/*
** Return true if two DLine elements are identical.
*/
static int same_dline(DLine *pA, DLine *pB){
  return pA->h==pB->h && memcmp(pA->z,pB->z,pA->h & LENGTH_MASK)==0;
}

/*
** Return true if the regular expression *pRe matches any of the
** N dlines
*/
static int re_dline_match(
  ReCompiled *pRe,    /* The regular expression to be matched */
  DLine *aDLine,      /* First of N DLines to compare against */
  int N               /* Number of DLines to check */
){
  while( N-- ){
    if( re_exec(pRe, aDLine->z, LENGTH(aDLine)) ){
      return 1;
    }
    aDLine++;
  }
  return 0;
}

/*
** Append a single line of context-diff output to pOut.
*/
static void appendDiffLine(
  Blob *pOut,         /* Where to write the line of output */
  char cPrefix,       /* One of " ", "+",  or "-" */
  DLine *pLine,       /* The line to be output */
  int html,           /* True if generating HTML.  False for plain text */
  ReCompiled *pRe     /* Colorize only if line matches this Regex */
){
  blob_append(pOut, &cPrefix, 1);
  if( html ){
    char *zHtml;
    if( pRe && re_dline_match(pRe, pLine, 1)==0 ){
      cPrefix = ' ';
    }else if( cPrefix=='+' ){
      blob_append(pOut, "<span class=\"diffadd\">", -1);
    }else if( cPrefix=='-' ){
      blob_append(pOut, "<span class=\"diffrm\">", -1);
    }
    zHtml = htmlize(pLine->z, (pLine->h & LENGTH_MASK));
    blob_append(pOut, zHtml, -1);
    fossil_free(zHtml);

    blob_appendf(pOut, "%6d  ", lnB);
  }else{
    blob_append(pOut, "        ", 8);
  }
  if( html ) blob_append(pOut, "</span>", -1);
}


/*
** Given a raw diff p[] in which the p->aEdit[] array has been filled
** in, compute a context diff into pOut.
*/
static void contextDiff(
  DContext *p,      /* The difference */
  Blob *pOut,       /* Output a context diff to here */
  ReCompiled *pRe,  /* Only show changes that match this regex */
  u64 diffFlags     /* Flags controlling the diff format */

){
  DLine *A;     /* Left side of the diff */
  DLine *B;     /* Right side of the diff */
  int a = 0;    /* Index of next line in A[] */
  int b = 0;    /* Index of next line in B[] */
  int *R;       /* Array of COPY/DELETE/INSERT triples */
  int r;        /* Index into R[] */
  int nr;       /* Number of COPY/DELETE/INSERT triples to process */
  int mxr;      /* Maximum value for r */
  int na, nb;   /* Number of lines shown from A and B */
  int i, j;     /* Loop counters */
  int m;        /* Number of lines to output */
  int skip;     /* Number of lines to skip */
  int nChunk = 0;  /* Number of diff chunks seen so far */
  int nContext;    /* Number of lines of context */
  int showLn;      /* Show line numbers */
  int html;        /* Render as HTML */
  int showDivider = 0;  /* True to show the divider between diff blocks */

  nContext = diff_context_lines(diffFlags);
  showLn = (diffFlags & DIFF_LINENO)!=0;
  html = (diffFlags & DIFF_HTML)!=0;
  A = p->aFrom;
  B = p->aTo;
  R = p->aEdit;
  mxr = p->nEdit;
  while( mxr>2 && R[mxr-1]==0 && R[mxr-2]==0 ){ mxr -= 3; }
  for(r=0; r<mxr; r += 3*nr){
    /* Figure out how many triples to show in a single block */
    for(nr=1; R[r+nr*3]>0 && R[r+nr*3]<nContext*2; nr++){}
    /* printf("r=%d nr=%d\n", r, nr); */

    /* If there is a regex, skip this block (generate no diff output)
    ** if the regex matches or does not match both insert and delete.
    ** Only display the block if one side matches but the other side does
    ** not.
    */
    if( pRe ){
      int hideBlock = 1;
      int xa = a, xb = b;
      for(i=0; hideBlock && i<nr; i++){
        int c1, c2;
        xa += R[r+i*3];
        xb += R[r+i*3];
        c1 = re_dline_match(pRe, &A[xa], R[r+i*3+1]);
        c2 = re_dline_match(pRe, &B[xb], R[r+i*3+2]);
        hideBlock = c1==c2;
        xa += R[r+i*3+1];
        xb += R[r+i*3+2];
      }
      if( hideBlock ){
        a = xa;
        b = xb;
        continue;
      }
    }
    
    /* For the current block comprising nr triples, figure out
    ** how many lines of A and B are to be displayed
    */
    if( R[r]>nContext ){
      na = nb = nContext;
      skip = R[r] - nContext;
    }else{

    /* Show the header for this block, or if we are doing a modified
    ** context diff that contains line numbers, show the separator from
    ** the previous block.
    */
    nChunk++;
    if( showLn ){
      if( !showDivider ){
        /* Do not show a top divider */
        showDivider = 1;
      }else if( html ){
        blob_appendf(pOut, "<span class=\"diffhr\">%.80c</span>\n", '.');
        blob_appendf(pOut, "<a name=\"chunk%d\"></a>\n", nChunk);
      }else{
        blob_appendf(pOut, "%.80c\n", '.');
      }
    }else{

    /* Show the initial common area */
    a += skip;
    b += skip;
    m = R[r] - skip;
    for(j=0; j<m; j++){
      if( showLn ) appendDiffLineno(pOut, a+j+1, b+j+1, html);
      appendDiffLine(pOut, ' ', &A[a+j], html, 0);
    }
    a += m;
    b += m;

    /* Show the differences */
    for(i=0; i<nr; i++){
      m = R[r+i*3+1];
      for(j=0; j<m; j++){
        char cMark = '-';
        if( showLn ) appendDiffLineno(pOut, a+j+1, 0, html);
        if( pRe && re_dline_match(pRe, &A[a+j], 1)==0 ) cMark = ' ';
        appendDiffLine(pOut, '-', &A[a+j], html, pRe);
      }
      a += m;
      m = R[r+i*3+2];
      for(j=0; j<m; j++){
        if( showLn ) appendDiffLineno(pOut, 0, b+j+1, html);
        appendDiffLine(pOut, '+', &B[b+j], html, pRe);
      }
      b += m;
      if( i<nr-1 ){
        m = R[r+i*3+3];
        for(j=0; j<m; j++){
          if( showLn ) appendDiffLineno(pOut, a+j+1, b+j+1, html);
          appendDiffLine(pOut, ' ', &B[b+j], html, 0);
        }
        b += m;
        a += m;
      }
    }

    /* Show the final common area */
    assert( nr==i );
    m = R[r+nr*3];
    if( m>nContext ) m = nContext;
    for(j=0; j<m; j++){
      if( showLn ) appendDiffLineno(pOut, a+j+1, b+j+1, html);
      appendDiffLine(pOut, ' ', &B[b+j], html, 0);
    }
  }
}

/*
** Status of a single output line
*/
typedef struct SbsLine SbsLine;
struct SbsLine {
  char *zLine;             /* The output line under construction */
  int n;                   /* Index of next unused slot in the zLine[] */
  int width;               /* Maximum width of a column in the output */
  unsigned char escHtml;   /* True to escape html characters */
  int iStart;              /* Write zStart prior to character iStart */
  const char *zStart;      /* A <span> tag */
  int iEnd;                /* Write </span> prior to character iEnd */
  int iStart2;             /* Write zStart2 prior to character iStart2 */
  const char *zStart2;     /* A <span> tag */
  int iEnd2;               /* Write </span> prior to character iEnd2 */
  ReCompiled *pRe;         /* Only colorize matching lines, if not NULL */
};

/*
** Flags for sbsWriteText()
*/
#define SBS_NEWLINE      0x0001   /* End with \n\000 */
#define SBS_PAD          0x0002   /* Pad output to width spaces */

  int i;   /* Number of input characters consumed */
  int j;   /* Number of output characters generated */
  int k;   /* Cursor position */
  int needEndSpan = 0;
  const char *zIn = pLine->z;
  char *z = &p->zLine[p->n];
  int w = p->width;
  int colorize = p->escHtml;
  if( colorize && p->pRe && re_dline_match(p->pRe, pLine, 1)==0 ){
    colorize = 0;
  }
  for(i=j=k=0; k<w && i<n; i++, k++){
    char c = zIn[i];
    if( colorize ){
      if( i==p->iStart ){
        int x = strlen(p->zStart);
        memcpy(z+j, p->zStart, x);
        j += x;
        needEndSpan = 1;
        if( p->iStart2 ){
          p->iStart = p->iStart2;

/*
** Given a diff context in which the aEdit[] array has been filled
** in, compute a side-by-side diff into pOut.
*/
static void sbsDiff(
  DContext *p,       /* The computed diff */
  Blob *pOut,        /* Write the results here */
  ReCompiled *pRe,   /* Only show changes that match this regex */
  u64 diffFlags      /* Flags controlling the diff */

){
  DLine *A;     /* Left side of the diff */
  DLine *B;     /* Right side of the diff */
  int a = 0;    /* Index of next line in A[] */
  int b = 0;    /* Index of next line in B[] */
  int *R;       /* Array of COPY/DELETE/INSERT triples */
  int r;        /* Index into R[] */
  int nr;       /* Number of COPY/DELETE/INSERT triples to process */
  int mxr;      /* Maximum value for r */
  int na, nb;   /* Number of lines shown from A and B */
  int i, j;     /* Loop counters */
  int m, ma, mb;/* Number of lines to output */
  int skip;     /* Number of lines to skip */
  int nChunk = 0; /* Number of chunks of diff output seen so far */
  SbsLine s;    /* Output line buffer */
  int nContext; /* Lines of context above and below each change */
  int showDivider = 0;  /* True to show the divider */

  memset(&s, 0, sizeof(s));
  s.width = diff_width(diffFlags);
  s.zLine = fossil_malloc( 15*s.width + 200 );
  if( s.zLine==0 ) return;
  nContext = diff_context_lines(diffFlags);
  s.escHtml = (diffFlags & DIFF_HTML)!=0;
  s.pRe = pRe;
  s.iStart = -1;
  s.iStart2 = 0;
  s.iEnd = -1;
  A = p->aFrom;
  B = p->aTo;
  R = p->aEdit;
  mxr = p->nEdit;
  while( mxr>2 && R[mxr-1]==0 && R[mxr-2]==0 ){ mxr -= 3; }
  for(r=0; r<mxr; r += 3*nr){
    /* Figure out how many triples to show in a single block */
    for(nr=1; R[r+nr*3]>0 && R[r+nr*3]<nContext*2; nr++){}
    /* printf("r=%d nr=%d\n", r, nr); */

    /* If there is a regex, skip this block (generate no diff output)
    ** if the regex matches or does not match both insert and delete.
    ** Only display the block if one side matches but the other side does
    ** not.
    */
    if( pRe ){
      int hideBlock = 1;
      int xa = a, xb = b;
      for(i=0; hideBlock && i<nr; i++){
        int c1, c2;
        xa += R[r+i*3];
        xb += R[r+i*3];
        c1 = re_dline_match(pRe, &A[xa], R[r+i*3+1]);
        c2 = re_dline_match(pRe, &B[xb], R[r+i*3+2]);
        hideBlock = c1==c2;
        xa += R[r+i*3+1];
        xb += R[r+i*3+2];
      }
      if( hideBlock ){
        a = xa;
        b = xb;
        continue;
      }
    }

    /* For the current block comprising nr triples, figure out
    ** how many lines of A and B are to be displayed
    */
    if( R[r]>nContext ){
      na = nb = nContext;
      skip = R[r] - nContext;

    }
    for(i=1; i<nr; i++){
      na += R[r+i*3];
      nb += R[r+i*3];
    }

    /* Draw the separator between blocks */
    if( showDivider ){
      if( s.escHtml ){
        blob_appendf(pOut, "<span class=\"diffhr\">%.*c</span>\n",
                           s.width*2+16, '.');
      }else{
        blob_appendf(pOut, "%.*c\n", s.width*2+16, '.');
      }
    }
    showDivider = 1;
    nChunk++;
    if( s.escHtml ){
      blob_appendf(pOut, "<a name=\"chunk%d\"></a>\n", nChunk);
    }

    /* Show the initial common area */
    a += skip;
    b += skip;
    m = R[r] - skip;

          /* Delete one line from the left */
          s.n = 0;
          sbsWriteLineno(&s, a);
          s.iStart = 0;
          s.zStart = "<span class=\"diffrm\">";
          s.iEnd = s.width;
          sbsWriteText(&s, &A[a], SBS_PAD);
          if( s.escHtml ){
            sbsWrite(&s, " &lt;\n", 6);
          }else{
            sbsWrite(&s, " <\n", 3);
          }
          blob_append(pOut, s.zLine, s.n);
          assert( ma>0 );
          ma--;

          ma--;
          mb--;
          a++;
          b++;
        }else if( alignment[j]==2 ){
          /* Insert one line on the right */
          s.n = 0;
          sbsWriteSpace(&s, s.width + 7);
          if( s.escHtml ){
            sbsWrite(&s, " > ", 6);
          }else{
            sbsWrite(&s, " > ", 3);
          }
          sbsWriteLineno(&s, b);
          s.iStart = 0;
          s.zStart = "<span class=\"diffadd\">";

** file is encountered, 0 is returned and pOut is written with
** text "cannot compute difference between binary files".
*/
int *text_diff(
  Blob *pA_Blob,   /* FROM file */
  Blob *pB_Blob,   /* TO file */
  Blob *pOut,      /* Write diff here if not NULL */
  ReCompiled *pRe, /* Only output changes where this Regexp matches */
  u64 diffFlags    /* DIFF_* flags defined above */
){
  int ignoreEolWs; /* Ignore whitespace at the end of lines */
  int nContext;    /* Amount of context to display */
  DContext c;

  if( diffFlags & DIFF_INVERT ){
    Blob *pTemp = pA_Blob;
    pA_Blob = pB_Blob;
    pB_Blob = pTemp;
  }

  ignoreEolWs = (diffFlags & DIFF_IGNORE_EOLWS)!=0;

  /* Prepare the input files */
  memset(&c, 0, sizeof(c));
  c.aFrom = break_into_lines(blob_str(pA_Blob), blob_size(pA_Blob),
                             &c.nFrom, ignoreEolWs);
  c.aTo = break_into_lines(blob_str(pB_Blob), blob_size(pB_Blob),

  /* Compute the difference */
  diff_all(&c);
  if( (diffFlags & DIFF_NOOPT)==0 ) diff_optimize(&c);

  if( pOut ){
    /* Compute a context or side-by-side diff into pOut */

    if( diffFlags & DIFF_SIDEBYSIDE ){

      sbsDiff(&c, pOut, pRe, diffFlags);
    }else{

      contextDiff(&c, pOut, pRe, diffFlags);
    }
    fossil_free(c.aFrom);
    fossil_free(c.aTo);
    fossil_free(c.aEdit);
    return 0;
  }else{
    /* If a context diff is not requested, then return the

  int *R;
  u64 diffFlags = diff_options();
  if( g.argc<4 ) usage("FILE1 FILE2 ...");
  blob_read_from_file(&a, g.argv[2]);
  for(i=3; i<g.argc; i++){
    if( i>3 ) fossil_print("-------------------------------\n");
    blob_read_from_file(&b, g.argv[i]);
    R = text_diff(&a, &b, 0, 0, diffFlags);
    for(r=0; R[r] || R[r+1] || R[r+2]; r += 3){
      fossil_print(" copy %4d  delete %4d  insert %4d\n", R[r], R[r+1], R[r+2]);
    }
    /* free(R); */
    blob_reset(&b);
  }
}

/*
** COMMAND: test-diff
**
** Usage: %fossil [options] FILE1 FILE2
**
** Print the difference between two files.  The usual diff options apply.
*/
void test_diff_cmd(void){
  Blob a, b, out;
  u64 diffFlag;
  const char *zRe;           /* Regex filter for diff output */
  ReCompiled *pRe = 0;       /* Regex filter for diff output */

  if( find_option("tk",0,0)!=0 ){
    diff_tk("test-diff", 2);
    return;
  }
  find_option("i",0,0);
  zRe = find_option("regexp","e",1);
  if( zRe ){
    const char *zErr = re_compile(&pRe, zRe, 0);
    if( zErr ) fossil_fatal("regex error: %s", zErr);
  }
  diffFlag = diff_options();
  verify_all_options();
  if( g.argc!=4 ) usage("FILE1 FILE2");
  diff_print_filenames(g.argv[2], g.argv[3], diffFlag);
  blob_read_from_file(&a, g.argv[2]);
  blob_read_from_file(&b, g.argv[3]);
  blob_zero(&out);
  text_diff(&a, &b, &out, pRe, diffFlag);
  blob_write_to_file(&out, "-");
  re_free(pRe);
}

/**************************************************************************
** The basic difference engine is above.  What follows is the annotation
** engine.  Both are in the same file since they share many components.
*/

    /* Compute and output the differences */
    if( diffFlags & DIFF_BRIEF ){
      if( blob_compare(pFile1, &file2) ){
        fossil_print("CHANGED  %s\n", zName);
      }
    }else{
      blob_zero(&out);
      text_diff(pFile1, &file2, &out, 0, diffFlags);
      if( blob_size(&out) ){
        diff_print_filenames(zName, zName2, diffFlags);
        fossil_print("%s\n", blob_str(&out));
      }
      blob_reset(&out);
    }

  u64 diffFlags             /* Diff flags */
){
  if( diffFlags & DIFF_BRIEF ) return;
  if( zDiffCmd==0 ){
    Blob out;      /* Diff output text */

    blob_zero(&out);
    text_diff(pFile1, pFile2, &out, 0, diffFlags);
    diff_print_filenames(zName, zName, diffFlags);
    fossil_print("%s\n", blob_str(&out));

    /* Release memory resources */
    blob_reset(&out);
  }else{
    Blob cmd;

    toid = uuid_to_rid(zTo, 0);
    content_get(toid, &to);
  }else{
    blob_zero(&to);
  }
  blob_zero(&out);
  if( diffFlags & DIFF_SIDEBYSIDE ){
    text_diff(&from, &to, &out, 0, diffFlags | DIFF_HTML);
    @ <div class="sbsdiff">
    @ %s(blob_str(&out))
    @ </div>
  }else{
    text_diff(&from, &to, &out, 0, diffFlags | DIFF_LINENO | DIFF_HTML);
    @ <div class="udiff">
    @ %s(blob_str(&out))
    @ </div>
  }
  blob_reset(&from);
  blob_reset(&to);
  blob_reset(&out);

    }else{
      diffFlags |= DIFF_LINENO;
      zStyle = "udiff";
    }
  }
  content_get(v1, &c1);
  content_get(v2, &c2);
  text_diff(&c1, &c2, pOut, 0, diffFlags);
  blob_reset(&c1);
  blob_reset(&c2);
  if( !isPatch ){
    style_header("Diff");
    style_submenu_element("Patch", "Patch", "%s/fdiff?v1=%T&v2=%T&patch",
                          g.zTop, P("v1"), P("v2"));
    if( !sideBySide ){

      json_set_err(FSL_JSON_E_UNRESOLVED_UUID,
                   "Could not resolve 'to' ID.");
      return NULL;
  }
  content_get(fromid, &from);
  content_get(toid, &to);
  blob_zero(&out);
  text_diff(&from, &to, &out, 0, flags);
  blob_reset(&from);
  blob_reset(&to);
  outLen = blob_size(&out);
  if(outLen>=0){
    rc = cson_value_new_string(blob_buffer(&out),
                               (unsigned int)blob_size(&out));
  }

  }

  blob_init(&w1, pW1->zWiki, -1);
  blob_zero(&w2);
  blob_init(&w2, pW2->zWiki, -1);
  blob_zero(&d);
  diffFlags = DIFF_IGNORE_EOLWS | DIFF_INLINE;
  text_diff(&w2, &w1, &d, 0, diffFlags);
  blob_reset(&w1);
  blob_reset(&w2);

  pay = cson_new_object();
  
  zUuid = json_wiki_get_uuid_for_rid( pW1->rid );
  cson_object_set(pay, "v1", json_new_string(zUuid) );

  ** and pPivot => pV2 (into aC2).  Each of the aC1 and aC2 arrays is
  ** an array of integer triples.  Within each triple, the first integer
  ** is the number of lines of text to copy directly from the pivot,
  ** the second integer is the number of lines of text to omit from the
  ** pivot, and the third integer is the number of lines of text that are
  ** inserted.  The edit array ends with a triple of 0,0,0.
  */
  aC1 = text_diff(pPivot, pV1, 0, 0, 0);
  aC2 = text_diff(pPivot, pV2, 0, 0, 0);
  if( aC1==0 || aC2==0 ){
    free(aC1);
    free(aC2);
    return -1;
  }

  blob_rewind(pV1);        /* Rewind inputs:  Needed to reconstruct output */

*/
typedef struct ReStateSet {
  unsigned nState;            /* Number of current states */
  ReStateNumber *aState;      /* Current states */
} ReStateSet;

#if INTERFACE
/* An input string read one character at a time.
*/
struct ReInput {
  const unsigned char *z;  /* All text */
  int i;                   /* Next byte to read */
  int mx;                  /* EOF when i>=mx */
};

/* A compiled NFA (or an NFA that is in the process of being compiled) is
** an instance of the following object.
*/
struct ReCompiled {
  ReInput sIn;                /* Regular expression text */
  const char *zErr;           /* Error message to return */
  char *aOp;                  /* Operators for the virtual machine */
  int *aArg;                  /* Arguments to each operator */
  unsigned (*xNextChar)(ReInput*);  /* Next character function */
  char zInit[12];             /* Initial text to match */
  int nInit;                  /* Number of characters in zInit */
  unsigned nState;            /* Number of entries in aOp[] and aArg[] */
  unsigned nAlloc;            /* Slots allocated for aOp[] and aArg[] */
};
#endif

/* Add a state to the given state set if it is not already there */
static void re_add_state(ReStateSet *pSet, int newState){
  unsigned i;
  for(i=0; i<pSet->nState; i++) if( pSet->aState[i]==newState ) return;
  pSet->aState[pSet->nState++] = newState;
}

/* Extract the next unicode character from *pzIn and return it.  Advance
** *pzIn to the first byte past the end of the character returned.  To
** be clear:  this routine converts utf8 to unicode.  This routine is 
** optimized for the common case where the next character is a single byte.
*/
static unsigned re_next_char(ReInput *p){
  unsigned c;
  if( p->i>=p->mx ) return 0;
  c = p->z[p->i++];
  if( c>0x80 ){
    if( (c&0xe0)==0xc0 && p->i<p->mx && (p->z[p->i]&0xc0)==0x80 ){
      c = (c&0x1f)<<6 | (p->z[p->i++]&0x3f);

      if( c<0x80 ) c = 0xfffd;
    }else if( (c&0xf0)==0xe0 && p->i+1<p->mx && (p->z[p->i]&0xc0)==0x80
           && (p->z[p->i+1]&0xc0)==0x80 ){
      c = (c&0x0f)<<12 | ((p->z[p->i]&0x3f)<<6) | (p->z[p->i+1]&0x3f);
      p->i += 2;
      if( c<0x3ff || (c>=0xd800 && c<=0xdfff) ) c = 0xfffd;
    }else if( (c&0xf8)==0xf0 && p->i+3<p->mx && (p->z[p->i]&0xc0)==0x80
           && (p->z[p->i+1]&0xc0)==0x80 && (p->z[p->i+2]&0xc0)==0x80 ){
      c = (c&0x07)<<18 | ((p->z[p->i]&0x3f)<<12) | ((p->z[p->i+1]&0x3f)<<6)
                       | (p->z[p->i+2]&0x3f);
      p->i += 3;
      if( c<0xffff ) c = 0xfffd;
    }else{
      c = 0xfffd;
    }
  }
  return c;
}
static unsigned re_next_char_nocase(ReInput *p){
  unsigned c = re_next_char(p);
  return unicode_fold(c,1);
}

/* Return true if c is a perl "word" character:  [A-Za-z0-9_] */
static int re_word_char(int c){
  return unicode_isalnum(c) || c=='_';
}

static int re_space_char(int c){
  return c==' ' || c=='\t' || c=='\n' || c=='\v' || c=='\f';
}

/* Run a compiled regular expression on the zero-terminated input
** string zIn[].  Return true on a match and false if there is no match.
*/
int re_exec(ReCompiled *pRe, const unsigned char *zIn, int nIn){
  ReStateSet aStateSet[2], *pThis, *pNext;
  ReStateNumber aSpace[100];
  ReStateNumber *pToFree;
  unsigned int i = 0;
  unsigned int iSwap = 0;
  int c = RE_EOF+1;
  int cPrev = 0;
  int rc = 0;
  ReInput in;

  in.z = zIn;
  in.i = 0;
  in.mx = nIn>=0 ? nIn : strlen(zIn);
  if( pRe->nInit ){
    unsigned char x = pRe->zInit[0];
    while( in.i+pRe->nInit<in.mx 
        && (zIn[in.i]!=x || memcmp(zIn+in.i, pRe->zInit, pRe->nInit)!=0)
    ){
      in.i++;
    }
    if( in.i+pRe->nInit>=in.mx ) return 0;
  }
  if( pRe->nState<=(sizeof(aSpace)/(sizeof(aSpace[0])*2)) ){
    pToFree = 0;
    aStateSet[0].aState = aSpace;
  }else{
    pToFree = fossil_malloc( sizeof(ReStateNumber)*2*pRe->nState );
    if( pToFree==0 ) return -1;
    aStateSet[0].aState = pToFree;
  }
  aStateSet[1].aState = &aStateSet[0].aState[pRe->nState];
  pNext = &aStateSet[1];
  pNext->nState = 0;
  re_add_state(pNext, 0);
  while( c!=RE_EOF && pNext->nState>0 ){
    cPrev = c;
    c = pRe->xNextChar(&in);
    pThis = pNext;
    pNext = &aStateSet[iSwap];
    iSwap = 1 - iSwap;
    pNext->nState = 0;
    for(i=0; i<pThis->nState; i++){
      int x = pThis->aState[i];
      switch( pRe->aOp[x] ){

/* A backslash character has been seen, read the next character and
** return its intepretation.
*/
static unsigned re_esc_char(ReCompiled *p){
  static const char zEsc[] = "afnrtv\\()*.+?[$^{|}]";
  static const char zTrans[] = "\a\f\n\r\t\v";
  int i, v = 0;
  char c;
  if( p->sIn.i>=p->sIn.mx ) return 0;
  c = p->sIn.z[0];
  if( c=='u' && p->sIn.i+5<p->sIn.mx ){
    v = 0;
    const unsigned char *zIn = p->sIn.z + p->sIn.i;
    if( re_hex(zIn[1],&v)
     && re_hex(zIn[2],&v)
     && re_hex(zIn[3],&v)
     && re_hex(zIn[4],&v)
    ){
      p->sIn.i += 5;
      return v;
    }
  }
  if( c=='x' ){
    v = 0;
    for(i=1; p->sIn.i<p->sIn.mx && re_hex(p->sIn.z[p->sIn.i+i], &v); i++){}
    if( i>1 ){
      p->sIn.i += i;
      return v;
    }
  }
  for(i=0; zEsc[i] && zEsc[i]!=c; i++){}
  if( zEsc[i] ){
    if( i<6 ) c = zTrans[i];
    p->sIn.i++;
  }else{
    p->zErr = "unknown \\ escape";
  }
  return c;
}

/* Forward declaration */
static const char *re_subcompile_string(ReCompiled*);

/* Peek at the next byte of input */
static unsigned char rePeek(ReCompiled *p){
  return p->sIn.i<p->sIn.mx ? p->sIn.z[p->sIn.i] : 0;
}

/* Compile RE text into a sequence of opcodes.  Continue up to the
** first unmatched ")" character, then return.  If an error is found,
** return a pointer to the error message string.
*/
static const char *re_subcompile_re(ReCompiled *p){
  const char *zErr;
  int iStart, iEnd, iGoto;
  iStart = p->nState;
  zErr = re_subcompile_string(p);
  if( zErr ) return zErr;
  while( rePeek(p)=='|' ){
    iEnd = p->nState;
    re_insert(p, iStart, RE_OP_FORK, iEnd + 2 - iStart);
    iGoto = re_append(p, RE_OP_GOTO, 0);
    p->sIn.i++;
    zErr = re_subcompile_string(p);
    if( zErr ) return zErr;
    p->aArg[iGoto] = p->nState - iGoto;
  }
  return 0;
}

/* Compile an element of regular expression text (anything that can be
** an operand to the "|" operator).  Return NULL on success or a pointer
** to the error message if there is a problem.
*/
static const char *re_subcompile_string(ReCompiled *p){
  int iPrev = -1;
  int iStart;
  unsigned c;
  const char *zErr;
  while( (c = p->xNextChar(&p->sIn))!=0 ){
    iStart = p->nState;
    switch( c ){
      case '|':
      case '$': 
      case ')': {
        p->sIn.i--;
        return 0;
      }
      case '(': {
        zErr = re_subcompile_re(p);
        if( zErr ) return zErr;
        if( rePeek(p)!=')' ) return "unmatched '('";
        p->sIn.i++;
        break;
      }
      case '.': {
        if( rePeek(p)=='*' ){
          re_append(p, RE_OP_ANYSTAR, 0);
          p->sIn.i++;
        }else{ 
          re_append(p, RE_OP_ANY, 0);
        }
        break;
      }
      case '*': {
        if( iPrev<0 ) return "'*' without operand";

        re_insert(p, iPrev, RE_OP_FORK, p->nState - iPrev+1);
        break;
      }
      case '{': {
        int m = 0, n = 0;
        int sz, j;
        if( iPrev<0 ) return "'{m,n}' without operand";
        while( (c=rePeek(p))>='0' && c<='9' ){ m = m*10 + c - '0'; p->sIn.i++; }
        n = m;
        if( c==',' ){
          p->sIn.i++;
          n = 0;
          while( (c=rePeek(p))>='0' && c<='9' ){ n = n*10 + c-'0'; p->sIn.i++; }
        }
        if( c!='}' ) return "unmatched '{'";
        if( n>0 && n<m ) return "n less than m in '{m,n}'";
        p->sIn.i++;
        sz = p->nState - iPrev;
        if( m==0 ){
          if( n==0 ) return "both m and n are zero in '{m,n}'";
          re_insert(p, iPrev, RE_OP_FORK, sz+1);
          n--;
        }else{
          for(j=1; j<m; j++) re_copy(p, iPrev, sz);
        }
        for(j=m; j<n; j++){
          re_append(p, RE_OP_FORK, sz+1);
          re_copy(p, iPrev, sz);
        }
        if( n==0 && m>0 ){
          re_append(p, RE_OP_FORK, -sz);
        }
        break;
      }
      case '[': {
        int iFirst = p->nState;
        if( rePeek(p)=='^' ){
          re_append(p, RE_OP_CC_EXC, 0);
          p->sIn.i++;
        }else{
          re_append(p, RE_OP_CC_INC, 0);
        }
        while( (c = p->xNextChar(&p->sIn))!=0 ){
          if( c=='[' && rePeek(p)==':' ){
            return "POSIX character classes not supported";
          }
          if( c=='\\' ) c = re_esc_char(p);
          if( rePeek(p)=='-' ){
            re_append(p, RE_OP_CC_RANGE, c);
            p->sIn.i++;
            c = p->xNextChar(&p->sIn);
            if( c=='\\' ) c = re_esc_char(p);
            re_append(p, RE_OP_CC_RANGE, c);
          }else{
            re_append(p, RE_OP_CC_VALUE, c);
          }
          if( rePeek(p)==']' ){ p->sIn.i++; break; }
        }
        if( c==0 ) return "unclosed '['";
        p->aArg[iFirst] = p->nState - iFirst;
        break;
      }
      case '\\': {
        int specialOp = 0;
        switch( rePeek(p) ){
          case 'b': specialOp = RE_OP_BOUNDARY;   break;
          case 'd': specialOp = RE_OP_DIGIT;      break;
          case 'D': specialOp = RE_OP_NOTDIGIT;   break;
          case 's': specialOp = RE_OP_SPACE;      break;
          case 'S': specialOp = RE_OP_NOTSPACE;   break;
          case 'w': specialOp = RE_OP_WORD;       break;
          case 'W': specialOp = RE_OP_NOTWORD;    break;
        }
        if( specialOp ){
          p->sIn.i++;
          re_append(p, specialOp, 0);
        }else{
          c = re_esc_char(p);
          re_append(p, RE_OP_MATCH, c);
        }
        break;
      }

    return "out of memory";
  }
  if( zIn[0]=='^' ){
    zIn++;
  }else{
    re_append(pRe, RE_OP_ANYSTAR, 0);
  }
  pRe->sIn.z = (unsigned char*)zIn;
  pRe->sIn.i = 0;
  pRe->sIn.mx = strlen(pRe->sIn.z);
  zErr = re_subcompile_re(pRe);
  if( zErr ){
    re_free(pRe);
    return zErr;
  }
  if( rePeek(pRe)=='$' && pRe->sIn.i+1==pRe->sIn.mx ){
    re_append(pRe, RE_OP_MATCH, RE_EOF);
    re_append(pRe, RE_OP_ACCEPT, 0);
    *ppRe = pRe;
  }else if( pRe->sIn.i>=pRe->sIn.mx ){
    re_append(pRe, RE_OP_ACCEPT, 0);
    *ppRe = pRe;
  }else{
    re_free(pRe);
    return "unrecognized character";
  }
  if( pRe->aOp[0]==RE_OP_ANYSTAR ){

      sqlite3_result_error_nomem(context);
      return;
    }
    sqlite3_set_auxdata(context, 0, pRe, (void(*)(void*))re_free);
  }
  zStr = (const unsigned char*)sqlite3_value_text(argv[1]);
  if( zStr!=0 ){
    sqlite3_result_int(context, re_exec(pRe, zStr, -1));
  }
}

/*
** Invoke this routine in order to install the REGEXP function in an
** SQLite database connection.
**

  int ln = 0;
  int n;
  char zLine[2000];
  while( fgets(zLine, sizeof(zLine), in) ){
    ln++;
    n = (int)strlen(zLine);
    while( n && (zLine[n-1]=='\n' || zLine[n-1]=='\r') ) n--;

    if( re_exec(pRe, (const unsigned char*)zLine, n) ){
      printf("%s:%d:%.*s\n", zFile, ln, n, zLine);
    }
  }
}

/*
** COMMAND: test-grep
**

  blob_init(&w1, pW1->zWiki, -1);
  blob_zero(&w2);
  if( rid2 && (pW2 = manifest_get(rid2, CFTYPE_WIKI))!=0 ){
    blob_init(&w2, pW2->zWiki, -1);
  }
  blob_zero(&d);
  diffFlags = construct_diff_flags(1,0);
  text_diff(&w2, &w1, &d, 0, diffFlags | DIFF_HTML | DIFF_LINENO);
  @ <div class="udiff">
  @ %s(blob_str(&d))
  @ </div>
  manifest_destroy(pW1);
  manifest_destroy(pW2);
  style_footer();
}