unsigned char getCharacterLanguage ( char *utf8Char ) {
// romantic?
char cs = getUtf8CharSize ( utf8Char );
// can't say what language it is
if ( cs == 1 ) return langUnknown;
// convert to 32 bit unicode
UChar32 c = utf8Decode ( utf8Char );
UCScript us = ucGetScript ( c );
// arabic? this also returns for persian!! fix?
if ( us == ucScriptArabic )
return langArabic;
if ( us == ucScriptCyrillic )
return langRussian;
if ( us == ucScriptHebrew )
return langHebrew;
if ( us == ucScriptGreek )
return langGreek;
return langUnknown;
}
bool Words::addWords(char *s,long nodeLen,bool computeWordIds, long niceness) {
long i = 0;
long j;
//long k = 0;
long wlen;
//unsigned long e;
//long skip;
long badCount = 0;
bool hadApostrophe = false;
UCScript oldScript = ucScriptCommon;
UCScript saved;
UCProps props;
uptop:
// bad utf8 can cause a breach
if ( i >= nodeLen ) goto done;
if ( ! s[i] ) goto done;
if ( ! is_alnum_utf8(s+i) ) { // && m_numWords < m_preCount ) {
if ( m_numWords >= m_preCount ) goto done;
// tag?
if ( s[i]=='<' && m_hasTags && isTagStart(s+i) ) {
// get the tag id
if ( s[i+1]=='/' ) {
// skip over /
m_tagIds [m_numWords] = ::getTagId(s+i+2);
m_tagIds [m_numWords] |= BACKBIT;
}
else
m_tagIds [m_numWords] = ::getTagId(s+i+1);
// word start
m_words [m_numWords] = s + i;
m_wordIds [m_numWords] = 0LL;
// skip till end
long tagLen = getTagLen(s+i); // ,niceness);
m_wordLens [m_numWords] = tagLen;
m_numWords++;
// advance
i += tagLen;
goto uptop;
}
// it is a punct word, find end of it
char *start = s+i;
//for (;s[i] && ! is_alnum_utf8(s+i);i+=getUtf8CharSize(s+i));
for ( ; s[i] ; i += getUtf8CharSize(s+i)){
// stop on < if we got tags
if ( s[i] == '<' && m_hasTags ) break;
// breathe
QUICKPOLL(niceness);
// if we are simple ascii, skip quickly
if ( is_ascii(s[i]) ) {
// accumulate NON-alnum chars
if ( ! is_alnum_a(s[i]) ) continue;
// update
oldScript = ucScriptCommon;
// otherwise, stop we got alnum
break;
}
// if we are utf8 we stop on special props
UChar32 c = utf8Decode ( s+i );
// stop if word char
if ( ! ucIsWordChar ( c ) ) continue;
// update first though
oldScript = ucGetScript ( c );
// then stop
break;
}
m_words [ m_numWords ] = start;
m_wordLens [ m_numWords ] = s+i - start;
m_wordIds [ m_numWords ] = 0LL;
if (m_tagIds) m_tagIds[m_numWords] = 0;
m_numWords++;
goto uptop;
}
// get an alnum word
j = i;
again:
//for ( ; is_alnum_utf8 (&s[i] ) ; i += getUtf8CharSize(s+i) );
for ( ; s[i] ; i += getUtf8CharSize(s+i) ) {
// breathe
QUICKPOLL(niceness);
// simple ascii?
if ( is_ascii(s[i]) ) {
// accumulate alnum chars
if ( is_alnum_a(s[i]) ) continue;
// update
oldScript = ucScriptCommon;
// otherwise, stop we got punct
break;
}
// get the code point of the utf8 char
UChar32 c = utf8Decode ( s+i );
// get props
props = ucProperties ( c );
// good stuff?
if ( props & (UC_IGNORABLE|UC_EXTEND) ) continue;
// stop? if UC_WORCHAR is set, that means its an alnum
if ( ! ( props & UC_WORDCHAR ) ) {
// reset script between words
oldScript = ucScriptCommon;
break;
}
// save it
saved = oldScript;
// update here
oldScript = ucGetScript(c);
// treat ucScriptLatin (30) as common so we can have latin1
// like char without breaking the word!
if ( oldScript == ucScriptLatin ) oldScript = ucScriptCommon;
// stop on this crap too i guess. like japanes chars?
if ( props & ( UC_IDEOGRAPH | UC_HIRAGANA | UC_THAI ) ) {
// include it
i += getUtf8CharSize(s+i);
// but stop
break;
}
// script change?
if ( saved != oldScript ) break;
}
// . java++, A++, C++ exception
// . A+, C+, exception
// . TODO: consider putting in Bits.cpp w/ D_CAN_BE_IN_PHRASE
if ( s[i]=='+' ) {
if ( s[i+1]=='+' && !is_alnum_utf8(&s[i+2]) ) i += 2;
else if ( !is_alnum_utf8(&s[i+1]) ) i++;
}
// . c#, j#, ...
if ( s[i]=='#' && !is_alnum_utf8(&s[i+1]) ) i++;
// allow for words like we're dave's and i'm
if(s[i]=='\''&&s[i+1]&&is_alnum_utf8(&s[i+1])&&!hadApostrophe){
i++;
hadApostrophe = true;
goto again;
}
hadApostrophe = false;
// get word length
wlen = i - j;
if ( m_numWords >= m_preCount ) goto done;
m_words [ m_numWords ] = &s[j];
m_wordLens[ m_numWords ] = wlen;
// . Lars says it's better to leave the accented chars intact
// . google agrees
// . but what about "re'sume"?
if ( computeWordIds ) {
long long h = hash64Lower_utf8(&s[j],wlen);
m_wordIds [m_numWords] = h;
// until we get an accent removal algo, comment this
// out and possibly use the query synonym pipeline
// to search without accents. MDW
//long long h2 = hash64AsciiLowerE(&s[j],wlen);
//if ( h2 != h ) m_stripWordIds [m_numWords] = h2;
//else m_stripWordIds [m_numWords] = 0LL;
//m_stripWordIds[m_numWords] = 0;
}
if (m_tagIds) m_tagIds[m_numWords] = 0;
m_numWords++;
m_numAlnumWords++;
// break on \0 or MAX_WORDS
//if ( ! s[i] ) goto done;
// get a punct word
goto uptop;
/*
j = i;
// delineate the "punctuation" word
for ( ; s[i] && !is_alnum_utf8(&s[i]);i+=getUtf8CharSize(s+i));
// bad utf8 could cause us to breach the node, so watch out!
if ( i > nodeLen ) {
badCount++;
i = nodeLen;
}
// get word length
wlen = i - j;
if ( m_numWords >= m_preCount ) goto done;
m_words [m_numWords ] = &s[j];
m_wordLens [m_numWords ] = wlen;
m_wordIds [m_numWords ] = 0LL;
if (m_tagIds) m_tagIds[m_numWords] = 0;
m_numWords++;
*/
done:
// bad programming warning
if ( m_numWords > m_preCount ) {
log(LOG_LOGIC,
"build: words: set: Fix counting routine.");
char *xx = NULL; *xx = 0;
}
// compute total length
if ( m_numWords <= 0 ) m_totalLen = 0;
else m_totalLen = m_words[m_numWords-1] - s + m_wordLens[m_numWords-1];
if ( badCount )
log("words: had %li bad utf8 chars",badCount);
return true;
}
bool Words::addWords( char *s, int32_t nodeLen, bool computeWordIds ) {
int32_t i = 0;
int32_t j;
int32_t wlen;
bool hadApostrophe = false;
UCScript oldScript = ucScriptCommon;
UCScript saved;
UCProps props;
uptop:
// bad utf8 can cause a breach
if ( i >= nodeLen ) {
goto done;
}
if ( ! s[i] ) {
goto done;
}
if ( !is_alnum_utf8( s + i ) ) {
if ( m_numWords >= m_preCount ) {
goto done;
}
// tag?
if ( s[i]=='<' && m_hasTags && isTagStart(s+i) ) {
// get the tag id
if( m_tagIds ) {
if ( s[i + 1] == '/' ) {
// skip over /
m_tagIds[m_numWords] = ::getTagId( s + i + 2 );
m_tagIds[m_numWords] |= BACKBIT;
} else {
m_tagIds[m_numWords] = ::getTagId( s + i + 1 );
}
}
m_words[m_numWords] = s + i;
m_wordIds[m_numWords] = 0LL;
// skip till end
int32_t tagLen = getTagLen( s + i );
m_wordLens[m_numWords] = tagLen;
m_nodes[m_numWords] = 0;
m_numWords++;
// advance
i += tagLen;
goto uptop;
}
// it is a punct word, find end of it
char *start = s+i;
for ( ; s[i] ; i += getUtf8CharSize(s+i)) {
// stop on < if we got tags
if ( s[i] == '<' && m_hasTags ) {
break;
}
// if we are simple ascii, skip quickly
if ( is_ascii(s[i]) ) {
// accumulate NON-alnum chars
if ( ! is_alnum_a(s[i]) ) {
continue;
}
// update
oldScript = ucScriptCommon;
// otherwise, stop we got alnum
break;
}
// if we are utf8 we stop on special props
UChar32 c = utf8Decode ( s+i );
// stop if word char
if ( ! ucIsWordChar ( c ) ) {
continue;
}
// update first though
oldScript = ucGetScript ( c );
// then stop
break;
}
m_words [ m_numWords ] = start;
m_wordLens [ m_numWords ] = s+i - start;
m_wordIds [ m_numWords ] = 0LL;
m_nodes [ m_numWords ] = 0;
if (m_tagIds) {
m_tagIds[m_numWords] = 0;
}
m_numWords++;
goto uptop;
}
// get an alnum word
j = i;
again:
for ( ; s[i] ; i += getUtf8CharSize(s+i) ) {
// simple ascii?
if ( is_ascii(s[i]) ) {
// accumulate alnum chars
if ( is_alnum_a(s[i]) ) continue;
// update
oldScript = ucScriptCommon;
// otherwise, stop we got punct
break;
}
// get the code point of the utf8 char
UChar32 c = utf8Decode ( s+i );
// get props
props = ucProperties ( c );
// good stuff?
if ( props & (UC_IGNORABLE|UC_EXTEND) ) continue;
// stop? if UC_WORCHAR is set, that means its an alnum
if ( ! ( props & UC_WORDCHAR ) ) {
// reset script between words
oldScript = ucScriptCommon;
break;
}
// save it
saved = oldScript;
// update here
oldScript = ucGetScript(c);
// treat ucScriptLatin (30) as common so we can have latin1
// like char without breaking the word!
if ( oldScript == ucScriptLatin ) oldScript = ucScriptCommon;
// stop on this crap too i guess. like japanes chars?
if ( props & ( UC_IDEOGRAPH | UC_HIRAGANA | UC_THAI ) ) {
// include it
i += getUtf8CharSize(s+i);
// but stop
break;
}
// script change?
if ( saved != oldScript ) break;
}
// . java++, A++, C++ exception
// . A+, C+, exception
// . TODO: consider putting in Bits.cpp w/ D_CAN_BE_IN_PHRASE
if ( s[i]=='+' ) {
if ( s[i+1]=='+' && !is_alnum_utf8(&s[i+2]) ) i += 2;
else if ( !is_alnum_utf8(&s[i+1]) ) i++;
}
// . c#, j#, ...
if ( s[i]=='#' && !is_alnum_utf8(&s[i+1]) ) i++;
// comma is ok if like ,ddd!d
if ( s[i]==',' &&
i-j <= 3 &&
is_digit(s[i-1]) ) {
// if word so far is 2 or 3 chars, make sure digits
if ( i-j >= 2 && ! is_digit(s[i-2]) ) goto nogo;
if ( i-j >= 3 && ! is_digit(s[i-3]) ) goto nogo;
// scan forward
while ( s[i] == ',' &&
is_digit(s[i+1]) &&
is_digit(s[i+2]) &&
is_digit(s[i+3]) &&
! is_digit(s[i+4]) ) {
i += 4;
}
}
// decimal point?
if ( s[i] == '.' &&
is_digit(s[i-1]) &&
is_digit(s[i+1]) ) {
// allow the decimal point
i++;
// skip over string of digits
while ( is_digit(s[i]) ) i++;
}
nogo:
// allow for words like we're dave's and i'm
if ( s[i] == '\'' && s[i + 1] && is_alnum_utf8( &s[i + 1] ) && !hadApostrophe ) {
i++;
hadApostrophe = true;
goto again;
}
hadApostrophe = false;
// get word length
wlen = i - j;
if ( m_numWords >= m_preCount ) goto done;
m_words [ m_numWords ] = &s[j];
m_wordLens[ m_numWords ] = wlen;
if ( computeWordIds ) {
int64_t h = hash64Lower_utf8(&s[j],wlen);
m_wordIds [m_numWords] = h;
}
m_nodes[m_numWords] = 0;
if (m_tagIds) m_tagIds[m_numWords] = 0;
m_numWords++;
m_numAlnumWords++;
// get a punct word
goto uptop;
done:
// bad programming warning
if ( m_numWords > m_preCount ) {
log(LOG_LOGIC, "build: words: set: Fix counting routine.");
gbshutdownLogicError();
}
return true;
}
