SQLite

#define FTS5_DATA_ZERO_PADDING 8
#define FTS5_DATA_PADDING 20

typedef struct Fts5Data Fts5Data;
typedef struct Fts5DlidxIter Fts5DlidxIter;
typedef struct Fts5DlidxLvl Fts5DlidxLvl;
typedef struct Fts5DlidxWriter Fts5DlidxWriter;
typedef struct Fts5NodeIter Fts5NodeIter;
typedef struct Fts5PageWriter Fts5PageWriter;
typedef struct Fts5SegIter Fts5SegIter;
typedef struct Fts5DoclistIter Fts5DoclistIter;
typedef struct Fts5SegWriter Fts5SegWriter;
typedef struct Fts5Structure Fts5Structure;
typedef struct Fts5StructureLevel Fts5StructureLevel;
typedef struct Fts5StructureSegment Fts5StructureSegment;

  i64 iSwitchRowid;               /* Firstest rowid of other than aFirst[1] */
  Fts5CResult *aFirst;            /* Current merge state (see above) */
  Fts5SegIter aSeg[1];            /* Array of segment iterators */
};


/*
** Object for iterating through the conents of a single internal node in 
** memory.
*/
struct Fts5NodeIter {
  /* Internal. Set and managed by fts5NodeIterXXX() functions. Except, 
  ** the EOF test for the iterator is (Fts5NodeIter.aData==0).  */
  const u8 *aData;
  int nData;
  int iOff;

  /* Output variables */
  Fts5Buffer term;
  int nEmpty;
  int iChild;
  int bDlidx;
};

/*
** An instance of the following type is used to iterate through the contents
** of a doclist-index record.
**
** pData:
**   Record containing the doclist-index data.
**

** Compare the contents of the pLeft buffer with the pRight/nRight blob.
**
** Return -ve if pLeft is smaller than pRight, 0 if they are equal or
** +ve if pRight is smaller than pLeft. In other words:
**
**     res = *pLeft - *pRight
*/

static int fts5BufferCompareBlob(
  Fts5Buffer *pLeft,              /* Left hand side of comparison */
  const u8 *pRight, int nRight    /* Right hand side of comparison */
){
  int nCmp = MIN(pLeft->n, nRight);
  int res = memcmp(pLeft->p, pRight, nCmp);
  return (res==0 ? (pLeft->n - nRight) : res);
}


/*
** Compare the contents of the two buffers using memcmp(). If one buffer
** is a prefix of the other, it is considered the lesser.
**
** Return -ve if pLeft is smaller than pRight, 0 if they are equal or
** +ve if pRight is smaller than pLeft. In other words:

      szPromote = szSeg;
    }
    fts5StructurePromoteTo(p, iPromote, szPromote, pStruct);
  }
}


/*
** If the pIter->iOff offset currently points to an entry indicating one
** or more term-less nodes, advance past it and set pIter->nEmpty to
** the number of empty child nodes.
*/
static void fts5NodeIterGobbleNEmpty(Fts5NodeIter *pIter){
  if( pIter->iOff<pIter->nData && 0==(pIter->aData[pIter->iOff] & 0xfe) ){
    pIter->bDlidx = pIter->aData[pIter->iOff] & 0x01;
    pIter->iOff++;
    pIter->iOff += fts5GetVarint32(&pIter->aData[pIter->iOff], pIter->nEmpty);
  }else{
    pIter->nEmpty = 0;
    pIter->bDlidx = 0;
  }
}

/*
** Advance to the next entry within the node.
*/
static void fts5NodeIterNext(int *pRc, Fts5NodeIter *pIter){
  if( pIter->iOff>=pIter->nData ){
    pIter->aData = 0;
    pIter->iChild += pIter->nEmpty;
  }else{
    int nPre, nNew;
    pIter->iOff += fts5GetVarint32(&pIter->aData[pIter->iOff], nPre);
    pIter->iOff += fts5GetVarint32(&pIter->aData[pIter->iOff], nNew);
    pIter->term.n = nPre-2;
    fts5BufferAppendBlob(pRc, &pIter->term, nNew, pIter->aData+pIter->iOff);
    pIter->iOff += nNew;
    pIter->iChild += (1 + pIter->nEmpty);
    fts5NodeIterGobbleNEmpty(pIter);
    if( *pRc ) pIter->aData = 0;
  }
}


/*
** Initialize the iterator object pIter to iterate through the internal
** segment node in pData.
*/
static void fts5NodeIterInit(const u8 *aData, int nData, Fts5NodeIter *pIter){
  memset(pIter, 0, sizeof(*pIter));
  pIter->aData = aData;
  pIter->nData = nData;
  pIter->iOff = fts5GetVarint32(aData, pIter->iChild);
  fts5NodeIterGobbleNEmpty(pIter);
}

/*
** Free any memory allocated by the iterator object.
*/
static void fts5NodeIterFree(Fts5NodeIter *pIter){
  fts5BufferFree(&pIter->term);
}

/*
** Advance the iterator passed as the only argument. If the end of the 
** doclist-index page is reached, return non-zero.
*/
static int fts5DlidxLvlNext(Fts5DlidxLvl *pLvl){
  Fts5Data *pData = pLvl->pData;

      iOff += nPos;
    }
  }

  pIter->pDlidx = fts5DlidxIterInit(p, bRev, iSeg, pIter->iTermLeafPgno);
}

#ifdef SQLITE_DEBUG
static void fts5AssertNodeSeekOk(
  Fts5Buffer *pNode,
  const u8 *pTerm, int nTerm,     /* Term to search for */
  int iExpectPg,
  int bExpectDlidx
){
  int bDlidx;
  int iPg;
  int rc = SQLITE_OK;
  Fts5NodeIter node;

  fts5NodeIterInit(pNode->p, pNode->n, &node);
  assert( node.term.n==0 );
  iPg = node.iChild;
  bDlidx = node.bDlidx;
  for(fts5NodeIterNext(&rc, &node);
      node.aData && fts5BufferCompareBlob(&node.term, pTerm, nTerm)<=0;
      fts5NodeIterNext(&rc, &node)
  ){
    iPg = node.iChild;
    bDlidx = node.bDlidx;
  }
  fts5NodeIterFree(&node);

  assert( rc!=SQLITE_OK || iPg==iExpectPg );
  assert( rc!=SQLITE_OK || bDlidx==bExpectDlidx );
}
#else
#define fts5AssertNodeSeekOk(v,w,x,y,z)
#endif

/*
** Argument pNode is an internal b-tree node. This function searches
** within the node for the largest term that is smaller than or equal
** to (pTerm/nTerm).
**
** It returns the associated page number. Or, if (pTerm/nTerm) is smaller
** than all terms within the node, the leftmost child page number. 
**
** Before returning, (*pbDlidx) is set to true if the last term on the
** returned child page number has a doclist-index. Or left as is otherwise.
*/
static int fts5NodeSeek(
  Fts5Buffer *pNode,              /* Node to search */
  const u8 *pTerm, int nTerm,     /* Term to search for */
  int *pbDlidx                    /* OUT: True if dlidx flag is set */
){
  int iPg;
  u8 *pPtr = pNode->p;
  u8 *pEnd = &pPtr[pNode->n];
  int nMatch = 0;                 /* Number of bytes of pTerm already matched */
  
  assert( *pbDlidx==0 );

  pPtr += fts5GetVarint32(pPtr, iPg);
  while( pPtr<pEnd ){
    int nEmpty = 0;
    int nKeep;
    int nNew;

    /* If there is a "no terms" record at pPtr, read it now. Store the
    ** number of termless pages in nEmpty. If it indicates a doclist-index, 
    ** set (*pbDlidx) to true.*/
    if( *pPtr<2 ){
      *pbDlidx = (*pPtr==0x01);
      pPtr++;
      pPtr += fts5GetVarint32(pPtr, nEmpty);
      if( pPtr>=pEnd ) break;
    }

    /* Read the next "term" pointer. Set nKeep to the number of bytes to
    ** keep from the previous term, and nNew to the number of bytes of
    ** new data that will be appended to it. */
    nKeep = (int)*pPtr++;
    nNew = (int)*pPtr++;
    if( (nKeep | nNew) & 0x0080 ){
      pPtr -= 2;
      pPtr += fts5GetVarint32(pPtr, nKeep);
      pPtr += fts5GetVarint32(pPtr, nNew);
    }
    nKeep -= 2;

    /* Compare (pTerm/nTerm) to the current term on the node (the one described
    ** by nKeep/nNew). If the node term is larger, break out of the while()
    ** loop. 
    **
    ** Otherwise, if (pTerm/nTerm) is larger or the two terms are equal, 
    ** leave variable nMatch set to the size of the largest prefix common to
    ** both terms in bytes.  */
    if( nKeep==nMatch ){
      int nTst = MIN(nNew, nTerm-nMatch);
      int i;
      for(i=0; i<nTst; i++){
        if( pTerm[nKeep+i]!=pPtr[i] ) break;
      }
      nMatch += i;
      assert( nMatch<=nTerm );

      if( i<nNew && (nMatch==nTerm || pPtr[i] > pTerm[nMatch]) ) break;
    }else if( nKeep<nMatch ){
      break;
    }

    iPg += 1 + nEmpty;
    *pbDlidx = 0;
    pPtr += nNew;
  }

  fts5AssertNodeSeekOk(pNode, pTerm, nTerm, iPg, *pbDlidx);
  return iPg;
}

#define fts5IndexGetVarint32(a, iOff, nVal) {     \
  nVal = a[iOff++];                               \
  if( nVal & 0x80 ){                              \
    iOff--;                                       \
    iOff += fts5GetVarint32(&a[iOff], nVal);      \
  }                                               \
}

  }else if( iSegid==0 ){
    if( iRowid==FTS5_AVERAGES_ROWID ){
      /* todo */
    }else{
      fts5DecodeStructure(&rc, &s, a, n);
    }
  }else{

    Fts5Buffer term;





    memset(&term, 0, sizeof(Fts5Buffer));

    if( iHeight==0 ){
      int iTermOff = 0;
      int iRowidOff = 0;
      int iOff;
      int nKeep = 0;

      if( n>=4 ){
        iRowidOff = fts5GetU16(&a[0]);
        iTermOff = fts5GetU16(&a[2]);
      }else{
        sqlite3Fts5BufferSet(&rc, &s, 8, (const u8*)"corrupt");
        goto decode_out;
      }

      if( iRowidOff ){
        iOff = iRowidOff;
      }else if( iTermOff ){
        iOff = iTermOff;
      }else{
        iOff = n;
      }
      fts5DecodePoslist(&rc, &s, &a[4], iOff-4);

      assert( iRowidOff==0 || iOff==iRowidOff );
      if( iRowidOff ){
        iOff += fts5DecodeDoclist(&rc, &s, &a[iOff], n-iOff);
      }

      assert( iTermOff==0 || iOff==iTermOff );
      while( iOff<n ){
        int nByte;
        iOff += fts5GetVarint32(&a[iOff], nByte);
        term.n= nKeep;
        fts5BufferAppendBlob(&rc, &term, nByte, &a[iOff]);
        iOff += nByte;

        sqlite3Fts5BufferAppendPrintf(
            &rc, &s, " term=%.*s", term.n, (const char*)term.p
        );
        iOff += fts5DecodeDoclist(&rc, &s, &a[iOff], n-iOff);
        if( iOff<n ){
          iOff += fts5GetVarint32(&a[iOff], nKeep);
        }
      }
      fts5BufferFree(&term);
    }else{
      Fts5NodeIter ss;
      for(fts5NodeIterInit(a, n, &ss); ss.aData; fts5NodeIterNext(&rc, &ss)){
        if( ss.term.n==0 ){
          sqlite3Fts5BufferAppendPrintf(&rc, &s, " left=%d", ss.iChild);
        }else{
          sqlite3Fts5BufferAppendPrintf(&rc,&s, " \"%.*s\"", 
              ss.term.n, ss.term.p
          );
        }
        if( ss.nEmpty ){
          sqlite3Fts5BufferAppendPrintf(&rc, &s, " empty=%d%s", ss.nEmpty,
              ss.bDlidx ? "*" : ""
          );
        }
      }
      fts5NodeIterFree(&ss);
    }
  }
  
 decode_out:
  sqlite3_free(a);
  if( rc==SQLITE_OK ){
    sqlite3_result_text(pCtx, (const char*)s.p, s.n, SQLITE_TRANSIENT);
  }else{