Fix

core/search/ngram_similarity_filter.cpp

@@ -26,6 +26,7 @@
 #include "index/index_reader.hpp"
 #include "search/collectors.hpp"
 #include "search/ngram_similarity_query.hpp"
+#include "search/phrase_filter.hpp"
 #include "search/terms_filter.hpp"
 #include "shared.hpp"
 
@@ -36,25 +37,37 @@ filter::prepared::ptr by_ngram_similarity::Prepare(
  const std::vector<irs::bstring>& ngrams, float_t threshold) {
  if (ngrams.empty() || field_name.empty()) {
  // empty field or terms or invalid threshold
- return filter::prepared::empty();
+ return prepared::empty();
  }
+ const auto terms_count = ngrams.size();
 
  threshold = std::clamp(threshold, 0.f, 1.f);
  const auto min_match_count =
- std::clamp(static_cast<size_t>(std::ceil(ngrams.size() * threshold)),
- size_t{1}, ngrams.size());
+ std::clamp(static_cast<size_t>(std::ceil(terms_count * threshold)),
+ size_t{1}, terms_count);
  if (ctx.scorers.empty() && 1 == min_match_count) {
  irs::by_terms disj;
- for (auto& terms = disj.mutable_options()->terms; auto& term : ngrams) {
+ auto& terms = disj.mutable_options()->terms;
+ for (const auto& term : ngrams) {
  terms.emplace(term, irs::kNoBoost);
  }
  *disj.mutable_field() = field_name;
  return disj.prepare(ctx);
  }
+
+ if (min_match_count == terms_count) {
+ irs::by_phrase phrase;
+ auto& options = *phrase.mutable_options();
+ for (size_t i = 0; const auto& ngram : ngrams) {
+ options.insert(by_term_options{ngram}, i++);
+ }
+ *phrase.mutable_field() = field_name;
+ return phrase.prepare(ctx);
+ }
+
  NGramStates query_states{ctx.memory, ctx.index.size()};
 
  // per segment terms states
- const auto terms_count = ngrams.size();
  ManagedVector<seek_cookie::ptr> term_states{{ctx.memory}};
  term_states.reserve(terms_count);
 
@@ -81,10 +94,9 @@ filter::prepared::ptr by_ngram_similarity::Prepare(
  field_stats.collect(segment, *field);
  size_t term_idx = 0;
  size_t count_terms = 0;
- seek_term_iterator::ptr term = field->iterator(SeekMode::NORMAL);
+ auto term = field->iterator(SeekMode::NORMAL);
  for (const auto& ngram : ngrams) {
- term_states.emplace_back();
- auto& state = term_states.back();
+ auto& state = term_states.emplace_back();
  if (term->seek(ngram)) {
  // read term attributes
  term->read();
@@ -110,6 +122,10 @@ filter::prepared::ptr by_ngram_similarity::Prepare(
  term_states.reserve(terms_count);
  }
 
+ if (query_states.empty()) {
+ return prepared::empty();
+ }
+
  bstring stats(ctx.scorers.stats_size(), 0);
  auto* stats_buf = stats.data();
 

core/search/phrase_filter.cpp

@@ -278,6 +278,10 @@ filter::prepared::ptr by_phrase::fixed_prepare_collect(
  phrase_terms.reserve(phrase_size);
  }
 
+ if (phrase_states.empty()) {
+ return prepared::empty();
+ }
+
  // offset of the first term in a phrase
  IRS_ASSERT(!options().empty());
  const size_t base_offset = options().begin()->first;
@@ -392,6 +396,10 @@ filter::prepared::ptr by_phrase::variadic_prepare_collect(
  num_terms.resize(phrase_size);
  }
 
+ if (phrase_states.empty()) {
+ return prepared::empty();
+ }
+
  // offset of the first term in a phrase
  IRS_ASSERT(!options().empty());
  const size_t base_offset = options().begin()->first;