src/yapf/blob.hpp
changeset 5884 be0c8467aeb4
parent 5883 78ad0a4f6d5a
child 5885 262234e81333
equal deleted inserted replaced
5883:78ad0a4f6d5a 5884:be0c8467aeb4
     1 /* $Id$ */
       
     2 
       
     3 #ifndef  BLOB_HPP
       
     4 #define  BLOB_HPP
       
     5 
       
     6 /** Type-safe version of memcpy().
       
     7  * @param d destination buffer
       
     8  * @param s source buffer
       
     9  * @param num_items number of items to be copied (!not number of bytes!) */
       
    10 template <class Titem_>
       
    11 FORCEINLINE void MemCpyT(Titem_* d, const Titem_* s, int num_items = 1)
       
    12 {
       
    13 	memcpy(d, s, num_items * sizeof(Titem_));
       
    14 }
       
    15 
       
    16 
       
    17 /** Base class for simple binary blobs.
       
    18  *  Item is byte.
       
    19  *  The word 'simple' means:
       
    20  *    - no configurable allocator type (always made from heap)
       
    21  *    - no smart deallocation - deallocation must be called from the same
       
    22  *        module (DLL) where the blob was allocated
       
    23  *    - no configurable allocation policy (how big blocks should be allocated)
       
    24  *    - no extra ownership policy (i.e. 'copy on write') when blob is copied
       
    25  *    - no thread synchronization at all
       
    26  *
       
    27  *  Internal member layout:
       
    28  *  1. The only class member is pointer to the first item (see union ptr_u).
       
    29  *  2. Allocated block contains the blob header (see CHdr) followed by the raw byte data.
       
    30  *     Always, when it allocates memory the allocated size is:
       
    31  *                                                      sizeof(CHdr) + <data capacity>
       
    32  *  3. Two 'virtual' members (m_size and m_max_size) are stored in the CHdr at beginning
       
    33  *     of the alloated block.
       
    34  *  4. The pointer (in ptr_u) points behind the header (to the first data byte).
       
    35  *     When memory block is allocated, the sizeof(CHdr) it added to it.
       
    36  *  5. Benefits of this layout:
       
    37  *     - items are accessed in the simplest possible way - just dereferencing the pointer,
       
    38  *       which is good for performance (assuming that data are accessed most often).
       
    39  *     - sizeof(blob) is the same as the size of any other pointer
       
    40  *  6. Drawbacks of this layout:
       
    41  *     - the fact, that pointer to the alocated block is adjusted by sizeof(CHdr) before
       
    42  *       it is stored can lead to several confusions:
       
    43  *         - it is not common pattern so the implementation code is bit harder to read
       
    44  *         - valgrind can generate warning that allocated block is lost (not accessible)
       
    45  * */
       
    46 class CBlobBaseSimple {
       
    47 protected:
       
    48 	/** header of the allocated memory block */
       
    49 	struct CHdr {
       
    50 		int    m_size;      ///< actual blob size in bytes
       
    51 		int    m_max_size;  ///< maximum (allocated) size in bytes
       
    52 	};
       
    53 
       
    54 	/** type used as class member */
       
    55 	union {
       
    56 		int8   *m_pData;    ///< pointer to the first byte of data
       
    57 		CHdr   *m_pHdr_1;   ///< pointer just after the CHdr holding m_size and m_max_size
       
    58 	} ptr_u;
       
    59 
       
    60 public:
       
    61 	static const int Ttail_reserve = 4; ///< four extra bytes will be always allocated and zeroed at the end
       
    62 
       
    63 	/** default constructor - initializes empty blob */
       
    64 	FORCEINLINE CBlobBaseSimple() { InitEmpty(); }
       
    65 	/** copy constructor */
       
    66 	FORCEINLINE CBlobBaseSimple(const CBlobBaseSimple& src)
       
    67 	{
       
    68 		InitEmpty();
       
    69 		AppendRaw(src);
       
    70 	}
       
    71 	/** destructor */
       
    72 	FORCEINLINE ~CBlobBaseSimple() { Free(); }
       
    73 protected:
       
    74 	/** initialize the empty blob by setting the ptr_u.m_pHdr_1 pointer to the static CHdr with
       
    75 	 *  both m_size and m_max_size containing zero */
       
    76 	FORCEINLINE void InitEmpty() { static CHdr hdrEmpty[] = {{0, 0}, {0, 0}}; ptr_u.m_pHdr_1 = &hdrEmpty[1]; }
       
    77 	/** initialize blob by attaching it to the given header followed by data */
       
    78 	FORCEINLINE void Init(CHdr* hdr) { ptr_u.m_pHdr_1 = &hdr[1]; }
       
    79 	/** blob header accessor - use it rather than using the pointer arithmetics directly - non-const version */
       
    80 	FORCEINLINE CHdr& Hdr() { return ptr_u.m_pHdr_1[-1]; }
       
    81 	/** blob header accessor - use it rather than using the pointer arithmetics directly - const version */
       
    82 	FORCEINLINE const CHdr& Hdr() const { return ptr_u.m_pHdr_1[-1]; }
       
    83 	/** return reference to the actual blob size - used when the size needs to be modified */
       
    84 	FORCEINLINE int& RawSizeRef() { return Hdr().m_size; };
       
    85 
       
    86 public:
       
    87 	/** return true if blob doesn't contain valid data */
       
    88 	FORCEINLINE bool IsEmpty() const { return RawSize() == 0; }
       
    89 	/** return the number of valid data bytes in the blob */
       
    90 	FORCEINLINE int RawSize() const { return Hdr().m_size; };
       
    91 	/** return the current blob capacity in bytes */
       
    92 	FORCEINLINE int MaxRawSize() const { return Hdr().m_max_size; };
       
    93 	/** return pointer to the first byte of data - non-const version */
       
    94 	FORCEINLINE int8* RawData() { return ptr_u.m_pData; }
       
    95 	/** return pointer to the first byte of data - const version */
       
    96 	FORCEINLINE const int8* RawData() const { return ptr_u.m_pData; }
       
    97 #if 0 // reenable when needed
       
    98 	/** return the 32 bit CRC of valid data in the blob */
       
    99 	FORCEINLINE uint32 Crc32() const {return CCrc32::Calc(RawData(), RawSize());}
       
   100 #endif //0
       
   101 	/** invalidate blob's data - doesn't free buffer */
       
   102 	FORCEINLINE void Clear() { RawSizeRef() = 0; }
       
   103 	/** free the blob's memory */
       
   104 	FORCEINLINE void Free() { if (MaxRawSize() > 0) {RawFree(&Hdr()); InitEmpty();} }
       
   105 	/** copy data from another blob - replaces any existing blob's data */
       
   106 	FORCEINLINE void CopyFrom(const CBlobBaseSimple& src) { Clear(); AppendRaw(src); }
       
   107 	/** overtake ownership of data buffer from the source blob - source blob will become empty */
       
   108 	FORCEINLINE void MoveFrom(CBlobBaseSimple& src) { Free(); ptr_u.m_pData = src.ptr_u.m_pData; src.InitEmpty(); }
       
   109 	/** swap buffers (with data) between two blobs (this and source blob) */
       
   110 	FORCEINLINE void Swap(CBlobBaseSimple& src) { int8 *tmp = ptr_u.m_pData; ptr_u.m_pData = src.ptr_u.m_pData; src.ptr_u.m_pData = tmp; }
       
   111 
       
   112 	/** append new bytes at the end of existing data bytes - reallocates if necessary */
       
   113 	FORCEINLINE void AppendRaw(int8 *p, int num_bytes)
       
   114 	{
       
   115 		assert(p != NULL);
       
   116 		if (num_bytes > 0) {
       
   117 			memcpy(GrowRawSize(num_bytes), p, num_bytes);
       
   118 		} else {
       
   119 			assert(num_bytes >= 0);
       
   120 		}
       
   121 	}
       
   122 
       
   123 	/** append bytes from given source blob to the end of existing data bytes - reallocates if necessary */
       
   124 	FORCEINLINE void AppendRaw(const CBlobBaseSimple& src)
       
   125 	{
       
   126 		if (!src.IsEmpty())
       
   127 			memcpy(GrowRawSize(src.RawSize()), src.RawData(), src.RawSize());
       
   128 	}
       
   129 
       
   130 	/** Reallocate if there is no free space for num_bytes bytes.
       
   131 	 *  @return pointer to the new data to be added */
       
   132 	FORCEINLINE int8* MakeRawFreeSpace(int num_bytes)
       
   133 	{
       
   134 		assert(num_bytes >= 0);
       
   135 		int new_size = RawSize() + num_bytes;
       
   136 		if (new_size > MaxRawSize()) SmartAlloc(new_size);
       
   137 		FixTail();
       
   138 		return ptr_u.m_pData + RawSize();
       
   139 	}
       
   140 
       
   141 	/** Increase RawSize() by num_bytes.
       
   142 	 *  @return pointer to the new data added */
       
   143 	FORCEINLINE int8* GrowRawSize(int num_bytes)
       
   144 	{
       
   145 		int8* pNewData = MakeRawFreeSpace(num_bytes);
       
   146 		RawSizeRef() += num_bytes;
       
   147 		return pNewData;
       
   148 	}
       
   149 
       
   150 	/** Decrease RawSize() by num_bytes. */
       
   151 	FORCEINLINE void ReduceRawSize(int num_bytes)
       
   152 	{
       
   153 		if (MaxRawSize() > 0 && num_bytes > 0) {
       
   154 			assert(num_bytes <= RawSize());
       
   155 			if (num_bytes < RawSize()) RawSizeRef() -= num_bytes;
       
   156 			else RawSizeRef() = 0;
       
   157 		}
       
   158 	}
       
   159 	/** reallocate blob data if needed */
       
   160 	void SmartAlloc(int new_size)
       
   161 	{
       
   162 		int old_max_size = MaxRawSize();
       
   163 		if (old_max_size >= new_size) return;
       
   164 		// calculate minimum block size we need to allocate
       
   165 		int min_alloc_size = sizeof(CHdr) + new_size + Ttail_reserve;
       
   166 		// ask allocation policy for some reasonable block size
       
   167 		int alloc_size = AllocPolicy(min_alloc_size);
       
   168 		// allocate new block
       
   169 		CHdr* pNewHdr = RawAlloc(alloc_size);
       
   170 		// setup header
       
   171 		pNewHdr->m_size = RawSize();
       
   172 		pNewHdr->m_max_size = alloc_size - (sizeof(CHdr) + Ttail_reserve);
       
   173 		// copy existing data
       
   174 		if (RawSize() > 0)
       
   175 			memcpy(pNewHdr + 1, ptr_u.m_pData, pNewHdr->m_size);
       
   176 		// replace our block with new one
       
   177 		CHdr* pOldHdr = &Hdr();
       
   178 		Init(pNewHdr);
       
   179 		if (old_max_size > 0)
       
   180 			RawFree(pOldHdr);
       
   181 	}
       
   182 	/** simple allocation policy - can be optimized later */
       
   183 	FORCEINLINE static int AllocPolicy(int min_alloc)
       
   184 	{
       
   185 		if (min_alloc < (1 << 9)) {
       
   186 			if (min_alloc < (1 << 5)) return (1 << 5);
       
   187 			return (min_alloc < (1 << 7)) ? (1 << 7) : (1 << 9);
       
   188 		}
       
   189 		if (min_alloc < (1 << 15)) {
       
   190 			if (min_alloc < (1 << 11)) return (1 << 11);
       
   191 			return (min_alloc < (1 << 13)) ? (1 << 13) : (1 << 15);
       
   192 		}
       
   193 		if (min_alloc < (1 << 20)) {
       
   194 			if (min_alloc < (1 << 17)) return (1 << 17);
       
   195 			return (min_alloc < (1 << 19)) ? (1 << 19) : (1 << 20);
       
   196 		}
       
   197 		min_alloc = (min_alloc | ((1 << 20) - 1)) + 1;
       
   198 		return min_alloc;
       
   199 	}
       
   200 
       
   201 	/** all allocation should happen here */
       
   202 	static FORCEINLINE CHdr* RawAlloc(int num_bytes) { return (CHdr*)malloc(num_bytes); }
       
   203 	/** all deallocations should happen here */
       
   204 	static FORCEINLINE void RawFree(CHdr* p) { free(p); }
       
   205 	/** fixing the four bytes at the end of blob data - useful when blob is used to hold string */
       
   206 	FORCEINLINE void FixTail()
       
   207 	{
       
   208 		if (MaxRawSize() > 0) {
       
   209 			int8 *p = &ptr_u.m_pData[RawSize()];
       
   210 			for (int i = 0; i < Ttail_reserve; i++) p[i] = 0;
       
   211 		}
       
   212 	}
       
   213 };
       
   214 
       
   215 /** Blob - simple dynamic Titem_ array. Titem_ (template argument) is a placeholder for any type.
       
   216  *  Titem_ can be any integral type, pointer, or structure. Using Blob instead of just plain C array
       
   217  *  simplifies the resource management in several ways:
       
   218  *  1. When adding new item(s) it automatically grows capacity if needed.
       
   219  *  2. When variable of type Blob comes out of scope it automatically frees the data buffer.
       
   220  *  3. Takes care about the actual data size (number of used items).
       
   221  *  4. Dynamically constructs only used items (as opposite of static array which constructs all items) */
       
   222 template <class Titem_, class Tbase_ = CBlobBaseSimple>
       
   223 class CBlobT : public CBlobBaseSimple {
       
   224 	// make template arguments public:
       
   225 public:
       
   226 	typedef Titem_ Titem;
       
   227 	typedef Tbase_ Tbase;
       
   228 
       
   229 	static const int Titem_size = sizeof(Titem);
       
   230 
       
   231 	/** Default constructor - makes new Blob ready to accept any data */
       
   232 	FORCEINLINE CBlobT() : Tbase() {}
       
   233 	/** Copy constructor - make new blob to become copy of the original (source) blob */
       
   234 	FORCEINLINE CBlobT(const Tbase& src) : Tbase(src) {assert((RawSize() % Titem_size) == 0);}
       
   235 	/** Destructor - ensures that allocated memory (if any) is freed */
       
   236 	FORCEINLINE ~CBlobT() { Free(); }
       
   237 	/** Check the validity of item index (only in debug mode) */
       
   238 	FORCEINLINE void CheckIdx(int idx) { assert(idx >= 0); assert(idx < Size()); }
       
   239 	/** Return pointer to the first data item - non-const version */
       
   240 	FORCEINLINE Titem* Data() { return (Titem*)RawData(); }
       
   241 	/** Return pointer to the first data item - const version */
       
   242 	FORCEINLINE const Titem* Data() const { return (const Titem*)RawData(); }
       
   243 	/** Return pointer to the idx-th data item - non-const version */
       
   244 	FORCEINLINE Titem* Data(int idx) { CheckIdx(idx); return (Data() + idx); }
       
   245 	/** Return pointer to the idx-th data item - const version */
       
   246 	FORCEINLINE const Titem* Data(int idx) const { CheckIdx(idx); return (Data() + idx); }
       
   247 	/** Return number of items in the Blob */
       
   248 	FORCEINLINE int Size() const { return (RawSize() / Titem_size); }
       
   249 	/** Free the memory occupied by Blob destroying all items */
       
   250 	FORCEINLINE void Free()
       
   251 	{
       
   252 		assert((RawSize() % Titem_size) == 0);
       
   253 		int old_size = Size();
       
   254 		if (old_size > 0) {
       
   255 			// destroy removed items;
       
   256 			Titem* pI_last_to_destroy = Data(0);
       
   257 			for (Titem* pI = Data(old_size - 1); pI >= pI_last_to_destroy; pI--) pI->~Titem_();
       
   258 		}
       
   259 		Tbase::Free();
       
   260 	}
       
   261 	/** Grow number of data items in Blob by given number - doesn't construct items */
       
   262 	FORCEINLINE Titem* GrowSizeNC(int num_items) { return (Titem*)GrowRawSize(num_items * Titem_size); }
       
   263 	/** Grow number of data items in Blob by given number - constructs new items (using Titem_'s default constructor) */
       
   264 	FORCEINLINE Titem* GrowSizeC(int num_items)
       
   265 	{
       
   266 		Titem* pI = GrowSizeNC(num_items);
       
   267 		for (int i = num_items; i > 0; i--, pI++) new (pI) Titem();
       
   268 	}
       
   269 	/** Destroy given number of items and reduce the Blob's data size */
       
   270 	FORCEINLINE void ReduceSize(int num_items)
       
   271 	{
       
   272 		assert((RawSize() % Titem_size) == 0);
       
   273 		int old_size = Size();
       
   274 		assert(num_items <= old_size);
       
   275 		int new_size = (num_items <= old_size) ? (old_size - num_items) : 0;
       
   276 		// destroy removed items;
       
   277 		Titem* pI_last_to_destroy = Data(new_size);
       
   278 		for (Titem* pI = Data(old_size - 1); pI >= pI_last_to_destroy; pI--) pI->~Titem();
       
   279 		// remove them
       
   280 		ReduceRawSize(num_items * Titem_size);
       
   281 	}
       
   282 	/** Append one data item at the end (calls Titem_'s default constructor) */
       
   283 	FORCEINLINE Titem* AppendNew()
       
   284 	{
       
   285 		Titem& dst = *GrowSizeNC(1); // Grow size by one item
       
   286 		Titem* pNewItem = new (&dst) Titem(); // construct the new item by calling in-place new operator
       
   287 		return pNewItem;
       
   288 	}
       
   289 	/** Append the copy of given item at the end of Blob (using copy constructor) */
       
   290 	FORCEINLINE Titem* Append(const Titem& src)
       
   291 	{
       
   292 		Titem& dst = *GrowSizeNC(1); // Grow size by one item
       
   293 		Titem* pNewItem = new (&dst) Titem(src); // construct the new item by calling in-place new operator with copy ctor()
       
   294 		return pNewItem;
       
   295 	}
       
   296 	/** Add given items (ptr + number of items) at the end of blob */
       
   297 	FORCEINLINE Titem* Append(const Titem* pSrc, int num_items)
       
   298 	{
       
   299 		Titem* pDst = GrowSizeNC(num_items);
       
   300 		Titem* pDstOrg = pDst;
       
   301 		Titem* pDstEnd = pDst + num_items;
       
   302 		while (pDst < pDstEnd) new (pDst++) Titem(*(pSrc++));
       
   303 		return pDstOrg;
       
   304 	}
       
   305 	/** Remove item with the given index by replacing it by the last item and reducing the size by one */
       
   306 	FORCEINLINE void RemoveBySwap(int idx)
       
   307 	{
       
   308 		CheckIdx(idx);
       
   309 		// destroy removed item
       
   310 		Titem* pRemoved = Data(idx);
       
   311 		RemoveBySwap(pRemoved);
       
   312 	}
       
   313 	/** Remove item given by pointer replacing it by the last item and reducing the size by one */
       
   314 	FORCEINLINE void RemoveBySwap(Titem* pItem)
       
   315 	{
       
   316 		Titem* pLast = Data(Size() - 1);
       
   317 		assert(pItem >= Data() && pItem <= pLast);
       
   318 		// move last item to its new place
       
   319 		if (pItem != pLast) {
       
   320 			pItem->~Titem_();
       
   321 			new (pItem) Titem_(*pLast);
       
   322 		}
       
   323 		// destroy the last item
       
   324 		pLast->~Titem_();
       
   325 		// and reduce the raw blob size
       
   326 		ReduceRawSize(Titem_size);
       
   327 	}
       
   328 	/** Ensures that given number of items can be added to the end of Blob. Returns pointer to the
       
   329 	 *  first free (unused) item */
       
   330 	FORCEINLINE Titem* MakeFreeSpace(int num_items) { return (Titem*)MakeRawFreeSpace(num_items * Titem_size); }
       
   331 };
       
   332 
       
   333 // simple string implementation
       
   334 struct CStrA : public CBlobT<char>
       
   335 {
       
   336 	typedef CBlobT<char> base;
       
   337 	CStrA(const char* str = NULL) {Append(str);}
       
   338 	FORCEINLINE CStrA(const CBlobBaseSimple& src) : base(src) {}
       
   339 	void Append(const char* str) {if (str != NULL && str[0] != '\0') base::Append(str, (int)strlen(str));}
       
   340 };
       
   341 
       
   342 #endif /* BLOB_HPP */