SQLite

The SQL database used for ACD contains 113 tables and indices implemented
in GDBM.  The following are statistics on the sizes of keys and data
within these tables and indices.

Entries:      962080
Size:         45573853
Avg Size:     48
Key Size:     11045299
Avg Key Size: 12
Max Key Size: 99

    0..8            266    0%
    9..12          5485    0%
   13..16         73633    8%
   17..24        180918   27%
   25..32        209823   48%
   33..40        148995   64%
   41..48         76304   72%
   49..56         14346   73%
   57..64         15725   75%
   65..80         44916   80%
   81..96        127815   93%
   97..112        34769   96%
  113..128        13314   98%
  129..144         8098   99%
  145..160         3355   99%
  161..176         1159   99%
  177..192          629   99%
  193..208          221   99%
  209..224          210   99%
  225..240          129   99%
  241..256           57   99%
  257..288          496   99%
  289..320           60   99%
  321..352           37   99%
  353..384           46   99%
  385..416           22   99%
  417..448           24   99%
  449..480           26   99%
  481..512           27   99%
  513..1024         471   99%
 1025..2048         389   99%
 2049..4096         182   99%
 4097..8192          74   99%
 8193..16384         34   99%
16385..32768         17   99%
32769..65536          5   99%
65537..131073         3  100%

      - Able to delete without disturbing scan order
      - Now keeps a count of number of table entries
         + Special processing for count(*)
         + Better selection of indices on a select
      - Transactions
  *  Modify sqlite_master to store the table number.
  *  Add a cache in DbCursor to speed up the sqliteDbReadOvfl() routine.
  *  Add cache information to speed up sqliteDbCursorMoveTo().

Longer term:
  *  Document all the changes and release Sqlite 2.0.
  *  Techniques for optimizing querys by grouping data with similar
     indices.
  *  "OPTIMIZE select" statement to automatically create and/or tune
     indices.
  *  "CREATE INDEX FOR select" to automatically generate needed indices.
  *  "VACUUM table USING index".
  *  Parse and use constraints.

** Boston, MA  02111-1307, USA.
**
** Author contact information:
**   drh@hwaci.com
**   http://www.hwaci.com/drh/
**
*************************************************************************
** $Id: db.c,v 1.3 2001/01/22 00:31:53 drh Exp $
*/
#include "sqliteInt.h"
#include "pg.h"

/*
** Everything we need to know about an open database
*/

** Leaf blocks:
**
**     0.   BLOCK_MAGIC | BLOCK_LEAF 
**     1.   number of table entries  (only used if a table root block)
**     entries....
**         0.  size of this entry (measured in u32's)
**         1.  hash
**         2.  keysize  (in bytes)
**         3.  datasize (in bytes)
**         4.  payload

**
** Payload area:
**
**     *   up to LOCAL_PAYLOAD bytes of data
**     *   10 page number of direct blocks
**     *   1 indirect block
**     *   1 double-indirect block
**
** Index block:
**
**     0.   BLOCK_MAGIC | BLOCK_INDEX
**     1.   number of table entries  (only used if a table root block)
**     2.   entries in this index block
**     entries...
**         0.  largest hash value for pgno
**         1.  pgno of subblock
**
** Contents block:  (The first page in the file)
**     0.   BLOCK_MAGIC | BLOCK_CONTENTS
**     1.   zero
**     2.   number of bytes of payload
**     3.   freelist
**     4... root pages numbers of tables
*/

#define U32_PER_PAGE  (SQLITE_PAGE_SIZE/sizeof(u32))
#deifne LOCAL_PAYLOAD  (SQLITE_PAGE_SIZE - 18*sizeof(u32))

/*
** Byte swapping code.
*/
#ifdef BIG_ENDIAN
# SWB(x) (x)
#else
# SWB(x) sqliteDbSwapBytes(x)

  d[1] = s[2];
  d[2] = s[1];
  d[3] = s[0];
  return r;
}

#endif

/*
** Return the number of bytes of payload storage required on the leaf
** node to hold the key and data given.  Overflow pages do not count.
** The argument is the total size of the payload.
*/
static int payloadLocalSize(int nTotal){
  int nLocal, i;
  if( nTotal<0 ) nTotal = 0;
  if( nTotal <= LOCAL_PAYLOAD ){
    /* All the data fits on the leaf page */
    return (nTotal + 3)/4;
  }
  nLocal = LOCAL_PAYLOAD;
  nTotal -= LOCAL_PAYLOAD;
  if( nTotal < 10*SQLITE_PAGE_SIZE ){
    return nLocal + ((nTotal+SQLITE_PAGE_SIZE-1)/SQLITE_PAGE_SIZE)*sizeof(u32);
  }
  nLocal += 10*sizeof(u32);
  nTotal -= 10*SQLITE_PAGE_SIZE;
  if( nTotal < U32_PER_PAGE*SQLITE_PAGE_SIZE ){
    return nLocal + sizeof(u32);
  }
  nLocal += sizeof(u32);
  nTotal -= U32_PER_PAGE*SQLITE_PAGE_SIZE;
  if( nTotal < U32_PER_PAGE*U32_PER_PAGE*SQLITE_PAGE_SIZE ){
    return nLocal + sizeof(u32);
  }
  return -1;  /* This payload will not fit. */
}

/*
** Read data from the payload area.
**
** aPage points directly at the beginning of the payload.  No bounds 
** checking is done on offset or amt -- it is assumed that the payload
** area is big enough to accomodate.
*/
static int payloadRead(Db *pDb, u32 *aPage, int offset, int amt, void *pBuf){
  int rc;
  int toread, more;

  assert( offset>=0 && amt>=0 );
  if( offset < LOCAL_PAYLOAD ){
    /* Data stored directly in the leaf block of the BTree */
    if( amt+offset>LOCAL_PAYLOAD ){
      toread = LOCAL_PAYLOAD - offset;
      more = 1;
    }else{
      toread = amt;
      more = 0;
    }
    memcpy(pBuf, &((char*)aPage)[offset], toread);
    if( !more ) return SQLITE_OK;
    pBuf = &((char*)pBuf)[toread];
    offset += toread;
    amt -= toread;
  }
  offset -= LOCAL_PAYLOAD;
  aPage += LOCAL_PAYLOAD/sizeof(aPage[0]);
  while( offset < 10*SQLITE_PAGE_SIZE ){
    /* Data stored in one of 10 direct pages */
    int iDir;
    char *aData;
    iDir = offset/SQLITE_PAGE_SIZE;
    base = offset - iDir*SQLITE_PAGE_SIZE;
    rc = sqlitePgGet(pDb->pPgr, aPage[iDir], &aData);
    if( rc!=SQLITE_OK ) return rc;
    if( amt+base > SQLITE_PAGE_SIZE ){
      toread = SQLITE_PAGE_SIZE - base;
      more = 1;
    }else{
      toread = amt;
      more = 0;
    }
    memcpy(pBuf, &aData[base], toread);
    sqlitePgUnref(aData);
    if( !more ) return SQLITE_OK;
    pBuf = &((char*)pBuf)[toread];
    amt -= toread;
    offset += toread;
  }
  offset -= 10*SQLITE_PAGE_SIZE;
  aPage += 10;
  if( offset < U32_PER_PAGE*SQLITE_PAGE_SIZE ){
    /* Data stored in an indirect page */ 
    u32 *indirPage;
    rc = sqlitePgGet(pDb->pPgr, aPage[0], &indirPage);
    if( rc!=SQLITE_OK ) return rc;
    while( amt>0 && offset < U32_PER_PAGE*SQLITE_PAGE_SIZE ){
      int idx, base;
      char *aData;
      idx = offset/SQLITE_PAGE_SIZE;
      base = offset - idx*SQLITE_PAGE_SIZE;
      rc = sqlitePgGet(pDb->pPgr, indirPage[idx], &aData);
      if( rc!=SQLITE_OK ) break;
      if( amt+base > SQLITE_PAGE_SIZE ){
        toread = SQLITE_PAGE_SIZE - base;
      }else{
        toread = amt;
      }
      memcpy(pBuf, &aData[base], toread);
      sqlitePgUnref(aData);
      pBuf = &((char*)pBuf)[toread];
      amt -= toread;
      offset += toread;
    }
    sqlitePgUnref(indirPage);
    if( rc!=SQLITE_OK ) return rc;
  }
  offset -= U32_PER_PAGE*SQLITE_PAGE_SIZE;
  aPage++;
  if( offset < U32_PER_PAGE*U32_PER_PAGE*SQLITE_PAGE_SIZE ){
    /* Data stored in a double-indirect page */
    u32 *dblIndirPage;
    rc = sqlitePgGet(pDb->pPgr, aPage[0], &dblIndirPage);
    if( rc!=SQLITE_OK ) return rc;
    while( amt>0 && offset < U32_PER_PAGE*U32_PER_PAGE*SQLITE_PAGE_SIZE ){
      int dblidx;
      u32 *indirPage;
      int basis;
      dblidx = offset/(U32_PER_PAGE*SQLITE_PAGE_SIZE);
      rc = sqlitePgGet(pDb->pPgr, dblIndirPage[dblidx], &indirPage);
      if( rc!=SQLITE_OK ) break;
      basis = dblidx*U32_PER_PAGE*SQLITE_PAGE_SIZE;
      while( amt>0 && offset < basis + U32_PER_PAGE*SQLITE_PAGE_SIZE ){
        int idx, base;
        char *aData;
        idx = (offset - basis)/SQLITE_PAGE_SIZE;
        base = (offset - basis) - idx*SQLITE_PAGE_SIZE;
        rc = sqlitePgGet(pDb->pPgr, indirPage[idx], &aData);
        if( rc!=SQLITE_OK ) break;
        if( amt+base > SQLITE_PAGE_SIZE ){
          toread = SQLITE_PAGE_SIZE - base;
        }else{
          toread = amt;
        }
        memcpy(pBuf, &aData[base], toread);
        sqlitePgUnref(aData);
        pBuf = &((char*)pBuf)[toread];
        amt -= toread;
        offset += toread;
      }
      sqlitePgUnref(indirPage);
      if( rc!=SQLITE_OK ) break;
    }
    sqlitePgUnref(dblIndirPage);
    return rc;
  }
  memset(pBuf, 0, amt);
  return SQLITE_OK;
}

/*
** Write data into the payload area.
**
** If pages have already been allocated for the payload, they are
** simply overwritten.  New pages are allocated as necessary to
** fill in gaps.  sqlitePgTouch() is called on all overflow pages,
** but the calling function must invoke sqlitePgTouch() for aPage
** itself.
*/
static int payloadWrite(Db *pDb, u32 *aPage, int offset, int amt, void *pBuf){
  assert( offset>=0 && amt>=0 );
  if( offset < LOCAL_PAYLOAD ){
    if( amt+offset>LOCAL_PAYLOAD ){
      towrite = LOCAL_PAYLOAD - offset;
      more = 1;
    }else{
      towrite = amt;
      more = 0;
    }
    memcpy(&((char*)aPage)[offset], pBuf, towrite);
    if( !more ) return SQLITE_OK;
    pBuf = &((char*)pBuf)[towrite];
    offset += toread;
    amt -= toread;
  }
  offset -= LOCAL_PAYLOAD;
  aPage += LOCAL_PAYLOAD/sizeof(aPage[0]);
  while( offset < 10*SQLITE_PAGE_SIZE ){
    int iDir;
    char *aData;
    iDir = offset/SQLITE_PAGE_SIZE;
    base = offset - iDir*SQLITE_PAGE_SIZE;
    if( aPage[iDir] ){
      rc = sqliteGet(pDb->pPgr, aPage[iDir], &aData);
    }else{
      rc = sqliteDbAllocPage(pDb, &aPage[iDir], &aData);
    }
    if( rc!=SQLITE_OK ) return rc;
    if( amt+base > SQLITE_PAGE_SIZE ){
      towrite = SQLITE_PAGE_SIZE - base;
      more = 1;
    }else{
      towrite = amt;
      more = 0;
    }
    memcpy(&aData[base], pBuf, towrite);
    sqlitePgUnref(aData);
    if( !more ) return SQLITE_OK;
    pBuf = &((char*)pBuf)[towrite];
    amt -= towrite;
    offset += towrite;
  }
  /* TBD.... */
}

/*
** Release any and all overflow pages associated with data starting
** with byte "newSize" up to but not including "oldSize".
*/
static int payloadFree(Db *pDb, u32 *aPage, int newSize, int oldSize){
  int i;

  if( newSize>=oldSize ) return;
  oldSize -= LOCAL_PAYLOAD;
  if( oldSize<=0 ) return SQLITE_OK;
  newSize -= LOCAL_PAYLOAD;
  if( newSize<0 ) newSize = 0;
  aPage += LOCAL_PAYLOAD/sizeof(u32);
*************
  for(i=0; i<10; i++){
    sqliteDbFreePage(pDb, aPage[0], 0);
    amt -= SQLITE_PAGE_SIZE;
    if( amt<=0 ) return SQLITE_OK;
    aPage++;
  }
  rc = sqlitePgGet(pDb->pPgr, aPage[0], &indirPage);
  if( rc!=SQLITE_OK ) return rc;
  for(i=0; i<U32_PER_PAGE; i++){
    if( indirPage[i]==0 ) continue;
    sqliteDbFreePage(pDb, indirPage[i], 0);
  }
  sqliteDbFreePage(pDb, aPage[0], indirPage);
  sqlitePgUnref(indirPage);
  amt -= U32_PER_PAGE*SQLITE_PAGE_SIZE;
  if( amt<=0 ) return SQLITE_OK;
  aPage++;
  rc = sqlitePgGet(pDb->pPgr, aPage[0], &dblIndirPage);
  if( rc!=SQLITE_OK ) return rc;
  for(i=0; i<U32_PER_PAGE; i++){
    if( dblIndirPage[i]==0 ) continue;
    rc = sqlitePgGet(pDb->pPgr, dblIndirPage[i], &indirPage);
    if( rc!=SQLITE_OK ) break;
    for(j=0; j<U32_PER_PAGE; j++){
      if( indirPage[j]==0 ) continue;
      sqliteDbFreePage(pDb, dblIndirPage[i], 0);
    }
    sqliteDbFreePage(pDb, dblIndirPage[i], indirPage);
    sqlitePgUnder(indirPage);
  }
  sqliteDbFreePage(pDb, aPage[0], dblIndirPage);
  sqlitePgUnref(dblIndirPage);
  return SQLITE_OK;    
}

/*
** Allocate space for the content table in the given Db structure.
** return SQLITE_OK on success and SQLITE_NOMEM if it fails.
*/
static int sqliteDbExpandContent(Db *pDb, int newSize){
  if( pDb->nAlloc>=newSize ) return SQLITE_OK;

  aPage[1] = pDb->aContent[0];
  memset(&aPage[2], 0, SQLITE_PAGE_SIZE - 2*sizeof(u32));
  pDb->aContent[0] = SWB(pgno);
  sqlitePgTouch(aPage);
  sqlitePgUnref(aPage);
}































/*
** Open a database.
*/
int sqliteDbOpen(const char *filename, Db **ppDb){
  Db *pDb = 0;
  Pgr *pPgr = 0;
  u32 *aPage1;

  pDb->pCursor = 0;
  pDb->inTransaction = 0;
  sqlitePgCount(pDb->pPgr, &nPage);
  rc = sqlitePgGet(pDb->pPgr, 1, &aPage1);
  if( rc!=0 ) goto open_err;
  if( nPage==0 ){
    sqlitePgBeginTransaction(pDb->pPgr);
    aPage1[0] = BLOCK_MAGIC|BLOCK_CONTENT;
    aPage1[2] = sizeof(u32)*10;
    sqlitePgTouch(aPage1);
    sqlitePgCommit(pDb->pPgr);
  }
  pDb->nContent = aPage1[2]/sizeof(u32);
  pDb->nAlloc = 0;
  rc = sqliteDbExpandContent(pDb, pDb->nContent);
  if( rc!=SQLITE_OK ) goto open_err;

  rc = payloadRead(pDb, &aPage1[3], 0, aPage[2], pDb->aContent);

  sqlitePgUnref(aPage1);
  if( rc!=SQLITE_OK ) goto open_err;
  *ppDb = pDb;
  return SQLITE_OK;

open_err:
  *ppDb = 0;
  if( pPgr ) sqlitePgClose(pPgr);
  if( pDb && pDb->aContent ) sqliteFree(pDb->aContent);

  return SQLITE_OK;
}

/*
** Commit changes to the database
*/ 
int sqliteDbCommit(Db *pDb){
  u32 *aPage1;
  int rc;
  if( !pDb->inTransaction ){
    return SQLITE_OK;
  }
  rc = sqlitePgGet(pDb->pPgr, 1, &aPage1);
  if( rc!=SQLITE_OK ) return rc;
  aPage1[2] = pDb->nContent*sizeof(u32);
  payloadWrite(pDb, 0, aPage1[2], pDb->aContent);
  sqlitePgUnref(aPage1);
  rc = sqlitePgCommit(pDb->pPgr);
  if( rc!=SQLITE_OK ) return rc;
  pDb->inTransaction = 0;
  return SQLITE_OK;
}

/*

#   drh@hwaci.com
#   http://www.hwaci.com/drh/
#
#***********************************************************************
# This file implements some common TCL routines used for regression
# testing the SQLite library
#
# $Id: tester.tcl,v 1.9 2001/01/22 00:31:53 drh Exp $

# Create a test database
#
if {![info exists dbprefix]} {
  if {[info exists env(SQLITE_PREFIX)]} {
    set dbprefix $env(SQLITE_PREFIX):
  } else {

  puts "$nErr errors out of $nTest tests"
  exit $nErr
}

# A procedure to execute SQL
#
proc execsql {sql} {
  # puts "SQL = $sql"
  return [db eval $sql]
}

# Another procedure to execute SQL.  This one includes the field
# names in the returned list.
#
proc execsql2 {sql} {