Phrase fix and other improvements (#584)

* Fix phrase levenshtein part

* Fix std::swap UB and remove bad hash

* Make filter/prepared smaller, important for removes

core/analysis/analyzers.cpp

@@ -65,17 +65,14 @@ struct value {
 
 }  // namespace
 
-namespace std {
-
 template<>
-struct hash<::key> {
+struct std::hash<::key> {
   size_t operator()(const ::key& value) const noexcept {
     return irs::hash_combine(
       std::hash<irs::type_info::type_id>()(value.args_format.id()), value.type);
   }
 };
 
-}  // namespace std
 namespace irs::analysis {
 namespace {
 

core/analysis/pipeline_token_stream.hpp

@@ -70,22 +70,15 @@ class pipeline_token_stream final : public TypedAnalyzer<pipeline_token_stream>,
   template<typename Visitor>
   bool visit_members(Visitor&& visitor) const {
     for (const auto& sub : pipeline_) {
-      if (sub.get_stream().type() ==
-          type()) {  // pipe inside pipe - forward visiting
-#if IRESEARCH_DEBUG
-        const auto& sub_pipe =
-          dynamic_cast<const pipeline_token_stream&>(sub.get_stream());
-#else
-        const auto& sub_pipe =
-          static_cast<const pipeline_token_stream&>(sub.get_stream());
-#endif
+      const auto& stream = sub.get_stream();
+      if (stream.type() == type()) {
+        // pipe inside pipe - forward visiting
+        const auto& sub_pipe = DownCast<pipeline_token_stream>(stream);
         if (!sub_pipe.visit_members(visitor)) {
           return false;
         }
-      } else {
-        if (!visitor(sub.get_stream())) {
-          return false;
-        }
+      } else if (!visitor(sub.get_stream())) {
+        return false;
       }
     }
     return true;

core/search/all_docs_provider.cpp

@@ -26,7 +26,7 @@
 
 namespace irs {
 
-filter::ptr AllDocsProvider::Default(score_t boost) {
+AllDocsProvider::Ptr AllDocsProvider::Default(score_t boost) {
   auto filter = std::make_unique<all>();
   filter->boost(boost);
   return filter;

core/search/all_docs_provider.hpp

@@ -28,13 +28,12 @@ namespace irs {
 
 class AllDocsProvider {
  public:
-  using ProviderFunc = std::function<filter::ptr(score_t)>;
+  using Ptr = std::unique_ptr<FilterWithBoost>;
+  using ProviderFunc = std::function<Ptr(score_t)>;
 
-  static filter::ptr Default(score_t boost);
+  static Ptr Default(score_t boost);
 
-  filter::ptr MakeAllDocsFilter(score_t boost) const {
-    return all_docs_(boost);
-  }
+  Ptr MakeAllDocsFilter(score_t boost) const { return all_docs_(boost); }
 
   void SetProvider(ProviderFunc&& provider) {
     all_docs_ = provider ? std::move(provider) : ProviderFunc{&Default};

core/search/all_filter.cpp

@@ -30,21 +30,24 @@ namespace irs {
 class all_query : public filter::prepared {
  public:
   explicit all_query(bstring&& stats, score_t boost)
-    : filter::prepared(boost), stats_(std::move(stats)) {}
+    : stats_{std::move(stats)}, boost_{boost} {}
 
   doc_iterator::ptr execute(const ExecutionContext& ctx) const final {
     auto& rdr = ctx.segment;
 
     return memory::make_managed<AllIterator>(rdr, stats_.c_str(), ctx.scorers,
-                                             rdr.docs_count(), boost());
+                                             rdr.docs_count(), boost_);
   }
 
   void visit(const SubReader&, PreparedStateVisitor&, score_t) const final {
     // No terms to visit
   }
 
+  score_t boost() const noexcept final { return boost_; }
+
  private:
   bstring stats_;
+  score_t boost_;
 };
 
 filter::prepared::ptr all::prepare(const PrepareContext& ctx) const {

core/search/all_filter.hpp

@@ -27,9 +27,9 @@
 namespace irs {
 
 // Filter returning all documents
-class all : public filter {
+class all : public FilterWithBoost {
  public:
-  filter::prepared::ptr prepare(const PrepareContext& ctx) const final;
+  prepared::ptr prepare(const PrepareContext& ctx) const final;
 
   irs::type_info::type_id type() const noexcept final {
     return irs::type<all>::id();

core/search/boolean_filter.cpp

@@ -22,55 +22,36 @@
 
 #include "boolean_filter.hpp"
 
-#include <boost/container_hash/hash.hpp>
-
 #include "conjunction.hpp"
 #include "disjunction.hpp"
 #include "exclusion.hpp"
 #include "min_match_disjunction.hpp"
 #include "prepared_state_visitor.hpp"
 #include "search/boolean_query.hpp"
 
+namespace irs {
 namespace {
 
-// first - pointer to the innermost not "not" node
-// second - collapsed negation mark
-std::pair<const irs::filter*, bool> optimize_not(const irs::Not& node) {
+std::pair<const filter*, bool> optimize_not(const Not& node) {
   bool neg = true;
-  const irs::filter* inner = node.filter();
-  while (inner && inner->type() == irs::type<irs::Not>::id()) {
+  const auto* inner = node.filter();
+  while (inner != nullptr && inner->type() == type<Not>::id()) {
     neg = !neg;
-    inner = static_cast<const irs::Not*>(inner)->filter();
+    inner = DownCast<Not>(inner)->filter();
   }
 
-  return std::make_pair(inner, neg);
+  return std::pair{inner, neg};
 }
 
 }  // namespace
-
-namespace irs {
-
-size_t boolean_filter::hash() const noexcept {
-  size_t seed = 0;
-
-  ::boost::hash_combine(seed, filter::hash());
-  std::for_each(
-    filters_.begin(), filters_.end(),
-    [&seed](const filter::ptr& f) { ::boost::hash_combine(seed, *f); });
-
-  return seed;
-}
-
 bool boolean_filter::equals(const filter& rhs) const noexcept {
   if (!filter::equals(rhs)) {
     return false;
   }
   const auto& typed_rhs = DownCast<boolean_filter>(rhs);
-  return filters_.size() == typed_rhs.size() &&
-         std::equal(begin(), end(), typed_rhs.begin(),
-                    [](const filter::ptr& lhs, const filter::ptr& rhs) {
-                      return *lhs == *rhs;
-                    });
+  return std::equal(
+    begin(), end(), typed_rhs.begin(), typed_rhs.end(),
+    [](const auto& lhs, const auto& rhs) { return *lhs == *rhs; });
 }
 
 filter::prepared::ptr boolean_filter::prepare(const PrepareContext& ctx) const {
@@ -94,8 +75,8 @@ filter::prepared::ptr boolean_filter::prepare(const PrepareContext& ctx) const {
   std::vector<const filter*> incl;
   std::vector<const filter*> excl;
 
-  irs::filter::ptr all_docs_zero_boost;
-  irs::filter::ptr all_docs_no_boost;
+  AllDocsProvider::Ptr all_docs_zero_boost;
+  AllDocsProvider::Ptr all_docs_no_boost;
 
   group_filters(all_docs_zero_boost, incl, excl);
 
@@ -108,29 +89,29 @@ filter::prepared::ptr boolean_filter::prepare(const PrepareContext& ctx) const {
   return PrepareBoolean(incl, excl, ctx);
 }
 
-void boolean_filter::group_filters(filter::ptr& all_docs_zero_boost,
+void boolean_filter::group_filters(AllDocsProvider::Ptr& all_docs_zero_boost,
                                    std::vector<const filter*>& incl,
                                    std::vector<const filter*>& excl) const {
   incl.reserve(size() / 2);
   excl.reserve(incl.capacity());
 
-  const irs::filter* empty_filter{nullptr};
+  const filter* empty_filter = nullptr;
   const auto is_or = type() == irs::type<Or>::id();
-  for (auto begin = this->begin(), end = this->end(); begin != end; ++begin) {
-    if (irs::type<irs::empty>::id() == (*begin)->type()) {
-      empty_filter = begin->get();
+  for (const auto& filter : *this) {
+    if (irs::type<Empty>::id() == filter->type()) {
+      empty_filter = filter.get();
       continue;
     }
-    if (irs::type<Not>::id() == (*begin)->type()) {
-      const auto res = optimize_not(DownCast<Not>(**begin));
+    if (irs::type<Not>::id() == filter->type()) {
+      const auto res = optimize_not(DownCast<Not>(*filter));
 
       if (!res.first) {
         continue;
       }
 
       if (res.second) {
         if (!all_docs_zero_boost) {
-          all_docs_zero_boost = MakeAllDocsFilter(0.f);
+          all_docs_zero_boost = MakeAllDocsFilter(0.F);
         }
 
         if (*all_docs_zero_boost == *res.first) {
@@ -148,7 +129,7 @@ void boolean_filter::group_filters(filter::ptr& all_docs_zero_boost,
         incl.push_back(res.first);
       }
     } else {
-      incl.push_back(begin->get());
+      incl.push_back(filter.get());
     }
   }
   if (empty_filter != nullptr) {
@@ -162,7 +143,7 @@ filter::prepared::ptr And::PrepareBoolean(std::vector<const filter*>& incl,
   // optimization step
   //  if include group empty itself or has 'empty' -> this whole conjunction is
   //  empty
-  if (incl.empty() || incl.back()->type() == irs::type<irs::empty>::id()) {
+  if (incl.empty() || incl.back()->type() == irs::type<Empty>::id()) {
     return prepared::empty();
   }
 
@@ -180,7 +161,7 @@ filter::prepared::ptr And::PrepareBoolean(std::vector<const filter*>& incl,
   for (auto filter : incl) {
     if (*filter == *cumulative_all) {
       all_count++;
-      all_boost += filter->boost();
+      all_boost += DownCast<FilterWithBoost>(*filter).boost();
     }
   }
   if (all_count != 0) {
@@ -204,7 +185,7 @@ filter::prepared::ptr And::PrepareBoolean(std::vector<const filter*>& incl,
       // resulting boost will be: new_boost * OR_BOOST * LEFT_BOOST. If we
       // substitute new_boost back we will get ( boost * OR_BOOST * ALL_BOOST +
       // boost * OR_BOOST * LEFT_BOOST) - original non-optimized boost value
-      auto left_boost = (*incl.begin())->boost();
+      auto left_boost = (*incl.begin())->BoostImpl();
       if (boost() != 0 && left_boost != 0 && !sub_ctx.scorers.empty()) {
         sub_ctx.boost = (sub_ctx.boost * boost() * all_boost +
                          sub_ctx.boost * boost() * left_boost) /
@@ -247,7 +228,7 @@ filter::prepared::ptr Or::PrepareBoolean(std::vector<const filter*>& incl,
     return MakeAllDocsFilter(kNoBoost)->prepare(sub_ctx);
   }
 
-  if (!incl.empty() && incl.back()->type() == irs::type<irs::empty>::id()) {
+  if (!incl.empty() && incl.back()->type() == irs::type<Empty>::id()) {
     incl.pop_back();
   }
 
@@ -270,7 +251,7 @@ filter::prepared::ptr Or::PrepareBoolean(std::vector<const filter*>& incl,
   for (auto filter : incl) {
     if (*filter == *cumulative_all) {
       all_count++;
-      all_boost += filter->boost();
+      all_boost += DownCast<FilterWithBoost>(*filter).boost();
       incl_all = filter;
     }
   }
@@ -353,15 +334,6 @@ filter::prepared::ptr Not::prepare(const PrepareContext& ctx) const {
   return res.first->prepare(sub_ctx);
 }
 
-size_t Not::hash() const noexcept {
-  size_t seed = 0;
-  ::boost::hash_combine(seed, filter::hash());
-  if (filter_) {
-    ::boost::hash_combine<const irs::filter&>(seed, *filter_);
-  }
-  return seed;
-}
-
 bool Not::equals(const irs::filter& rhs) const noexcept {
   if (!filter::equals(rhs)) {
     return false;