Merge branch 'f-977-hash-table-size-m'. Closes krlmlr#17. Closes #977.

- Using larger hash tables gives slightly better performance for `n_distinct()` and ordering of character vectors (#977).

inst/include/dplyr/Result/Count_Distinct.h

@@ -17,11 +17,12 @@ namespace dplyr {
     typedef dplyr_hash_set<int, Hash, Pred > Set;
 
     Count_Distinct(Visitor v_):
-      v(v_), set(1024, Hash(v), Pred(v))
+      v(v_), set(0, Hash(v), Pred(v))
     {}
 
     inline int process_chunk(const SlicingIndex& indices) {
       set.clear();
+      set.rehash(indices.size());
       int n = indices.size();
       for (int i=0; i<n; i++) {
         set.insert(indices[i]);
@@ -42,11 +43,12 @@ namespace dplyr {
     typedef dplyr_hash_set<int, Hash, Pred > Set;
 
     Count_Distinct_Narm(Visitor v_):
-      v(v_), set(1024, Hash(v), Pred(v))
+      v(v_), set(0, Hash(v), Pred(v))
     {}
 
     inline int process_chunk(const SlicingIndex& indices) {
       set.clear();
+      set.rehash(indices.size());
       int n = indices.size();
       for (int i=0; i<n; i++) {
         int index=indices[i];

inst/include/dplyr/Result/In.h

@@ -13,10 +13,9 @@ namespace dplyr {
     typedef typename Rcpp::Vector<RTYPE> Vec;
     typedef typename Rcpp::traits::storage_type<RTYPE>::type STORAGE;
 
-    In(Vec data_, Vec table_) :
+    In(Vec data_, const Vec& table_) :
       data(data_),
-      table(table_),
-      set(table.begin(), table.end())
+      set(table_.begin(), table_.end())
     {}
 
     void process_slice(LogicalVector& out, const SlicingIndex& index, const SlicingIndex& out_index) {
@@ -32,7 +31,7 @@ namespace dplyr {
     }
 
   private:
-    Vec data, table;
+    Vec data;
     dplyr_hash_set<STORAGE> set;
 
   };

src/api.cpp

@@ -293,7 +293,7 @@ namespace dplyr {
 
   CharacterVectorOrderer::CharacterVectorOrderer(const CharacterVector& data_) :
     data(data_),
-    set(),
+    set(data.size()),
     orders(no_init(data.size()))
   {
     int n = data.size();