Added murmur hash and finished porting all c code.

cxxmph/Makefile.am

@@ -1,8 +1,11 @@
-bin_PROGRAMS = cmph_hash_map_test
+bin_PROGRAMS = cmph_hash_map_test mphtable_test
 lib_LTLIBRARIES = libcxxmph.la
 
-libcxxmph_la_SOURCES = trigragh.h trigraph.cc
+libcxxmph_la_SOURCES = stringpiece.h MurmurHash2.h randomly_seeded_hash.h trigragh.h trigraph.cc mphtable.h mphtable.cc
 libcxxmph_la_LDFLAGS = -version-info 0:0:0
 
 cmph_hash_map_test_LDADD   = libcxxmph.la
 cmph_hash_map_test_SOURCES = cmph_hash_map_test.cc
+
+mphtable_test_LDADD   = libcxxmph.la
+mphtable_test_SOURCES = mphtable_test.cc

cxxmph/MurmurHash2.h

@@ -0,0 +1,64 @@
+//-----------------------------------------------------------------------------
+// MurmurHash2, by Austin Appleby
+
+// Note - This code makes a few assumptions about how your machine behaves -
+
+// 1. We can read a 4-byte value from any address without crashing
+// 2. sizeof(int) == 4
+
+// And it has a few limitations -
+
+// 1. It will not work incrementally.
+// 2. It will not produce the same results on little-endian and big-endian
+//    machines.
+
+unsigned int MurmurHash2 ( const void * key, int len, unsigned int seed )
+{
+	// 'm' and 'r' are mixing constants generated offline.
+	// They're not really 'magic', they just happen to work well.
+
+	const unsigned int m = 0x5bd1e995;
+	const int r = 24;
+
+	// Initialize the hash to a 'random' value
+
+	unsigned int h = seed ^ len;
+
+	// Mix 4 bytes at a time into the hash
+
+	const unsigned char * data = (const unsigned char *)key;
+
+	while(len >= 4)
+	{
+		unsigned int k = *(unsigned int *)data;
+
+		k *= m; 
+		k ^= k >> r; 
+		k *= m; 
+		
+		h *= m; 
+		h ^= k;
+
+		data += 4;
+		len -= 4;
+	}
+	
+	// Handle the last few bytes of the input array
+
+	switch(len)
+	{
+	case 3: h ^= data[2] << 16;
+	case 2: h ^= data[1] << 8;
+	case 1: h ^= data[0];
+	        h *= m;
+	};
+
+	// Do a few final mixes of the hash to ensure the last few
+	// bytes are well-incorporated.
+
+	h ^= h >> 13;
+	h *= m;
+	h ^= h >> 15;
+
+	return h;
+} 

cxxmph/cmph_hash_map.h

@@ -2,8 +2,6 @@
 #include <vector>
 #include <utility>  // for std::pair
 
-#include <cmph.h>
-
 // Save on repetitive typing.
 #define CMPH_TMPL_SPEC template <class Key, class Data, class HashFcn, class EqualKey, class Alloc> 
 #define CMPH_CLASS_SPEC cmph_hash_map<Key, Data, HashFcn, EqualKey, Alloc>

cxxmph/mphtable.cc

@@ -1,105 +1,213 @@
-#include <numerical_limits>
+#include <limits>
 
 #include "mphtable.h"
 
 using std::vector;
 
+namespace cxxmph {
+
 template <class Key, class HashFcn>
 template <class ForwardIterator>
-
+bool MPHTable<Key, HashFcn>::Reset(ForwardIterator begin, ForwardIterator end) {
-void MPHTable::Reset(ForwardIterator begin, ForwardIterator end) {
+  TableBuilderState<ForwardIterator> st;
-  TableBuilderState st;
+  m_ = end - begin;
-  st.c = 1.23;
+  r_ = static_cast<cmph_uint32>(ceil((c_*m_)/3));
-  st.b = 7;
+  if (r_ % 2) == 0) r_ += 1;
-  st.m = end - begin;
+  n_ = 3*r_;
-  st.r = static_cast<cmph_uint32>(ceil((st.c*st.m)/3));
+  k_ = 1U << b_;
-  if ((st.r % 2) == 0) st.r += 1;
-  st.n = 3*st.r;
-  st.k = 1U << st.b;
-  st.ranktablesize = static_cast<cmph_uint32>(
-      ceil(st.n / static_cast<double>(st.k)));
-  st.graph_builder = TriGraph(st.m, st.n);  // giant copy
-  st.edges_queue.resize(st.m)
 
   int iterations = 1000;
   while (1) {
-    hasher hasher0 = HashFcn(); 
+    for (int i = 0; i < 3; ++i) hash_function_[i] = hasher();
-    ok = Mapping(st.graph_builder, st.edges_queue);
+    vector<Edge> edges;
-    if (ok) break;
+    vector<cmph_uint32> queue;
+    if (Mapping(begin, end, &edges, &queue)) break;
     else --iterations;
     if (iterations == 0) break;
   }
   if (iterations == 0) return false;
-  vector<ConnectedEdge> graph;
+  vector<Edge>& edges;
-  st.graph_builder.ExtractEdgesAndClear(&graph);
+  graph->ExtractEdgesAndClear(&edges);
-  Assigning(graph, st.edges_queue);
+  Assigning(queue, edges);
-  vector<cmph_uint32>().swap(st.edges_queue);
+  vector<cmph_uint32>().swap(edges);
-  Ranking(graph);
+  Ranking();
   
 }
 
 template <class Key, class HashFcn>
-int MPHTable::GenerateQueue(
+bool MPHTable<Key, HashFcn>::GenerateQueue(
-    cmph_uint32 nedges, cmph_uint32 nvertices,
+    TriGraph* graph, vector<cmph_uint32>* queue_output) {
-    TriGraph* graph, Queue* queue) {
   cmph_uint32 queue_head = 0, queue_tail = 0;
+  cmph_uint32 nedges = n_;
+  cmph_uint32 nvertices = m_;
   // Relies on vector<bool> using 1 bit per element
   vector<bool> marked_edge((nedges >> 3) + 1, false);
-  queue->swap(Queue(nvertices, 0));
+  Queue queue(nvertices, 0);
   for (int i = 0; i < nedges; ++i) {
-    TriGraph::Edge e = graph.edges[i].vertices;
+    const TriGraph::Edge& e = graph->edges()[i];
-    if (graph.vertex_degree_[e.vertices[0]] == 1 ||
+    if (graph->vertex_degree()[e[0]] == 1 ||
-        graph.vertex_degree_[e.vertices[1]] == 1 ||
+        graph->vertex_degree()[e[1]] == 1 ||
-        graph.vertex_degree[e.vertices[2]] == 1) {
+        graph->vertex_degree()[e[2]] == 1) {
       if (!marked_edge[i]) {
-        (*queue)[queue_head++] = i;
+        queue[queue_head++] = i;
         marked_edge[i] = true;
       }
     }
   }
   while (queue_tail != queue_head) {
-    cmph_uint32 current_edge = (*queue)[queue_tail++];
+    cmph_uint32 current_edge = queue[queue_tail++];
     graph->RemoveEdge(current_edge);
-    TriGraph::Edge e = graph->edges[current_edge];
+    const TriGraph::Edge& e = graph->edges()[current_edge];
     for (int i = 0; i < 3; ++i) {
-      cmph_uint32 v = e.vertices[i];
+      cmph_uint32 v = e[i];
-      if (graph->vertex_degree[v] == 1) {
+      if (graph->vertex_degree()[v] == 1) {
-        cmph_uint32 first_edge = graph->first_edge_[v];
+        cmph_uint32 first_edge = graph->first_edge()[v];
-        if (!marked_edge[first_edge) {
+        if (!marked_edge[first_edge]) {
           queue[queue_head++] = first_edge;
           marked_edge[first_edge] = true;
         }
       }
     }
   }
-  vector<bool>().swap(marked_edge);
+  int cycles = queue_head - nedges;
-  return queue_head - nedges;
+  if (cycles == 0) queue.swap(*queue_output);
-}
-
-template <class Key, class HashFcn>
-int MPHTable::Mapping(TriGraph* graph, Queue* queue) {
-  int cycles = 0;
-  graph->Reset(m, n);
-  for (ForwardIterator it = begin_; it != end_; ++it) { 
-    cmph_uint32 hash_values[3];
-    for (int i = 0; i < 3; ++i) {
-      hash_values[i] = hasher_(*it);
-    }
-    cmph_uint32 v0 = hash_values[0] % bdz->r;
-    cmph_uint32 v1 = hash_values[1] % bdz->r + bdz->r;
-    cmph_uint32 v2 = hash_values[2] % bdz->r + (bdz->r << 1);
-    graph->AddEdge(Edge(v0, v1, v2));
-  }
-  cycles = GenerateQueue(bdz->m, bdz->n, queue, graph);
   return cycles == 0;
 }
 
-void MPHTable::Assigning(TriGraph* graph, Queue* queue) {
+template <class Key, class HashFcn>
-}
+template <class ForwardIterator>
-void MPHTable::Ranking(TriGraph* graph, Queue* queue) {
+bool MPHTable<Key, HashFcn>::Mapping(
-}
+    ForwardIterator begin, ForwardIterator end,
-cmph_uint32 MPHTable::Search(const key_type& key) {
+    vector<Edge>* edges, vector<cmph_uint32> queue) {
+  int cycles = 0;
+  TriGraph graph(m, n);
+  for (ForwardIterator it = begin; it != end; ++it) { 
+    cmph_uint32 h[3];
+    for (int i = 0; i < 3; ++i) h[i] = hash_function_[i](*it);
+    cmph_uint32 v0 = h[0] % r_;
+    cmph_uint32 v1 = h[1] % r_ + r_;
+    cmph_uint32 v2 = h[2] % r_ + (r_ << 1);
+    graph.AddEdge(Edge(v0, v1, v2));
+  }
+  if (GenerateQueue(&graph, queue)) {
+     graph.ExtractEdgesAndClear(edges);
+     return true;
+  }
+  return false;
 }
 
-cmph_uint32 MPHTable::Rank(const key_type& key) {
+template <class Key, class HashFcn>
+void MPHTable<Key, HashFcn>::Assigning(
+    const vector<Edge>& edges, const vector<cmph_uint32>& queue) {
+  cmph_uint32 nedges = n_;
+  cmph_uint32 current_edge = 0;
+  vector<bool> marked_vertices(nedges + 1);
+  // TODO(davi) use half nibbles instead
+  // vector<cmph_uint8> g(static_cast<cmph_uint32>(ceil(nedges / 4.0)),
+  //                     std::numerical_limits<cmph_uint8>::max());
+  static const cmph_uint8 kUnassigned = 3;
+  vector<cmph_uint8>(nedges, kUnassigned).swap(g_);
+  for (int i = nedges - 1; i + 1 >= 1; --i) {
+    current_edge = queue[i];
+    const TriGraph::Edge& e = edges[current_edge];
+    if (!marked_vertices[e[0]]) {
+      if (!marked_vertices[e[1]]) {
+        g_[e[1]] = kUnassigned;
+        marked_vertices[e[1]] = true;
+      }
+      if (!marked_vertices[e[2]]) {
+        g_[e[2]] = kUnassigned;
+        marked_vertices[e[2]] = true;
+      }
+      g_[e[0]] = (6 - g_[e[1]] + g_[e2]) % 3;
+      marked_vertices[e[0]] = true;
+    } else if (!marked_vertices[e[1]])) {
+      if (!marked_vertices[e[2]])) {
+        g_[e[2]] = kUnassigned;
+        marked_vertices[e[2]] = true;
+      }
+      g_[e[1]] = 7 - (g_[e[0]] + g_[e[2]]) % 3;
+      marked_vertices[e[1]] = true;
+    } else {
+      g_[e[2]] = (8 - g_[e[0]] + g_[e[1]]) % 3;
+      marked_vertices[e[2]] = true;
+    }
+  }
 }
+
+// table used for looking up the number of assigned vertices to a 8-bit integer
+static cmph_uint8 kBdzLookupTable[] =
+{
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 3, 3, 3, 3, 2,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 2, 2, 2, 1,
+2, 2, 2, 1, 2, 2, 2, 1, 2, 2, 2, 1, 1, 1, 1, 0
+};
+
+template <class Key, class HashFcn>
+void MPHTable<Key, HashFcn>::Ranking() {
+  cmph_uint32 nbytes_total = static_cast<cmph_uint32>(ceil(st->n / 4.0));
+  cmph_uint32 size = k_ >> 2U;
+  ranktablesize = static_cast<cmph_uint32>(ceil(n_ / static_cast<double>(k_)));
+  // TODO(davi) Change swap of member classes for resize + memset to avoid fragmentation
+  vector<cmph_uint32> (ranktablesize).swap(ranktable_);;
+  cmph_uint32 offset = 0;
+  cmph_uint32 count = 0;
+  cmph_uint32 i = 0;
+  while (1) {
+    if (i == ranktable.size()) break;
+    cmph_uint32 nbytes = size < nbytes_total ? size : nbytes_total;
+    for (j = 0; j < nbytes; ++j) count += kBdzLookupTable[g_[offset + j]];
+    ranktable_[i] = count;
+    offset += nbytes;
+    nbytes_total -= size;
+    ++i;
+  }
+}
+
+template <class Key, class HashFcn>
+cmph_uint32 MPHTable<Key, HashFcn>::Search(const key_type& key) const {
+  cmph_uint32 vertex;
+  cmph_uint32 h[3];
+  for (int i = 0; i < 3; ++i) h[i] = hash_function_[i](key);
+  h[0] = h[0] % st->r;
+  h[1] = h[1] % st->r + st->r;
+  h[2] = h[2] % st->r + (st->r << 1);
+  cmph_uint32 vertex = h[(h[g_[h[0]] + g_[h[1]] + g_[h[2]]) % 3];
+  return Rank(st->b, st->ranktable, vertex);
+}
+
+template <class Key, class HashFcn>
+cmph_uint32 MPHTable<Key, HashFcn>::Rank(cmph_uint32 vertex) const {
+  cmph_uint32 index = vertex >> b_;
+  cmph_uint32 base_rank = ranktable_[index];
+  cmph_uint32 beg_idx_v = index << b;
+  cmph_uint32 beg_idx_b = index >> 2 
+  cmph_uint32 end_idx_b = index >> 2 
+  while (beg_idx_b < end_idx_b) base_rank += kBdzLookupTable[g_[beg_idx_b++]];
+  beg_idx_v = beg_idx_b << 2;
+  while (beg_idx_v < vertex) {
+    if (g_[beg_idx_v) != kUnassigned) ++base_rank;
+    ++beg_idx_v;
+  }
+  return base_rank;
+}
+
+template <class Key, class HashFcn>
+cmph_uint32 MPHTable<Key, HashFcn>::index(const key_type& key) const {
+  return Search(key);
+}
+
+}  // namespace cxxmph

cxxmph/mphtable.h

@@ -1,15 +1,22 @@
+#ifndef __CXXMPH_MPHTABLE_H__
+#define __CXXMPH_MPHTABLE_H__
+
 // Minimal perfect hash abstraction implementing the BDZ algorithm
 
 #include <vector>
 
+#include "randomly_seeded_hash.h"
+#include "stringpiece.h"
 #include "trigraph.h"
 
-template <class Key, class NewRandomlySeededHashFcn = __gnu_cxx::hash<Key> >
+namespace cxxmph {
+
+template <class Key, class NewRandomlySeededHashFcn = RandomlySeededMurmur2>
 class MPHTable {
  public:
   typedef Key key_type;
   typedef NewRandomlySeededHashFcn hasher;
-  MPHTable();
+  MPHTable(double c = 1.23, cmph_uint8 b = 7) : c_(c), b_(b) { }
   ~MPHTable();
 
   template <class ForwardIterator>
@@ -17,28 +24,38 @@ class MPHTable {
   cmph_uint32 index(const key_type& x) const;
 
  private:
-  typedef std::vector<cmph_uint32> Queue;
+  template <class ForwardIterator>
-  template<class ForwardIterator>
+  bool Mapping(ForwardIterator begin, ForwardIterator end,
-  struct TableBuilderState {
+               vector<Edge>* edges, vector<cmph_uint32> queue);
-    ForwardIterator begin;
+  bool GenerateQueue(TriGraph* graph, vector<cmph_uint32>* queue);
-    ForwardIterator end;
+  void Assigning(TriGraph* graph_builder, Queue* queue);
-    Queue edges_queue;
+  void Ranking(TriGraph* graph_builder, Queue* queue);
-    TriGraph graph_builder;
+  cmph_uint32 Search(const StringPiece& key);
-    double c;
+  cmph_uint32 Rank(const StringPiece& key);
-    cmph_uint32 m;
-    cmph_uint32 n;
-    cmph_uint32 k;
-    cmph_uint32 ranktablesize;
-  };
-  int GenerateQueue(
-     cmph_uint32 nedges, cmph_uint32 nvertices,
-     TriGraph* graph, Queue* queue);
-   void Assigning(TriGraph* graph, Queue* queue);
-   void Ranking(TriGraph* graph, Queue* queue);
-   cmph_uint32 Search(const StringPiece& key);
-   cmph_uint32 Rank(const StringPiece& key);
 
-  std::vector<ConnectedEdge> graph_;
+  // Algorithm parameters
+  cmph_uint8 b_;  // Number of bits of the kth index in the ranktable
+  double c_;  // Number of bits per key (? is it right)
+
+  // Values used during generation
+  cmph_uint32 m_;  // edges count
+  cmph_uint32 n_;  // vertex count
+  cmph_uint32 k_  // kth index in ranktable, $k = log_2(n=3r)\varepsilon$
+
+  // Values used during search
+
+  // Partition vertex count, derived from c parameter.
+  cmph_uint32 r_;
+  // The array containing the minimal perfect hash function graph.
+  std::vector<cmph_uint8> g_;
+  // The table used for the rank step of the minimal perfect hash function
+  std::vector<cmph_uint32> ranktable_;
+  // The selected hash function triplet for finding the edges in the minimal
+  // perfect hash function graph.
+  hasher hash_function_[3];
+  
 };
 
+}  // namespace cxxmph
 
+#define // __CXXMPH_MPHTABLE_H__

cxxmph/mphtable_test.cc

@@ -0,0 +1,22 @@
+#include <cassert>
+#include <vector>
+
+#include "mphtable.h"
+
+using std::vector;
+using cxxmph::MPHTable;
+
+int main(int argc, char** argv) {
+  vector<int> keys;
+  keys.push_back(10);
+  keys.push_back(4);
+  keys.push_back(3);
+
+  MPHTable<int> mphtable;
+  assert(mphtable.Reset(keys.begin(), keys.end()));
+  vector<int> ids;
+  for (int i = 0; i < keys.size(); ++i) ids.push_back(mphtable.index(keys[i]));
+  sort(ids.begin(), ids.end());
+  for (int i = 0; i < ids.size(); ++i) assert(ids[i] == i);
+}
+  

cxxmph/randomly_seeded_hash.h

@@ -0,0 +1,24 @@
+#ifndef __CXXMPH_RANDOMLY_SEEDED_HASH__
+#define __CXXMPH_RANDOMLY_SEEDED_HASH__
+
+// Helper to create randomly seeded hash functions out of existing hash
+// functions that take a seed as a parameter.
+
+#include <cstdlib>
+
+#include "../src/cmph_types.h"
+#include "MurmurHash2.h"
+
+namespace cxxmph {
+
+struct RandomlySeededMurmur2 {
+  RandomlySeededHashFunction() : seed(random()) { }
+  cmph_uint32 operator()(const StringPiece& key) {
+    return MurmurHash2(key.data(), key.length(), seed);
+  }
+  cmph_uint32 seed;
+};
+
+}  // namespace cxxmph
+
+#endif  // __CXXMPH_RANDOMLY_SEEDED_HASH__

cxxmph/stringpiece.h

@@ -0,0 +1,177 @@
+// Copyright 2001-2010 The RE2 Authors.  All Rights Reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// A string-like object that points to a sized piece of memory.
+//
+// Functions or methods may use const StringPiece& parameters to accept either
+// a "const char*" or a "string" value that will be implicitly converted to
+// a StringPiece.  The implicit conversion means that it is often appropriate
+// to include this .h file in other files rather than forward-declaring
+// StringPiece as would be appropriate for most other Google classes.
+//
+// Systematic usage of StringPiece is encouraged as it will reduce unnecessary
+// conversions from "const char*" to "string" and back again.
+//
+//
+// Arghh!  I wish C++ literals were "string".
+
+#ifndef CXXMPH_STRINGPIECE_H__
+#define CXXMPH_STRINGPIECE_H__
+
+#include <string.h>
+#include <iosfwd>
+#include <string>
+
+namespace cxxmph {
+
+class StringPiece {
+ private:
+  const char*   ptr_;
+  int           length_;
+
+ public:
+  // We provide non-explicit singleton constructors so users can pass
+  // in a "const char*" or a "string" wherever a "StringPiece" is
+  // expected.
+  StringPiece() : ptr_(NULL), length_(0) { }
+  StringPiece(const char* str)
+    : ptr_(str), length_((str == NULL) ? 0 : static_cast<int>(strlen(str))) { }
+  StringPiece(const std::string& str)
+    : ptr_(str.data()), length_(static_cast<int>(str.size())) { }
+  StringPiece(const char* offset, int len) : ptr_(offset), length_(len) { }
+
+  // data() may return a pointer to a buffer with embedded NULs, and the
+  // returned buffer may or may not be null terminated.  Therefore it is
+  // typically a mistake to pass data() to a routine that expects a NUL
+  // terminated string.
+  const char* data() const { return ptr_; }
+  int size() const { return length_; }
+  int length() const { return length_; }
+  bool empty() const { return length_ == 0; }
+
+  void clear() { ptr_ = NULL; length_ = 0; }
+  void set(const char* data, int len) { ptr_ = data; length_ = len; }
+  void set(const char* str) {
+    ptr_ = str;
+    if (str != NULL)
+      length_ = static_cast<int>(strlen(str));
+    else
+      length_ = 0;
+  }
+  void set(const void* data, int len) {
+    ptr_ = reinterpret_cast<const char*>(data);
+    length_ = len;
+  }
+
+  char operator[](int i) const { return ptr_[i]; }
+
+  void remove_prefix(int n) {
+    ptr_ += n;
+    length_ -= n;
+  }
+
+  void remove_suffix(int n) {
+    length_ -= n;
+  }
+
+  int compare(const StringPiece& x) const {
+    int r = memcmp(ptr_, x.ptr_, std::min(length_, x.length_));
+    if (r == 0) {
+      if (length_ < x.length_) r = -1;
+      else if (length_ > x.length_) r = +1;
+    }
+    return r;
+  }
+
+  std::string as_string() const {
+    return std::string(data(), size());
+  }
+  // We also define ToString() here, since many other string-like
+  // interfaces name the routine that converts to a C++ string
+  // "ToString", and it's confusing to have the method that does that
+  // for a StringPiece be called "as_string()".  We also leave the
+  // "as_string()" method defined here for existing code.
+  std::string ToString() const {
+    return std::string(data(), size());
+  }
+
+  void CopyToString(std::string* target) const;
+  void AppendToString(std::string* target) const;
+
+  // Does "this" start with "x"
+  bool starts_with(const StringPiece& x) const {
+    return ((length_ >= x.length_) &&
+            (memcmp(ptr_, x.ptr_, x.length_) == 0));
+  }
+
+  // Does "this" end with "x"
+  bool ends_with(const StringPiece& x) const {
+    return ((length_ >= x.length_) &&
+            (memcmp(ptr_ + (length_-x.length_), x.ptr_, x.length_) == 0));
+  }
+
+  // standard STL container boilerplate
+  typedef char value_type;
+  typedef const char* pointer;
+  typedef const char& reference;
+  typedef const char& const_reference;
+  typedef size_t size_type;
+  typedef ptrdiff_t difference_type;
+  static const size_type npos;
+  typedef const char* const_iterator;
+  typedef const char* iterator;
+  typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
+  typedef std::reverse_iterator<iterator> reverse_iterator;
+  iterator begin() const { return ptr_; }
+  iterator end() const { return ptr_ + length_; }
+  const_reverse_iterator rbegin() const {
+    return const_reverse_iterator(ptr_ + length_);
+  }
+  const_reverse_iterator rend() const {
+    return const_reverse_iterator(ptr_);
+  }
+  // STLS says return size_type, but Google says return int
+  int max_size() const { return length_; }
+  int capacity() const { return length_; }
+
+  int copy(char* buf, size_type n, size_type pos = 0) const;
+
+  int find(const StringPiece& s, size_type pos = 0) const;
+  int find(char c, size_type pos = 0) const;
+  int rfind(const StringPiece& s, size_type pos = npos) const;
+  int rfind(char c, size_type pos = npos) const;
+
+  StringPiece substr(size_type pos, size_type n = npos) const;
+};
+
+}  // namespace cxxmph
+
+bool operator==(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y);
+
+inline bool operator!=(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y) {
+  return !(x == y);
+}
+
+inline bool operator<(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y) {
+  const int r = memcmp(x.data(), y.data(),
+                       std::min(x.size(), y.size()));
+  return ((r < 0) || ((r == 0) && (x.size() < y.size())));
+}
+
+inline bool operator>(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y) {
+  return y < x;
+}
+
+inline bool operator<=(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y) {
+  return !(x > y);
+}
+
+inline bool operator>=(const cxxmph::StringPiece& x, const cxxmph::StringPiece& y) {
+  return !(x < y);
+}
+
+// allow StringPiece to be logged
+extern std::ostream& operator<<(std::ostream& o, const cxxmph::StringPiece& piece);
+
+#endif  // CXXMPH_STRINGPIECE_H__

cxxmph/trigraph.cc

@@ -1,3 +1,4 @@
+#include <cassert>
 #include <limits>
 
 #include "trigraph.h"
@@ -8,17 +9,51 @@ namespace {
 static const cmph_uint8 kInvalidEdge = std::numeric_limits<cmph_uint8>::max();
 } 
 
+namespace cxxmph {
+
 TriGraph::TriGraph(cmph_uint32 nedges, cmph_uint32 nvertices)
       : nedges_(0),
         edges_(nedges),
         first_edge_(nvertices, kInvalidEdge),
         vertex_degree_(nvertices, 0) { }
 
-void TriGraph::ExtractEdgesAndClear(vector<ConnectedEdge>* edges) {
+void TriGraph::ExtractEdgesAndClear(vector<Edge>* edges) {
+  vector<Edge>().swap(next_edge_);
   vector<cmph_uint32>().swap(first_edge_);
   vector<cmph_uint8>().swap(vertex_degree_);
   nedges_ = 0;
   edges->swap(edges_);
 }
-void TriGraph::AddEdge(const Edge& edge) { }
+void TriGraph::AddEdge(const Edge& edge) { 
-void TriGraph::RemoveEdge(cmph_uint32 current_edge) { }
+  edges_[nedges_] = edge;
+  next_edge_[nedges_] = Edge(
+      first_edge_[edge[0]], first_edge_[edge[1]], first_edge_[edge[2]]);
+   first_edge_[edge[0]] = first_edge_[edge[1]] = first_edge_[edge[2]] = nedges_;
+   ++vertex_degree_[edge[0]];
+   ++vertex_degree_[edge[1]];
+   ++vertex_degree_[edge[2]];
+   ++nedges_;
+}
+
+void TriGraph::RemoveEdge(cmph_uint32 current_edge) {
+  cmph_uint32 vertex, edge1, edge2;
+  for (int i = 0; i < 3; ++i) {
+    cmph_uint32 vertex = edges_[current_edge][i];
+    cmph_uint32 edge1 = first_edge_[vertex];
+    cmph_uint32 edge2 = kInvalidEdge;
+    cmph_uint32 j = 0;
+    while (edge1 != current_edge && edge1 != kInvalidEdge) {
+      edge2 = edge1;
+      if (edges_[edge1][0] == vertex) j = 0;
+      else if (edges_[edge1][1] == vertex) j = 1;
+      else j = 2;
+      edge1 = next_edge_[edge1][j];
+    }
+    assert(edge1 != kInvalidEdge);
+    if (edge2 != kInvalidEdge) next_edge_[edge2][j] = next_edge_[edge1][i];
+    else first_edge_[vertex] = next_edge_[edge1][i];
+    --vertex_degree_[vertex];
+  }
+}
+     
+}  // namespace cxxmph

cxxmph/trigraph.h

@@ -1,26 +1,43 @@
+#ifndef __CXXMPH_TRIGRAPH_H__
+#define __CXXMPH_TRIGRAPH_H__
+// Build a trigraph using a memory efficient representation.
+//
+// Prior knowledge of the number of edges and vertices for the graph is
+// required. For each vertex, we store how many edges touch it (degree) and the
+// index of the first edge in the vector of triples representing the edges.
+
+
 #include <vector>
 
 #include "../src/cmph_types.h"
 
+namespace cxxmph {
+
 class TriGraph {
   struct Edge {
     Edge() { }
     Edge(cmph_uint32 v0, cmph_uint32 v1, cmph_uint32 v2);
+    cmph_uint32& operator[](cmph_uint8 v) { return vertices[v]; }
+    const cmph_uint32& operator[](cmph_uint8 v) const { return vertices[v]; }
     cmph_uint32 vertices[3];
   };
-  struct ConnectedEdge {
-    Edge current;
-    Edge next;
-  };
-
   TriGraph(cmph_uint32 nedges, cmph_uint32 nvertices);
   void AddEdge(const Edge& edge);
-  void RemoveEdge(cmph_uint32 current_edge);
+  void RemoveEdge(cmph_uint32 edge_id);
-  void ExtractEdgesAndClear(std::vector<ConnectedEdge>* edges);
+  void ExtractEdgesAndClear(std::vector<Edge>* edges);
+
+  const std::vector<Edge>& edges() const { return edges_; }
+  const std::vector<cmph_uint8>& vertex_degree() const { return vertex_degree_; }
+  const std::vector<cmph_uint32>& first_edge() const { return first_edge_; }
 
  private:
-  cmph_uint32 nedges_;
+  cmph_uint32 nedges_;  // total number of edges
-  std::vector<ConnectedEdge> edges_;
+  std::vector<Edge> edges_;
-  std::vector<cmph_uint32> first_edge_;
+  std::vector<Edge> next_edge_;  // for implementing removal
-  std::vector<cmph_uint8> vertex_degree_;
+  std::vector<cmph_uint32> first_edge_;  // the first edge for this vertex
+  std::vector<cmph_uint8> vertex_degree_;  // number of edges for this vertex
 };
+
+}  // namespace cxxmph
+
+#endif  // __CXXMPH_TRIGRAPH_H__