TweetFollow Us on Twitter

Jul 00 Challenge

Volume Number: 16 (2000)
Issue Number: 7
Column Tag: Programming

Programmer's Challenge

by Bob Boonstra, Westford, MA

RAID-5+

Those of you in the Information Systems business have certainly heard of Redundant Array of Independent (or Inexpensive) Disks, or RAID technology. As the name suggests, RAID stores data across multiple disks to provide improved performance and some level of protection against disk failure. Several RAID levels have been implemented, and one of the most common, RAID Level 5, employs disk striping (spreading blocks of data across multiple disks) and parity to optimize disk access for applications that perform random small read/write operations, and to protect against the failure of a single disk.

Our Challenge application, however, is a little more demanding. We're operating a mission critical application, one that cannot afford to lose data even with a double hardware failure. Your Challenge is to implement a RAID-5+ system that has the performance advantages of RAID 5, but can continue functioning when two disk drives fail.

The prototype for the code you should write is:

/*
 * ReadProc and WriteProc are callbacks that allow you to write to multiple drives
 * simultaneously, simulating the effect of striping.
 */
typedef void (* WriteProc) (/* conduct physical writes to multiple drives */
	long startByte[], /* start write from startbyte[n] physical byte on disk n */
	long numBytes[], 	/* write numbytes[n] bytes from disk n */
	char *writeBuffer[],	/* write to buffer[n] from disk n */
	Boolean readErr[]	
	/* returns writeErr[n]==true if disk n has a write error or parameters were bad */
);
typedef void (* ReadProc) (/* conduct physical reads from multiple drives */
	long startByte[], 	/* start read from startbyte[n] physical byte on disk n */
			/* bytes startByte..startByte+numBytes-1 must be within 0..diskSize-1 */
	long numBytes[], 		/* read numbytes[n] bytes from disk n */
	char *readBuffer[],	/* read into buffer[n] from disk n */
	Boolean readErr[]
		/* returns readErr[n]==true if disk n has a read error or parameters were bad */
);

/*
 * InitRaid provides the problem parameters
 */
void InitRaid(
	long numDisks,	
		/* problem size, you will have numDisks of real data, plus 2 disks for parity */
	long diskSize,			/* number of bytes in each disk */
	long failureRate, /* expect 1 failure in each failureRate read/write attempts */
	WriteProc physicalWrite,
		/* procedure that allows you to write to numDisks+2 disks of size diskSize */
	ReadProc physicalRead
		/* procedure that allows you to read from numDisks+2 disks of size diskSize */
);

void RepairDisk(
	long whichDisk /* index of disk that has been repaired, no more than 2 at one time */
);

/*
 *  RaidWrite and RaidRead ask you to write to the numDisks*diskSize bytes of 
 *  storage. You use WriteProc and Read proc to actually write to the (numDisks+2) 
 *  physical devices, using redundant writes to compensate for the loss of up to two 
 *	 disks. RaidWrite and RaidRead bytes are numbered 0..numDisks*diskSize-1.
 *  RaidWrite and RaidRead return true unless there is a problem performing the 
 * write/read.
 */
Boolean RaidWrite(
	long startByte,			/* write starting at this byte */
	long numBytes,			/* number of bytes to write */
	char *buffer				/* write bytes from this buffer */
);

Boolean RaidRead(
	long startByte,			/* read starting at this byte */
	long numBytes,			/* number of bytes to read */
	char *buffer				/* read bytes into this buffer */
);

void TermRaid(void);

This Challenge starts with a call to your InitRaid routine. InitRaid is provided with the number of disks (numDisks) available to your Raid 5+ implementation. Actually, numDisks represents the amount of data you need to store and protect; you actually have numDisks+2 disks available, with the additional 2 disks used to provide protection against disk failures. InitRaid is also provided with the size of the identically sized disks in bytes (numBytes), and the approximate failure rate of the disks (1 failure in failureRate read/write attempts). Finally, InitRaid is provided with two callback routines that provide you with read and write access to the physical disks in our simulated disk array, which we'll discuss below.

The Challenge evaluation consists of a large number of RaidWrite and RaidRead operations. Each RaidWrite call provides a logical address to write to (startByte) in the range 0..numDisks*diskSize-1, a number of bytes to write (numBytes), and a buffer from which to write. Your code must write the data to the simulated disks provided using the WriteProc callback, providing for error correction by writing redundant information to other disks. If WriteProc returns a writeErr[n] value of true, the write to disk n failed, and you may need to compensate. RaidRead provides a logical address to read from, a number of bytes to read, and a buffer to read into. If ReadProc returns a readErr[n] value of true, the corresponding read operation failed, and you will need to use the redundant disks to reconstruct the lost information.

At the end of the evaluation, your TermRaid routine will be called. You should free any memory you have allocated.

From time to time, a failed disk may be repaired. You'll be notified of such a repair by a call to RepairDisk. When a disk is repaired, you ought to reconstruct any necessary error correction information, because another disk may fail at any time. Up to two disks may be in "failed" mode at any given time.

WriteProc allows you to perform a write operation on all of the disks in the array simultaneously, writing numBytes[n] from writeBuffer[n] starting at physical location startByte[n] on disk n. Similarly, ReadProc allows you to perform a simultaneous read operation on all of the disks, reading numBytes[n] into readBuffer[n] starting from physical location startByte[n] on disk n. Unfortunately, the disk controllers in our simulated system are not very sophisticated: while disk I/O occurs in parallel, our disks are not able to chain sequential I/O operations. An I/O operation on the array requires read/write time proportional to the largest numBytes[n] value passed to any disk.

Our disks will have, say, 5msec average seek time, and 30MB/sec transfer rates. Your program's score will be the sum of the seek and transfer times for each read/write operation, plus the execution time required by your program. The winner will be the solution that correctly performs all read/write operations with the lowest program score.

There are no specific memory restrictions on this Challenge, but you may not allocate memory to store all of the data written to the simulated disk array, or to persistently store any error correction information. You must use the simulated disks and the WriteProc and ReadProc callbacks for that purpose.

This will be a native PowerPC Challenge, using the CodeWarrior Pro 5 environment. Solutions may be coded in C, C++, or Pascal.

Three Months Ago Winner

Congratulations to Ernst Munter (Kanata, Ontario) for taking first place in the April Text Compression Challenge. This Challenge required contestants to process input text and compress it into as few bytes as possible, while minimizing the execution time used to perform the compression. The input text consisted of English-language text; computer programs written in C, C++, or Pascal; and web pages written in html. Scoring was based on the number of characters of compressed text, with a 10% penalty added for each 100msec of execution time. Ernst's entry was the not the fastest of the 5 correct entries, but it did produce the most highly compressed output.

I used 17 test cases to evaluate the entries, including 8 text files, 3 computer programs, and 6 web pages, totaling ~5.6MB in length. Individual inputs ranged in size from less than 1000 bytes to just over 1MB. Ernst's solution compressed these inputs into about 35% of their original size. While I didn't independently verify this, Ernst observes that his code produces a smaller result for a large text file than a popular Mac compression utility, and does so in less time.

The three entries producing the smallest output all used a variant of Huffman encoding. This technique assigns bit representations of varying lengths to individual tokens in the input, with shorter Huffman codes used for tokens that occur more frequently. Ernst took the additional step of assigning a token code to each input token, and then compressing the token codes using Huffman encoding. This approach resulted in ~50% better compression than that achieved by Jan Schotsman's next best entry (although Jan's entry was submitted after the Challenge deadline and not eligible to win).

The table below lists, for each of the solutions submitted, the cumulative size of the input text and output texts, the execution time in milliseconds, the total score achieved for all test cases. It also provides the code size, data size, and programming language used for each entry. As usual, the number in parentheses after the entrant's name is the total number of Challenge points earned in all Challenges prior to this one.

NameInput Size (KB)Output Size (KB)Time (msec)ScoreCode SizeData SizeLang
Ernst Munter (587)559519121863.80242703787641216C++
Randy Boring (123)55954896256.355186714150456C++
Sebastian Maurer (101)559551421482.87551264653281394C++
Armin Schmich559544072112.0858248962400164C
Jan Schotsman (late entry)559522075845.33392806513060320C
J. T. crash1634813586C

Top Contestants

Listed here are the Top Contestants for the Programmer's Challenge, including everyone who has accumulated 10 or more points during the past two years. The numbers below include points awarded over the 24 most recent contests, including points earned by this month's entrants.

Rank Name Points
1. Munter, Ernst 243
2. Saxton, Tom 139
3. Maurer, Sebastian 78
4. Boring, Randy 56
5. Shearer, Rob 47
6. Rieken, Willeke 41
7. Heathcock, JG 33
8. Taylor, Jonathan 26
9. Brown, Pat 20
10. Downs, Andrew 12
11. Jones, Dennis 12
12. Duga, Brady 10
13. Fazekas, Miklos 10
14. Hewett, Kevin 10
15. Murphy, ACC 10
16. Selengut, Jared 10
17. Strout, Joe 10

There are three ways to earn points: (1) scoring in the top 5 of any Challenge, (2) being the first person to find a bug in a published winning solution or, (3) being the first person to suggest a Challenge that I use. The points you can win are:

1st place 20 points
2nd place 10 points
3rd place 7 points
4th place 4 points
5th place 2 points
finding bug 2 points
suggesting Challenge 2 points

Here is Ernst's winning Text Compression solution:

TextCompression.cp
Copyright © 2000
Ernst Munter

/*
  Task
  ——
  Find and implement an efficient algorithm to compress and expand text. Efficiency 
  is a compromise between compression factor and speed.
  
  Algorithm
  ————-
  The algorithm is based on the recognition that text will consist of alternating "word" 
  and "link" fragments.  Word fragments are character sequences from one character 
  set, links are sequences from a non-overlapping different set.  
  
  Each of the two sets contain 64 characters: alphanumerics + two extra characters 
  ( _ and @ ) to make up the word set, all others fall into the link set.
  
  Many of the fragments will be encountered only once, others many times.
  
  The compressed text is primarily a sequence of tokens, where each token represents 
  a fragment, or introduces a new fragment.  Each fragment is defined explicitely only 
  the first time it is encountered.  If the same lexical fragment occurs multiple times in 
  the text, a token code definition is generated for this fragment.  Subsequent 
  occurrences of the same fragment are then "quoted" by their code only.
  
  Token codes are Huffman encoded.  As a result the codes for frequently occurring 
  fragments (such as the single space between words) are coded with fewer bits than 
  less frequently repeating fragments. Two codes are reserved as flags, one for 
  introducing a fragment which only occurs once in the text; a second for introducing 
  a multiply occurring fragment plus its associated code.
  
  Implementation
  ———————
  Compressor:
  
  The text is parsed into words and links. A token cache tracks unique fragments, and 
  records their frequency. At the same time, a quotes list is built, one quote (token 
  pointer) for each fragment.
  
  Actually two caches and token sets are created, one for word fragments and a 
  separate one for link fragments.
  
  The next step is the creation of the Huffman codes for each token. To this end, all 
  tokens with a frequency > 1 are pushed on a heap (priority queue).  A Huffman tree 
  is built with intermediate nodes each holding pointers to two child nodes.  The leaf 
  nodes are the token records.  Traversing the finished tree from the root then
  provides the mechanism of assigning each leaf node (token) its code.
  
  Finally, the list of quotes is scanned, and each token or token code placed in the 
  compressed text array.
  
  A "BitPump" object provides a method of dealing with arbitrary bit size chunks.  
  The bit pump also provides the packing (and unpacking)  of the original 8-bit 
  characters into 6-bits.
  
  Expander:
  
  The expander reads a few length parameters from the compressed text  header and 
  then procedes to read the compressed text bit by bit to decode token codes.  At the 
  start it is given only the codes for the two flag tokens.  Then, each previously unseen 
  code and fragment is introduced as needed, preceded by a flag code.
  
  The two Huffman decode trees (for decoding word and link token codes) are thus 
  built one code at time, just-in-time, and linked to the decoded fragments which are 
  placed directly into the expanded text. 
  
  Optimizations
  ——————-
  List of possible optimizations considered but not applied:
  - more efficient Huffman decoder
  - recognize capitalized words as "the same" and flag
  - use a fixed dictionary of likey words, and provided default
    token codes for them which then do not need to be made explicit
  - more efficient bit pump
  - Huffman code the fragment characters
  - do run-length coding where it pays off (e.g. multiple spaces)
  
  All of these can bring additional small percentage improvements in compression at 
  the expense of added complexity and time. I have decided to keep it simple.
  
  Compiler Notes
  ———————
  One would expect the compiler to inline simple class methods such as
  
  	int SingleFlag() {return &token[0];}
  
  But the currently latest version of CodeWarrior (CW-5.3) does not do it with 
  inlining set to "Smart" and branches to a 2-instruction function when SingleFlag() is 
  called. I have replaced such instances with #define macros to get inlining. 
  
  Assumptions
  —————-
  Fragments (i.e. a word, or a sequence of non-word characters) should be less than 
  64K characters in length.  This is pretty well guaranteed in text files.
  
  Input text characters are limited to the range 0x00 to 0x7F inclusive.
  Characters above 0x7F will give incorrect results.
*/

#include <stdlib.h>
#include <string.h>
#include <ctype.h>
#include "TextCompression.h"
#include "Heap.h"
#include "Utilities.h"

//*************************  Compressor Code **********************//
//
//***************************************************************//
extern Tables gTables;

enum {
	kTokenSize		= 14,
	kMaxUniqueToken = 1<<kTokenSize,
	// = max different tokens in a single block
	
	kHashBits		= 11,
	kMaxHashTable	= 1<<kHashBits,
	kHashMask		= kMaxHashTable-1
};

struct CRC

// Standard CRC based hash method. Good hashing at some expense in time
static struct CRC {
  	enum {POLYNOMIAL=0x04c11db7L};
  	ulong table[256];
  	CRC() 
  	{
    	long i,j,x;
    	for (i=0;i<256;i++) {
      		x=i<<24;
      		for (j=0;j<8;j++) {
        		if (x<0) x=(x<<1) ^ POLYNOMIAL;
        		else x=(x<<1); 
      		}
      		table[i]=x;
    	}
  	}
	ulong HashFunction(uchar* ufrg,int frgLen)
	{
// Uses CRC on length and up to the first 4 chars of a fragment
  		ulong accum=0;
    	accum=(accum<<8) ^ table[(accum>>24) ^ frgLen]; 
    	accum=(accum<<8) ^ table[(accum>>24) ^ ufrg[0]];
    
    	if (frgLen>1) switch (frgLen)
    	{
default:
case 4:	accum=(accum<<8) ^ table[(accum>>24) ^ ufrg[3]];
case 3:	accum=(accum<<8) ^ table[(accum>>24) ^ ufrg[2]];
case 2:	accum=(accum<<8) ^ table[(accum>>24) ^ ufrg[1]];
    	}
 // returns 24 bits of the CRC + 8 bits form the fragment length
		return (accum & 0xFFFFFF) | (frgLen << 24);
	}
} crc;

struct Token

// Each Token describes a unique fragment of text 
// which occurs freq times in a block
// Fragments alternate as words and links, each with a
// different character set of 64 chars.
struct Token {
	Token*	next;		// linked list
	union	{
		ulong	code;		// codebits at LSB, code length at 5 MSB
		ulong	hashValue;
	};
	uchar*	frag;		// fragment
	ushort	fragLength;
	ushort 	freq;		// number of times in block, 
											// set to 0 after code is published
						
	Token(){}
	
	Token(uchar* frg,int frgLen,int n) :
		frag(frg),
		fragLength(frgLen),
		freq(n),
		next(0),
		code(0),
		hashValue((frg)?crc.HashFunction(frg,frgLen):0)
	{}
	
	void 	SetFrequency(int x){freq=x;}
	
//	int		CodeLength(){return code >> 27;}
#define CodeLength() code >> 27
	
	ulong 	MisMatch(uchar* frg,int frgLen,ulong hvalue)
	{
		ulong d = hvalue ^ hashValue;
		if (d == 0) 
		{
			uchar* 	f1=frg;
			uchar*	f2=frag;
			d=*f1 ^ *f2;
			if (d == 0) for (int i=1;i<frgLen;i++)
			{
				d=*++f1 ^ *++f2;
				if (d) break;
			}
		}	
		return d;
	}
};
typedef Token* TokenPtr;

struct TokenNode

// A TokenNode is used in bulding the Huffman tree of tokens
static long gMaxLength;
struct TokenNode
{
	Token* t;
	int	weight;
	TokenNode*	parent;			// 
	TokenNode*	child[2];	// subtrees
	
	TokenNode(){}
	
	TokenNode(Token* token) :
		t(token),weight(t->freq),parent(0)
	{
		child[0]=child[1]=0;
	}
	
	TokenNode(TokenNode & ch1,TokenNode & ch2) :
		t(0),
		weight(ch1.weight + ch2.weight),
		parent(0)
	{
		child[0]=&ch1;ch1.parent=this;
		child[1]=&ch2;ch2.parent=this;
	}
	
	int Freq(){return weight;}
	
	void Traverse(ulong code,int length)
// Builds the Huffman codes for each leaf child recursively.
	{
		if (!child[0] && !child[1])	// leaf
		{
			if (length==0) length=1;
			t->code=code | (length<<27);
			if (gMaxLength < length) gMaxLength = length;
		} else
		{
			assert(child[0] && child[1]);
 			child[0]->Traverse(code<<1,length+1);
			child[1]->Traverse((code<<1) | 1,length+1);
			delete child[0];
			delete child[1];
		}
	}
};
typedef TokenNode* TokenNodePtr;

struct HashBucket

struct HashBucket {
	Token*	first;
	Token*	last;
};

struct TokenSet

// TokenSet holds all tokens of one kind (word or link)
struct TokenSet {
	int		fill;					// number of tokens defined
	int		multi;				// number of tokens used multiple times
	int		maxFragLen;	// longest fragment encountered
	int		fragLenSize;// number of bits to represent longest frag length
	int 	codeLenSize;	// number of bits to represent longest code
	int		numCodes;		// number of Huffman codes defined
// the maximum number of unique tokens, and the hash table size
// are defined statically to some reasonable values
	Token 	token[kMaxUniqueToken];
	HashBucket	hashTable[kMaxHashTable];
	
	TokenSet():
		fill(2),multi(0),maxFragLen(0),fragLenSize(0),
		codeLenSize(0)		
	{
		token[0]=Token(0,0,1);// single occurrence flag
		token[1]=Token(0,0,2);// multiple occurrrence flag
		memset(hashTable,0,sizeof(hashTable));
	}
	
	Token*	Cache(uchar* frg,int frgLen)
// Finds the fragment in the cache, and increments its frequency.
// If fragment is not found, a new token is defined and cached. 
	{
		int hash=crc.HashFunction(frg,frgLen);
		HashBucket* HB=&hashTable[kHashMask & hash];	
		Token* t=HB->first;							
		while (t)
		{											
			if (0 == t->MisMatch(frg,frgLen,hash))
			{
				if (t->freq == 1)
					multi++;
				t->freq++;							
													
				return t;// found cached token
			}
			t=t->next;
		}
		// add a new token to the cache
		if (fill<kMaxUniqueToken-1)
		{
			if (frgLen>maxFragLen)maxFragLen=frgLen;
			token[fill]=Token(frg,frgLen,1);
			t=&token[fill++];
			
			if (HB->last)
			{
				HB->last->next=t;
			} else HB->first=t;
			HB->last=t;
		}
		return t;// could be 0!
	}
	
// The first two token spaces are reserved for the flags
//	Token* SingleFlag() {return &token[0];}
//	Token* MultiFlag()  {return &token[1];}

#define SingleFlag() (&token[0])
#define MultiFlag() (&token[1])

	void MakeTokenCodes()
// Pushes all relevant tokens on a priority queue, generates
// the Huffman tree, and then traverses it to assign each token a code.
	{
		Finish();
	// collect multi-use tokens on heap
		Heap<TokenNodePtr,int> qmap(3+multi);
		Token* t=token;
		for (int i=0;i<fill;i++,t++){
			if (t->freq > 1)
			qmap.Insert(new TokenNode(t));
		}
		numCodes=qmap.heapSize;
		
	// build the Huffman tree for link codes
		while(qmap.heapSize > 1) 
		{
			TokenNodePtr ch1=qmap.Pop();
			TokenNodePtr ch2=qmap.Pop();
			TokenNodePtr parent=new TokenNode(*ch1,*ch2);
			qmap.Insert(parent);
		}
	
		TokenNode* root=qmap.Pop();
	
		gMaxLength=0;
		root->Traverse(0,0);
		codeLenSize=BitsNeeded(gMaxLength);
		
		delete root;	
	}
	
	void Finish()
// Updates flag tokens with their frequencies and calculates the number
// of bits needed to send the fragment length when sending fragments
// explicitely.  The longest actually occurring fragment determines this.
	{
		int f0=fill-multi;
		if (f0<2) f0=2;
		token[0].SetFrequency(f0);
		int f1=multi;
		if (f1<2) f1=2;
		token[1].SetFrequency(f1);
		fragLenSize=BitsNeeded(maxFragLen);
	}
	
	void SendParms(BitPump & B)
// Sends the compressed text header, defining all constants needed
// by the expander before decoding can begin.
	{
		B.Send(fragLenSize,5);
		B.Send(codeLenSize,5);
		int numCodesSize=BitsNeeded(numCodes);
		B.Send(numCodesSize,5);
		B.Send(numCodes,numCodesSize);
		B.Send(token[0].CodeLength(),codeLenSize);
		B.Send(token[0].code,token[0].CodeLength());
		B.Send(token[1].CodeLength(),codeLenSize);
		B.Send(token[1].code,token[1].CodeLength());
	}
	
	void Send(Token* t,BitPump & B)
// Sends a single token, either:
//		just the token code (2nd or later occurrence)
//		"SingleFlag" and a single fragment which will not recur
//		"MultiFlag" followed by a fragment and its token code
	{
		if (t->freq==0)
		{
			B.Send(t->code,t->CodeLength());
		} else if (t->freq==1)
		{
			B.Send(SingleFlag()->code,SingleFlag()->CodeLength());
			B.CompressFragment((uchar*)t->frag,
				t->fragLength,fragLenSize); 
		} else
		{
			B.Send(MultiFlag()->code,MultiFlag()->CodeLength());
			B.CompressFragment((uchar*)t->frag,
				t->fragLength,fragLenSize); 
			B.Send(t->CodeLength(),codeLenSize);
			B.Send(t->code,t->CodeLength());
		// set frq=0 so this token will trigger code-only from now on
			t->freq=0;
		}
	}
};

NextLink

// NextLink and NextWord parse the next link or word respectively
// from the input text and return the length of the fragment.
inline int NextLink(uchar* inText,uchar* endText)
{
	uchar* wordStart=inText;
	while ((inText<endText) && 
		!gTables.IsWordChar(*inText)) inText++;
	return inText-wordStart;
}

NextWord

inline int NextWord(uchar* inText,uchar* endText)
{
	uchar* wordStart=inText;
	while ((inText<endText) && 
		gTables.IsWordChar(*inText)) inText++;
	return inText-wordStart;
}

Compress

static uchar* Compress(uchar *text,long length,
	uchar* compressedText,long & compressedLength)
{
// Compresses a block of text, until we either run out of text to compress, 
// unique quote space (fragLimit, allocated as a function of text size),
// or the number of unique tokens required exceeds the static limit.
// In the unlikely case we do run out, we return the compressed block
// and caller will call Compress again to compress the remaining text.
	uchar* ptext=text;
	uchar* textEnd=text+length;
	TokenSet* linkTokens=new TokenSet;
	TokenSet* wordTokens=new TokenSet;
	int estNumQuotePairs=length/5;
	int fragLimit=2*estNumQuotePairs+128;
	TokenPtr* quotes=new TokenPtr[fragLimit+1];
	TokenPtr* q=quotes;
	
// Text might start with a word or a link fragment:
	bool linkFirst=gTables.IsLinkChar(*text);
	int fragPair=0;
	
// scan text and gather all tokens 	
	if (linkFirst)
	{
		int fragLength=NextLink(ptext,textEnd);
		*q++=linkTokens->Cache(ptext,fragLength);
		ptext+=fragLength;
	}
	
	while ((ptext<textEnd)&&(fragPair<fragLimit))
	{
		int fragLength=NextWord(ptext,textEnd);
		
		Token* 
		token=wordTokens->Cache(ptext,fragLength);
		if (0==token)
			break;
		*q++=token;
		if ((ptext+=fragLength) >= textEnd)
			break;
			
		fragLength=NextLink(ptext,textEnd);
		
		token=linkTokens->Cache(ptext,fragLength);
		if (0==token)
			break;
		*q++=token;
		ptext+=fragLength;
		
		fragPair++;
	}
	
	int numQuotes=q-quotes;
	
// Make the codes

	linkTokens->MakeTokenCodes();
	wordTokens->MakeTokenCodes();

// Send all quotes ...	
	q=quotes;
	TokenPtr* endQuotes=quotes+numQuotes;
// using a bit pump	
	BitPump B((uchar*)compressedText,0);

// send the header of a few parameters	
	B.Send(linkFirst,1);
	int numQuotesSize=BitsNeeded(numQuotes);
	B.Send(numQuotesSize,5);
	B.Send(numQuotes,numQuotesSize);
	linkTokens->SendParms(B);
	wordTokens->SendParms(B);

// scan the quotes array wich will alternate between word and link tokens	
	if (linkFirst)
	{
		Token* t=*q++;
		assert(t);
		linkTokens->Send(t,B);
	}

	while (q<endQuotes)
	{
		Token* t=*q++;
		assert(t);
		wordTokens->Send(t,B);
		
		if (q>=endQuotes)
			break;
		
		t=*q++;
		assert(t);
		linkTokens->Send(t,B);
	}
	
 	uchar* endOutText=B.Close();
	compressedLength=endOutText-compressedText;

// Clean up explicitely allocated memory.
// Automatic objects (Heap, BitPump) will automatically destruct
	delete [] quotes;
	delete wordTokens;	
	delete linkTokens;		
	
	return ptext;
}

//*************************  Expander Code ************************//
//
//*****************************************************************//

struct Node

struct Node {				// Node in decode tree
	long	fragLength;	// fragLength=0 for non-terminal nodes
	uchar*	frag;			// frag=0 for FLAG nodes
	Node*	child[2];
};

struct Expander

struct Expander {
	int		fragLenSize;
	int 	codeLenSize;
	int		numCodesSize;
	int		numCodes;
	int		flag0size;
	int		flag0code;
	int		flag1size;
	int		flag1code;
	int 	numNodes;
	Node* 	node;
	Node*	nextNode;
	Node*	endNode;
	int		numQuotesRcvd;
	bool	link;
	
// Constructor parses compressed file header to establish needed
// parameters and define the bootstrap nodes of the Huffman tree  
	Expander(BitPump & B,bool lnk) :
		fragLenSize(B.Receive(5)),
		codeLenSize(B.Receive(5)),
		numCodesSize(B.Receive(5)),
		numCodes(B.Receive(numCodesSize)),
		flag0size(B.Receive(codeLenSize)),
		flag0code(B.Receive(flag0size)),
		flag1size(B.Receive(codeLenSize)),
		flag1code(B.Receive(flag1size)),
		numNodes(2*numCodes-1),
		node(new Node[numNodes]),
		nextNode(node),
		endNode(node+numNodes),
		numQuotesRcvd(0),
		link(lnk)
	{
		memset(node,0,numNodes*sizeof(Node));
		nextNode=node+1;
		MakeNode(flag0size,flag0code,0,1);// single flag
		MakeNode(flag1size,flag1code,0,2);// multi flag
	}
	
	~Expander(){delete [] node;}
	
	void MakeNode(int size,ulong code,uchar* frag,int fragLength)
// Traces from root to the node define by the code, assigning missing
// intermediate nodes as needed.
// Installs fragment length and text pointer in the leaf node.
	{
		Node* n=node;
		int bit;
		for (int i=size-1;i>0;—i)
		{
			bit=1 & (code >> i);
			if (0==n->child[bit]) 
			{
				assert(nextNode < endNode);
				n->child[bit]=nextNode++;
			}
			n=n->child[bit];
		}
		bit=1 & code;
		assert(nextNode < endNode);
		Node* terminalNode=nextNode++;
		assert(0==n->child[bit]);
		n->child[bit]=terminalNode;
		terminalNode->fragLength=fragLength;
		terminalNode->frag=frag;
	}
	
	int GetLinkQuote(BitPump & B,uchar* dest)
// Parses one link quote from the compressed text.
// returns the number of expanded characters decoded
	{
		int bit=B.Receive(1);
		Node* n=node->child[bit];
		int fragLength;
		for (;;)
		{
			int xflag=n->fragLength;
			if (xflag)
			{
				uchar* frag=n->frag;
				if (frag)	// cached quote
				{
					fragLength=xflag;
					
					*dest=*frag;
					while (—xflag) *++dest = *++frag;
						
					break;
				} else	// new quote
				{
					fragLength=B.Receive(fragLenSize);
					B.ExpandLink(dest,fragLength);
					if (xflag==1)	// SingleFlag
						break;
					// else MultiFlag: cache the code
					int codeLength=B.Receive(codeLenSize);
					int code=B.Receive(codeLength);
					MakeNode(codeLength,code,dest,fragLength);
					break;
				}
			} else
			{
				bit=B.Receive1Bit();// trace path in tree, bit by bit
				n=n->child[bit];
				assert(n);
			}
		} 
		return fragLength;
	}
	
	int GetWordQuote(BitPump & B,uchar* dest)
// Parses one word quote from the compressed text.
// returns the number of expanded characters decoded 
	{
		int bit=B.Receive(1);
		Node* n=node->child[bit];
		int fragLength;
		for (;;)
		{
			int xflag=n->fragLength;
			if (xflag)
			{
				uchar* frag=n->frag;
				if (frag)	// cached quote
				{
					fragLength=xflag;
					
					*dest=*frag;
					while (—xflag) *++dest = *++frag;
					
					break;
				} else	// new quote
				{
					fragLength=B.Receive(fragLenSize);
					B.ExpandWord(dest,fragLength);
					if (xflag==1)	// SingleFlag
						break;
					// else MultiFlag: cache the code
					int codeLength=B.Receive(codeLenSize);
					int code=B.Receive(codeLength);
					MakeNode(codeLength,code,dest,fragLength);
					break;
				}
			} else
			{
				bit=B.Receive1Bit();// trace path in tree, bit by bit
				n=n->child[bit];
				assert(n);
			}
		} 
		return fragLength;
	}
};

Expand
static uchar* Expand(
	uchar* compressedText,long compressedLength,
	uchar* expandedText,long & expandedLength)
{	
// Creates expanders for both word and link fragments,
// reads the compressed text header fields, and expands the text
	BitPump B(
		(uchar*)compressedText,
		(uchar*)compressedText+compressedLength);
	bool linkFirst=B.Receive(1);
	int numQuotesSize=B.Receive(5);
	int numQuotes=B.Receive(numQuotesSize);
	Expander linkCodes(B,true);
	Expander wordCodes(B,false);
	uchar* etext=expandedText;
	
// Starting with either a link or a word fragment, get alternating
// link and word quotes from the bit pump, decode into plain text,
// and place results into the expandedText array. 
	if (linkFirst)
	{
		int fragLength=linkCodes.GetLinkQuote(B,etext);
		etext += fragLength;
		numQuotes—;
	}
	
	while (numQuotes)
	{
		int fragLength=wordCodes.GetWordQuote(B,etext);
		etext += fragLength;
		numQuotes—;
		if (numQuotes<=0)
			break;
		
		fragLength=linkCodes.GetLinkQuote(B,etext);
		etext += fragLength;
		numQuotes—;
	}

	expandedLength = etext-expandedText;
// return the state of the compressed text pointer, in case
// the whole text was not compressed as a single block.
	return B.Buffer();	
}

//*********************  External Functions ***********************//
//
//*****************************************************************//

InitCompression
void * /* yourStorage */ InitCompression(void)
{
// No persistent storage needed
	return 0;
}

CompressText
long /* compressedLength */ CompressText(
 	char *inputText,						/* text to be compressed */
	long numInputChars,				/* length of inputText in bytes */
	char *compressedText,		/* return compressedText here */
	const void *yourStorage	/* storage returned by InitCompression */
) {
#pragma unused(yourStorage)

	uchar* it=(uchar*)inputText;
	uchar* ct=(uchar*)compressedText;
	long lit=numInputChars,lct;
	long compressedLength=0;
// Usually, text should compress in a single block.  
// Just in case it does not, this loop will compress text in blocks
	do
	{
		uchar* t2=Compress(it,lit,ct,lct);
		compressedLength += lct;
		lit -= t2 - it;
		it = t2;
		ct += lct;
	} while (lit>0);
	
	return compressedLength;
}

ExpandText
long /* expandedLength */ ExpandText(
	char *compressedText,		/* encoded text to be expanded */
	long compressedLength,		/* length of encoded text in bytes */
	char *expandedText,				/* return expanded text here */
	const void *yourStorage	/* storage returned by InitCompression */
) {
#pragma unused(yourStorage)
	
	uchar* ct=(uchar*)compressedText;
	uchar* et=(uchar*)expandedText;
	long  lct=compressedLength,let;
	long  expandedLength=0;
// Usually, text should come as a single compressed block.  
// Just in case it does not, this loop will expand multiple blocks
	do
	{
		uchar* t2=Expand(ct,lct,et,let);
		expandedLength += let;
		lct -= t2 - ct;
		ct = t2;
		et += let;
	} while (lct > 0);	
	
	return expandedLength;
}

TermCompression
void TermCompression(
	void *yourStorage		/* storage returned by InitCompression */
) {
#pragma unused(yourStorage)
// No persistent storage to destroy
}

Utilities.h

#ifndef UTILITIES_H
#define UTILITIES_H

#define NDEBUG
#include <assert.h>

typedef unsigned char uchar;
typedef uchar* ucharPtr;
typedef unsigned short ushort;
typedef unsigned long ulong;


inline int Caps(uchar* w1,uchar* w2)
// returns true if capitalization of w1 and w2 differs.
{
	return ((*w1 ^ *w2) & 0x20);
}

inline int BitsNeeded(ulong x)
// returns number of bits needed to encode range 0 to x
{
	if (x==0) return 1;
	int n=0;
	do { x >>= 1; n++;} while(x);
	return n;
}

class Tables {
public:
	Tables();
	bool 	IsWordChar(uchar c){return textchars[c];}
	bool 	IsLinkChar(uchar c){return !textchars[c];}
	uchar 	Ascii2Code(uchar c){return ascii2code[c];}
	uchar 	Code2Link(uchar c){return code2link[c];}
	uchar 	Code2Alnum(uchar c){return code2alnum[c];}
private:
	bool 	textchars[256];
	uchar	ascii2code[256];
	uchar	code2link[64];
	uchar	code2alnum[64];	
};


class BitPump {
public:
	BitPump(uchar* text,uchar* end):
		buffer(text),
		endBuffer(end),
		acc(0),
		fill(0)
	{}

	void Send(ulong x,int numBits)
	{
		acc <<= numBits;
		acc |= x & ((1<<numBits)-1);
		fill+=numBits;
		while (fill >= 8)
		{
			fill -= 8;
			*buffer++ = (acc >> fill) & 0xFF;
		}
	}
	
	ulong Receive(int numBits)
	{
		while (fill < numBits)
		{
			if (buffer>=endBuffer) return -1;
			acc = (acc<<8) | *buffer++;
			fill+=8;
		}
		fill -= numBits;
		ulong x = (acc>>fill) & ((1<<numBits)-1);
		return x;
	}
	
	ulong Receive1Bit()
	{
		if (!fill)
		{
			if (buffer>=endBuffer) return -1;
			acc = (acc<<8) | *buffer++;
			fill+=8;
		}
		fill —;
		return (acc>>fill) & 1;
	}
	
	void CompressFragment(uchar* frag,int length,int numBits);
	uchar* ExpandLink(uchar* outText,int length);
	uchar* ExpandWord(uchar* outText,int length);
	uchar* Buffer(){return buffer;}
	uchar* Close();
	
private:
	ulong 	Receive6Bits();
	uchar*	buffer;
	uchar*	endBuffer;
	ulong	acc;
	long	fill;
};

#endif // UTILITIES_H
Utilities.CP
#include <ctype.h>
#include "Utilities.h"

//********************  Utilities Implementation ******************//
//
//*****************************************************************//

Tables gTables;

Tables::Tables()
{
	int numAlnum=0;
	for (int c=0;c<128;c++)
	{
		if (isalnum(c))
		{
			textchars[c] = true;
			numAlnum++;
		} else textchars[c] = false;
	}
		 
	if (!textchars['@']){textchars['@']=true;numAlnum++;}
	if (!textchars['_']){textchars['_']=true;numAlnum++;}
	assert(numAlnum == 64);
		
	int alnumCode=0,linkCode=0;
	for (int c=0;c<128;c++)
	{
		if (textchars[c])
		{
			code2alnum[alnumCode]=c;
			ascii2code[c]=alnumCode++;
		} else
		{
			code2link[linkCode]=c;
			ascii2code[c]=linkCode++;
		}
		textchars[128+c] = textchars[c];
		ascii2code[128+c] = ascii2code[c];
	}
}

inline ulong BitPump::Receive6Bits()
// return -1U if buffer empty
{
	if (fill < 6)
	{
		if (buffer>=endBuffer) return -1;
		acc = (acc<<8) | *buffer++;
		fill+=8;
	}
	fill -= 6;
	return (acc>>fill) & 63;
}

void BitPump::CompressFragment(uchar* frag,int length,int numBits)
{
	uchar* s=frag-1;
	Send(length,numBits);		
	for (int i=0;i<length;i++)
	{
		int c=*++s;
		Send(gTables.Ascii2Code(c),6);// xlate ASCII to 6 bits 
	}
}

uchar* BitPump::ExpandLink(uchar* outText,int length)
{
	for (int i=0;i<length;i++)
	{
		*outText++ =
		gTables.Code2Link(Receive6Bits());// xlate to ASCII
	}
	return outText;
}
uchar* BitPump::ExpandWord(uchar* outText,int length)
{
	for (int i=0;i<length;i++)
	{
		*outText++ =
		gTables.Code2Alnum(Receive6Bits());// xlate to ASCII
	}
	return outText;
}
	
uchar* BitPump::Close()
{
	if (fill)
		Send(0,7&(32-fill));// zero pad last byte
	return buffer;
}
Heap.h
#ifndef HEAP_H
#define HEAP_H

template <class Base,class Value>
// Note: class Base must be a pointer to a class
//		having a member Freq() of type Value
// Value must have the operator > defined
struct Heap {
	Base* 	heapBase;
	int		heapSize;
	int 	maxHeapSize;
	Heap(int size) :
		heapBase(new Base[size+1]),
		heapSize(0),
		maxHeapSize(size){}
	~Heap(){Clear();}
	void Clear(){delete[] heapBase;}
	void Init(int size){
		heapBase=new Base[size+1];
		heapSize=0;
		maxHeapSize=size;
	}
	void Insert(Base k) 
	{
    	int i=++heapSize;
    	Value wk=k->Freq();
    	int j=i>>1;
    	Base z;
    	while (j && (wk < (z=heapBase[j])->Freq())) 
    	{
      		heapBase[i]=z;     
       		i=j;
      		j=i>>1;
    	}
    	heapBase[i]=k;    
  	}
  
  	Base Pop() 
  	{
//the node at heapBase[1] is the lowest weight
//it is removed from the heap and returned
//the heap is readjusted.
    	Base rc=heapBase[1];
    	Base k=heapBase[heapSize—];
    	if (heapSize<=1) {
      		heapBase[1]=k;            
      		return rc;
    	}
    	int i=1,j=2;
    	Value wk=k->Freq();
    	while (j<=heapSize) 
    	{
      		if ((j<heapSize)
      		&& (heapBase[j]->Freq() > heapBase[j+1]->Freq()))
        	j++;
      		if (heapBase[j]->Freq() > wk)
        		break;
      		heapBase[i]=heapBase[j];  
      		i=j;j+=j;
    	}
    	heapBase[i]=k;        
    	return rc;
  	}
};

#endif
 
AAPL
$567.34
Apple Inc.
+42.59
MSFT
$39.37
Microsoft Corpora
-0.32
GOOG
$527.22
Google Inc.
+0.28

MacTech Search:
Community Search:

Software Updates via MacUpdate

Cyberduck 4.4.4 - FTP and SFTP browser....
Cyberduck is a robust FTP/FTP-TLS/SFTP browser for the Mac whose lack of visual clutter and cleverly intuitive features make it easy to use. Support for external editors and system technologies such... Read more
TechTool Pro 7.0.3 - Hard drive and syst...
TechTool Pro is now 7, and this is the most advanced version of the acclaimed Macintosh troubleshooting utility created in its 20-year history. Micromat has redeveloped TechTool Pro 7 to be fully 64... Read more
MacFamilyTree 7.1.6 - Create and explore...
MacFamilyTree gives genealogy a facelift: it's modern, interactive, incredibly fast, and easy to use. We're convinced that generations of chroniclers would have loved to trade in their genealogy... Read more
EtreCheck 1.9.9 - For troubleshooting yo...
EtreCheck is a simple little app to display the important details of your system configuration and allow you to copy that information to the Clipboard. It is meant to be used with Apple Support... Read more
TeamViewer 9.0.28116 - Establish remote...
TeamViewer gives you remote control of any computer or Mac over the Internet within seconds, or can be used for online meetings. Find out why more than 200 million users trust TeamViewer! Free for... Read more
Viber 4.1.0 - Send messages and make cal...
Viber lets you send free messages and make free calls to other Viber users, on any device and network, in any country! Viber syncs your contacts, messages and call history with your mobile device,... Read more
Apple iOS 7.1.1 - The latest version of...
The latest version of iOS can be downloaded through iTunes. Apple iOS 7 brings an all-new design and all-new features. Simplicity Simplicity is often equated with minimalism. Yet true simplicity is... Read more
1Password 4.3 - Powerful password manage...
1Password is a password manager that uniquely brings you both security and convenience. It is the only program that provides anti-phishing protection and goes beyond password management by adding Web... Read more
Lens Blur 1.3.0 - True out-of-focus boke...
Let Lens Blur transform your existing photo into true SLR-quality out-of-focus bokeh effect! Everyone needs a gorgeous personalized background for a social profile, blog, Web/UI design, presentation... Read more
VMware Fusion 6.0.3 - Run Windows apps a...
VMware Fusion allows you to create a Virtual Machine on your Mac and run Windows (including Windows 8.1) and Windows software on your Mac. Run your favorite Windows applications alongside Mac... Read more

Latest Forum Discussions

See All

Galaxy Conquerors Review
Galaxy Conquerors Review By Jennifer Allen on April 24th, 2014 Our Rating: :: RETRO SHOOTINGUniversal App - Designed for iPhone and iPad Old school shooting is fun but inaccurate in Galaxy Conquerors.   | Read more »
Yomi Review
Yomi Review By Rob Thomas on April 24th, 2014 Our Rating: :: C-C-C-COMBO BREAKERiPad Only App - Designed for the iPad Round One – Fight! No quarters required for this iOS adaptation of a tabletop adaptation of the arcade fighting... | Read more »
Injustice: Gods Among Us Updated with Ne...
Injustice: Gods Among Us Updated with New Characters, Leaderboards, Gear, and Online Multiplayer Posted by Rob Rich on April 24th, 2014 [ | Read more »
Spin It Review
Spin It Review By Jordan Minor on April 24th, 2014 Our Rating: :: SPIN ME RIGHT ROUNDUniversal App - Designed for iPhone and iPad Spin It has a fine puzzle game model, but its execution lacks energy.   | Read more »
Productivity App NoteSuite is Having its...
Productivity App NoteSuite is Having its Biggest Sale Ever, Just for One Week Posted by Rob Rich on April 24th, 2014 [ permalink ] | Read more »
Wayward Souls Review
Wayward Souls Review By Carter Dotson on April 24th, 2014 Our Rating: :: CARRY ON, WAYWARD SONUniversal App - Designed for iPhone and iPad Wayward Souls is a roguelike-inspired action-RPG that sets a high bar for other games to... | Read more »
The Sandbox Gets Update, Receives New Ca...
The Sandbox Gets Update, Receives New Campaign and New Elements Posted by Tre Lawrence on April 24th, 2014 [ permalink ] Universal App - Designed for iPhone and iPad | Read more »
Football Management Simulator One For El...
Football Management Simulator One For Eleven Released Worldwide Today for iOS Posted by Simon Reed on April 24th, 2014 [ permalink ] Free-To-Play football management title One For E | Read more »
Leo’s Fortune Review
Leo’s Fortune Review By Jordan Minor on April 24th, 2014 Our Rating: :: FORTUNATE SONUniversal App - Designed for iPhone and iPad Leo’s Fortune delivers a platforming experience as creative and refined as any console game.   | Read more »
Suited Up (Games)
Suited Up 1.0 Device: iOS Universal Category: Games Price: $1.99, Version: 1.0 (iTunes) Description: Suited Up is a difficult, one-touch platformer that requires players to visualize each jump. The controls in Suited Up are simple,... | Read more »

Price Scanner via MacPrices.net

16GB 1st generation iPad mini available for $...
Radio Shack has a select number of refurbished 1st generation 16GB WiFi iPad minis available for $199.99 on their online store. Choose free shipping or free ship-to-store. We expect these to sell out... Read more
13-inch 2.5GHz MacBook Pro on sale for $100 o...
B&H Photo has the 13″ 2.5GHz MacBook Pro on sale for $1099 including free shipping plus NY sales tax only. Their price is $100 off MSRP. Read more
iPad Sales “Lull” A Reality Correction Of Unm...
I have lots of time for Jean-Louis Gassée, the former Apple Computer executive (1981 to 1990) who succeeded Steve Jobs as head of Macintosh development when the latter was dismissed in 1985. Mr.... Read more
Apple Makes OS X Betas Available To All – Wit...
Apple’s OS X Beta Seed Program, which lets you install the latest pre-release builds, try it out, and submit your feedback, is now open to anyone who wants to sign on rather than to developers and... Read more
Apple Releases iOS 7.1.1 Update
The latest iOS 7.1.1 update contains improvements, bug fixes and security updates, including: • Further improvements to Touch ID fingerprint recognition • Fixes a bug that could impact keyboard... Read more
Logitech Announces Thinner, Lighter, More Fle...
Logitech has announced an update to its Ultrathin for iPad Air, iPad mini and iPad mini with Retina display, improving the flexibility and design of its award-winning predecessor with an even thinner... Read more
Logitech Introduces Hinge, Big Bang and Turna...
Logitech has announced expansion of its tablet product line with three new cases – the Logitech Hinge, the Logitech Big Bang and the Logitech Turnaround – each for the iPad Air, iPad mini and iPad... Read more
WaterField’s Rough Rider Leather Messenger Ba...
WaterField Designs have announced the new 15-inch size of their popular Rough Rider leather messenger bag, a vintage-looking bag that combines Old West charm and ruggedness with distinctly modern... Read more
New Mac Pro on sale, save $100 on the 4-Core...
J&R has the new 4-Core Mac Pro in stock today and on sale for $2899 including free shipping plus NY sales tax only. Their price is $100 off MSRP, and it’s the lowest price available for this... Read more
Apple refurbished iMacs available for up to $...
The Apple Store has Apple Certified Refurbished 2013 iMacs available for up to $300 off the cost of new models. Apple’s one-year warranty is standard, and shipping is free. - 27″ 3.4GHz iMac – $1699... Read more

Jobs Board

*Apple* Solutions Consultant (ASC) - Apple (...
**Job Summary** The ASC is an Apple employee who serves as an Apple brand ambassador and influencer in a Reseller's store. The ASC's role is to grow Apple Read more
*Apple* Solutions Consultant (ASC) - Apple (...
**Job Summary** The ASC is an Apple employee who serves as an Apple brand ambassador and influencer in a Reseller's store. The ASC's role is to grow Apple Read more
Position Opening at *Apple* - Apple (United...
…customers purchase our products, you're the one who helps them get more out of their new Apple technology. Your day in the Apple Store is filled with a range of Read more
*Apple* Solutions Consultant (ASC) - Apple (...
**Job Summary** The ASC is an Apple employee who serves as an Apple brand ambassador and influencer in a Reseller's store. The ASC's role is to grow Apple Read more
*Apple* Inc. Research Data Specialist - Appl...
…of Worldwide Market Research & Intelligence. The team is responsible for conducting Apple branded consumer market research. It is also responsible for analyzing data Read more
All contents are Copyright 1984-2011 by Xplain Corporation. All rights reserved. Theme designed by Icreon.