third_party/sqlite/patches/0013-fts2-Interpret-foo-as-a-prefix-search.patch

   1 From b3310ec4e29ca2ef45e36a7b188503cd2637c8ba Mon Sep 17 00:00:00 2001
   2 From: Matt Perry <mpcomplete@google.com>
   3 Date: Wed, 17 Dec 2008 00:50:50 +0000
   4 Subject: [PATCH 13/16] [fts2] Interpret "foo*" as a prefix search.
   5
   6 By default it interpreted it as "foo *" (two tokens).
   7
   8 Original review URL: http://codereview.chromium.org/14176
   9 ---
  10  third_party/sqlite/src/ext/fts2/fts2.c | 15 +++++++++++++++
  11  1 file changed, 15 insertions(+)
  12
  13 diff --git a/third_party/sqlite/src/ext/fts2/fts2.c b/third_party/sqlite/src/ext/fts2/fts2.c
  14 index 944f324..1c68af5 100644
  15 --- a/third_party/sqlite/src/ext/fts2/fts2.c
  16 +++ b/third_party/sqlite/src/ext/fts2/fts2.c
  17 @@ -3558,6 +3558,7 @@ static int tokenizeSegment(
  18    int firstIndex = pQuery->nTerms;
  19    int iCol;
  20    int nTerm = 1;
  21 +  int iEndLast = -1;
  22
  23    int rc = pModule->xOpen(pTokenizer, pSegment, nSegment, &pCursor);
  24    if( rc!=SQLITE_OK ) return rc;
  25 @@ -3582,6 +3583,20 @@ static int tokenizeSegment(
  26        pQuery->nextIsOr = 1;
  27        continue;
  28      }
  29 +
  30 +    /*
  31 +     * The ICU tokenizer considers '*' a break character, so the code below
  32 +     * sets isPrefix correctly, but since that code doesn't eat the '*', the
  33 +     * ICU tokenizer returns it as the next token.  So eat it here until a
  34 +     * better solution presents itself.
  35 +     */
  36 +    if( pQuery->nTerms>0 && nToken==1 && pSegment[iBegin]=='*' &&
  37 +        iEndLast==iBegin){
  38 +      pQuery->pTerms[pQuery->nTerms-1].isPrefix = 1;
  39 +      continue;
  40 +    }
  41 +    iEndLast = iEnd;
  42 +
  43      queryAdd(pQuery, pToken, nToken);
  44      if( !inPhrase && iBegin>0 && pSegment[iBegin-1]=='-' ){
  45        pQuery->pTerms[pQuery->nTerms-1].isNot = 1;
  46 --
  47 2.2.1
  48