM_OK || lsmFsIntegrityCheck(pDb) );
#endif
  return rc;
}

    int iIn = CKPT_HDR_SIZE + CKPT_APPENDLIST_SIZE + CKPT_LOGPTR_SIZE;

    pNew->iId = lsmCheckpointId(aCkpt, 0);
    pNew->nBlock = aCkpt[CKPT_HDR_NBLOCK];
    pNew->nWrite = aCkpt[CKPT_HDR_NWRITE];
    rc = ckptLoadLevels(pDb, aCkpt, &iIn, nLevel, &pNew->pLevel);
    pNew->iLogOff = lsmCheckpointLogOffset(aCkpt);
    pNew->iCmpId = aCkpt[CKPT_HDR_CMPID];

    /* Make a copy of the append-list */
    for(i=0; i<LSM_APPLIST_SZ; i++){
      u32 *a = &aCkpt[CKPT_HDR_SIZE + CKPT_LOGPTR_SIZE + i*2];
      pNew->aiAppend[i] = ckptRead64(a);
    }

      if( nCkpt<(LSM_META_PAGE_SIZE/sizeof(u32)) ){
        u32 *aCopy = lsmMallocRc(pDb->pEnv, sizeof(u32) * nCkpt, &rc);
        if( aCopy ){
          memcpy(aCopy, aData, nCkpt*sizeof(u32));
          ckptChangeEndianness(aCopy, nCkpt);
          if( ckptChecksumOk(aCopy) ){
            if( piId ) *piId = lsmCheckpointId(aCopy, 0);
            if( piLog ) *piLog = (lsmCheckpointLogOffset(aCopy) >> 1);
            if( pnWrite ) *pnWrite = aCopy[CKPT_HDR_NWRITE];
          }
          lsmFree(pDb->pEnv, aCopy);
        }
      }
      lsmFsMetaPageRelease(pPg);
    }

int lsmCheckpointBlksz(u32 *aCkpt){ return (int)aCkpt[CKPT_HDR_BLKSZ]; }

void lsmCheckpointLogoffset(
  u32 *aCkpt,
  DbLog *pLog
){ 


  pLog->aRegion[2].iStart = (lsmCheckpointLogOffset(aCkpt) >> 1);

  pLog->cksum0 = aCkpt[CKPT_HDR_LO_CKSUM1];
  pLog->cksum1 = aCkpt[CKPT_HDR_LO_CKSUM2];
  pLog->iSnapshotId = lsmCheckpointId(aCkpt, 0);
}

void lsmCheckpointZeroLogoffset(lsm_db *pDb){
  u32 nCkpt;

      &pDb->aSnapshot[nCkpt-2], &pDb->aSnapshot[nCkpt-1]
  );

  memcpy(pDb->pShmhdr->aSnap1, pDb->aSnapshot, nCkpt*sizeof(u32));
  memcpy(pDb->pShmhdr->aSnap2, pDb->aSnapshot, nCkpt*sizeof(u32));
}

/*
** Set the output variable to the number of KB of data written into the
** database file since the most recent checkpoint.
*/
int lsmCheckpointSize(lsm_db *db, int *pnKB){
  ShmHeader *pShm = db->pShmhdr;
  int rc = LSM_OK;
  u32 nSynced;

  /* Set nSynced to the number of pages that had been written when the 
  ** database was last checkpointed. */
  rc = lsmCheckpointSynced(db, 0, 0, &nSynced);

  if( rc==LSM_OK ){
    u32 nPgsz = db->pShmhdr->aSnap1[CKPT_HDR_PGSZ];
    u32 nWrite = db->pShmhdr->aSnap1[CKPT_HDR_NWRITE];
    *pnKB = (int)(( ((i64)(nWrite - nSynced) * nPgsz) + 1023) / 1024);
  }

  return rc;
}

**   are carrying pointers into the database file mapping (pMap/nMap). If the
**   file has to be unmapped and then remapped (required to grow the mapping
**   as the file grows), the Page.aData pointers are updated by iterating
**   through the contents of this list.
**
**   In non-mmap() mode, this list is an LRU list of cached pages with 
**   nRef==0.
**
** apHash, nHash:
*/
struct FileSystem {
  lsm_db *pDb;                    /* Database handle that owns this object */
  lsm_env *pEnv;                  /* Environment pointer */
  char *zDb;                      /* Database file name */
  char *zLog;                     /* Database file name */
  int nMetasize;                  /* Size of meta pages in bytes */

**     lsmEnvSync()
**     lsmEnvSectorSize()
**     lsmEnvClose()
**     lsmEnvTruncate()
**     lsmEnvUnlink()
**     lsmEnvRemap()
*/
int lsmEnvOpen(lsm_env *pEnv, const char *zFile, int flags, lsm_file **ppNew){
  return pEnv->xOpen(pEnv, zFile, flags, ppNew);
}
static int lsmEnvRead(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  lsm_i64 iOff, 
  void *pRead, 
  int nRead

  return pEnv->xMap(pFile, iOff, szMin, ppMap, pszMap);
}

int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock){
  if( pFile==0 ) return LSM_OK;
  return pEnv->xLock(pFile, iLock, eLock);
}

int lsmEnvTestLock(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iLock, 
  int nLock, 
  int eLock
){
  return pEnv->xTestLock(pFile, iLock, nLock, eLock);
}

int lsmEnvShmMap(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iChunk, 
  int sz, 
  void **ppOut

*/
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte){
  if( pFS->fdLog==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdLog, nByte);
}

/*
** Truncate the db file to nByte bytes in size.
*/
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte){
  if( pFS->fdDb==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdDb, nByte);
}

/*

/*
** This is a helper function for lsmFsOpen(). It opens a single file on
** disk (either the database or log file).
*/
static lsm_file *fsOpenFile(
  FileSystem *pFS,                /* File system object */
  int bReadonly,                  /* True to open this file read-only */
  int bLog,                       /* True for log, false for db */
  int *pRc                        /* IN/OUT: Error code */
){
  lsm_file *pFile = 0;
  if( *pRc==LSM_OK ){
    int flags = (bReadonly ? LSM_OPEN_READONLY : 0);
    const char *zPath = (bLog ? pFS->zLog : pFS->zDb);

    *pRc = lsmEnvOpen(pFS->pEnv, zPath, flags, &pFile);
  }
  return pFile;
}


static void fsGrowMapping(
  FileSystem *pFS,

**
** The log file must be opened before any of the following may be called:
**
**     lsmFsWriteLog
**     lsmFsSyncLog
**     lsmFsReadLog
*/
int lsmFsOpenLog(lsm_db *db, int *pbOpen){
  int rc = LSM_OK;
  FileSystem *pFS = db->pFS;

  if( 0==pFS->fdLog ){ 
    pFS->fdLog = fsOpenFile(pFS, db->bReadonly, 1, &rc); 

    if( rc==LSM_IOERR_NOENT && db->bReadonly ){
      rc = LSM_OK;
    }
  }

  if( pbOpen ) *pbOpen = (pFS->fdLog!=0);
  return rc;
}

void lsmFsCloseLog(lsm_db *db){
  FileSystem *pFS = db->pFS;
  if( pFS->fdLog ){
    lsmEnvClose(pFS->pEnv, pFS->fdLog);
    pFS->fdLog = 0;
  }
}

/*
** Open a connection to a database stored within the file-system (the
** "system of files").
**
** If parameter bReadonly is true, then open a read-only file-descriptor
** on the database file. It is possible that bReadonly will be false even
** if the user requested that pDb be opened read-only. This is because the
** file-descriptor may later on be recycled by a read-write connection.
** If the db file can be opened for read-write access, it always is. Parameter
** bReadonly is only ever true if it has already been determined that the
** db can only be opened for read-only access.
*/
int lsmFsOpen(
  lsm_db *pDb,                    /* Database connection to open fd for */
  const char *zDb,                /* Full path to database file */
  int bReadonly                   /* True to open db file read-only */
){
  FileSystem *pFS;
  int rc = LSM_OK;
  int nDb = strlen(zDb);
  int nByte;

  assert( pDb->pFS==0 );
  assert( pDb->pWorker==0 && pDb->pClient==0 );

  nByte = sizeof(FileSystem) + nDb+1 + nDb+4+1;
  pFS = (FileSystem *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
  if( pFS ){
    LsmFile *pLsmFile;
    pFS->zDb = (char *)&pFS[1];
    pFS->zLog = &pFS->zDb[nDb+1];
    pFS->nPagesize = LSM_DFLT_PAGE_SIZE;
    pFS->nBlocksize = LSM_DFLT_BLOCK_SIZE;
    pFS->nMetasize = 4 * 1024;
    pFS->pDb = pDb;
    pFS->pEnv = pDb->pEnv;
    if( !pDb->compress.xCompress ){


      pFS->mmapmgr.eUseMmap = pDb->eMmap;
      pFS->mmapmgr.nMapsz = 1*1024*1024;

      pFS->mmapmgr.nMapsz = 4*1024;
    }

    /* Make a copy of the database and log file names. */

    if( pLsmFile ){
      pFS->pLsmFile = pLsmFile;
      pFS->fdDb = pLsmFile->pFile;
      memset(pLsmFile, 0, sizeof(LsmFile));
    }else{
      pFS->pLsmFile = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmFile), &rc);
      if( rc==LSM_OK ){
        pFS->fdDb = fsOpenFile(pFS, bReadonly, 0, &rc);
      }
    }

    if( rc!=LSM_OK ){
      lsmFsClose(pFS);
      pFS = 0;
    }else{
      pFS->szSector = lsmEnvSectorSize(pFS->pEnv, pFS->fdDb);
    }
  }

  pDb->pFS = pFS;
  return rc;
}

/*
** Configure the file-system object according to the current values of
** the LSM_CONFIG_MMAP and LSM_CONFIG_SET_COMPRESSION options.
*/
int lsmFsConfigure(lsm_db *db){
  FileSystem *pFS = db->pFS;
  if( pFS ){
    lsm_env *pEnv = pFS->pEnv;
    Page *pPg;

    assert( pFS->nOut==0 );
    assert( pFS->pWaiting==0 );

    /* Reset any compression/decompression buffers already allocated */
    lsmFree(pEnv, pFS->aIBuffer);
    lsmFree(pEnv, pFS->aOBuffer);
    pFS->nBuffer = 0;

    /* Unmap the file, if it is currently mapped */
    if( pFS->pMap ){
      lsmEnvRemap(pEnv, pFS->fdDb, -1, &pFS->pMap, &pFS->nMap);
      pFS->bUseMmap = 0;
    }

    /* Free all allocate page structures */
    pPg = pFS->pLruFirst;
    while( pPg ){
      Page *pNext = pPg->pLruNext;
      if( pPg->flags & PAGE_FREE ) lsmFree(pEnv, pPg->aData);
      lsmFree(pEnv, pPg);
      pPg = pNext;
    }

    /* Zero pointers that point to deleted page objects */
    pFS->nCacheAlloc = 0;
    pFS->pLruFirst = 0;
    pFS->pLruLast = 0;
    pFS->pFree = 0;

    /* Configure the FileSystem object */
    if( db->compress.xCompress ){
      pFS->pCompress = &db->compress;
      pFS->bUseMmap = 0;
    }else{
      pFS->pCompress = 0;
      pFS->bUseMmap = db->bMmap;
    }
  }

  return LSM_OK;
}

/*
** Close and destroy a FileSystem object.
*/
void lsmFsClose(FileSystem *pFS){
  if( pFS ){
    Page *pPg;

    p->aData = fsMmapRef(pFS, iOff, pFS->nPagesize, &p->pRef, &rc);
    if( rc!=LSM_OK ){
      p->pHashNext = pFS->pFree;
      pFS->pFree = p;
      p = 0;
    }

    assert( (p->flags & PAGE_FREE)==0 );
  }else{

    /* Search the hash-table for the page */
    iHash = fsHashKey(pFS->nHash, iReal);
    for(p=pFS->apHash[iHash]; p; p=p->pHashNext){
      if( p->iPg==iReal) break;
    }

#define LSM_LOG_WRITE_CKSUM  0x07
#define LSM_LOG_DELETE       0x08
#define LSM_LOG_DELETE_CKSUM 0x09

/* Require a checksum every 32KB. */
#define LSM_CKSUM_MAXDATA (32*1024)

/* Do not wrap a log file smaller than this in bytes. */
#define LSM_MIN_LOGWRAP      (128*1024)

/*
** szSector:
**   Commit records must be aligned to end on szSector boundaries. If
**   the safety-mode is set to NORMAL or OFF, this value is 1. Otherwise,
**   if the safety-mode is set to FULL, it is the size of the file-system
**   sectors as reported by lsmFsSectorSize().
*/

/*
** If possible, reclaim log file space. Log file space is reclaimed after
** a snapshot that points to the same data in the database file is synced
** into the db header.
*/
static int logReclaimSpace(lsm_db *pDb){
  int rc;
  int iMeta;
  int bRotrans;                   /* True if there exists some ro-trans */

  /* Test if there exists some other connection with a read-only transaction
  ** open. If there does, then log file space may not be reclaimed.  */
  rc = lsmDetectRoTrans(pDb, &bRotrans);
  if( rc!=LSM_OK || bRotrans ) return rc;

  iMeta = (int)pDb->pShmhdr->iMetaPage;
  if( iMeta==1 || iMeta==2 ){
    DbLog *pLog = &pDb->treehdr.log;
    i64 iSyncedId;

    /* Read the snapshot-id of the snapshot stored on meta-page iMeta. Note

  LogWriter *pNew;
  LogRegion *aReg;

  if( pDb->bUseLog==0 ) return LSM_OK;

  /* If the log file has not yet been opened, open it now. Also allocate
  ** the LogWriter structure, if it has not already been allocated.  */
  rc = lsmFsOpenLog(pDb, 0);
  if( pDb->pLogWriter==0 ){
    pNew = lsmMallocZeroRc(pDb->pEnv, sizeof(LogWriter), &rc);
    if( pNew ){
      lsmStringInit(&pNew->buf, pDb->pEnv);
      rc = lsmStringExtend(&pNew->buf, 2);
    }
  }else{

  }else{
    pNew->szSector = 1;
  }

  /* There are now three scenarios:
  **
  **   1) Regions 0 and 1 are both zero bytes in size and region 2 begins
  **      at a file offset greater than LSM_MIN_LOGWRAP. In this case, wrap

  **      around to the start and write data into the start of the log file. 
  **
  **   2) Region 1 is zero bytes in size and region 2 occurs earlier in the 
  **      file than region 0. In this case, append data to region 2, but
  **      remember to jump over region 1 if required.
  **
  **   3) Region 2 is the last in the file. Append to it.
  */
  aReg = &pDb->treehdr.log.aRegion[0];

  assert( aReg[0].iEnd==0 || aReg[0].iEnd>aReg[0].iStart );
  assert( aReg[1].iEnd==0 || aReg[1].iEnd>aReg[1].iStart );

  pNew->cksum0 = pDb->treehdr.log.cksum0;
  pNew->cksum1 = pDb->treehdr.log.cksum1;

  if( aReg[0].iEnd==0 && aReg[1].iEnd==0 && aReg[2].iStart>=LSM_MIN_LOGWRAP ){
    /* Case 1. Wrap around to the start of the file. Write an LSM_LOG_JUMP 
    ** into the log file in this case. Pad it out to 8 bytes using a PAD2
    ** record so that the checksums can be updated immediately.  */
    u8 aJump[] = { 
      LSM_LOG_PAD2, 0x04, 0x00, 0x00, 0x00, 0x00, LSM_LOG_JUMP, 0x00 
    };

  LsmString buf2;                 /* Value buffer */
  LogReader reader;               /* Log reader object */
  int rc = LSM_OK;                /* Return code */
  int nCommit = 0;                /* Number of transactions to recover */
  int iPass;
  int nJump = 0;                  /* Number of LSM_LOG_JUMP records in pass 0 */
  DbLog *pLog;
  int bOpen;

  rc = lsmFsOpenLog(pDb, &bOpen);
  if( rc!=LSM_OK ) return rc;

  rc = lsmTreeInit(pDb);
  if( rc!=LSM_OK ) return rc;

  pLog = &pDb->treehdr.log;
  lsmCheckpointLogoffset(pDb->pShmhdr->aSnap2, pLog);

  logReaderInit(pDb, pLog, 1, &reader);
  lsmStringInit(&buf1, pDb->pEnv);
  lsmStringInit(&buf2, pDb->pEnv);

  /* The outer for() loop runs at most twice. The first iteration is to 
  ** count the number of committed transactions in the log. The second 
  ** iterates through those transactions and updates the in-memory tree 
  ** structure with their contents.  */
  if( bOpen ){
  for(iPass=0; iPass<2 && rc==LSM_OK; iPass++){
    int bEof = 0;

    while( rc==LSM_OK && !bEof ){
      u8 eType = 0;
      logReaderByte(&reader, &eType, &rc);

          lsmCheckpointZeroLogoffset(pDb);
        }
      }
      logReaderInit(pDb, pLog, 0, &reader);
      nCommit = nCommit * -1;
    }
  }
  }

  /* Initialize DbLog object */
  if( rc==LSM_OK ){
    pLog->aRegion[2].iEnd = reader.iOff - reader.buf.n + reader.iBuf;
    pLog->cksum0 = reader.cksum0;
    pLog->cksum1 = reader.cksum1;
  }

  if( rc==LSM_OK ){
    rc = lsmFinishRecovery(pDb);
  }else{
    lsmFinishRecovery(pDb);
  }

  if( pDb->bRoTrans ){
    lsmFsCloseLog(pDb);
  }

  lsmStringClear(&buf1);
  lsmStringClear(&buf2);
  lsmStringClear(&reader.buf);
  return rc;
}

*/
static void assert_db_state(lsm_db *pDb){

  /* If there is at least one cursor or a write transaction open, the database
  ** handle must be holding a pointer to a client snapshot. And the reverse 
  ** - if there are no open cursors and no write transactions then there must 
  ** not be a client snapshot.  */
  
  assert( (pDb->pCsr!=0||pDb->nTransOpen>0)==(pDb->iReader>=0||pDb->bRoTrans) );

  assert( (pDb->iReader<0 && pDb->bRoTrans==0) || pDb->pClient!=0 );

  assert( pDb->nTransOpen>=0 );
}
#else
# define assert_db_state(x) 
#endif

  /* Initialize the new object */
  pDb->pEnv = pEnv;
  pDb->nTreeLimit = LSM_DFLT_AUTOFLUSH;
  pDb->nAutockpt = LSM_DFLT_AUTOCHECKPOINT;
  pDb->bAutowork = LSM_DFLT_AUTOWORK;
  pDb->eSafety = LSM_DFLT_SAFETY;
  pDb->xCmp = xCmp;

  pDb->nDfltPgsz = LSM_DFLT_PAGE_SIZE;
  pDb->nDfltBlksz = LSM_DFLT_BLOCK_SIZE;
  pDb->nMerge = LSM_DFLT_AUTOMERGE;
  pDb->nMaxFreelist = LSM_MAX_FREELIST_ENTRIES;
  pDb->bUseLog = LSM_DFLT_USE_LOG;
  pDb->iReader = -1;
  pDb->iRwclient = -1;
  pDb->bMultiProc = LSM_DFLT_MULTIPLE_PROCESSES;
  pDb->eMmap = LSM_DFLT_MMAP;
  pDb->xLog = xLog;
  pDb->compress.iId = LSM_COMPRESSION_NONE;
  return LSM_OK;
}

lsm_env *lsm_get_env(lsm_db *pDb){
  assert( pDb->pEnv );
  return pDb->pEnv;
}

  if( rc!=LSM_OK ){
    lsmFree(pEnv, zAlloc);
    zAlloc = 0;
  }
  *pzAbs = zAlloc;
  return rc;
}

/*
** Check that the bits in the db->mLock mask are consistent with the
** value stored in db->iRwclient. An assert shall fail otherwise.
*/
static void assertRwclientLockValue(lsm_db *db){
#ifndef NDEBUG
  u64 msk;                        /* Mask of mLock bits for RWCLIENT locks */
  u64 rwclient = 0;               /* Bit corresponding to db->iRwclient */

  if( db->iRwclient>=0 ){
    rwclient = ((u64)1 << (LSM_LOCK_RWCLIENT(db->iRwclient)-1));
  }
  msk  = ((u64)1 << (LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT)-1)) - 1;
  msk -= (((u64)1 << (LSM_LOCK_RWCLIENT(0)-1)) - 1);

  assert( (db->mLock & msk)==rwclient );
#endif
}

/*
** Open a new connection to database zFilename.
*/
int lsm_open(lsm_db *pDb, const char *zFilename){
  int rc;

    ** than one purpose - to open both the database and log files, and 
    ** perhaps to unlink the log file during disconnection. An absolute
    ** path is required to ensure that the correct files are operated
    ** on even if the application changes the cwd.  */
    rc = getFullpathname(pDb->pEnv, zFilename, &zFull);
    assert( rc==LSM_OK || zFull==0 );

    /* Connect to the database. */
    if( rc==LSM_OK ){
      rc = lsmDbDatabaseConnect(pDb, zFull);
    }

    if( pDb->bReadonly==0 ){
    /* Configure the file-system connection with the page-size and block-size
    ** of this database. Even if the database file is zero bytes in size
      ** on disk, these values have been set in shared-memory by now, and so 
      ** are guaranteed not to change during the lifetime of this connection.  
      */
    if( rc==LSM_OK && LSM_OK==(rc = lsmCheckpointLoad(pDb, 0)) ){
      lsmFsSetPageSize(pDb->pFS, lsmCheckpointPgsz(pDb->aSnapshot));
      lsmFsSetBlockSize(pDb->pFS, lsmCheckpointBlksz(pDb->aSnapshot));
    }
    }

    lsmFree(pDb->pEnv, zFull);
    assertRwclientLockValue(pDb);
  }

  assert( pDb->bReadonly==0 || pDb->bReadonly==1 );
  assert( rc!=LSM_OK || (pDb->pShmhdr==0)==(pDb->bReadonly==1) );

  return rc;
}


int lsm_close(lsm_db *pDb){
  int rc = LSM_OK;
  if( pDb ){
    assert_db_state(pDb);
    if( pDb->pCsr || pDb->nTransOpen ){
      rc = LSM_MISUSE_BKPT;
    }else{
      lsmMCursorFreeCache(pDb);
      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
      pDb->pClient = 0;

      assertRwclientLockValue(pDb);

      lsmDbDatabaseRelease(pDb);
      lsmLogClose(pDb);
      lsmFsClose(pDb->pFS);
      assert( pDb->mLock==0 );
      
      /* Invoke any destructors registered for the compression or 
      ** compression factory callbacks.  */
      if( pDb->factory.xFree ) pDb->factory.xFree(pDb->factory.pCtx);
      if( pDb->compress.xFree ) pDb->compress.xFree(pDb->compress.pCtx);

      lsmFree(pDb->pEnv, pDb->rollback.aArray);
      lsmFree(pDb->pEnv, pDb->aTrans);
      lsmFree(pDb->pEnv, pDb->apShm);
      lsmFree(pDb->pEnv, pDb);
    }
  }
  return rc;
}

int lsm_config(lsm_db *pDb, int eParam, ...){
  int rc = LSM_OK;
  va_list ap;
  va_start(ap, eParam);

  switch( eParam ){
    case LSM_CONFIG_AUTOFLUSH: {
      /* This parameter is read and written in KB. But all internal 
      ** processing is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      int iVal = *piVal;
      if( iVal>=0 && iVal<=(1024*1024) ){
        pDb->nTreeLimit = iVal*1024;
      }
      *piVal = (pDb->nTreeLimit / 1024);
      break;
    }

    case LSM_CONFIG_AUTOWORK: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 ){
        pDb->bAutowork = *piVal;
      }
      *piVal = pDb->bAutowork;
      break;
    }

    case LSM_CONFIG_AUTOCHECKPOINT: {
      /* This parameter is read and written in KB. But all internal processing
      ** (including the lsm_db.nAutockpt variable) is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 ){
        int iVal = *piVal;
        pDb->nAutockpt = (i64)iVal * 1024;
      }
      *piVal = (int)(pDb->nAutockpt / 1024);









      break;
    }

    case LSM_CONFIG_PAGE_SIZE: {
      int *piVal = va_arg(ap, int *);
      if( pDb->pDatabase ){
        /* If lsm_open() has been called, this is a read-only parameter. 

          *piVal = pDb->nDfltPgsz;
        }
      }
      break;
    }

    case LSM_CONFIG_BLOCK_SIZE: {
      /* This parameter is read and written in KB. But all internal 
      ** processing is done in bytes.  */
      int *piVal = va_arg(ap, int *);
      if( pDb->pDatabase ){
        /* If lsm_open() has been called, this is a read-only parameter. 
        ** Set the output variable to the block-size in KB according to the 
        ** FileSystem object.  */
        *piVal = lsmFsBlockSize(pDb->pFS) / 1024;
      }else{
        int iVal = *piVal;
        if( iVal>=64 && iVal<=65536 && ((iVal-1) & iVal)==0 ){
          pDb->nDfltBlksz = iVal * 1024;
        }else{
          *piVal = pDb->nDfltBlksz / 1024;
        }
      }
      break;
    }

    case LSM_CONFIG_SAFETY: {
      int *piVal = va_arg(ap, int *);
      if( *piVal>=0 && *piVal<=2 ){
        pDb->eSafety = *piVal;
      }
      *piVal = pDb->eSafety;
      break;
    }

    case LSM_CONFIG_MMAP: {
      int *piVal = va_arg(ap, int *);
      if( pDb->iReader<0 && *piVal>=0 && *piVal<=1 ){
        pDb->bMmap = *piVal;
        rc = lsmFsConfigure(pDb);
      }
      *piVal = pDb->eMmap;
      break;
    }

    case LSM_CONFIG_USE_LOG: {
      int *piVal = va_arg(ap, int *);

        ** in multi-process mode.  */
        *piVal = lsmDbMultiProc(pDb);
      }else{
        pDb->bMultiProc = *piVal = (*piVal!=0);
      }
      break;
    }

    case LSM_CONFIG_READONLY: {
      int *piVal = va_arg(ap, int *);
      /* If lsm_open() has been called, this is a read-only parameter. */
      if( pDb->pDatabase==0 && *piVal>=0 ){
        pDb->bReadonly = *piVal = (*piVal!=0);
      }
      *piVal = pDb->bReadonly;
      break;
    }

    case LSM_CONFIG_SET_COMPRESSION: {
      lsm_compress *p = va_arg(ap, lsm_compress *);
      if( pDb->iReader>=0 && pDb->bInFactory==0 ){
        /* May not change compression schemes with an open transaction */
        rc = LSM_MISUSE_BKPT;
      }else{
        if( pDb->compress.xFree ){
          /* Invoke any destructor belonging to the current compression. */
          pDb->compress.xFree(pDb->compress.pCtx);
        }
        if( p->xBound==0 ){
          memset(&pDb->compress, 0, sizeof(lsm_compress));
          pDb->compress.iId = LSM_COMPRESSION_NONE;
        }else{
        memcpy(&pDb->compress, p, sizeof(lsm_compress));
      }
        rc = lsmFsConfigure(pDb);
      }
      break;
    }

    case LSM_CONFIG_SET_COMPRESSION_FACTORY: {
      lsm_compress_factory *p = va_arg(ap, lsm_compress_factory *);
      if( pDb->factory.xFree ){
        /* Invoke any destructor belonging to the current factory. */
        pDb->factory.xFree(pDb->factory.pCtx);
      }
      memcpy(&pDb->factory, p, sizeof(lsm_compress_factory));
      break;
    }

    case LSM_CONFIG_GET_COMPRESSION: {
      lsm_compress *p = va_arg(ap, lsm_compress *);
      memcpy(p, &pDb->compress, sizeof(lsm_compress));
      break;

  return 0;
}

int lsmInfoFreelist(lsm_db *pDb, char **pzOut){
  Snapshot *pWorker;              /* Worker snapshot */
  int bUnlock = 0;
  LsmString s;

  int rc;

  /* Obtain the worker snapshot */
  rc = infoGetWorker(pDb, &pWorker, &bUnlock);
  if( rc!=LSM_OK ) return rc;

  lsmStringInit(&s, pDb->pEnv);
  rc = lsmWalkFreelist(pDb, 0, infoFreelistCb, &s);
  if( rc!=LSM_OK ){
    lsmFree(pDb->pEnv, s.z);
  }else{
    *pzOut = s.z;
  }

  /* Release the snapshot and return */
  infoFreeWorker(pDb, bUnlock);
  return rc;
}




static int infoTreeSize(lsm_db *db, int *pnOldKB, int *pnNewKB){
  ShmHeader *pShm = db->pShmhdr;
  TreeHeader *p = &pShm->hdr1;

  /* The following code suffers from two race conditions, as it accesses and
  ** trusts the contents of shared memory without verifying checksums:
  **
  **   * The two values read - TreeHeader.root.nByte and oldroot.nByte - are 

  **     for the size of the "old" tree may reflect the size of an "old"
  **     tree that was recently flushed to disk.
  **
  ** Given the context in which this function is called (as a result of an
  ** lsm_info(LSM_INFO_TREE_SIZE) request), neither of these are considered to
  ** be problems.
  */
  *pnNewKB = ((int)p->root.nByte + 1023) / 1024;
  if( p->iOldShmid ){
    if( p->iOldLog==lsmCheckpointLogOffset(pShm->aSnap1) ){
      *pnOldKB = 0;
    }else{
      *pnOldKB = ((int)p->oldroot.nByte + 1023) / 1024;
    }
  }else{
    *pnOldKB = 0;
  }

  return LSM_OK;
}

int lsm_info(lsm_db *pDb, int eParam, ...){
  int rc = LSM_OK;

    case LSM_INFO_FREELIST: {
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoFreelist(pDb, pzVal);
      break;
    }

    case LSM_INFO_CHECKPOINT_SIZE: {
      int *pnKB = va_arg(ap, int *);
      rc = lsmCheckpointSize(pDb, pnKB);
      break;
    }

    case LSM_INFO_TREE_SIZE: {
      int *pnOld = va_arg(ap, int *);
      int *pnNew = va_arg(ap, int *);
      rc = infoTreeSize(pDb, pnOld, pnNew);
      break;
    }

    case LSM_INFO_COMPRESSION_ID: {
      unsigned int *piOut = va_arg(ap, unsigned int *);
      if( pDb->pClient ){
        *piOut = pDb->pClient->iCmpId;
      }else{
        rc = lsmInfoCompressionId(pDb, piOut);
      }
      break;
    }

    default:
      rc = LSM_MISUSE;
      break;
  }

  va_end(ap);

/*
** Open a new cursor handle. 
**
** If there are currently no other open cursor handles, and no open write
** transaction, open a read transaction here.
*/
int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr){
  int rc = LSM_OK;                /* Return code */
  MultiCursor *pCsr = 0;          /* New cursor object */

  /* Open a read transaction if one is not already open. */
  assert_db_state(pDb);

  if( pDb->pShmhdr==0 ){
    assert( pDb->bReadonly );
    rc = lsmBeginRoTrans(pDb);
  }else if( pDb->iReader<0 ){
  rc = lsmBeginReadTrans(pDb);
  }

  /* Allocate the multi-cursor. */
  if( rc==LSM_OK ){
    rc = lsmMCursorNew(pDb, &pCsr);
  }

  /* If an error has occured, set the output to NULL and delete any partially
  ** allocated cursor. If this means there are no open cursors, release the
  ** client snapshot.  */
  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
    dbReleaseClientSnapshot(pDb);
  }

  assert_db_state(pDb);
  *ppCsr = (lsm_cursor *)pCsr;
  return rc;
}

/*
** Close a cursor opened using lsm_csr_open().
*/
int lsm_csr_close(lsm_cursor *p){
  if( p ){
    lsm_db *pDb = lsmMCursorDb((MultiCursor *)p);
    assert_db_state(pDb);
    lsmMCursorClose((MultiCursor *)p, 1);
    dbReleaseClientSnapshot(pDb);
    assert_db_state(pDb);
  }
  return LSM_OK;
}

/*

    va_end(ap2);
    pDb->xLog(pDb->pLogCtx, rc, s.z);
    lsmStringClear(&s);
  }
}

int lsm_begin(lsm_db *pDb, int iLevel){
  int rc;

  assert_db_state( pDb );
  rc = (pDb->bReadonly ? LSM_READONLY : LSM_OK);

  /* A value less than zero means open one more transaction. */
  if( iLevel<0 ) iLevel = pDb->nTransOpen + 1;

  if( iLevel>pDb->nTransOpen ){
    int i;

    /* Extend the pDb->aTrans[] array if required. */
    if( rc==LSM_OK && pDb->nTransAlloc<iLevel ){
      TransMark *aNew;            /* New allocation */
      int nByte = sizeof(TransMark) * (iLevel+1);

  assert_db_state( pDb );

  /* A value less than zero means close the innermost nested transaction. */
  if( iLevel<0 ) iLevel = LSM_MAX(0, pDb->nTransOpen - 1);

  if( iLevel<pDb->nTransOpen ){
    if( iLevel==0 ){


      /* Commit the transaction to disk. */
      if( rc==LSM_OK ) rc = lsmLogCommit(pDb);
      if( rc==LSM_OK && pDb->eSafety==LSM_SAFETY_FULL ){
        rc = lsmFsSyncLog(pDb->pFS);
      }
      lsmFinishWriteTrans(pDb, (rc==LSM_OK));
    }

  /* Protected by the global mutex (enterGlobalMutex/leaveGlobalMutex): */
  char *zName;                    /* Canonical path to database file */
  int nName;                      /* strlen(zName) */
  int nDbRef;                     /* Number of associated lsm_db handles */
  Database *pDbNext;              /* Next Database structure in global list */

  /* Protected by the local mutex (pClientMutex) */
  int bReadonly;                  /* True if Database.pFile is read-only */
  int bMultiProc;                 /* True if running in multi-process mode */
  lsm_file *pFile;                /* Used for locks/shm in multi-proc mode */
  LsmFile *pLsmFile;              /* List of deferred closes */
  lsm_mutex *pClientMutex;        /* Protects the apShmChunk[] and pConn */
  int nShmChunk;                  /* Number of entries in apShmChunk[] array */
  void **apShmChunk;              /* Array of "shared" memory regions */
  lsm_db *pConn;                  /* List of connections to this db. */

** to as small a size as possible without truncating away any blocks that
** contain data.
*/
static int dbTruncateFile(lsm_db *pDb){
  int rc;

  assert( pDb->pWorker==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) );
  rc = lsmCheckpointLoadWorker(pDb);

  if( rc==LSM_OK ){
    DbTruncateCtx ctx;

    /* Walk the database free-block-list in reverse order. Set ctx.nBlock
    ** to the block number of the last block in the database that actually

  pDb->pWorker = 0;
  return rc;
}

static void doDbDisconnect(lsm_db *pDb){
  int rc;

  if( pDb->bReadonly ){
    lsmShmLock(pDb, LSM_LOCK_DMS3, LSM_LOCK_UNLOCK, 0);
  }else{
  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
  if( rc==LSM_OK ){

    /* Try an exclusive lock on DMS2. If successful, this is the last
    ** connection to the database. In this case flush the contents of the
    ** in-memory tree to disk and write a checkpoint.  */
      rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 1, LSM_LOCK_EXCL);
    if( rc==LSM_OK ){
        rc = lsmShmTestLock(pDb, LSM_LOCK_CHECKPOINTER, 1, LSM_LOCK_EXCL);
      }
      if( rc==LSM_OK ){
        int bReadonly = 0;        /* True if there exist read-only conns. */

      /* Flush the in-memory tree, if required. If there is data to flush,
      ** this will create a new client snapshot in Database.pClient. The
      ** checkpoint (serialization) of this snapshot may be written to disk
      ** by the following block.  
      **
        ** There is no need to take a WRITER lock here. That there are no 
        ** other locks on DMS2 guarantees that there are no other read-write
        ** connections at this time (and the lock on DMS1 guarantees that
        ** no new ones may appear).
      */
      rc = lsmTreeLoadHeader(pDb, 0);
      if( rc==LSM_OK && (lsmTreeHasOld(pDb) || lsmTreeSize(pDb)>0) ){
        rc = lsmFlushTreeToDisk(pDb);
      }

        /* Now check if there are any read-only connections. If there are,
        ** then do not truncate the db file or unlink the shared-memory 
        ** region.  */
        if( rc==LSM_OK ){
          rc = lsmShmTestLock(pDb, LSM_LOCK_DMS3, 1, LSM_LOCK_EXCL);
          if( rc==LSM_BUSY ){
            bReadonly = 1;
            rc = LSM_OK;
          }
        }

      /* Write a checkpoint to disk. */
      if( rc==LSM_OK ){
          rc = lsmCheckpointWrite(pDb, (bReadonly==0), 0);
      }

      /* If the checkpoint was written successfully, delete the log file
      ** and, if possible, truncate the database file.  */
      if( rc==LSM_OK ){
          int bRotrans = 0;
        Database *p = pDb->pDatabase;

          /* The log file may only be deleted if there are no clients 
          ** read-only clients running rotrans transactions.  */
          rc = lsmDetectRoTrans(pDb, &bRotrans);
          if( rc==LSM_OK && bRotrans==0 ){
        lsmFsCloseAndDeleteLog(pDb->pFS);
          }

          /* The database may only be truncated if there exist no read-only
          ** clients - either connected or running rotrans transactions. */
          if( bReadonly==0 && bRotrans==0 ){
            dbTruncateFile(pDb);
            if( p->pFile && p->bMultiProc ){
              lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
      }
    }
  }
      }
    }

    if( pDb->iRwclient>=0 ){
      lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
      pDb->iRwclient = -1;
    }

  lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
  lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
  }
  pDb->pShmhdr = 0;
}

static int doDbConnect(lsm_db *pDb){
  const int nUsMax = 100000;      /* Max value for nUs */
  int nUs = 1000;                 /* us to wait between DMS1 attempts */
  int rc;

  /* Obtain a pointer to the shared-memory header */
  assert( pDb->pShmhdr==0 );
  assert( pDb->bReadonly==0 );
  rc = lsmShmCacheChunks(pDb, 1);
  if( rc!=LSM_OK ) return rc;
  pDb->pShmhdr = (ShmHeader *)pDb->apShm[0];

  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  while( 1 ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
    if( rc!=LSM_BUSY ) break;
    lsmEnvSleep(pDb->pEnv, nUs);
    nUs = nUs * 2;
    if( nUs>nUsMax ) nUs = nUsMax;
  }
  if( rc!=LSM_OK ){
    pDb->pShmhdr = 0;
    return rc;
  }

  /* Try an exclusive lock on DMS2/DMS3. If successful, this is the first 
  ** and only connection to the database. In this case initialize the 
  ** shared-memory and run log file recovery.  */
  assert( LSM_LOCK_DMS3==1+LSM_LOCK_DMS2 );
  rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 2, LSM_LOCK_EXCL);
  if( rc==LSM_OK ){
    memset(pDb->pShmhdr, 0, sizeof(ShmHeader));
    rc = lsmCheckpointRecover(pDb);
    if( rc==LSM_OK ){
      rc = lsmLogRecover(pDb);
    }
    if( rc==LSM_OK ){
      ShmHeader *pShm = pDb->pShmhdr;
      pShm->aReader[0].iLsmId = lsmCheckpointId(pShm->aSnap1, 0);
      pShm->aReader[0].iTreeId = pDb->treehdr.iUsedShmid;
    }
  }else if( rc==LSM_BUSY ){
    rc = LSM_OK;
  }

  /* Take a shared lock on DMS2. In multi-process mode this lock "cannot" 
  ** fail, as connections may only hold an exclusive lock on DMS2 if they 
  ** first hold an exclusive lock on DMS1. And this connection is currently 
  ** holding the exclusive lock on DSM1. 
  **
  ** However, if some other connection has the database open in single-process
  ** mode, this operation will fail. In this case, return the error to the
  ** caller - the attempt to connect to the db has failed.
  */
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_SHARED, 0);
  }

  /* If anything went wrong, unlock DMS2. Otherwise, try to take an exclusive
  ** lock on one of the LSM_LOCK_RWCLIENT() locks. Unlock DMS1 in any case. */
  if( rc!=LSM_OK ){

    pDb->pShmhdr = 0;
  }else{
    int i;
    for(i=0; i<LSM_LOCK_NRWCLIENT; i++){
      int rc2 = lsmShmLock(pDb, LSM_LOCK_RWCLIENT(i), LSM_LOCK_EXCL, 0);
      if( rc2==LSM_OK ) pDb->iRwclient = i;
      if( rc2!=LSM_BUSY ){
        rc = rc2;
        break;
  }
    }
  }
  lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

  return rc;
}

static int dbOpenSharedFd(lsm_env *pEnv, Database *p, int bRoOk){
  int rc;

  rc = lsmEnvOpen(pEnv, p->zName, 0, &p->pFile);
  if( rc==LSM_IOERR && bRoOk ){
    rc = lsmEnvOpen(pEnv, p->zName, LSM_OPEN_READONLY, &p->pFile);
    p->bReadonly = 1;
  }

  return rc;
}

/*
** Return a reference to the shared Database handle for the database 
** identified by canonical path zName. If this is the first connection to
** the named database, a new Database object is allocated. Otherwise, a

        rc = lsmMutexNew(pEnv, &p->pClientMutex);
      }

      /* If nothing has gone wrong so far, open the shared fd. And if that
      ** succeeds and this connection requested single-process mode, 
      ** attempt to take the exclusive lock on DMS2.  */
      if( rc==LSM_OK ){
        int bReadonly = (pDb->bReadonly && pDb->bMultiProc);
        rc = dbOpenSharedFd(pDb->pEnv, p, bReadonly);
      }

      if( rc==LSM_OK && p->bMultiProc==0 ){
        assert( p->bReadonly==0 );
        rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS2, LSM_LOCK_EXCL);
      }

      if( rc==LSM_OK ){
        p->pDbNext = gShared.pDatabase;
        gShared.pDatabase = p;
      }else{

      lsmMutexLeave(pDb->pEnv, p->pClientMutex);
    }
  }

  pDb->pDatabase = p;
  if( rc==LSM_OK ){
    assert( p );
    rc = lsmFsOpen(pDb, zName, p->bReadonly);
  }

  /* If the db handle is read-write, then connect to the system now. Run
  ** recovery as necessary. Or, if this is a read-only database handle,
  ** defer attempting to connect to the system until a read-transaction
  ** is opened.  */
  if( pDb->bReadonly==0 ){
  if( rc==LSM_OK ){
    rc = doDbConnect(pDb);
    }
    if( rc==LSM_OK ){
      rc = lsmFsConfigure(pDb);
    }
  }

  return rc;
}

static void dbDeferClose(lsm_db *pDb){
  if( pDb->pFS ){

    lsmLogMessage(pDb, 0, "lsmBlockAllocate(): "
        "snapshot-in-use: %lld (iSynced=%lld) (client-id=%lld)", 
        iInUse, iSynced, (pDb->iReader>=0 ? pDb->pClient->iId : 0)
    );
  }
#endif


  /* Unless there exists a read-only transaction (which prevents us from
  ** recycling any blocks regardless, query the free block list for a 
  ** suitable block to reuse. 
  **
  ** It might seem more natural to check for a read-only transaction at
  ** the start of this function. However, it is better do wait until after
  ** the call to lsmCheckpointSynced() to do so.
  */
  if( rc==LSM_OK ){
    int bRotrans;
    rc = lsmDetectRoTrans(pDb, &bRotrans);

    if( rc==LSM_OK && bRotrans==0 ){
      rc = findFreeblock(pDb, iInUse, (iBefore>0), &iRet);
    }
  }

  if( iBefore>0 && (iRet<=0 || iRet>=iBefore) ){
    iRet = 0;

  }else if( rc==LSM_OK ){
    /* If a block was found in the free block list, use it and remove it from 
    ** the list. Otherwise, if no suitable block was found, allocate one from

** but then not used. This function is used to push the block back onto
** the freelist. Refreeing a block is different from freeing is, as a refreed
** block may be reused immediately. Whereas a freed block can not be reused 
** until (at least) after the next checkpoint.
*/
int lsmBlockRefree(lsm_db *pDb, int iBlk){
  int rc = LSM_OK;                /* Return code */


#ifdef LSM_LOG_FREELIST
  lsmLogMessage(pDb, LSM_OK, "lsmBlockRefree(): Refree block %d", iBlk);
#endif

  rc = freelistAppend(pDb, iBlk, 0);
  return rc;

  if( p ){
    lsmSortedFreeLevel(pEnv, p->pLevel);
    lsmFree(pEnv, p->freelist.aEntry);
    lsmFree(pEnv, p->redirect.a);
    lsmFree(pEnv, p);
  }
}

/*
** Attempt to populate one of the read-lock slots to contain lock values
** iLsm/iShm. Or, if such a slot exists already, this function is a no-op.
**
** It is not an error if no slot can be populated because the write-lock
** cannot be obtained. If any other error occurs, return an LSM error code.
** Otherwise, LSM_OK.
**
** This function is called at various points to try to ensure that there
** always exists at least one read-lock slot that can be used by a read-only
** client. And so that, in the usual case, there is an "exact match" available
** whenever a read transaction is opened by any client. At present this
** function is called when:
**
**    * A write transaction that called lsmTreeDiscardOld() is committed, and
**    * Whenever the working snapshot is updated (i.e. lsmFinishWork()).
*/
static int dbSetReadLock(lsm_db *db, i64 iLsm, u32 iShm){
  int rc = LSM_OK;
  ShmHeader *pShm = db->pShmhdr;
  int i;

  /* Check if there is already a slot containing the required values. */
  for(i=0; i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId==iLsm && p->iTreeId==iShm ) return LSM_OK;
  }

  /* Iterate through all read-lock slots, attempting to take a write-lock
  ** on each of them. If a write-lock succeeds, populate the locked slot
  ** with the required values and break out of the loop.  */
  for(i=0; rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
    if( rc==LSM_BUSY ){
      rc = LSM_OK;
    }else{
      ShmReader *p = &pShm->aReader[i];
      p->iLsmId = iLsm;
      p->iTreeId = iShm;
      lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
      break;
    }
  }

  return rc;
}

/*
** Release the read-lock currently held by connection db.
*/
int dbReleaseReadlock(lsm_db *db){
  int rc = LSM_OK;
  if( db->iReader>=0 ){
    rc = lsmShmLock(db, LSM_LOCK_READER(db->iReader), LSM_LOCK_UNLOCK, 0);
    db->iReader = -1;
  }
  db->bRoTrans = 0;
  return rc;
}


/*
** Argument bFlush is true if the contents of the in-memory tree has just
** been flushed to disk. The significance of this is that once the snapshot
** created to hold the updated state of the database is synced to disk, log
** file space can be recycled.
*/
void lsmFinishWork(lsm_db *pDb, int bFlush, int *pRc){
  int rc = *pRc;
  assert( rc!=0 || pDb->pWorker );
  if( pDb->pWorker ){
    /* If no error has occurred, serialize the worker snapshot and write
    ** it to shared memory.  */
    if( rc==LSM_OK ){
      rc = lsmSaveWorker(pDb, bFlush);
    }

    /* Assuming no error has occurred, update a read lock slot with the
    ** new snapshot id (see comments above function dbSetReadLock()).  */
    if( rc==LSM_OK ){
      if( pDb->iReader<0 ){
        rc = lsmTreeLoadHeader(pDb, 0);
      }
      if( rc==LSM_OK ){
        rc = dbSetReadLock(pDb, pDb->pWorker->iId, pDb->treehdr.iUsedShmid);
      }
    }

    /* Free the snapshot object. */
    lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);
    pDb->pWorker = 0;
  }

  lsmShmLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK, 0);
  *pRc = rc;
}


/*
** Called when recovery is finished.
*/
int lsmFinishRecovery(lsm_db *pDb){
  lsmTreeEndTransaction(pDb, 1);
  return LSM_OK;
}

/*
** Check if the currently configured compression functions
** (LSM_CONFIG_SET_COMPRESSION) are compatible with a database that has its
** compression id set to iReq. Compression routines are compatible if iReq
** is zero (indicating the database is empty), or if it is equal to the 
** compression id of the configured compression routines.
**
** If the check shows that the current compression are incompatible and there
** is a compression factory registered, give it a chance to install new
** compression routines.
**
** If, after any registered factory is invoked, the compression functions
** are still incompatible, return LSM_MISMATCH. Otherwise, LSM_OK.
*/
int lsmCheckCompressionId(lsm_db *pDb, u32 iReq){
  if( iReq!=LSM_COMPRESSION_EMPTY && pDb->compress.iId!=iReq ){
    if( pDb->factory.xFactory ){
      pDb->bInFactory = 1;
      pDb->factory.xFactory(pDb->factory.pCtx, pDb, iReq);
      pDb->bInFactory = 0;
    }
    if( pDb->compress.iId!=iReq ){
      /* Incompatible */
      return LSM_MISMATCH;
    }
  }
  /* Compatible */
  return LSM_OK;
}

/*
** Begin a read transaction. This function is a no-op if the connection
** passed as the only argument already has an open read transaction.
*/
int lsmBeginReadTrans(lsm_db *pDb){
  const int MAX_READLOCK_ATTEMPTS = 10;
  const int nMaxAttempt = (pDb->bRoTrans ? 1 : MAX_READLOCK_ATTEMPTS);

  int rc = LSM_OK;                /* Return code */
  int iAttempt = 0;

  assert( pDb->pWorker==0 );

  while( rc==LSM_OK && pDb->iReader<0 && (iAttempt++)<nMaxAttempt ){
    int iTreehdr = 0;
    int iSnap = 0;
    assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

    /* Load the in-memory tree header. */
    rc = lsmTreeLoadHeader(pDb, &iTreehdr);

    /* Load the database snapshot */
    if( rc==LSM_OK ){
      if( lsmCheckpointClientCacheOk(pDb)==0 ){
        lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
        pDb->pClient = 0;
        lsmMCursorFreeCache(pDb);
        rc = lsmCheckpointLoad(pDb, &iSnap);
      }else{
        iSnap = 1;
      }
    }

    /* Take a read-lock on the tree and snapshot just loaded. Then check

          ** lsm_sorted.c is changed to work directly from the serialized
          ** version of the snapshot.  */
          if( pDb->pClient==0 ){
            rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot,&pDb->pClient);
          }
          assert( (rc==LSM_OK)==(pDb->pClient!=0) );
          assert( pDb->iReader>=0 );

          /* Check that the client has the right compression hooks loaded.
          ** If not, set rc to LSM_MISMATCH.  */
          if( rc==LSM_OK ){
            rc = lsmCheckCompressionId(pDb, pDb->pClient->iCmpId);
          }
        }else{
          rc = dbReleaseReadlock(pDb);
        }
      }

      if( rc==LSM_BUSY ){
        rc = LSM_OK;
      }
    }
#if 0
if( rc==LSM_OK && pDb->pClient ){
  fprintf(stderr, 

#endif
  }

  if( rc==LSM_OK ){
    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
  }
  if( rc!=LSM_OK ){
    dbReleaseReadlock(pDb);
  }
  if( pDb->pClient==0 && rc==LSM_OK ) rc = LSM_BUSY;
  return rc;
}

/*
** This function is used by a read-write connection to determine if there
** are currently one or more read-only transactions open on the database
** (in this context a read-only transaction is one opened by a read-only
** connection on a non-live database).
**
** If no error occurs, LSM_OK is returned and *pbExists is set to true if
** some other connection has a read-only transaction open, or false 
** otherwise. If an error occurs an LSM error code is returned and the final
** value of *pbExist is undefined.
*/
int lsmDetectRoTrans(lsm_db *db, int *pbExist){
  int rc;

  /* Only a read-write connection may use this function. */
  assert( db->bReadonly==0 );

  rc = lsmShmTestLock(db, LSM_LOCK_ROTRANS, 1, LSM_LOCK_EXCL);
  if( rc==LSM_BUSY ){
    *pbExist = 1;
    rc = LSM_OK;
  }else{
    *pbExist = 0;
  }

  return rc;
}

/*
** db is a read-only database handle in the disconnected state. This function
** attempts to open a read-transaction on the database. This may involve
** connecting to the database system (opening shared memory etc.).
*/
int lsmBeginRoTrans(lsm_db *db){
  int rc = LSM_OK;

  assert( db->bReadonly && db->pShmhdr==0 );
  assert( db->iReader<0 );

  if( db->bRoTrans==0 ){

    /* Attempt a shared-lock on DMS1. */
    rc = lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_SHARED, 0);
    if( rc!=LSM_OK ) return rc;

    rc = lsmShmTestLock(
        db, LSM_LOCK_RWCLIENT(0), LSM_LOCK_NREADER, LSM_LOCK_SHARED
    );
    if( rc==LSM_OK ){
      /* System is not live. Take a SHARED lock on the ROTRANS byte and
      ** release DMS1. Locking ROTRANS tells all read-write clients that they
      ** may not recycle any disk space from within the database or log files,
      ** as a read-only client may be using it.  */
      rc = lsmShmLock(db, LSM_LOCK_ROTRANS, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

      if( rc==LSM_OK ){
        db->bRoTrans = 1;
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
          memset(db->pShmhdr, 0, sizeof(ShmHeader));
          rc = lsmCheckpointRecover(db);
          if( rc==LSM_OK ){
            rc = lsmLogRecover(db);
          }
        }
      }
    }else if( rc==LSM_BUSY ){
      /* System is live! */
      rc = lsmShmLock(db, LSM_LOCK_DMS3, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
      if( rc==LSM_OK ){
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
        }
      }
    }

    if( rc==LSM_OK ){
      rc = lsmBeginReadTrans(db);
    }
  }

  return rc;
}

/*
** Close the currently open read transaction.
*/
void lsmFinishReadTrans(lsm_db *pDb){

  /* Worker connections should not be closing read transactions. And
  ** read transactions should only be closed after all cursors and write
  ** transactions have been closed. Finally pClient should be non-NULL
  ** only iff pDb->iReader>=0.  */
  assert( pDb->pWorker==0 );
  assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

  if( pDb->bRoTrans ){
    int i;
    for(i=0; i<pDb->nShm; i++){
      lsmFree(pDb->pEnv, pDb->apShm[i]);

  }
    lsmFree(pDb->pEnv, pDb->apShm);
    pDb->apShm = 0;
    pDb->nShm = 0;
    pDb->pShmhdr = 0;

    lsmShmLock(pDb, LSM_LOCK_ROTRANS, LSM_LOCK_UNLOCK, 0);

}

  dbReleaseReadlock(pDb);
}

/*
** Open a write transaction.
*/
int lsmBeginWriteTrans(lsm_db *pDb){
  int rc = LSM_OK;                /* Return code */
  ShmHeader *pShm = pDb->pShmhdr; /* Shared memory header */

  assert( pDb->nTransOpen==0 );
  assert( pDb->bDiscardOld==0 );
  assert( pDb->bReadonly==0 );

  /* If there is no read-transaction open, open one now. */
  if( pDb->iReader<0 ){
  rc = lsmBeginReadTrans(pDb);
  }

  /* Attempt to take the WRITER lock */
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_EXCL, 0);
  }

  /* If the previous writer failed mid-transaction, run emergency rollback. */

  ** WRITER lock and return an error code.  */
  if( rc==LSM_OK ){
    TreeHeader *p = &pDb->treehdr;
    pShm->bWriter = 1;
    p->root.iTransId++;
    if( lsmTreeHasOld(pDb) && p->iOldLog==pDb->pClient->iLogOff ){
      lsmTreeDiscardOld(pDb);
      pDb->bDiscardOld = 1;
    }
  }else{
    lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);
    if( pDb->pCsr==0 ) lsmFinishReadTrans(pDb);
  }
  return rc;
}

  lsmLogEnd(pDb, bCommit);
  if( rc==LSM_OK && bCommit && lsmTreeSize(pDb)>pDb->nTreeLimit ){
    bFlush = 1;
    lsmTreeMakeOld(pDb);
  }
  lsmTreeEndTransaction(pDb, bCommit);

  if( rc==LSM_OK ){
    if( bFlush && pDb->bAutowork ){
    rc = lsmSortedAutoWork(pDb, 1);
    }else if( bCommit && pDb->bDiscardOld ){
      rc = dbSetReadLock(pDb, pDb->pClient->iId, pDb->treehdr.iUsedShmid);
  }
  }
  pDb->bDiscardOld = 0;
  lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);

  if( bFlush && pDb->bAutowork==0 && pDb->xWork ){
    pDb->xWork(pDb, pDb->pWorkCtx);
  }
  return rc;
}

int lsmReadlock(lsm_db *db, i64 iLsm, u32 iShmMin, u32 iShmMax){
  int rc = LSM_OK;
  ShmHeader *pShm = db->pShmhdr;
  int i;

  assert( db->iReader<0 );
  assert( shm_sequence_ge(iShmMax, iShmMin) );

  /* This is a no-op if the read-only transaction flag is set. */
  if( db->bRoTrans ){
    db->iReader = 0;
    return LSM_OK;
  }

  /* Search for an exact match. */
  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
    ShmReader *p = &pShm->aReader[i];
    if( p->iLsmId==iLsm && p->iTreeId==iShmMax ){
      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
      if( rc==LSM_OK && p->iLsmId==iLsm && p->iTreeId==iShmMax ){

  if( db->pClient && db->pClient->iId<=iLsmId ){
    *pbInUse = 1;
    return LSM_OK;
  }
  return isInUse(db, iLsmId, 0, pbInUse);
}













/*
** This function may only be called after a successful call to
** lsmDbDatabaseConnect(). It returns true if the connection is in
** multi-process mode, or false otherwise.
*/
int lsmDbMultiProc(lsm_db *pDb){
  return pDb->pDatabase && pDb->pDatabase->bMultiProc;

** Ensure that database connection db has cached pointers to at least the 
** first nChunk chunks of shared memory.
*/
int lsmShmCacheChunks(lsm_db *db, int nChunk){
  int rc = LSM_OK;
  if( nChunk>db->nShm ){
    static const int NINCR = 16;

    Database *p = db->pDatabase;
    lsm_env *pEnv = db->pEnv;
    int nAlloc;
    int i;

    /* Ensure that the db->apShm[] array is large enough. If an attempt to
    ** allocate memory fails, return LSM_NOMEM immediately. The apShm[] array
    ** is always extended in multiples of 16 entries - so the actual allocated
    ** size can be inferred from nShm.  */ 
    nAlloc = ((db->nShm + NINCR - 1) / NINCR) * NINCR;
    while( nChunk>=nAlloc ){
      void **apShm;
      nAlloc += NINCR;
      apShm = lsmRealloc(pEnv, db->apShm, sizeof(void*)*nAlloc);
      if( !apShm ) return LSM_NOMEM_BKPT;
      db->apShm = apShm;
    }

    if( db->bRoTrans ){
      for(i=db->nShm; rc==LSM_OK && i<nChunk; i++){
        db->apShm[i] = lsmMallocZeroRc(pEnv, LSM_SHM_CHUNK_SIZE, &rc);
        db->nShm++;
      }

    }else{

    /* Enter the client mutex */
    lsmMutexEnter(pEnv, p->pClientMutex);

    /* Extend the Database objects apShmChunk[] array if necessary. Using the
    ** same pattern as for the lsm_db.apShm[] array above.  */
    nAlloc = ((p->nShmChunk + NINCR - 1) / NINCR) * NINCR;

        db->nShm++;
      }
    }

    /* Release the client mutex */
    lsmMutexLeave(pEnv, p->pClientMutex);
  }
  }

  return rc;
}

static int lockSharedFile(lsm_env *pEnv, Database *p, int iLock, int eOp){
  int rc = LSM_OK;
  if( p->bMultiProc ){
    rc = lsmEnvLock(pEnv, p->pFile, iLock, eOp);
  }
  return rc;
}

/*
** Test if it would be possible for connection db to obtain a lock of type
** eType on the nLock locks starting at iLock. If so, return LSM_OK. If it
** would not be possible to obtain the lock due to a lock held by another
** connection, return LSM_BUSY. If an IO or other error occurs (i.e. in the 
** lsm_env.xTestLock function), return some other LSM error code.
**
** Note that this function never actually locks the database - it merely
** queries the system to see if there exists a lock that would prevent
** it from doing so.
*/
int lsmShmTestLock(
  lsm_db *db,
  int iLock,
  int nLock,
  int eOp
){
  int rc = LSM_OK;
  lsm_db *pIter;
  Database *p = db->pDatabase;
  int i;
  u64 mask = 0;

  for(i=iLock; i<(iLock+nLock); i++){
    mask |= ((u64)1 << (iLock-1));
    if( eOp==LSM_LOCK_EXCL ) mask |= ((u64)1 << (iLock+32-1));
  }

  lsmMutexEnter(db->pEnv, p->pClientMutex);
  for(pIter=p->pConn; pIter; pIter=pIter->pNext){
    if( pIter!=db && (pIter->mLock & mask) ) break;
  }

  if( pIter ){
    rc = LSM_BUSY;
  }else if( p->bMultiProc ){
    rc = lsmEnvTestLock(db->pEnv, p->pFile, iLock, nLock, eOp);
  }

  lsmMutexLeave(db->pEnv, p->pClientMutex);
  return rc;
}

/*
** Attempt to obtain the lock identified by the iLock and bExcl parameters.
** If successful, return LSM_OK. If the lock cannot be obtained because 
** there exists some other conflicting lock, return LSM_BUSY. If some other
** error occurs, return an LSM error code.
**
** Parameter iLock must be one of LSM_LOCK_WRITER, WORKER or CHECKPOINTER,
** or else a value returned by the LSM_LOCK_READER macro.
*/
int lsmShmLock(
  lsm_db *db, 
  int iLock,
  int eOp,                        /* One of LSM_LOCK_UNLOCK, SHARED or EXCL */
  int bBlock                      /* True for a blocking lock */
){
  lsm_db *pIter;
  const u64 me = ((u64)1 << (iLock-1));
  const u64 ms = ((u64)1 << (iLock+32-1));
  int rc = LSM_OK;
  Database *p = db->pDatabase;

  assert( eOp!=LSM_LOCK_EXCL || db->bReadonly==0 );
  assert( iLock>=1 && iLock<=LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1) );
  assert( LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1)<=32 );
  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );

  /* Check for a no-op. Proceed only if this is not one of those. */
  if( (eOp==LSM_LOCK_UNLOCK && (db->mLock & (me|ms))!=0)
   || (eOp==LSM_LOCK_SHARED && (db->mLock & (me|ms))!=ms)
   || (eOp==LSM_LOCK_EXCL   && (db->mLock & me)==0)
  ){

      case LSM_LOCK_SHARED:
        if( nExcl ){
          rc = LSM_BUSY;
        }else{
          if( nShared==0 ){
            rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_SHARED);
          }
          if( rc==LSM_OK ){
          db->mLock |= ms;
          db->mLock &= ~me;
        }
        }
        break;

      default:
        assert( eOp==LSM_LOCK_EXCL );
        if( nExcl || nShared ){
          rc = LSM_BUSY;
        }else{
          rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_EXCL);
          if( rc==LSM_OK ){
          db->mLock |= (me|ms);
        }
        }
        break;
    }

    lsmMutexLeave(db->pEnv, p->pClientMutex);
  }

  return rc;
}

#ifdef LSM_DEBUG

int shmLockType(lsm_db *db, int iLock){
  const u64 me = ((u64)1 << (iLock-1));
  const u64 ms = ((u64)1 << (iLock+32-1));

  if( db->mLock & me ) return LSM_LOCK_EXCL;
  if( db->mLock & ms ) return LSM_LOCK_SHARED;
  return LSM_LOCK_UNLOCK;
}

/*

}
#endif

void lsmShmBarrier(lsm_db *db){
  lsmEnvShmBarrier(db->pEnv);
}

int lsm_checkpoint(lsm_db *pDb, int *pnKB){
  int rc;                         /* Return code */
  u32 nWrite = 0;                 /* Number of pages checkpointed */

  /* Attempt the checkpoint. If successful, nWrite is set to the number of
  ** pages written between this and the previous checkpoint.  */
  rc = lsmCheckpointWrite(pDb, 0, &nWrite);

  /* If required, calculate the output variable (KB of data checkpointed). 
  ** Set it to zero if an error occured.  */
  if( pnKB ){
    int nKB = 0;
    if( rc==LSM_OK && nWrite ){
      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;
    }
    *pnKB = nKB;
  }

  return rc;
}

  pCsr->aTree = 0;
  pCsr->pSystemVal = 0;
  pCsr->apTreeCsr[0] = 0;
  pCsr->apTreeCsr[1] = 0;
  pCsr->pBtCsr = 0;
}

void lsmMCursorFreeCache(lsm_db *pDb){
  MultiCursor *p;
  MultiCursor *pNext;
  for(p=pDb->pCsrCache; p; p=pNext){
    pNext = p->pNext;
    lsmMCursorClose(p, 0);
  }
  pDb->pCsrCache = 0;
}

/*
** Close the cursor passed as the first argument.
**
** If the bCache parameter is true, then shift the cursor to the pCsrCache
** list for possible reuse instead of actually deleting it.
*/
void lsmMCursorClose(MultiCursor *pCsr, int bCache){
  if( pCsr ){
    lsm_db *pDb = pCsr->pDb;
    MultiCursor **pp;             /* Iterator variable */

    /* The cursor may or may not be currently part of the linked list 
    ** starting at lsm_db.pCsr. If it is, extract it.  */
    for(pp=&pDb->pCsr; *pp; pp=&((*pp)->pNext)){
      if( *pp==pCsr ){
        *pp = pCsr->pNext;
        break;
      }
    }

    if( bCache ){
      int i;                      /* Used to iterate through segment-pointers */

      /* Release any page references held by this cursor. */
      assert( !pCsr->pBtCsr );
      for(i=0; i<pCsr->nPtr; i++){
        SegmentPtr *pPtr = &pCsr->aPtr[i];
        lsmFsPageRelease(pPtr->pPg);
        pPtr->pPg = 0;
      }

      /* Reset the tree cursors */
      lsmTreeCursorReset(pCsr->apTreeCsr[0]);
      lsmTreeCursorReset(pCsr->apTreeCsr[1]);

      /* Add the cursor to the pCsrCache list */
      pCsr->pNext = pDb->pCsrCache;
      pDb->pCsrCache = pCsr;
    }else{
    /* Free the allocation used to cache the current key, if any. */
    sortedBlobFree(&pCsr->key);
    sortedBlobFree(&pCsr->val);

    /* Free the component cursors */
    mcursorFreeComponents(pCsr);

    /* Free the cursor structure itself */
    lsmFree(pDb->pEnv, pCsr);
  }
}
}

#define TREE_NONE 0
#define TREE_OLD  1
#define TREE_BOTH 2

/*
** Parameter eTree is one of TREE_OLD or TREE_BOTH.

  pCsr->flags |= CURSOR_FLUSH_FREELIST;
  pCsr->pSystemVal = lsmMallocRc(pCsr->pDb->pEnv, 4 + 8, &rc);
  return rc;
}

/*
** Allocate and return a new database cursor.
**
** This method should only be called to allocate user cursors. As it may
** recycle a cursor from lsm_db.pCsrCache.
*/
int lsmMCursorNew(
  lsm_db *pDb,                    /* Database handle */
  MultiCursor **ppCsr             /* OUT: Allocated cursor */
){
  MultiCursor *pCsr = 0;
  int rc = LSM_OK;

  if( pDb->pCsrCache ){
    int bOld;                     /* True if there is an old in-memory tree */

    /* Remove a cursor from the pCsrCache list and add it to the open list. */
    pCsr = pDb->pCsrCache;
    pDb->pCsrCache = pCsr->pNext;
    pCsr->pNext = pDb->pCsr;
    pDb->pCsr = pCsr;

    /* The cursor can almost be used as is, except that the old in-memory
    ** tree cursor may be present and not required, or required and not
    ** present. Fix this if required.  */
    bOld = (lsmTreeHasOld(pDb) && pDb->treehdr.iOldLog!=pDb->pClient->iLogOff);
    if( !bOld && pCsr->apTreeCsr[1] ){
      lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);
      pCsr->apTreeCsr[1] = 0;
    }else if( bOld && !pCsr->apTreeCsr[1] ){
      rc = lsmTreeCursorNew(pDb, 1, &pCsr->apTreeCsr[1]);
    }

  }else{
  pCsr = multiCursorNew(pDb, &rc);
  if( rc==LSM_OK ) rc = multiCursorInit(pCsr, pDb->pClient);
  }

  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
    pCsr = 0;
  }
  assert( (rc==LSM_OK)==(pCsr!=0) );
  *ppCsr = pCsr;
  return rc;
}

        iSnap = (i64)lsmGetU64((u8 *)pVal);
        if( x(pCtx, iBlk, iSnap) ) break;
        rc = multiCursorAdvance(pCsr, !bReverse);
      }
    }
  }

  lsmMCursorClose(pCsr, 0);
  if( pSnap!=pDb->pWorker ){
    lsmFreeSnapshot(pDb->pEnv, pSnap);
  }

  return rc;
}

        if( *ppVal ){
          memcpy(*ppVal, pVal, nVal);
          *pnVal = nVal;
        }
      }
    }

    lsmMCursorClose(pCsr, 0);
  }

  return rc;
}

static int multiCursorAllocTree(MultiCursor *pCsr){
  int rc = LSM_OK;

    }else{
      btreeCursorSplitkey(pCsr->pBtCsr, &pMerge->splitkey);
    }
    
    pMerge->iOutputOff = -1;
  }

  lsmMCursorClose(pCsr, 0);

  /* Persist and release the output page. */
  if( rc==LSM_OK ) rc = mergeWorkerPersistAndRelease(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerBtreeIndirect(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerFinishHierarchy(pMW);
  if( rc==LSM_OK ) rc = mergeWorkerAddPadding(pMW);
  lsmFsFlushWaiting(pMW->pDb->pFS, &rc);

    ** markers present in the in-memory tree.  */
    if( pNext==0 ){
      multiCursorIgnoreDelete(pCsr);
    }
  }

  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
  }else{
    Pgno iLeftPtr = 0;
    Merge merge;                  /* Merge object used to create new level */
    MergeWorker mergeworker;      /* MergeWorker object for the same purpose */

    memset(&merge, 0, sizeof(Merge));
    memset(&mergeworker, 0, sizeof(MergeWorker));

  rc = lsmFsMoveBlock(pDb->pFS, &pLvl->lhs, iTo, iFrom);
  if( rc==LSM_OK ){
    if( p->redirect.a==0 ){
      int nByte = sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS;
      p->redirect.a = lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
    }
    if( rc==LSM_OK ){

      /* Check if the block just moved was already redirected. */
      int i;
      for(i=0; i<p->redirect.n; i++){
        if( p->redirect.a[i].iTo==iFrom ) break;
      }

      if( i==p->redirect.n ){
        /* Block iFrom was not already redirected. Add a new array entry. */
      memmove(&p->redirect.a[1], &p->redirect.a[0], 
          sizeof(struct RedirectEntry) * p->redirect.n
      );
      p->redirect.a[0].iFrom = iFrom;
      p->redirect.a[0].iTo = iTo;
      p->redirect.n++;
      }else{
        /* Block iFrom was already redirected. Overwrite existing entry. */
        p->redirect.a[i].iTo = iTo;
      }

      rc = lsmBlockFree(pDb, iFrom);

      *pnWrite = lsmFsBlockSize(pDb->pFS) / lsmFsPageSize(pDb->pFS);
      pLvl->lhs.pRedirect = &p->redirect;
    }
  }

static int doLsmWork(lsm_db *pDb, int nMerge, int nPage, int *pnWrite){
  int rc = LSM_OK;                /* Return code */
  int nWrite = 0;                 /* Number of pages written */

  assert( nMerge>=1 );

  if( nPage!=0 ){
    int bCkpt = 0;
    do {
      int nThis = 0;
      int nReq = (nPage>=0) ? (nPage-nWrite) : ((int)0x7FFFFFFF);

      bCkpt = 0;
      rc = doLsmSingleWork(pDb, 0, nMerge, nReq, &nThis, &bCkpt);
      nWrite += nThis;
      if( rc==LSM_OK && bCkpt ){
        rc = lsm_checkpoint(pDb, 0);
      }
    }while( rc==LSM_OK && bCkpt && (nWrite<nPage || nPage<0) );
  }

  if( pnWrite ){
    if( rc==LSM_OK ){
      *pnWrite = nWrite;
    }else{
      *pnWrite = 0;
    }
  }
  return rc;
}

/*
** Perform work to merge database segments together.
*/
int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite){
  int rc;                         /* Return code */
  int nPgsz;                      /* Nominal page size in bytes */
  int nPage;                      /* Equivalent of nKB in pages */
  int nWrite = 0;                 /* Number of pages written */

  /* This function may not be called if pDb has an open read or write
  ** transaction. Return LSM_MISUSE if an application attempts this.  */
  if( pDb->nTransOpen || pDb->pCsr ) return LSM_MISUSE_BKPT;
  if( nMerge<=0 ) nMerge = pDb->nMerge;

  /* Convert from KB to pages */
  nPgsz = lsmFsPageSize(pDb->pFS);
  if( nKB>=0 ){
    nPage = ((i64)nKB * 1024 + nPgsz - 1) / nPgsz;
  }else{
    nPage = -1;
  }

  rc = doLsmWork(pDb, nMerge, nPage, &nWrite);
  
  if( pnWrite ){
    /* Convert back from pages to KB */
    *pnWrite = (int)(((i64)nWrite * 1024 + nPgsz - 1) / nPgsz);
  }
  return rc;
}

int lsm_flush(lsm_db *db){
  int rc;

  if( db->nTransOpen>0 || db->pCsr ){
    rc = LSM_MISUSE_BKPT;

    int nRemaining;               /* Units of work to do before returning */

    nRemaining = nUnit * nDepth;
#ifdef LSM_LOG_WORK
    lsmLogMessage(pDb, rc, "lsmSortedAutoWork(): %d*%d = %d pages", 
        nUnit, nDepth, nRemaining);
#endif
    assert( nRemaining>=0 );
    rc = doLsmWork(pDb, pDb->nMerge, nRemaining, 0);
    if( rc==LSM_BUSY ) rc = LSM_OK;

    if( bRestore && pDb->pCsr ){
      lsmMCursorFreeCache(pDb);
      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
      pDb->pClient = 0;
      rc = lsmCheckpointLoad(pDb, 0);
      if( rc==LSM_OK ){
        rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot, &pDb->pClient);
      }
      if( rc==LSM_OK ){

** discarded.
*/
static void intArrayTruncate(IntArray *p, int nVal){
  p->nArray = nVal;
}
/* End of IntArray methods.
***********************************************************************/

static int treeKeycmp(void *p1, int n1, void *p2, int n2){
  int res;
  res = memcmp(p1, p2, LSM_MIN(n1, n2));
  if( res==0 ) res = (n1-n2);
  return res;
}

/*
** The pointer passed as the first argument points to an interior node,
** not a leaf. This function returns the offset of the iCell'th child
** sub-tree of the node.
*/
static u32 getChildPtr(TreeNode *p, int iVersion, int iCell){
  assert( iCell>=0 && iCell<=array_size(p->aiChildPtr) );
  if( p->iV2 && p->iV2<=iVersion && iCell==p->iV2Child ) return p->iV2Ptr;
  return p->aiChildPtr[iCell];
}

/*
** Given an offset within the *-shm file, return the associated chunk number.
*/
static int treeOffsetToChunk(u32 iOff){
  assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
  return (int)(iOff>>15);
}

#define treeShmptrUnsafe(pDb, iPtr) \
(&((u8*)((pDb)->apShm[(iPtr)>>15]))[(iPtr) & (LSM_SHM_CHUNK_SIZE-1)])

/*
** Return a pointer to the mapped memory location associated with *-shm 
** file offset iPtr.
*/
static void *treeShmptr(lsm_db *pDb, u32 iPtr){

  assert( (iPtr>>15)<pDb->nShm );
  assert( pDb->apShm[iPtr>>15] );

  return iPtr ? treeShmptrUnsafe(pDb, iPtr) : 0;
}

static ShmChunk * treeShmChunk(lsm_db *pDb, int iChunk){
  return (ShmChunk *)(pDb->apShm[iChunk]);
}

static ShmChunk * treeShmChunkRc(lsm_db *pDb, int iChunk, int *pRc){

}

/*
** Return a pointer to the mapping of the TreeKey object that the cursor
** is pointing to. 
*/
static TreeKey *csrGetKey(TreeCursor *pCsr, TreeBlob *pBlob, int *pRc){
  TreeKey *pRet;
  lsm_db *pDb = pCsr->pDb;
  u32 iPtr = pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]];

  assert( iPtr );
  pRet = treeShmptrUnsafe(pDb, iPtr);
  if( !(pRet->flags & LSM_CONTIGUOUS) ){
    pRet = treeShmkey(pDb, iPtr, TKV_LOADVAL, pBlob, pRc);
  }


  return pRet;
}

/*
** Save the current position of tree cursor pCsr.
*/
int lsmTreeCursorSave(TreeCursor *pCsr){

  u32 *piPtr, 
  void *pKey, int nKey,           /* Key data */
  void *pVal, int nVal,           /* Value data (or nVal<0 for delete) */
  int *pRc
){
  TreeKey *p;
  u32 iPtr;
  u32 iEnd;
  int nRem;
  u8 *a;
  int n;

  /* Allocate space for the TreeKey structure itself */
  *piPtr = iPtr = treeShmalloc(pDb, 1, sizeof(TreeKey), pRc);
  p = treeShmptr(pDb, iPtr);

      memcpy(aAlloc, &a[n-nRem], nAlloc);
      nRem -= nAlloc;
    }
    a = pVal;
    n = nRem = nVal;
    pVal = 0;
  }

  iEnd = iPtr + sizeof(TreeKey) + nKey + LSM_MAX(0, nVal);
  if( (iPtr & ~(LSM_SHM_CHUNK_SIZE-1))!=(iEnd & ~(LSM_SHM_CHUNK_SIZE-1)) ){
    p->flags = 0;
  }else{
    p->flags = LSM_CONTIGUOUS;
  }

  if( *pRc ) return 0;
#if 0
  printf("store: %d %s\n", (int)iPtr, (char *)pKey);
#endif
  return p;
}

    }
  }

  return rc;
}

void lsmTreeMakeOld(lsm_db *pDb){

  /* A write transaction must be open. Otherwise the code below that
  ** assumes (pDb->pClient->iLogOff) is current may malfunction. 
  **
  ** Update: currently this assert fails due to lsm_flush(), which does
  ** not set nTransOpen.
  */
  assert( /* pDb->nTransOpen>0 && */ pDb->iReader>=0 );

  if( pDb->treehdr.iOldShmid==0 ){
    pDb->treehdr.iOldLog = (pDb->treehdr.log.aRegion[2].iEnd << 1);
    pDb->treehdr.iOldLog |= (~(pDb->pClient->iLogOff) & (i64)0x0001);

    pDb->treehdr.oldcksum0 = pDb->treehdr.log.cksum0;
    pDb->treehdr.oldcksum1 = pDb->treehdr.log.cksum1;
    pDb->treehdr.iOldShmid = pDb->treehdr.iNextShmid-1;
    memcpy(&pDb->treehdr.oldroot, &pDb->treehdr.root, sizeof(TreeRoot));

    pDb->treehdr.root.iTransId = 1;
    pDb->treehdr.root.iRoot = 0;

** is initialized here - it will be copied into shared memory if log file
** recovery is successful.
*/
int lsmTreeInit(lsm_db *pDb){
  ShmChunk *pOne;
  int rc = LSM_OK;

  memset(&pDb->treehdr, 0, sizeof(TreeHeader));
  pDb->treehdr.root.iTransId = 1;
  pDb->treehdr.iFirst = 1;
  pDb->treehdr.nChunk = 2;
  pDb->treehdr.iWrite = LSM_SHM_CHUNK_SIZE + LSM_SHM_CHUNK_HDR;
  pDb->treehdr.iNextShmid = 2;
  pDb->treehdr.iUsedShmid = 1;

  int res;                        /* Result of seek operation on csr */

  assert( nVal>=0 || pVal==0 );
  assert_tree_looks_ok(LSM_OK, pTree);
  assert( flags==LSM_INSERT       || flags==LSM_POINT_DELETE 
       || flags==LSM_START_DELETE || flags==LSM_END_DELETE 
  );
  assert( (flags & LSM_CONTIGUOUS)==0 );
#if 0
  dump_tree_contents(pDb, "before");
#endif

  if( p->iRoot ){
    TreeKey *pRes;                /* Key at end of seek operation */
    treeCursorInit(pDb, 0, &csr);

  }else{
    memset(&csr, 0, sizeof(TreeCursor));
  }

  /* Allocate and populate a new key-value pair structure */
  pTreeKey = newTreeKey(pDb, &iTreeKey, pKey, nKey, pVal, nVal, &rc);
  if( rc!=LSM_OK ) return rc;
  assert( pTreeKey->flags==0 || pTreeKey->flags==LSM_CONTIGUOUS );
  pTreeKey->flags |= flags;

  if( p->iRoot==0 ){
    /* The tree is completely empty. Add a new root node and install
    ** (pKey/nKey) as the middle entry. Even though it is a leaf at the
    ** moment, use newTreeNode() to allocate the node (i.e. allocate enough
    ** space for the fields used by interior nodes). This is because the
    ** treeInsert() routine may convert this node to an interior node. */

){
  int rc = LSM_OK;
  int bDone = 0;
  TreeRoot *p = &db->treehdr.root;
  TreeBlob blob = {0, 0};

  /* The range must be sensible - that (key1 < key2). */
  assert( treeKeycmp(pKey1, nKey1, pKey2, nKey2)<0 );
  assert( assert_delete_ranges_match(db) );

#if 0
  static int nCall = 0;
  printf("\n");
  nCall++;
  printf("%d delete %s .. %s\n", nCall, (char *)pKey1, (char *)pKey2);

    /* If there is no such entry, or if it is greater than pKey2, then the
    ** tree now contains no keys in the range being deleted. In this case
    ** break out of the loop.  */
    bDone = 1;
    if( lsmTreeCursorValid(&csr) ){
      lsmTreeCursorKey(&csr, 0, &pDel, &nDel);
      if( treeKeycmp(pDel, nDel, pKey2, nKey2)<0 ) bDone = 0;
    }

    if( bDone==0 ){
      if( csr.iNode==(p->nHeight-1) ){
        /* The element to delete already lies on a leaf node */
        rc = treeDeleteEntry(db, &csr, 0);
      }else{

static int treeCsrCompare(TreeCursor *pCsr, void *pKey, int nKey){
  TreeKey *p;
  int cmp = 0;
  int rc = LSM_OK;
  assert( pCsr->iNode>=0 );
  p = csrGetKey(pCsr, &pCsr->blob, &rc);
  if( p ){
    cmp = treeKeycmp(TKV_KEY(p), p->nKey, pKey, nKey);
  }
  return cmp;
}
#endif


/*

**
**   * If the tree is empty, leave the cursor at EOF and set *pRes to -1.
*/
int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes){
  int rc = LSM_OK;                /* Return code */
  lsm_db *pDb = pCsr->pDb;
  TreeRoot *pRoot = pCsr->pRoot;


  u32 iNodePtr;                   /* Location of current node in search */

  /* Discard any saved position data */
  treeCursorRestore(pCsr, 0);

  iNodePtr = pRoot->iRoot;
  if( iNodePtr==0 ){
    /* Either an error occurred or the tree is completely empty. */
    assert( rc!=LSM_OK || pRoot->iRoot==0 );
    *pRes = -1;
    pCsr->iNode = -1;
  }else{
    TreeBlob b = {0, 0};
    int res = 0;                  /* Result of comparison function */
    int iNode = -1;
    while( iNodePtr ){
      TreeNode *pNode;            /* Node at location iNodePtr */
      int iTest;                  /* Index of second key to test (0 or 2) */
      u32 iTreeKey;
      TreeKey *pTreeKey;          /* Key to compare against */

      pNode = (TreeNode *)treeShmptrUnsafe(pDb, iNodePtr);
      iNode++;
      pCsr->apTreeNode[iNode] = pNode;

      /* Compare (pKey/nKey) with the key in the middle slot of B-tree node
      ** pNode. The middle slot is never empty. If the comparison is a match,
      ** then the search is finished. Break out of the loop. */
      pTreeKey = treeShmptrUnsafe(pDb, pNode->aiKeyPtr[1]);
      if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
      pTreeKey = treeShmkey(pDb, pNode->aiKeyPtr[1], TKV_LOADKEY, &b, &rc);
      if( rc!=LSM_OK ) break;
      }
      res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
      if( res==0 ){
        pCsr->aiCell[iNode] = 1;
        break;
      }

      /* Based on the results of the previous comparison, compare (pKey/nKey)
      ** to either the left or right key of the B-tree node, if such a key
      ** exists. */
      iTest = (res>0 ? 0 : 2);
      iTreeKey = pNode->aiKeyPtr[iTest];
      if( iTreeKey ){
        pTreeKey = treeShmptrUnsafe(pDb, iTreeKey);
        if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
          pTreeKey = treeShmkey(pDb, iTreeKey, TKV_LOADKEY, &b, &rc);
      if( rc ) break;



        }
        res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
        if( res==0 ){
          pCsr->aiCell[iNode] = iTest;
          break;
        }
      }else{
        iTest = 1;
      }

      if( iNode<(pRoot->nHeight-1) ){
        iNodePtr = getChildPtr(pNode, pRoot->iTransId, iTest + (res<0));
      }else{
        iNodePtr = 0;
      }

      if( iCell<3 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
    }
  }

#ifndef NDEBUG
  if( pCsr->iNode>=0 ){
    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
    assert( rc||treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)>=0 );
  }
  tblobFree(pDb, &key1);
#endif

  return rc;
}

    }while( (--pCsr->iNode)>=0 );
    pCsr->aiCell[pCsr->iNode] = iCell;
  }

#ifndef NDEBUG
  if( pCsr->iNode>=0 ){
    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
    assert( rc || treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)<0 );
  }
  tblobFree(pDb, &key1);
#endif

  return rc;
}

  return rc;
}

int lsmTreeCursorFlags(TreeCursor *pCsr){
  int flags = 0;
  if( pCsr && pCsr->iNode>=0 ){
    int rc = LSM_OK;
    TreeKey *pKey = (TreeKey *)treeShmptrUnsafe(pCsr->pDb,
        pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]]
    );
    assert( rc==LSM_OK );
    flags = (pKey->flags & ~LSM_CONTIGUOUS);
  }
  return flags;
}

int lsmTreeCursorKey(TreeCursor *pCsr, int *pFlags, void **ppKey, int *pnKey){
  TreeKey *pTreeKey;
  int rc = LSM_OK;

    if( treeHeaderChecksumOk(&pDb->treehdr) ){
      if( piRead ) *piRead = 2;
      return LSM_OK;
    }

    lsmShmBarrier(pDb);
  }
  return LSM_PROTOCOL_BKPT;
}

int lsmTreeLoadHeaderOk(lsm_db *pDb, int iRead){
  TreeHeader *p = (iRead==1) ? &pDb->pShmhdr->hdr1 : &pDb->pShmhdr->hdr2;
  assert( iRead==1 || iRead==2 );
  return (0==memcmp(pDb->treehdr.aCksum, p->aCksum, sizeof(u32)*2));
}

#include <unistd.h>
#include <errno.h>

#include <sys/mman.h>
#include "lsmInt.h"

/* There is no fdatasync() call on Android */
#ifdef __ANDROID__
# define fdatasync(x) fsync(x)
#endif

/*
** An open file is an instance of the following object
*/
typedef struct PosixFile PosixFile;
struct PosixFile {
  lsm_env *pEnv;                  /* The run-time environment */
  const char *zName;              /* Full path to file */
  int fd;                         /* The open file descriptor */
  int shmfd;                      /* Shared memory file-descriptor */
  void *pMap;                     /* Pointer to mapping of file fd */
  off_t nMap;                     /* Size of mapping at pMap in bytes */
  int nShm;                       /* Number of entries in array apShm[] */
  void **apShm;                   /* Array of 32K shared memory segments */
};



static char *posixShmFile(PosixFile *p){
  char *zShm;
  int nName = strlen(p->zName);
  zShm = (char *)lsmMalloc(p->pEnv, nName+4+1);
  if( zShm ){
    memcpy(zShm, p->zName, nName);
    memcpy(&zShm[nName], "-shm", 5);
  }
  return zShm;
}

static int lsmPosixOsOpen(
  lsm_env *pEnv,
  const char *zFile, 
  int flags,
  lsm_file **ppFile
){
  int rc = LSM_OK;
  PosixFile *p;

  p = lsm_malloc(pEnv, sizeof(PosixFile));
  if( p==0 ){
    rc = LSM_NOMEM;
  }else{
    int bReadonly = (flags & LSM_OPEN_READONLY);
    int oflags = (bReadonly ? O_RDONLY : (O_RDWR|O_CREAT));
    memset(p, 0, sizeof(PosixFile));
    p->zName = zFile;
    p->pEnv = pEnv;
    p->fd = open(zFile, oflags, 0644);
    if( p->fd<0 ){
      lsm_free(pEnv, p);
      p = 0;
      if( errno==ENOENT ){
        rc = lsmErrorBkpt(LSM_IOERR_NOENT);
      }else{
        rc = LSM_IOERR_BKPT;
      }
    }
  }

  *ppFile = (lsm_file *)p;
  return rc;
}

static int lsmPosixOsWrite(
  lsm_file *pFile,                /* File to write to */
  lsm_i64 iOff,                   /* Offset to write to */
  void *pData,                    /* Write data from this buffer */
  int nData                       /* Bytes of data to write */
){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  off_t offset;

  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
  if( offset!=iOff ){
    rc = LSM_IOERR_BKPT;
  }else{
    ssize_t prc = write(p->fd, pData, (size_t)nData);
    if( prc<0 ) rc = LSM_IOERR_BKPT;
  }

  return rc;
}

static int lsmPosixOsTruncate(
  lsm_file *pFile,                /* File to write to */
  lsm_i64 nSize                   /* Size to truncate file to */
){
  PosixFile *p = (PosixFile *)pFile;
  int rc = LSM_OK;                /* Return code */
  int prc;                        /* Posix Return Code */

  struct stat sStat;              /* Result of fstat() invocation */

  prc = fstat(p->fd, &sStat);
  if( prc==0 && sStat.st_size>nSize ){
  prc = ftruncate(p->fd, (off_t)nSize);
  }
  if( prc<0 ) rc = LSM_IOERR_BKPT;

  return rc;
}

static int lsmPosixOsRead(
  lsm_file *pFile,                /* File to read from */
  lsm_i64 iOff,                   /* Offset to read from */
  void *pData,                    /* Read data into this buffer */
  int nData                       /* Bytes of data to read */
){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  off_t offset;

  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
  if( offset!=iOff ){
    rc = LSM_IOERR_BKPT;
  }else{
    ssize_t prc = read(p->fd, pData, (size_t)nData);
    if( prc<0 ){ 
      rc = LSM_IOERR_BKPT;
    }else if( prc<nData ){
      memset(&((u8 *)pData)[prc], 0, nData - prc);
    }

  }

  return rc;
}

static int lsmPosixOsSync(lsm_file *pFile){
  int rc = LSM_OK;

#ifndef LSM_NO_SYNC
  PosixFile *p = (PosixFile *)pFile;
  int prc = 0;

  if( p->pMap ){
    prc = msync(p->pMap, p->nMap, MS_SYNC);
  }
  if( prc==0 ) prc = fdatasync(p->fd);
  if( prc<0 ) rc = LSM_IOERR_BKPT;
#else
  (void)pFile;
#endif

  return rc;
}

  if( p->pMap ){
    munmap(p->pMap, p->nMap);
    *ppOut = p->pMap = 0;
    *pnOut = p->nMap = 0;
  }

  if( iMin>=0 ){
  memset(&buf, 0, sizeof(buf));
  prc = fstat(p->fd, &buf);
  if( prc!=0 ) return LSM_IOERR_BKPT;
  iSz = buf.st_size;
  if( iSz<iMin ){
    iSz = ((iMin + (2<<20) - 1) / (2<<20)) * (2<<20);
    prc = ftruncate(p->fd, iSz);
    if( prc!=0 ) return LSM_IOERR_BKPT;
  }

  p->pMap = mmap(0, iSz, PROT_READ|PROT_WRITE, MAP_SHARED, p->fd, 0);
  p->nMap = iSz;
  }

  *ppOut = p->pMap;
  *pnOut = p->nMap;
  return LSM_OK;
}

static int lsmPosixOsFullpath(

  static const short aType[3] = { F_UNLCK, F_RDLCK, F_WRLCK };
  struct flock lock;

  assert( aType[LSM_LOCK_UNLOCK]==F_UNLCK );
  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
  assert( eType>=0 && eType<array_size(aType) );
  assert( iLock>0 && iLock<=32 );

  memset(&lock, 0, sizeof(lock));
  lock.l_whence = SEEK_SET;
  lock.l_len = 1;
  lock.l_type = aType[eType];
  lock.l_start = (4096-iLock);

  if( fcntl(p->fd, F_SETLK, &lock) ){
    int e = errno;
    if( e==EACCES || e==EAGAIN ){
      rc = LSM_BUSY;
    }else{
      rc = LSM_IOERR_BKPT;
    }
  }

  return rc;
}

int lsmPosixOsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  static const short aType[3] = { 0, F_RDLCK, F_WRLCK };
  struct flock lock;

  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
  assert( eType>=0 && eType<array_size(aType) );
  assert( iLock>0 && iLock<=32 );

  memset(&lock, 0, sizeof(lock));
  lock.l_whence = SEEK_SET;
  lock.l_len = nLock;
  lock.l_type = aType[eType];
  lock.l_start = (4096-iLock);

  if( fcntl(p->fd, F_GETLK, &lock) ){
    rc = LSM_IOERR_BKPT;
  }else if( lock.l_type!=F_UNLCK ){
    rc = LSM_BUSY;
  }

  return rc;
}

int lsmPosixOsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  PosixFile *p = (PosixFile *)pFile;

    p->nShm = nNew;
  }

  if( p->apShm[iChunk]==0 ){
    p->apShm[iChunk] = mmap(0, LSM_SHM_CHUNK_SIZE, 
        PROT_READ|PROT_WRITE, MAP_SHARED, p->shmfd, iChunk*LSM_SHM_CHUNK_SIZE
    );
    if( p->apShm[iChunk]==0 ) return LSM_IOERR_BKPT;
  }

  *ppShm = p->apShm[iChunk];
  return LSM_OK;
}

void lsmPosixOsShmBarrier(void){

   close(p->fd);
   lsm_free(p->pEnv, p->apShm);
   lsm_free(p->pEnv, p);
   return LSM_OK;
}

static int lsmPosixOsSleep(lsm_env *pEnv, int us){
#if 0
  /* Apparently on Android usleep() returns void */
  if( usleep(us) ) return LSM_IOERR;

#endif
  usleep(us);
  return LSM_OK;
}

/****************************************************************************
** Memory allocation routines.
*/
#define BLOCK_HDR_SIZE ROUND8( sizeof(sqlite4_size_t) )

#endif
    lsmPosixOsMap,           /* xMap */
    lsmPosixOsUnmap,         /* xUnmap */
    lsmPosixOsFileid,        /* xFileid */
    lsmPosixOsClose,         /* xClose */
    lsmPosixOsUnlink,        /* xUnlink */
    lsmPosixOsLock,          /* xLock */
    lsmPosixOsTestLock,      /* xTestLock */
    lsmPosixOsShmMap,        /* xShmMap */
    lsmPosixOsShmBarrier,    /* xShmBarrier */
    lsmPosixOsShmUnmap,      /* xShmUnmap */
    /***** memory allocation *********/
    0,                       /* pMemCtx */
    lsmPosixOsMalloc,        /* xMalloc */
    lsmPosixOsRealloc,       /* xRealloc */

#ifdef SQLITE4_ENABLE_ICU
# include "sqliteicu.h"
#endif

/*
** Dummy function used as a unique symbol for SQLITE4_DYNAMIC
*/
void sqlite4_dynamic(void *pArg,void *p){ (void)pArg; (void)p; }








/* IMPLEMENTATION-OF: R-53536-42575 The sqlite4_libversion() function returns
** a pointer to the to the sqlite4_version[] string constant. 
*/
const char *sqlite4_libversion(void){ return SQLITE4_VERSION; }

/* IMPLEMENTATION-OF: R-63124-39300 The sqlite4_sourceid() function returns a

** SQLITE4_ENABLE_IOTRACE is enabled, then messages describing
** I/O active are written using this function.  These messages
** are intended for debugging activity only.
*/
void (*sqlite4IoTrace)(const char*, ...) = 0;
#endif

















































































































































































































































































































































































/*
** Set up the lookaside buffers for a database connection.
** Return SQLITE4_OK on success.  
** If lookaside is already active, return SQLITE4_BUSY.
**
** The sz parameter is the number of bytes in each lookaside slot.
** The cnt parameter is the number of slots.  If pStart is NULL the

    for(i=0; i<nIn; i++){
      aOut[i] = sqlite4_tolower(aIn[i]);
    }
  }
  return nIn;
}








/*
** Return the number of changes in the most recent call to sqlite4_exec().
*/
int sqlite4_changes(sqlite4 *db){
  return db->nChange;
}

  sqlite4_mutex_enter(db->mutex);
  if( db->mallocFailed ){
    z = (void *)outOfMem;
  }else{
    z = sqlite4_value_text16(db->pErr);
    if( z==0 ){
      sqlite4ValueSetStr(db->pErr, -1, sqlite4ErrStr(db->errCode),
           SQLITE4_UTF8, SQLITE4_STATIC, 0);
      z = sqlite4_value_text16(db->pErr);
    }
    /* A malloc() may have failed within the call to sqlite4_value_text16()
    ** above. If this is the case, then the db->mallocFailed flag needs to
    ** be cleared before returning. Do this directly, instead of via
    ** sqlite4ApiExit(), to avoid setting the database handle error message.
    */

  db->xCollNeeded16 = xCollNeeded16;
  db->pCollNeededArg = pCollNeededArg;
  sqlite4_mutex_leave(db->mutex);
  return SQLITE4_OK;
}
#endif /* SQLITE4_OMIT_UTF16 */











/*
** Test to see whether or not the database connection is in autocommit
** mode.  Return TRUE if it is and FALSE if not.  Autocommit mode is on
** by default.  Autocommit is disabled by a BEGIN statement and reenabled
** by the next COMMIT or ROLLBACK.


*/
int sqlite4_get_autocommit(sqlite4 *db){
  return (db->pSavepoint==0);
}

/*
** The following routines are subtitutes for constants SQLITE4_CORRUPT,

  r.sign = A.sign ^ B.sign;
  r.approx = A.approx | B.approx;
  if( r.approx==0 && A.m%B.m!=0 ) r.approx = 1;
  r.m = A.m/B.m;
  r.e = A.e - B.e;
  return r;
}

/*
** Test if A is infinite.
*/
int sqlite4_num_isinf(sqlite4_num A){
  return A.e>SQLITE4_MX_EXP && A.m!=0;
}

/*
** Test if A is NaN.
*/
int sqlite4_num_isnan(sqlite4_num A){
  return A.e>SQLITE4_MX_EXP && A.m==0; 
}

/*
** Compare numbers A and B.  Return:
**
**    1     if A<B
**    2     if A==B
**    3     if A>B

    return A.sign ? 1 : 3;
  }
  if( B.e>SQLITE4_MX_EXP ){
    if( B.m==0 ) return 0;
    return B.sign ? 3 : 1;
  }
  if( A.sign!=B.sign ){
    if ( A.m==0 && B.m==0 ) return 2;
    return A.sign ? 1 : 3;
  }
  adjustExponent(&A, &B);
  if( A.sign ){
    sqlite4_num t = A;
    A = B;
    B = t;

    i = incr;
  }else if( zIn[0]=='+' ){
    i = incr;
  }else{
    i = 0;
  }
  if( nIn<=0 ) goto not_a_valid_number;
  if( nIn>=incr*3
   && ((c=zIn[i])=='i' || c=='I')
   && ((c=zIn[i+incr])=='n' || c=='N')
   && ((c=zIn[i+incr*2])=='f' || c=='F')
  ){
    r.e = SQLITE4_MX_EXP+1;
    r.m = nIn<=i+incr*3 || zIn[i+incr*3]==0;
    return r;
  }
  while( i<nIn && (c = zIn[i])!=0 ){
    i += incr;
    if( c>='0' && c<='9' ){
      if( c=='0' && nDigit==0 ){
        if( seenRadix && r.e > -(SQLITE4_MX_EXP+1000) ) r.e--;
        continue;
      }
      nDigit++;
      if( nDigit<=18 ){
        r.m = (r.m*10) + c - '0';
        if( seenRadix ) r.e--;
      }else{
        if( c!='0' ) r.approx = 1;
        if( !seenRadix ) r.e++;
      }
    }else if( c=='.' ){
      seenRadix = 1;




    }else if( c=='e' || c=='E' ){
    int exp = 0;
    int expsign = 0;
    int nEDigit = 0;
    if( zIn[i]=='-' ){
      expsign = 1;
      i += incr;
    }else if( zIn[i]=='+' ){
      i += incr;
    }
    if( i>=nIn ) goto not_a_valid_number;
    while( i<nIn && (c = zIn[i])!=0  ){
      i += incr;
        if( c<'0' || c>'9' ) goto not_a_valid_number;
      if( c=='0' && nEDigit==0 ) continue;
      nEDigit++;
      if( nEDigit>3 ) goto not_a_valid_number;
      exp = exp*10 + c - '0';
    }
    if( expsign ) exp = -exp;
    r.e += exp;
      break;
    }else{
      goto not_a_valid_number;
    }
  }
  return r;
  
not_a_valid_number:
  r.e = SQLITE4_MX_EXP+1;
  r.m = 0;
  return r;  
}

/*
** Convert an sqlite4_int64 to a number and return that number.
*/
sqlite4_num sqlite4_num_from_int64(sqlite4_int64 n){
  sqlite4_num r;
  r.approx = 0;
  r.e = 0;
  r.sign = n < 0;
  if( n>=0 ){
    r.m = n;
  }else if( n!=SMALLEST_INT64 ){
    r.m = -n;
  }else{
    r.m = 1+(u64)LARGEST_INT64;
  }
  return r;
}

/*
** Convert an integer into text in the buffer supplied. The
** text is zero-terminated and right-justified in the buffer.
** A pointer to the first character of text is returned.
**
** The buffer needs to be at least 21 bytes in length.