Merge branch 'master' into var-subset

src/hb-algs.hh

@@ -40,12 +40,12 @@ static const struct
   //{ return hb_deref_pointer (v).hash (); }
   /* Instead, the following ugly soution: */
   template <typename T,
-	    hb_enable_if (!hb_is_integer (hb_remove_reference (T)) && !hb_is_pointer (T))>
+	    hb_enable_if (!hb_is_integer (hb_remove_const (hb_remove_reference (T))) && !hb_is_pointer (T))>
   uint32_t operator () (T&& v) const { return v.hash (); }
 
   template <typename T>
   uint32_t operator () (const T *v) const
-  { return hb_hash (v); }
+  { return operator() (*v); }
 
   template <typename T,
 	    hb_enable_if (hb_is_integer (T))>

src/hb-array.hh

@@ -79,6 +79,8 @@ struct hb_array_t : hb_iter_with_fallback_t<hb_array_t<Type>, Type&>
   operator hb_array_t<const Type> () { return hb_array_t<const Type> (arrayZ, length); }
   template <typename T> operator T * () const { return arrayZ; }
 
+  uint32_t hash () const;
+
   /*
    * Compare, Sort, and Search.
    */
@@ -273,9 +275,20 @@ template <typename T, unsigned int length_> inline hb_sorted_array_t<T>
 hb_sorted_array (T (&array_)[length_])
 { return hb_sorted_array_t<T> (array_); }
 
+template <typename T>
+uint32_t hb_array_t<T>::hash () const
+{
+  uint32_t h = 0;
+  for (unsigned i = 0; i < length; i++)
+    h ^= hb_hash (arrayZ[i]);
+  return h;
+}
 
 typedef hb_array_t<const char> hb_bytes_t;
 typedef hb_array_t<const unsigned char> hb_ubytes_t;
 
+/* TODO Specialize hashing for hb_bytes_t and hb_ubytes_t. */
+//template <>
+//uint32_t hb_array_t<const char>::hash () const { return 0; }
 
 #endif /* HB_ARRAY_HH */

src/hb-map.hh

@@ -34,6 +34,8 @@
  * hb_hashmap_t
  */
 
+/* TODO if K/V is signed integer, -1 is not a good default.
+ * Don't know how to get to -MAX using bit work. */
 template <typename K, typename V,
 	  K kINVALID = hb_is_pointer (K) ? 0 : (K) -1,
 	  V vINVALID = hb_is_pointer (V) ? 0 : (V) -1>
@@ -46,6 +48,10 @@ struct hb_hashmap_t
   static_assert (hb_is_integer (K) || hb_is_pointer (K), "");
   static_assert (hb_is_integer (V) || hb_is_pointer (V), "");
 
+  /* TODO If key type is a pointer, keep hash in item_t and use to:
+   * 1. avoid rehashing when resizing table, and
+   * 2. compare hash before comparing keys, for speed.
+   */
   struct item_t
   {
     K key;
@@ -82,14 +88,21 @@ struct hb_hashmap_t
   {
     free (items);
     items = nullptr;
+    population = occupancy = 0;
   }
   void fini ()
   {
-    population = occupancy = 0;
     hb_object_fini (this);
     fini_shallow ();
   }
 
+  void reset ()
+  {
+    /* TODO Keep array? */
+    fini_shallow ();
+    init_shallow ();
+  }
+
   bool in_error () const { return !successful; }
 
   bool resize ()
@@ -248,6 +261,10 @@ struct hb_hashmap_t
   }
 };
 
+/*
+ * hb_map_t
+ */
+
 struct hb_map_t : hb_hashmap_t<hb_codepoint_t,
 			       hb_codepoint_t,
 			       HB_MAP_VALUE_INVALID,

src/hb-meta.hh

@@ -72,6 +72,17 @@ static const struct
 } hb_deref_pointer HB_UNUSED;
 
 
+/* std::move and std::forward */
+
+template <typename T>
+hb_remove_reference (T)&& hb_move (T&& t) { return (hb_remove_reference (T)&&) (t); }
+
+template <typename T>
+T&& hb_forward (hb_remove_reference (T)& t) { return (T&&) t; }
+template <typename T>
+T&& hb_forward (hb_remove_reference (T)&& t) { return (T&&) t; }
+
+
 /* Void!  For when we need a expression-type of void. */
 struct hb_void_t { typedef void value; };
 

src/hb-serialize.hh

@@ -31,6 +31,7 @@
 
 #include "hb.hh"
 #include "hb-blob.hh"
+#include "hb-map.hh"
 
 
 /*
@@ -39,20 +40,70 @@
 
 struct hb_serialize_context_t
 {
+  typedef unsigned objidx_t;
+
+  struct range_t
+  {
+    char *head, *tail;
+  };
+
+  struct object_t : range_t
+  {
+    void fini () { links.fini (); }
+
+    bool operator == (const object_t &o) const
+    {
+      return (tail - head == o.tail - o.head)
+	  && (links.length != o.links.length)
+	  && 0 == memcmp (head, o.head, tail - head)
+	  && 0 == memcmp (&links, &o.links, links.get_size ());
+    }
+    uint32_t hash () const
+    {
+      return hb_bytes_t (head, tail - head).hash () ^
+	     links.as_bytes ().hash ();
+    }
+
+    struct link_t
+    {
+      bool wide: 1;
+      unsigned offset : 31;
+      objidx_t objidx;
+    };
+
+    hb_vector_t<link_t> links;
+  };
+
+  range_t snapshot () { range_t s = {head, tail} ; return s; }
+
+
   hb_serialize_context_t (void *start_, unsigned int size)
   {
     this->start = (char *) start_;
     this->end = this->start + size;
     reset ();
   }
+  ~hb_serialize_context_t ()
+  {
+    current.fini_deep ();
+    packed.fini_deep ();
+    packed_map.fini ();
+  }
 
   bool in_error () const { return !this->successful; }
 
   void reset ()
   {
     this->successful = true;
+    this->ran_out_of_room = false;
     this->head = this->start;
+    this->tail = this->end;
     this->debug_depth = 0;
+
+    this->current.reset ();
+    this->packed.reset ();
+    this->packed.push ()->head = this->end;
+    this->packed_map.reset ();
   }
 
   bool propagate_error (bool e)
@@ -61,15 +112,10 @@ struct hb_serialize_context_t
   { return this->successful = this->successful && !obj.in_error (); }
   template <typename T> bool propagate_error (const T *obj)
   { return this->successful = this->successful && !obj->in_error (); }
-  template <typename T1, typename T2> bool propagate_error (T1 &o1, T2 &o2)
+  template <typename T1, typename T2> bool propagate_error (T1 &&o1, T2 &&o2)
-  { return propagate_error (o1) && propagate_error (o2); }
-  template <typename T1, typename T2> bool propagate_error (T1 *o1, T2 *o2)
   { return propagate_error (o1) && propagate_error (o2); }
   template <typename T1, typename T2, typename T3>
-  bool propagate_error (T1 &o1, T2 &o2, T3 &o3)
+  bool propagate_error (T1 &&o1, T2 &&o2, T3 &&o3)
-  { return propagate_error (o1) && propagate_error (o2, o3); }
-  template <typename T1, typename T2, typename T3>
-  bool propagate_error (T1 *o1, T2 *o2, T3 *o3)
   { return propagate_error (o1) && propagate_error (o2, o3); }
 
   /* To be called around main operation. */
@@ -81,18 +127,103 @@ struct hb_serialize_context_t
 		     this->start, this->end,
 		     (unsigned long) (this->end - this->start));
 
-    return start_embed<Type> ();
+    assert (!current.length);
+    return push<Type> ();
   }
   void end_serialize ()
   {
     DEBUG_MSG_LEVEL (SERIALIZE, this->start, 0, -1,
-		     "end [%p..%p] serialized %d bytes; %s",
+		     "end [%p..%p] serialized %u bytes; %s",
 		     this->start, this->end,
-		     (int) (this->head - this->start),
+		     (unsigned) (this->head - this->start),
 		     this->successful ? "successful" : "UNSUCCESSFUL");
+
+    /* TODO Propagate errors. */
+
+    assert (current.length == 1);
+
+    /* Only "pack" if there exist other objects... Otherwise, don't bother.
+     * Saves a move. */
+    if (packed.length == 1)
+      return;
+
+    pop_pack ();
+
+    link ();
   }
 
-  unsigned int length () const { return this->head - this->start; }
+  template <typename Type>
+  Type *push ()
+  {
+    object_t obj;
+    obj.head = head;
+    obj.tail = tail;
+    current.push (obj);
+    return start_embed<Type> ();
+  }
+  void pop_discard ()
+  {
+    revert (current.pop ());
+  }
+  objidx_t pop_pack ()
+  {
+    object_t obj = current.pop ();
+    obj.tail = head;
+    unsigned len = obj.tail - obj.head;
+
+    objidx_t objidx = packed_map.get (&obj);
+    if (objidx)
+    {
+      obj.fini ();
+      return objidx;
+    }
+
+    tail -= len;
+    memmove (tail, obj.head, len);
+    head = obj.head;
+
+    if (!len)
+      return 0;
+
+    obj.head = tail;
+    obj.tail = tail + len;
+
+    object_t *key = packed.push (hb_move (obj));
+
+    /* TODO Handle error. */
+    if (unlikely (packed.in_error ()))
+      return 0;
+
+    objidx = packed.length - 1;
+
+    packed_map.set (key, objidx);
+
+    return objidx;
+  }
+
+  void revert (range_t snap)
+  {
+    assert (snap.head <= head);
+    assert (tail <= snap.tail);
+    head = snap.head;
+    tail = snap.tail;
+    discard_stale_objects ();
+  }
+
+  void discard_stale_objects ()
+  {
+    while (packed.length > 1 &&
+	   packed.tail ().head < tail)
+      packed.pop ();
+    assert (packed.tail ().head == tail);
+  }
+
+  void link ()
+  {
+    // XXX
+  }
+
+  unsigned int length () const { return this->head - current.tail ().head; }
 
   void align (unsigned int alignment)
   {
@@ -111,7 +242,11 @@ struct hb_serialize_context_t
   template <typename Type>
   Type *allocate_size (unsigned int size)
   {
-    if (unlikely (!this->successful || this->end - this->head < ptrdiff_t (size))) {
+    if (unlikely (!this->successful)) return nullptr;
+
+    if (this->tail - this->head < ptrdiff_t (size))
+    {
+      this->ran_out_of_room = true;
       this->successful = false;
       return nullptr;
     }
@@ -156,40 +291,49 @@ struct hb_serialize_context_t
   Type *extend (Type &obj) { return extend_size (obj, obj.get_size ()); }
 
   /* Output routines. */
-  template <typename Type>
-  Type *copy () const
-  {
-    assert (this->successful);
-    unsigned int len = this->head - this->start;
-    void *p = malloc (len);
-    if (p)
-      memcpy (p, this->start, len);
-    return reinterpret_cast<Type *> (p);
-  }
   hb_bytes_t copy_bytes () const
   {
     assert (this->successful);
-    unsigned int len = this->head - this->start;
+    /* Copy both items from head side and tail side... */
-    void *p = malloc (len);
+    unsigned int len = (this->head - this->start)
+		     + (this->end  - this->tail);
+    char *p = (char *) malloc (len);
     if (p)
-      memcpy (p, this->start, len);
+    {
+      memcpy (p, this->start, this->head - this->start);
+      memcpy (p + (this->head - this->start), this->tail, this->end - this->tail);
+    }
     else
       return hb_bytes_t ();
-    return hb_bytes_t ((char *) p, len);
+    return hb_bytes_t (p, len);
   }
+  template <typename Type>
+  Type *copy () const
+  { return reinterpret_cast<Type *> ((char *) copy_bytes ().arrayZ); }
   hb_blob_t *copy_blob () const
   {
-    assert (this->successful);
+    hb_bytes_t b = copy_bytes ();
-    return hb_blob_create (this->start,
+    return hb_blob_create (b.arrayZ, b.length,
-			   this->head - this->start,
+			   HB_MEMORY_MODE_WRITABLE,
-			   HB_MEMORY_MODE_DUPLICATE,
+			   (char *) b.arrayZ, free);
-			   nullptr, nullptr);
   }
 
-  public:
+  public: /* TODO Make private. */
+  char *start, *head, *tail, *end;
   unsigned int debug_depth;
-  char *start, *end, *head;
   bool successful;
+  bool ran_out_of_room;
+
+  private:
+
+  /* Stack of currently under construction objects. */
+  hb_vector_t<object_t> current;
+
+  /* Stack of packed objects.  Object 0 is always nil object. */
+  hb_vector_t<object_t> packed;
+
+  /* Map view of packed objects. */
+  hb_hashmap_t<const object_t *, objidx_t, nullptr, 0> packed_map;
 };
 
 

src/hb-subset.cc

@@ -85,7 +85,7 @@ _subset2 (hb_subset_plan_t *plan)
     hb_serialize_context_t serializer ((void *) buf, buf_size);
     hb_subset_context_t c (plan, &serializer);
     result = table->subset (&c);
-    if (serializer.in_error ())
+    if (serializer.ran_out_of_room)
     {
       buf_size += (buf_size >> 1) + 32;
       DEBUG_MSG(SUBSET, nullptr, "OT::%c%c%c%c ran out of room; reallocating to %u bytes.", HB_UNTAG (tag), buf_size);
@@ -96,6 +96,11 @@ _subset2 (hb_subset_plan_t *plan)
       }
       goto retry;
     }
+    if (serializer.in_error ())
+    {
+      abort ();
+    }
+
     if (result)
     {
       hb_blob_t *dest_blob = serializer.copy_blob ();

src/hb-vector.hh

@@ -38,8 +38,20 @@ struct hb_vector_t
   typedef Type item_t;
   static constexpr unsigned item_size = hb_static_size (Type);
 
-  HB_NO_COPY_ASSIGN_TEMPLATE (hb_vector_t, Type);
   hb_vector_t ()  { init (); }
+  hb_vector_t (const hb_vector_t &o)
+  {
+    init ();
+    alloc (o.length);
+    hb_iter (o) | hb_sink (this);
+  }
+  hb_vector_t (hb_vector_t &&o)
+  {
+    allocated = o.allocated;
+    length = o.length;
+    arrayZ_ = o.arrayZ_;
+    o.init ();
+  }
   ~hb_vector_t () { fini (); }
 
   unsigned int length;
@@ -69,6 +81,30 @@ struct hb_vector_t
     fini ();
   }
 
+  void reset () { resize (0); }
+
+  hb_vector_t& operator = (const hb_vector_t &o)
+  {
+    reset ();
+    alloc (o.length);
+    hb_iter (o) | hb_sink (this);
+    return *this;
+  }
+  hb_vector_t& operator = (hb_vector_t &&o)
+  {
+    fini ();
+    allocated = o.allocated;
+    length = o.length;
+    arrayZ_ = o.arrayZ_;
+    o.init ();
+    return *this;
+  }
+
+  hb_bytes_t as_bytes () const { return hb_bytes_t ((const char *) arrayZ_,
+						    length * item_size); }
+
+  uint32_t hash () const { return as_bytes ().hash (); }
+
   const Type * arrayZ () const { return arrayZ_; }
         Type * arrayZ ()       { return arrayZ_; }
 
@@ -87,11 +123,15 @@ struct hb_vector_t
     return arrayZ()[i];
   }
 
+  Type& tail () { return (*this)[length - 1]; }
+  const Type& tail () const { return (*this)[length - 1]; }
+
   explicit operator bool () const { return length; }
+  unsigned get_size () const { return length * item_size; }
 
   /* Sink interface. */
   template <typename T>
-  hb_vector_t& operator << (const T& v) { push (v); return *this; }
+  hb_vector_t& operator << (T&& v) { push (hb_forward<T> (v)); return *this; }
 
   hb_array_t<      Type> as_array ()       { return hb_array (arrayZ(), length); }
   hb_array_t<const Type> as_array () const { return hb_array (arrayZ(), length); }
@@ -131,10 +171,10 @@ struct hb_vector_t
     return &arrayZ()[length - 1];
   }
   template <typename T>
-  Type *push (const T& v)
+  Type *push (T&& v)
   {
     Type *p = push ();
-    *p = v;
+    *p = hb_forward<T> (v);
     return p;
   }
 
@@ -188,10 +228,10 @@ struct hb_vector_t
     return true;
   }
 
-  void pop ()
+  Type pop ()
   {
-    if (!length) return;
+    if (!length) return Null(Type);
-    length--;
+    return hb_move (arrayZ()[--length]); /* Does this move actually work? */
   }
 
   void remove (unsigned int i)