inline SearchBuffer::SearchBuffer(const String& target, FindOptions options)
: m_options(options)
, m_prefixLength(0)
, m_numberOfCharactersJustAppended(0)
, m_atBreak(true)
, m_needsMoreContext(options & AtWordStarts)
, m_targetRequiresKanaWorkaround(containsKanaLetters(target))
{
ASSERT(!target.isEmpty());
target.appendTo(m_target);
// FIXME: We'd like to tailor the searcher to fold quote marks for us instead
// of doing it in a separate replacement pass here, but ICU doesn't offer a way
// to add tailoring on top of the locale-specific tailoring as of this writing.
foldQuoteMarksAndSoftHyphens(m_target.data(), m_target.size());
size_t targetLength = m_target.size();
m_buffer.reserveInitialCapacity(std::max(targetLength * 8, minimumSearchBufferSize));
m_overlap = m_buffer.capacity() / 4;
if ((m_options & AtWordStarts) && targetLength) {
UChar32 targetFirstCharacter;
U16_GET(m_target.data(), 0, 0, targetLength, targetFirstCharacter);
// Characters in the separator category never really occur at the beginning of a word,
// so if the target begins with such a character, we just ignore the AtWordStart option.
if (isSeparator(targetFirstCharacter)) {
m_options &= ~AtWordStarts;
m_needsMoreContext = false;
}
}
// Grab the single global searcher.
// If we ever have a reason to do more than once search buffer at once, we'll have
// to move to multiple searchers.
lockSearcher();
UStringSearch* searcher = blink::searcher();
UCollator* collator = usearch_getCollator(searcher);
UCollationStrength strength = m_options & CaseInsensitive ? UCOL_PRIMARY : UCOL_TERTIARY;
if (ucol_getStrength(collator) != strength) {
ucol_setStrength(collator, strength);
usearch_reset(searcher);
}
UErrorCode status = U_ZERO_ERROR;
usearch_setPattern(searcher, m_target.data(), targetLength, &status);
ASSERT(status == U_ZERO_ERROR);
// The kana workaround requires a normalized copy of the target string.
if (m_targetRequiresKanaWorkaround)
normalizeCharactersIntoNFCForm(m_target.data(), m_target.size(), m_normalizedTarget);
}
inline bool SearchBuffer::isBadMatch(const UChar* match, size_t matchLength) const
{
// This function implements the kana workaround. If usearch treats
// it as a match, but we do not want to, then it's a "bad match".
if (!m_targetRequiresKanaWorkaround)
return false;
// Normalize into a match buffer. We reuse a single buffer rather than
// creating a new one each time.
normalizeCharactersIntoNFCForm(match, matchLength, m_normalizedMatch);
return !checkOnlyKanaLettersInStrings(m_normalizedTarget.begin(), m_normalizedTarget.size(), m_normalizedMatch.begin(), m_normalizedMatch.size());
}
inline SearchBuffer::SearchBuffer(const String& target, FindOptions options)
: m_options(options),
m_prefixLength(0),
m_numberOfCharactersJustAppended(0),
m_atBreak(true),
m_needsMoreContext(options & AtWordStarts),
m_targetRequiresKanaWorkaround(containsKanaLetters(target)) {
DCHECK(!target.isEmpty()) << target;
target.appendTo(m_target);
// FIXME: We'd like to tailor the searcher to fold quote marks for us instead
// of doing it in a separate replacement pass here, but ICU doesn't offer a
// way to add tailoring on top of the locale-specific tailoring as of this
// writing.
foldQuoteMarksAndSoftHyphens(m_target.data(), m_target.size());
size_t targetLength = m_target.size();
m_buffer.reserveInitialCapacity(
std::max(targetLength * 8, kMinimumSearchBufferSize));
m_overlap = m_buffer.capacity() / 4;
if ((m_options & AtWordStarts) && targetLength) {
const UChar32 targetFirstCharacter =
getCodePointAt(m_target.data(), 0, targetLength);
// Characters in the separator category never really occur at the beginning
// of a word, so if the target begins with such a character, we just ignore
// the AtWordStart option.
if (isSeparator(targetFirstCharacter)) {
m_options &= ~AtWordStarts;
m_needsMoreContext = false;
}
}
m_textSearcher = WTF::makeUnique<TextSearcherICU>();
m_textSearcher->setPattern(StringView(m_target.data(), m_target.size()),
!(m_options & CaseInsensitive));
// The kana workaround requires a normalized copy of the target string.
if (m_targetRequiresKanaWorkaround)
normalizeCharactersIntoNFCForm(m_target.data(), m_target.size(),
m_normalizedTarget);
}