Change the position list format so that its size in bytes is stored at the start of the list itself.

FossilOrigin-Name: 62f2ff20418702ed0fbf708369edf5638445b51b

ext/fts5/fts5Int.h

@@ -75,7 +75,6 @@ void sqlite3Fts5Dequote(char *z);
 typedef struct Fts5Index Fts5Index;
 typedef struct Fts5IndexIter Fts5IndexIter;
 
-
 /*
 ** Values used as part of the flags argument passed to IndexQuery().
 */
@@ -117,18 +116,9 @@ void sqlite3Fts5IterNext(Fts5IndexIter*, i64 iMatch);
 i64  sqlite3Fts5IterRowid(Fts5IndexIter*);
 
 /*
-** Position list iteration.
+** Obtain the position list that corresponds to the current position.
-**
-**   for(
-**     iPos=sqlite3Fts5IterFirstPos(pIter, iCol); 
-**     iPos>=0; 
-**     iPos=sqlite3Fts5IterNextPos(pIter)
-**   ){
-**     // token appears at position iPos of column iCol of the current document
-**   }
 */
-// int sqlite3Fts5IterFirstPos(Fts5IndexIter*, int iCol);
+const u8 *sqlite3Fts5IterPoslist(Fts5IndexIter*, int *pn);
-// int sqlite3Fts5IterNextPos(Fts5IndexIter*);
 
 /*
 ** Close an iterator opened by sqlite3Fts5IndexQuery().

ext/fts5/fts5_index.c

@@ -110,13 +110,13 @@
 **
 **     poslist format:
 **
+**         varint: size of poslist in bytes. not including this field.
 **         collist: collist for column 0
 **         zero-or-more {
 **           0x01 byte
 **           varint: column number (I)
 **           collist: collist for column I
 **         }
-**         0x00 byte
 **
 **     collist format:
 **
@@ -255,6 +255,7 @@ static int fts5Corrupt() { return SQLITE_CORRUPT_VTAB; }
 typedef struct Fts5BtreeIter Fts5BtreeIter;
 typedef struct Fts5BtreeIterLevel Fts5BtreeIterLevel;
 typedef struct Fts5Buffer Fts5Buffer;
+typedef struct Fts5ChunkIter Fts5ChunkIter;
 typedef struct Fts5Data Fts5Data;
 typedef struct Fts5MultiSegIter Fts5MultiSegIter;
 typedef struct Fts5NodeIter Fts5NodeIter;
@@ -297,13 +298,6 @@ struct Fts5Index {
   sqlite3_stmt *pDeleter;         /* "DELETE FROM %_data ... id>=? AND id<=?" */
 };
 
-struct Fts5IndexIter {
-  Fts5Index *pIndex;
-  Fts5Structure *pStruct;
-  Fts5MultiSegIter *pMulti;
-};
-
-
 /*
 ** Buffer object for the incremental building of string data.
 */
@@ -313,6 +307,13 @@ struct Fts5Buffer {
   int nSpace;
 };
 
+struct Fts5IndexIter {
+  Fts5Index *pIndex;
+  Fts5Structure *pStruct;
+  Fts5MultiSegIter *pMulti;
+  Fts5Buffer poslist;             /* Buffer containing current poslist */
+};
+
 /*
 ** A single record read from the %_data table.
 */
@@ -422,9 +423,9 @@ struct Fts5MultiSegIter {
 **
 ** iLeafOffset:
 **   Byte offset within the current leaf that is one byte past the end of the
-**   rowid field of the current entry. Usually this is the first byte of 
+**   rowid field of the current entry. Usually this is the size field of the
-**   the position list data. The exception is if the rowid for the current 
+**   position list data. The exception is if the rowid for the current entry 
-**   entry is the last thing on the leaf page.
+**   is the last thing on the leaf page.
 **
 ** pLeaf:
 **   Buffer containing current leaf page data. Set to NULL at EOF.
@@ -453,13 +454,25 @@ struct Fts5SegIter {
   i64 iRowid;                     /* Current rowid */
 };
 
+/*
+** Object for iterating through paginated data.
+*/
+struct Fts5ChunkIter {
+  Fts5Data *pLeaf;                /* Current leaf data. NULL -> EOF. */
+  i64 iLeafRowid;                 /* Absolute rowid of current leaf */
+  int nRem;                       /* Remaining bytes of data to read */
+
+  /* Output parameters */
+  u8 *p;                          /* Pointer to chunk of data */
+  int n;                          /* Size of buffer p in bytes */
+};
+
 /*
 ** Object for iterating through a single position list.
 */
 struct Fts5PosIter {
-  Fts5Data *pLeaf;                /* Current leaf data. NULL -> EOF. */
+  Fts5ChunkIter chunk;            /* Current chunk of data */
-  i64 iLeafRowid;                 /* Absolute rowid of current leaf */
+  int iOff;                       /* Offset within chunk data */
-  int iLeafOffset;                /* Current offset within leaf */
 
   int iCol;
   int iPos;
@@ -1107,6 +1120,10 @@ static void fts5SegIterNextPage(
   }
 }
 
+/*
+** Leave pIter->iLeafOffset as the offset to the size field of the first
+** position list. The position list belonging to document pIter->iRowid.
+*/
 static void fts5SegIterLoadTerm(Fts5Index *p, Fts5SegIter *pIter, int nKeep){
   u8 *a = pIter->pLeaf->p;        /* Buffer to read data from */
   int iOff = pIter->iLeafOffset;  /* Offset to read at */
@@ -1218,14 +1235,17 @@ static void fts5SegIterSeekInit(
 
       while( (res = fts5BufferCompareBlob(&pIter->term, pTerm, nTerm)) ){
         if( res<0 && pIter->iLeafPgno==iPg ){
-          /* Search for the end of the current doclist within the current
+          int iOff = pIter->iLeafOffset;
-          ** page. The end of a doclist is marked by a pair of successive
+          while( iOff<n ){
-          ** 0x00 bytes.  */
+            int nPos;
-          int iOff;
+            iOff += getVarint32(&a[iOff], nPos);
-          for(iOff=pIter->iLeafOffset+1; iOff<n; iOff++){
+            iOff += nPos;
-            if( a[iOff]==0 && a[iOff-1]==0 ) break;
+            if( iOff<n ){
+              u64 rowid;
+              iOff += getVarint(&a[iOff], &rowid);
+              if( rowid==0 ) break;
+            }
           }
-          iOff++;
 
           /* If the iterator is not yet at the end of the next page, load
           ** the next term and jump to the next iteration of the while()
@@ -1267,8 +1287,13 @@ static void fts5SegIterNext(
     /* Search for the end of the position list within the current page. */
     u8 *a = pLeaf->p;
     int n = pLeaf->n;
-    for(iOff=pIter->iLeafOffset; iOff<n && a[iOff]; iOff++);
+
-    iOff++;
+    iOff = pIter->iLeafOffset;
+    if( iOff<=n ){
+      int nPoslist;
+      iOff += getVarint32(&a[iOff], nPoslist);
+      iOff += nPoslist;
+    }
 
     if( iOff<n ){
       /* The next entry is on the current page */
@@ -1532,6 +1557,70 @@ static const u8 *fts5MultiIterTerm(Fts5MultiSegIter *pIter, int *pn){
   return p->term.p;
 }
 
+/*
+** Return true if the chunk iterator passed as the second argument is
+** at EOF. Or if an error has already occurred. Otherwise, return false.
+*/
+static int fts5ChunkIterEof(Fts5Index *p, Fts5ChunkIter *pIter){
+  return (p->rc || pIter->pLeaf==0);
+}
+
+/*
+** Advance the chunk-iterator to the next chunk of data to read.
+*/
+static void fts5ChunkIterNext(Fts5Index *p, Fts5ChunkIter *pIter){
+  assert( pIter->nRem>=pIter->n );
+  pIter->nRem -= pIter->n;
+  fts5DataRelease(pIter->pLeaf);
+  pIter->pLeaf = 0;
+  pIter->p = 0;
+  if( pIter->nRem>0 ){
+    Fts5Data *pLeaf;
+    pIter->iLeafRowid++;
+    pLeaf = pIter->pLeaf = fts5DataRead(p, pIter->iLeafRowid);
+    if( pLeaf ){
+      pIter->n = MIN(pIter->nRem, pLeaf->n-4);
+      pIter->p = pLeaf->p+4;
+    }
+  }
+}
+
+/*
+** Intialize the chunk iterator to read the position list data for which 
+** the size field is at offset iOff of leaf pLeaf. 
+*/
+static void fts5ChunkIterInit(
+  Fts5Index *p,                   /* FTS5 backend object */
+  Fts5SegIter *pSeg,              /* Segment iterator to read poslist from */
+  Fts5ChunkIter *pIter            /* Initialize this object */
+){
+  int iId = pSeg->pSeg->iSegid;
+  i64 rowid = FTS5_SEGMENT_ROWID(pSeg->iIdx, iId, 0, pSeg->iLeafPgno);
+  Fts5Data *pLeaf = pSeg->pLeaf;
+  int iOff = pSeg->iLeafOffset;
+
+  memset(pIter, 0, sizeof(*pIter));
+  pIter->iLeafRowid = rowid;
+  if( iOff<pLeaf->n ){
+    fts5DataReference(pLeaf);
+    pIter->pLeaf = pLeaf;
+  }else{
+    pIter->nRem = 1;
+    fts5ChunkIterNext(p, pIter);
+    if( p->rc ) return;
+    iOff = 4;
+    pLeaf = pIter->pLeaf;
+  }
+
+  iOff += getVarint32(&pLeaf->p[iOff], pIter->nRem);
+  pIter->n = MIN(pLeaf->n - iOff, pIter->nRem);
+  pIter->p = pLeaf->p + iOff;
+
+  if( pIter->n==0 ){
+    fts5ChunkIterNext(p, pIter);
+  }
+}
+
 /*
 ** Read and return the next 32-bit varint from the position-list iterator 
 ** passed as the second argument.
@@ -1543,17 +1632,12 @@ static const u8 *fts5MultiIterTerm(Fts5MultiSegIter *pIter, int *pn){
 static int fts5PosIterReadVarint(Fts5Index *p, Fts5PosIter *pIter){
   int iVal = 0;
   if( p->rc==SQLITE_OK ){
-    int iOff = pIter->iLeafOffset;
+    if( pIter->iOff>=pIter->chunk.n ){
-    if( iOff < pIter->pLeaf->n ){
+      fts5ChunkIterNext(p, &pIter->chunk);
-      pIter->iLeafOffset += getVarint32(&pIter->pLeaf->p[iOff], iVal);
+      if( fts5ChunkIterEof(p, &pIter->chunk) ) return 0;
-    }else{
+      pIter->iOff = 0;
-      fts5DataRelease(pIter->pLeaf);
-      pIter->iLeafRowid++;
-      pIter->pLeaf = fts5DataRead(p, pIter->iLeafRowid);
-      if( pIter->pLeaf ){
-        pIter->iLeafOffset = 4 + getVarint32(&pIter->pLeaf->p[4], iVal);
-      }
     }
+    pIter->iOff += getVarint32(&pIter->chunk.p[pIter->iOff], iVal);
   }
   return iVal;
 }
@@ -1563,18 +1647,17 @@ static int fts5PosIterReadVarint(Fts5Index *p, Fts5PosIter *pIter){
 */
 static void fts5PosIterNext(Fts5Index *p, Fts5PosIter *pIter){
   int iVal;
+  assert( fts5ChunkIterEof(p, &pIter->chunk)==0 );
   iVal = fts5PosIterReadVarint(p, pIter);
-  if( iVal==0 ){
+  if( fts5ChunkIterEof(p, &pIter->chunk)==0 ){
-    fts5DataRelease(pIter->pLeaf);
+    if( iVal==1 ){
-    pIter->pLeaf = 0;
-  }
-  else if( iVal==1 ){
       pIter->iCol = fts5PosIterReadVarint(p, pIter);
       pIter->iPos = fts5PosIterReadVarint(p, pIter) - 2;
     }else{
       pIter->iPos += (iVal - 2);
     }
   }
+}
 
 /*
 ** Initialize the Fts5PosIter object passed as the final argument to iterate
@@ -1588,23 +1671,20 @@ static void fts5PosIterInit(
 ){
   if( p->rc==SQLITE_OK ){
     Fts5SegIter *pSeg = &pMulti->aSeg[ pMulti->aFirst[1] ];
-    int iId = pSeg->pSeg->iSegid;
-
     memset(pIter, 0, sizeof(*pIter));
-    pIter->pLeaf = pSeg->pLeaf;
+    fts5ChunkIterInit(p, pSeg, &pIter->chunk);
-    pIter->iLeafOffset = pSeg->iLeafOffset;
+    if( fts5ChunkIterEof(p, &pIter->chunk)==0 ){
-    pIter->iLeafRowid = FTS5_SEGMENT_ROWID(pSeg->iIdx, iId, 0, pSeg->iLeafPgno);
-    fts5DataReference(pIter->pLeaf);
       fts5PosIterNext(p, pIter);
     }
   }
+}
 
 /*
 ** Return true if the position iterator passed as the second argument is
 ** at EOF. Or if an error has already occurred. Otherwise, return false.
 */
 static int fts5PosIterEof(Fts5Index *p, Fts5PosIter *pIter){
-  return (p->rc || pIter->pLeaf==0);
+  return (p->rc || pIter->chunk.pLeaf==0);
 }
 
 
@@ -2106,15 +2186,15 @@ static void fts5WritePendingDoclist(
     /* Append the rowid itself */
     fts5WriteAppendRowid(p, pWriter, pPoslist->iRowid);
 
+    /* Append the size of the position list in bytes */
+    fts5WriteAppendPoslistInt(p, pWriter, pPoslist->buf.n);
+
     /* Copy the position list to the output segment */
     while( i<pPoslist->buf.n){
       int iVal;
       i += getVarint32(&pPoslist->buf.p[i], iVal);
       fts5WriteAppendPoslistInt(p, pWriter, iVal);
     }
-
-    /* Write the position list terminator */
-    fts5WriteAppendZerobyte(p, pWriter);
   }
 
   /* Write the doclist terminator */
@@ -2297,9 +2377,8 @@ fflush(stdout);
       fts5MultiIterEof(p, pIter)==0;
       fts5MultiIterNext(p, pIter)
   ){
-    Fts5PosIter sPos;             /* Used to iterate through position list */
+    Fts5SegIter *pSeg = &pIter->aSeg[ pIter->aFirst[1] ];
-    int iCol = 0;                 /* Current output column */
+    Fts5ChunkIter sPos;           /* Used to iterate through position list */
-    int iPos = 0;                 /* Current output position */
     int nTerm;
     const u8 *pTerm = fts5MultiIterTerm(pIter, &nTerm);
 
@@ -2319,20 +2398,16 @@ fflush(stdout);
     fts5WriteAppendRowid(p, &writer, fts5MultiIterRowid(pIter));
 
     /* Copy the position list from input to output */
-    for(fts5PosIterInit(p, pIter, &sPos);
+    fts5ChunkIterInit(p, pSeg, &sPos);
-        fts5PosIterEof(p, &sPos)==0;
+    fts5WriteAppendPoslistInt(p, &writer, sPos.nRem);
-        fts5PosIterNext(p, &sPos)
+    for(/* noop */; fts5ChunkIterEof(p, &sPos)==0; fts5ChunkIterNext(p, &sPos)){
-    ){
+      int iOff = 0;
-      if( sPos.iCol!=iCol ){
+      while( iOff<sPos.n ){
-        fts5WriteAppendPoslistInt(p, &writer, 1);
+        int iVal;
-        fts5WriteAppendPoslistInt(p, &writer, sPos.iCol);
+        iOff += getVarint32(&sPos.p[iOff], iVal);
-        iCol = sPos.iCol;
+        fts5WriteAppendPoslistInt(p, &writer, iVal);
-        iPos = 0;
       }
-      fts5WriteAppendPoslistInt(p, &writer, (sPos.iPos-iPos) + 2);
-      iPos = sPos.iPos;
     }
-    fts5WriteAppendZerobyte(p, &writer);
   }
 
   /* Flush the last leaf page to disk. Set the output segment b-tree height
@@ -2910,7 +2985,6 @@ static int fts5DecodePoslist(int *pRc, Fts5Buffer *pBuf, const u8 *a, int n){
     int iVal;
     iOff += getVarint32(&a[iOff], iVal);
     fts5BufferAppendPrintf(pRc, pBuf, " %d", iVal);
-    if( iVal==0 ) break;
   }
   return iOff;
 }
@@ -2932,7 +3006,9 @@ static int fts5DecodeDoclist(int *pRc, Fts5Buffer *pBuf, const u8 *a, int n){
     fts5BufferAppendPrintf(pRc, pBuf, " rowid=%lld", iDocid);
   }
   while( iOff<n ){
-    iOff += fts5DecodePoslist(pRc, pBuf, &a[iOff], n-iOff);
+    int nPos;
+    iOff += getVarint32(&a[iOff], nPos);
+    iOff += fts5DecodePoslist(pRc, pBuf, &a[iOff], MIN(n-iOff, nPos));
     if( iOff<n ){
       i64 iDelta;
       iOff += sqlite3GetVarint(&a[iOff], (u64*)&iDelta);
@@ -2989,11 +3065,16 @@ static void fts5DecodeFunction(
 
       iRowidOff = fts5GetU16(&a[0]);
       iTermOff = fts5GetU16(&a[2]);
-      iOff = 4;
+
-      if( iTermOff!=4 && iRowidOff!=4 ){
+      if( iRowidOff ){
-        iOff += fts5DecodePoslist(&rc, &s, &a[iOff], n-iOff);
+        iOff = iRowidOff;
-        if( iRowidOff==0 ) iOff++;
+      }else if( iTermOff ){
+        iOff = iTermOff;
+      }else{
+        iOff = n;
       }
+      fts5DecodePoslist(&rc, &s, &a[4], iOff-4);
+
 
       assert( iRowidOff==0 || iOff==iRowidOff );
       if( iRowidOff ){
@@ -3115,6 +3196,7 @@ int sqlite3Fts5IterEof(Fts5IndexIter *pIter){
 ** Move to the next matching rowid. 
 */
 void sqlite3Fts5IterNext(Fts5IndexIter *pIter, i64 iMatch){
+  fts5BufferZero(&pIter->poslist);
   fts5MultiIterNext(pIter->pIndex, pIter->pMulti);
 }
 
@@ -3125,6 +3207,21 @@ i64 sqlite3Fts5IterRowid(Fts5IndexIter *pIter){
   return fts5MultiIterRowid(pIter->pMulti);
 }
 
+/*
+** Return a pointer to a buffer containing a copy of the position list for
+** the current entry. Output variable *pn is set to the size of the buffer 
+** in bytes before returning.
+**
+** The returned buffer does not include the 0x00 terminator byte stored on
+** disk.
+*/
+const u8 *sqlite3Fts5IterPoslist(Fts5IndexIter *pIter, int *pn){
+  assert( sqlite3Fts5IterEof(pIter)==0 );
+
+  *pn = pIter->poslist.n;
+  return pIter->poslist.p;
+}
+
 /*
 ** Close an iterator opened by an earlier call to sqlite3Fts5IndexQuery().
 */

manifest

@@ -1,5 +1,5 @@
-C Fix\sminor\sproblems\sin\sterm\smatching.
+C Change\sthe\sposition\slist\sformat\sso\sthat\sits\ssize\sin\sbytes\sis\sstored\sat\sthe\sstart\sof\sthe\slist\sitself.
-D 2014-06-26T12:31:41.784
+D 2014-07-01T20:45:18.496
 F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f
 F Makefile.in b03432313a3aad96c706f8164fb9f5307eaf19f5
 F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23
@@ -104,10 +104,10 @@ F ext/fts3/unicode/CaseFolding.txt 8c678ca52ecc95e16bc7afc2dbf6fc9ffa05db8c
 F ext/fts3/unicode/UnicodeData.txt cd07314edb62d49fde34debdaf92fa2aa69011e7
 F ext/fts3/unicode/mkunicode.tcl dc6f268eb526710e2c6e496c372471d773d0c368
 F ext/fts5/fts5.c 1af3184dd9c0e5c1686f71202d6b6cac8f225f05
-F ext/fts5/fts5Int.h 3fd1ebeb58963727cae0ccc8e4e80751bd870296
+F ext/fts5/fts5Int.h 80f3d38a69a0c58ccc94428c8fc8adbcf7561a2d
 F ext/fts5/fts5_config.c 94f1b4cb4de6a7cd5780c14adb0198e289df8cef
 F ext/fts5/fts5_expr.c 1874b17f10a38d0b21e0c38a28637f74e4d2570a
-F ext/fts5/fts5_index.c c8b21d12f15ca6fe028ede3f8e040dcb2aaef6f6
+F ext/fts5/fts5_index.c ea3dfe56a16813fcf59e03f6156965894b4b5e6f
 F ext/fts5/fts5_storage.c 7848d8f8528d798bba159900ea310a6d4a279da8
 F ext/icu/README.txt d9fbbad0c2f647c3fdf715fc9fd64af53aedfc43
 F ext/icu/icu.c d415ccf984defeb9df2c0e1afcfaa2f6dc05eacb
@@ -1188,7 +1188,7 @@ F tool/vdbe_profile.tcl 67746953071a9f8f2f668b73fe899074e2c6d8c1
 F tool/warnings-clang.sh f6aa929dc20ef1f856af04a730772f59283631d4
 F tool/warnings.sh 0abfd78ceb09b7f7c27c688c8e3fe93268a13b32
 F tool/win/sqlite.vsix deb315d026cc8400325c5863eef847784a219a2f
-P 47a9f3cc92deefe163108e3507bd4614bf1f5da7
+P 94eeb077d08a1d2607f3ff3a9fbf18229ba475bb
-R b3c853e0de2f65908d06e4caafc2bc71
+R 5d9b8f6933c58725a24e426a963b0d97
 U dan
-Z 840b486e1cfb6171cd4015bc26a5f123
+Z bb8816e0d501865bff7c4c8da87350cb

manifest.uuid

@@ -1 +1 @@
-94eeb077d08a1d2607f3ff3a9fbf18229ba475bb
+62f2ff20418702ed0fbf708369edf5638445b51b