Improving IdentifierCompleter performance

We limit the number of candidates returned to Vim to 20 and also make sure that
we are not returning any duplicate candidates. This provides a noticeable
improvement in latency.

cpp/ycm/CandidateRepository.cpp

@@ -43,6 +43,13 @@ CandidateRepository& CandidateRepository::Instance()
 }
 
 
+int CandidateRepository::NumStoredCandidates()
+{
+  boost::lock_guard< boost::mutex > locker( holder_mutex_ );
+  return candidate_holder_.size();
+}
+
+
 std::vector< const Candidate* > CandidateRepository::GetCandidatesForStrings(
     const std::vector< std::string > &strings )
 {

cpp/ycm/CandidateRepository.h

@@ -39,6 +39,8 @@ class CandidateRepository : boost::noncopyable
 public:
   static CandidateRepository& Instance();
 
+  int NumStoredCandidates();
+
   std::vector< const Candidate* > GetCandidatesForStrings(
       const std::vector< std::string > &strings );
 

cpp/ycm/ClangCompleter.cpp

@@ -128,6 +128,8 @@ bool IsChunkKindForExtraMenuInfo( CXCompletionChunkKind kind )
 
 char CursorKindToVimKind( CXCursorKind kind )
 {
+  // TODO: actually it appears that Vim will show returned kinds even when they
+  // do not match the "approved" list, so let's use that
   switch ( kind )
   {
     case CXCursor_UnexposedDecl:

cpp/ycm/IdentifierCompleter.cpp

@@ -21,6 +21,7 @@
 #include "Candidate.h"
 #include "Utils.h"
 
+#include <boost/unordered_set.hpp>
 #include <boost/bind.hpp>
 #include <boost/make_shared.hpp>
 #include <algorithm>
@@ -173,11 +174,19 @@ void IdentifierCompleter::ResultsForQueryAndType(
 
   Bitset query_bitset = LetterBitsetFromString( query );
 
+  boost::unordered_set< const Candidate* > seen_candidates;
+  seen_candidates.reserve( candidate_repository_.NumStoredCandidates() );
+
   foreach ( const FilepathToCandidates::value_type &path_and_candidates,
             *it->second )
   {
     foreach ( const Candidate* candidate, *path_and_candidates.second )
     {
+      if ( ContainsKey( seen_candidates, candidate ) )
+        continue;
+      else
+        seen_candidates.insert( candidate );
+
       if ( !candidate->MatchesQueryBitset( query_bitset ) )
         continue;
 

cpp/ycm/tests/IdentifierCompleter_test.cpp

@@ -73,6 +73,15 @@ TEST( IdentifierCompleterTest, EmptyQueryNoResults )
 	             ElementsAre() );
 }
 
+TEST( IdentifierCompleterTest, NoDuplicatesReturned )
+{
+	EXPECT_THAT( IdentifierCompleter( Candidates(
+               "foobar",
+               "foobar",
+               "foobar" ) ).CandidatesForQuery( "foo" ),
+	             ElementsAre( "foobar" ) );
+}
+
 
 TEST( IdentifierCompleterTest, OneCandidate )
 {

python/ycm.py

@@ -21,8 +21,10 @@ import re
 import vim
 import indexer
 
-min_num_chars = int( vim.eval( "g:ycm_min_num_of_chars_for_completion" ) )
+MIN_NUM_CHARS = int( vim.eval( "g:ycm_min_num_of_chars_for_completion" ) )
-clang_filetypes = set( [ 'c', 'cpp', 'objc', 'objcpp' ] )
+CLANG_FILETYPES = set( [ 'c', 'cpp', 'objc', 'objcpp' ] )
+MAX_IDENTIFIER_COMPLETIONS_RETURNED = 20
+
 
 class Completer( object ):
   def __init__( self ):
@@ -89,6 +91,11 @@ class IdentifierCompleter( Completer ):
                                             filepath,
                                             True )
 
+  def CandidatesFromStoredRequest( self ):
+    if not self.future:
+      return []
+    return self.future.GetResults()[ : MAX_IDENTIFIER_COMPLETIONS_RETURNED ]
+
 
 class ClangCompleter( Completer ):
   def __init__( self ):
@@ -181,7 +188,7 @@ def CurrentLineAndColumn():
 
 def ShouldUseClang( start_column ):
   filetype = vim.eval( "&filetype" )
-  if filetype not in clang_filetypes:
+  if filetype not in CLANG_FILETYPES:
     return False
 
   line = vim.current.line
@@ -253,7 +260,7 @@ def PreviousIdentifier():
   while start_column > 0 and IsIdentifierChar( line[ start_column - 1 ] ):
     start_column -= 1
 
-  if end_column - start_column < min_num_chars:
+  if end_column - start_column < MIN_NUM_CHARS:
     return ""
 
   return line[ start_column : end_column ]