Another bzip2 update and speedup from Manuel Novoa III, with some

additional changes (primarily lots of comments) from Rob Landley.

1 files changed, 267 insertions, 223 deletions

diff --git a/archival/libunarchive/decompress_bunzip2.c b/archival/libunarchive/decompress_bunzip2.c
index b4bcb0b49..f00c43913 100644
--- a/archival/libunarchive/decompress_bunzip2.c
+++ b/archival/libunarchive/decompress_bunzip2.c
@@ -10,6 +10,36 @@
 		LGPL (http://www.gnu.org/copyleft/lgpl.html
 */
 
+/*
+	Size and speed optimizations by Manuel Novoa III  (mjn3@codepoet.org).
+
+	More efficient reading of huffman codes, a streamlined read_bunzip()
+	function, and various other tweaks.  In (limited) tests, approximately
+	20% faster than bzcat on x86 and about 10% faster on arm.
+
+	Note that about 2/3 of the time is spent in read_unzip() reversing
+	the Burrows-Wheeler transformation.  Much of that time is delay
+	resulting from cache misses.
+
+	I would ask that anyone benefiting from this work, especially those
+	using it in commercial products, consider making a donation to my local
+	non-profit hospice organization in the name of the woman I loved, who
+	passed away Feb. 12, 2003.
+
+		In memory of Toni W. Hagan
+
+		Hospice of Acadiana, Inc.
+		2600 Johnston St., Suite 200
+		Lafayette, LA 70503-3240
+
+		Phone (337) 232-1234 or 1-800-738-2226
+		Fax   (337) 232-1297
+
+		http://www.hospiceacadiana.com/
+
+	Manuel
+ */
+
 #include <setjmp.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -38,39 +68,31 @@
 /* Other housekeeping constants */
 #define IOBUF_SIZE			4096
 
-static char * const bunzip_errors[]={NULL,"Bad file checksum","Not bzip data",
-		"Unexpected input EOF","Unexpected output EOF","Data error",
-		 "Out of memory","Obsolete (pre 0.9.5) bzip format not supported."};
-
 /* This is what we know about each huffman coding group */
 struct group_data {
 	/* We have an extra slot at the end of limit[] for a sentinal value. */
 	int limit[MAX_HUFCODE_BITS+1],base[MAX_HUFCODE_BITS],permute[MAX_SYMBOLS];
-	char minLen, maxLen;
+	int minLen, maxLen;
 };
 
 /* Structure holding all the housekeeping data, including IO buffers and
    memory that persists between calls to bunzip */
 typedef struct {
-	/* For I/O error handling */
-	jmp_buf jmpbuf;
-	/* Input stream, input buffer, input bit buffer */
-	int in_fd,inbufCount,inbufPos;
-	unsigned char *inbuf;
+	/* State for interrupting output loop */
+	int writeCopies,writePos,writeRunCountdown,writeCount,writeCurrent;
+	/* I/O tracking data (file handles, buffers, positions, etc.) */
+	int in_fd,out_fd,inbufCount,inbufPos /*,outbufPos*/;
+	unsigned char *inbuf /*,*outbuf*/;
 	unsigned int inbufBitCount, inbufBits;
-	/* Output buffer */
-	char outbuf[IOBUF_SIZE];
-	int outbufPos;
 	/* The CRC values stored in the block header and calculated from the data */
-	unsigned int crc32Table[256],headerCRC, dataCRC, totalCRC;
+	unsigned int crc32Table[256],headerCRC, totalCRC, writeCRC;
 	/* Intermediate buffer and its size (in bytes) */
 	unsigned int *dbuf, dbufSize;
-	/* State for interrupting output loop */
-	int writePos,writeRun,writeCount,writeCurrent;
-
 	/* These things are a bit too big to go on the stack */
 	unsigned char selectors[32768];			/* nSelectors=15 bits */
 	struct group_data groups[MAX_GROUPS];	/* huffman coding tables */
+	/* For I/O error handling */
+	jmp_buf jmpbuf;
 } bunzip_data;
 
 /* Return the next nnn bits of input.  All reads from the compressed input
@@ -106,39 +128,29 @@ static unsigned int get_bits(bunzip_data *bd, char bits_wanted)
 	return bits;
 }
 
-/* At certain times, it pays to have an optimized inline version of
- * get_bits() which gets a single bit. */
-#define GET_A_BIT(bd) \
-   ((bd->inbufBitCount > 0) \
-	? ((unsigned int)(((bd)->inbufBits >> --(bd)->inbufBitCount) & 1)) \
-	: get_bits((bd), 1))
-
-
-/* Decompress a block of text to into intermediate buffer */
+/* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
 
-extern int read_bunzip_data(bunzip_data *bd)
+static int get_next_block(bunzip_data *bd)
 {
 	struct group_data *hufGroup;
-	int dbufCount,nextSym,dbufSize,origPtr,groupCount,*base,*limit,selector,
+	int dbufCount,nextSym,dbufSize,groupCount,*base,*limit,selector,
 		i,j,k,t,runPos,symCount,symTotal,nSelectors,byteCount[256];
 	unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
-	unsigned int *dbuf;
-
-	/* Read in header signature (borrowing mtfSymbol for temp space). */
-	for(i=0;i<6;i++) mtfSymbol[i]=get_bits(bd,8);
-	mtfSymbol[6]=0;
-	/* Read CRC (which is stored big endian). */
-	bd->headerCRC=get_bits(bd,32);
-	/* Is this the last block (with CRC for file)? */
-	if(!strcmp(mtfSymbol,"\x17\x72\x45\x38\x50\x90"))
-		return RETVAL_LAST_BLOCK;
-	/* If it's not a valid data block, barf. */
-	if(strcmp(mtfSymbol,"\x31\x41\x59\x26\x53\x59"))
-		return RETVAL_NOT_BZIP_DATA;
+	unsigned int *dbuf,origPtr;
 
 	dbuf=bd->dbuf;
 	dbufSize=bd->dbufSize;
 	selectors=bd->selectors;
+	/* Reset longjmp I/O error handling */
+	i=setjmp(bd->jmpbuf);
+	if(i) return i;
+	/* Read in header signature and CRC, then validate signature.
+	   (last block signature means CRC is for whole file, return now) */
+	i = get_bits(bd,24);
+	j = get_bits(bd,24);
+	bd->headerCRC=get_bits(bd,32);
+	if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK;
+	if ((i != 0x314159) || (j != 0x265359)) return RETVAL_NOT_BZIP_DATA;
 	/* We can add support for blockRandomised if anybody complains.  There was
 	   some code for this in busybox 1.0.0-pre3, but nobody ever noticed that
 	   it didn't actually work. */
@@ -150,10 +162,6 @@ extern int read_bunzip_data(bunzip_data *bd)
 	   values were present.  We make a translation table to convert the symbols
 	   back to the corresponding bytes. */
 	t=get_bits(bd, 16);
-#if 0
-	/* I don't believe this is necessary.  Rob? */
-	memset(symToByte,0,256);
-#endif
 	symTotal=0;
 	for (i=0;i<16;i++) {
 		if(t&(1<<(15-i))) {
@@ -167,7 +175,8 @@ extern int read_bunzip_data(bunzip_data *bd)
 	if (groupCount<2 || groupCount>MAX_GROUPS) return RETVAL_DATA_ERROR;
 	/* nSelectors: Every GROUP_SIZE many symbols we select a new huffman coding
 	   group.  Read in the group selector list, which is stored as MTF encoded
-	   bit runs. */
+	   bit runs.  (MTF=Move To Front, as each value is used it's moved to the
+	   start of the list.) */
 	if(!(nSelectors=get_bits(bd, 15))) return RETVAL_DATA_ERROR;
 	for(i=0; i<groupCount; i++) mtfSymbol[i] = i;
 	for(i=0; i<nSelectors; i++) {
@@ -175,13 +184,7 @@ extern int read_bunzip_data(bunzip_data *bd)
 		for(j=0;get_bits(bd,1);j++) if (j>=groupCount) return RETVAL_DATA_ERROR;
 		/* Decode MTF to get the next selector */
 		uc = mtfSymbol[j];
-		/* A very small amount of data to move, so memmove is overkill
-		 * and bigger at least in my tests. */
-		k = j;
-		while (k) {
-			mtfSymbol[k] = mtfSymbol[k-1];
-			--k;
-		}
+		for(;j;j--) mtfSymbol[j] = mtfSymbol[j-1];
 		mtfSymbol[0]=selectors[i]=uc;
 	}
 	/* Read the huffman coding tables for each group, which code for symTotal
@@ -190,16 +193,30 @@ extern int read_bunzip_data(bunzip_data *bd)
 	for (j=0; j<groupCount; j++) {
 		unsigned char length[MAX_SYMBOLS],temp[MAX_HUFCODE_BITS+1];
 		int	minLen,	maxLen, pp;
-		/* Read lengths */
-		t=get_bits(bd, 5) - 1;	/* This lets us avoid a test in the loop. */
+		/* Read huffman code lengths for each symbol.  They're stored in
+		   a way similar to mtf; record a starting value for the first symbol,
+		   and an offset from the previous value for everys symbol after that.
+		   (Subtracting 1 before the loop and then adding it back at the end is
+		   an optimization that makes the test inside the loop simpler: symbol
+		   length 0 becomes negative, so an unsigned inequality catches it.) */
+		t=get_bits(bd, 5)-1;
 		for (i = 0; i < symCount; i++) {
 			for(;;) {
-				if (((unsigned)t) > (MAX_HUFCODE_BITS-1)) return RETVAL_DATA_ERROR;
-					if(!get_bits(bd, 1)) break;
-					/* We can avoid an if/else with a little arithmetic. */
-					t += (1 - 2*get_bits(bd, 1)); /* 0 -> t++ ; 1 -> t-- */
+				if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
+					return RETVAL_DATA_ERROR;
+				/* If first bit is 0, stop.  Else second bit indicates whether
+				   to increment or decrement the value.  Optimization: grab 2
+				   bits and unget the second if the first was 0. */
+				k = get_bits(bd,2);
+				if (k < 2) {
+					bd->inbufBitCount++;
+					break;
+				}
+				/* Add one if second bit 1, else subtract 1.  Avoids if/else */
+				t+=(((k+1)&2)-1);
 			}
-			length[i] = t + 1;	/* Correct for the initial -1 adjustment. */
+			/* Correct for the initial -1, to get the final symbol length */
+			length[i]=t+1;
 		}
 		/* Find largest and smallest lengths in this group */
 		minLen=maxLen=length[0];
@@ -214,11 +231,8 @@ extern int read_bunzip_data(bunzip_data *bd)
 		 * value of a huffman symbol of a given length when using permute[].
 		 *
 		 * limit[] indicates the largest numerical value a symbol with a given
-		 * number of bits can have.  It lets us know when to stop reading.
-		 *
-		 * To use these, keep reading bits until value<=limit[bitcount] or
-		 * you've read over 20 bits (error).  Then the decoded symbol
-		 * equals permute[hufcode_value-base[hufcode_bitcount]].
+		 * number of bits can have.  This is how the huffman codes can vary in
+		 * length: each code with a value>limit[length] needs another bit.
 		 */
 		hufGroup=bd->groups+j;
 		hufGroup->minLen = minLen;
@@ -228,22 +242,29 @@ extern int read_bunzip_data(bunzip_data *bd)
 		   entry.  We do this again when using them (during symbol decoding).*/
 		base=hufGroup->base-1;
 		limit=hufGroup->limit-1;
-		/* Calculate permute[] */
-		pp = 0;
-		for(i=minLen;i<=maxLen;i++) 
+		/* Calculate permute[].  Concurently, initialize temp[] and limit[]. */
+		pp=0;
+		for(i=minLen;i<=maxLen;i++) {
+			temp[i]=limit[i]=0;
 			for(t=0;t<symCount;t++) 
 				if(length[t]==i) hufGroup->permute[pp++] = t;
-		/* Count cumulative symbols coded for at each bit length */
-		for (i=minLen;i<=maxLen;i++) temp[i]=limit[i]=0;
+		}
+		/* Count symbols coded for at each bit length */
 		for (i=0;i<symCount;i++) temp[length[i]]++;
 		/* Calculate limit[] (the largest symbol-coding value at each bit
 		 * length, which is (previous limit<<1)+symbols at this level), and
 		 * base[] (number of symbols to ignore at each bit length, which is
-		 * limit-cumulative count of symbols coded for already). */
+		 * limit minus the cumulative count of symbols coded for already). */
 		pp=t=0;
 		for (i=minLen; i<maxLen; i++) {
 			pp+=temp[i];
-			limit[i]=pp-1;
+			/* We read the largest possible symbol size and then unget bits
+			   after determining how many we need, and those extra bits could
+			   be set to anything.  (They're noise from future symbols.)  At
+			   each level we're really only interested in the first few bits,
+			   so here we set all the trailing to-be-ignored bits to 1 so they
+			   don't affect the value>limit[length] comparison. */
+			limit[i]= (pp << (maxLen - i)) - 1;
 			pp<<=1;
 			base[i+1]=pp-(t+=temp[i]);
 		}
@@ -255,10 +276,12 @@ extern int read_bunzip_data(bunzip_data *bd)
 	   block's huffman coded symbols from the file and undo the huffman coding
 	   and run length encoding, saving the result into dbuf[dbufCount++]=uc */
 
-	/* Initialize symbol occurrence counters and symbol mtf table */
-	memset(byteCount,0,256*sizeof(int));
-	for(i=0;i<256;i++) mtfSymbol[i]=(unsigned char)i;
-	/* Loop through compressed symbols */
+	/* Initialize symbol occurrence counters and symbol Move To Front table */
+	for(i=0;i<256;i++) {
+		byteCount[i] = 0;
+		mtfSymbol[i]=(unsigned char)i;
+	}
+	/* Loop through compressed symbols. */
 	runPos=dbufCount=symCount=selector=0;
 	for(;;) {
 		/* Determine which huffman coding group to use. */
@@ -269,17 +292,41 @@ extern int read_bunzip_data(bunzip_data *bd)
 			base=hufGroup->base-1;
 			limit=hufGroup->limit-1;
 		}
-		/* Read next huffman-coded symbol */
-		i = hufGroup->minLen;
-		j=get_bits(bd, i);
-		while (j > limit[i]) { /* The sentinal allows us to avoid testing i. */
-			j = (j << 1) | GET_A_BIT(bd);
-			++i;
-		}
-		/* Huffman decode nextSym (with bounds checking) */
-		if ((i > hufGroup->maxLen) || (((unsigned)(j-=base[i])) >= MAX_SYMBOLS)) return RETVAL_DATA_ERROR;
+		/* Read next huffman-coded symbol. */
+		/* Note: It is far cheaper to read maxLen bits and back up than it is
+		   to read minLen bits and then an additional bit at a time, testing
+		   as we go.  Because there is a trailing last block (with file CRC),
+		   there is no danger of the overread causing an unexpected EOF for a
+		   valid compressed file.  As a further optimization, we do the read
+		   inline (falling back to a call to get_bits if the buffer runs
+		   dry).  The following (up to got_huff_bits:) is equivalent to
+		   j=get_bits(bd,hufGroup->maxLen);
+		 */
+		while (bd->inbufBitCount<hufGroup->maxLen) {
+			if(bd->inbufPos==bd->inbufCount) {
+				j = get_bits(bd,hufGroup->maxLen);
+				goto got_huff_bits;
+			}
+			bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
+			bd->inbufBitCount+=8;
+		};
+		bd->inbufBitCount-=hufGroup->maxLen;
+		j = (bd->inbufBits>>bd->inbufBitCount)&((1<<hufGroup->maxLen)-1);
+got_huff_bits:
+		/* Figure how how many bits are in next symbol and unget extras */
+		i=hufGroup->minLen;
+		while(j>limit[i]) ++i;
+		bd->inbufBitCount += (hufGroup->maxLen - i);
+		/* Huffman decode value to get nextSym (with bounds checking) */
+		if ((i > hufGroup->maxLen)
+			|| (((unsigned)(j=(j>>(hufGroup->maxLen-i))-base[i]))
+				>= MAX_SYMBOLS))
+			return RETVAL_DATA_ERROR;
 		nextSym = hufGroup->permute[j];
-		/* If this is a repeated run, loop collecting data */
+		/* We have now decoded the symbol, which indicates either a new literal
+		   byte, or a repeated run of the most recent literal byte.  First,
+		   check if nextSym indicates a repeated run, and if so loop collecting
+		   how many times to repeat the last literal. */
 		if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
 			/* If this is the start of a new run, zero out counter */
 			if(!runPos) {
@@ -311,19 +358,20 @@ extern int read_bunzip_data(bunzip_data *bd)
 		}
 		/* Is this the terminating symbol? */
 		if(nextSym>symTotal) break;
-		/* At this point, the symbol we just decoded indicates a new literal
-		   character.  Subtract one to get the position in the MTF array
-		   at which this literal is currently to be found.  (Note that the
-		   result can't be -1 or 0, because 0 and 1 are RUNA and RUNB.
-		   Another instance of the first symbol in the mtf array, position 0,
-		   would have been handled as part of a run.) */
+		/* At this point, nextSym indicates a new literal character.  Subtract
+		   one to get the position in the MTF array at which this literal is
+		   currently to be found.  (Note that the result can't be -1 or 0,
+		   because 0 and 1 are RUNA and RUNB.  But another instance of the
+		   first symbol in the mtf array, position 0, would have been handled
+		   as part of a run above.  Therefore 1 unused mtf position minus
+		   2 non-literal nextSym values equals -1.) */
 		if(dbufCount>=dbufSize) return RETVAL_DATA_ERROR;
 		i = nextSym - 1;
 		uc = mtfSymbol[i];
-		/* Since we typically expect to move only a small number of symbols,
-		 * and are bound by 256 in any case, using memmove here would
-		 * typically be slower due to function call overhead and other
-		 * assorted setup costs. */
+		/* Adjust the MTF array.  Since we typically expect to move only a
+		 * small number of symbols, and are bound by 256 in any case, using
+		 * memmove here would typically be bigger and slower due to function
+		 * call overhead and other assorted setup costs. */
 		do {
 			mtfSymbol[i] = mtfSymbol[i-1];
 		} while (--i);
@@ -333,14 +381,12 @@ extern int read_bunzip_data(bunzip_data *bd)
 		byteCount[uc]++;
 		dbuf[dbufCount++] = (unsigned int)uc;
 	}
-	/* At this point, we've finished reading huffman-coded symbols and
-	   compressed runs from the input stream.  There are dbufCount many of
-	   them in dbuf[].  Now undo the Burrows-Wheeler transform on dbuf.
+	/* At this point, we've read all the huffman-coded symbols (and repeated
+       runs) for this block from the input stream, and decoded them into the
+	   intermediate buffer.  There are dbufCount many decoded bytes in dbuf[].
+	   Now undo the Burrows-Wheeler transform on dbuf.
 	   See http://dogma.net/markn/articles/bwt/bwt.htm
 	 */
-
-	/* Now we know what dbufCount is, do a better sanity check on origPtr.  */
-	if (((unsigned)origPtr)>=dbufCount) return RETVAL_DATA_ERROR;
 	/* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
 	j=0;
 	for(i=0;i<256;i++) {
@@ -350,144 +396,137 @@ extern int read_bunzip_data(bunzip_data *bd)
 	}
 	/* Figure out what order dbuf would be in if we sorted it. */
 	for (i=0;i<dbufCount;i++) {
-		uc = (unsigned char)(dbuf[i] & 0xff);
+		uc=(unsigned char)(dbuf[i] & 0xff);
 		dbuf[byteCount[uc]] |= (i << 8);
 		byteCount[uc]++;
 	}
-	/* blockRandomised support would go here. */
-
-	/* Using i as position, j as previous character, t as current character,
-	   and uc as run count */
-	bd->dataCRC = 0xffffffffL;
 	/* Decode first byte by hand to initialize "previous" byte.  Note that it
 	   doesn't get output, and if the first three characters are identical
-	   it doesn't qualify as a run (hence uc=255, which will either wrap
-	   to 1 or get reset). */
+	   it doesn't qualify as a run (hence writeRunCountdown=5). */
 	if(dbufCount) {
+		if(origPtr>=dbufCount) return RETVAL_DATA_ERROR;
 		bd->writePos=dbuf[origPtr];
 	    bd->writeCurrent=(unsigned char)(bd->writePos&0xff);
 		bd->writePos>>=8;
-		bd->writeRun=-1;
+		bd->writeRunCountdown=5;
 	}
 	bd->writeCount=dbufCount;
 
 	return RETVAL_OK;
 }
 
-/* Flush output buffer to disk */
-extern void flush_bunzip_outbuf(bunzip_data *bd, int out_fd)
-{
-	if(bd->outbufPos) {
-		if(write(out_fd, bd->outbuf, bd->outbufPos) != bd->outbufPos)
-			longjmp(bd->jmpbuf,RETVAL_UNEXPECTED_OUTPUT_EOF);
-		bd->outbufPos=0;
-	}
-}
-
-
 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
-   If !len, write up to len bytes of data to buf.  Otherwise write to out_fd.
-   Returns len ? bytes written : RETVAL_OK.  Notice all errors negative #'s. */
-extern int write_bunzip_data(bunzip_data *bd, int out_fd, char *outbuf, int len)
+   If start_bunzip was initialized with out_fd=-1, then up to len bytes of
+   data are written to outbuf.  Return value is number of bytes written or
+   error (all errors are negative numbers).  If out_fd!=-1, outbuf and len
+   are ignored, data is written to out_fd and return is RETVAL_OK or error.
+*/
+
+extern int read_bunzip(bunzip_data *bd, char *outbuf, int len)
 {
-	unsigned int *dbuf=bd->dbuf;
-	int count,pos,current, run,copies,outbyte,previous,gotcount=0;
+	const unsigned int *dbuf;
+	int pos,current,previous,gotcount;
 
-	for(;;) {
-		/* If last read was short due to end of file, return last block now */
-		if(bd->writeCount<0) return bd->writeCount;
-		/* If we need to refill dbuf, do it. */
-		if(!bd->writeCount) {
-			int i=read_bunzip_data(bd);
-			if(i) {
-				if(i==RETVAL_LAST_BLOCK) {
-					bd->writeCount=i;
-					return gotcount;
-				} else return i;
+	/* If last read was short due to end of file, return last block now */
+	if(bd->writeCount<0) return bd->writeCount;
+
+	gotcount = 0;
+	dbuf=bd->dbuf;
+	pos=bd->writePos;
+	current=bd->writeCurrent;
+
+	/* We will always have pending decoded data to write into the output
+	   buffer unless this is the very first call (in which case we haven't
+	   huffman-decoded a block into the intermediate buffer yet). */
+
+	if (bd->writeCopies) {
+		/* Inside the loop, writeCopies means extra copies (beyond 1) */
+		--bd->writeCopies;
+		/* Loop outputting bytes */
+		for(;;) {
+			/* If the output buffer is full, snapshot state and return */
+			if(gotcount >= len) {
+				bd->writePos=pos;
+				bd->writeCurrent=current;
+				bd->writeCopies++;
+				return len;
 			}
-		}
-		/* Loop generating output */
-		count=bd->writeCount;
-		pos=bd->writePos;
-		current=bd->writeCurrent;
-		run=bd->writeRun;
-		while(count) {
-			/* If somebody (like busybox tar) wants a certain number of bytes of
-			data from memory instead of written to a file, humor them */
-			if(len && bd->outbufPos>=len) goto dataus_interruptus;
-			count--;
+			/* Write next byte into output buffer, updating CRC */
+			outbuf[gotcount++] = current;
+			bd->writeCRC=(((bd->writeCRC)<<8)
+						  ^bd->crc32Table[((bd->writeCRC)>>24)^current]);
+			/* Loop now if we're outputting multiple copies of this byte */
+			if (bd->writeCopies) {
+				--bd->writeCopies;
+				continue;
+			}
+decode_next_byte:
+			if (!bd->writeCount--) break;
 			/* Follow sequence vector to undo Burrows-Wheeler transform */
 			previous=current;
 			pos=dbuf[pos];
 			current=pos&0xff;
 			pos>>=8;
-			/* Whenever we see 3 consecutive copies of the same byte,
-			   the 4th is a repeat count */
-			if(run++==3) {
-				copies=current;
-				outbyte=previous;
-				current=-1;
+			/* After 3 consecutive copies of the same byte, the 4th is a repeat
+			   count.  We count down from 4 instead
+			 * of counting up because testing for non-zero is faster */
+			if(--bd->writeRunCountdown) {
+				if(current!=previous) bd->writeRunCountdown=4;
 			} else {
-				copies=1;
-				outbyte=current;
-			}
-			/* Output bytes to buffer, flushing to file if necessary */
-			while(copies--) {
-				if(bd->outbufPos == IOBUF_SIZE) flush_bunzip_outbuf(bd,out_fd);
-				bd->outbuf[bd->outbufPos++] = outbyte;
-				bd->dataCRC = (bd->dataCRC << 8)
-								^ bd->crc32Table[(bd->dataCRC >> 24) ^ outbyte];
+				/* We have a repeated run, this byte indicates the count */
+				bd->writeCopies=current;
+				current=previous;
+				bd->writeRunCountdown=5;
+				/* Sometimes there are just 3 bytes (run length 0) */
+				if(!bd->writeCopies) goto decode_next_byte;
+				/* Subtract the 1 copy we'd output anyway to get extras */
+				--bd->writeCopies;
 			}
-			if(current!=previous) run=0;
 		}
 		/* Decompression of this block completed successfully */
-		bd->dataCRC=~(bd->dataCRC);
-		bd->totalCRC=((bd->totalCRC << 1) | (bd->totalCRC >> 31)) ^ bd->dataCRC;
+		bd->writeCRC=~bd->writeCRC;
+		bd->totalCRC=((bd->totalCRC<<1) | (bd->totalCRC>>31)) ^ bd->writeCRC;
 		/* If this block had a CRC error, force file level CRC error. */
-		if(bd->dataCRC!=bd->headerCRC) {
+		if(bd->writeCRC!=bd->headerCRC) {
 			bd->totalCRC=bd->headerCRC+1;
 			return RETVAL_LAST_BLOCK;
 		}
-dataus_interruptus:
-		bd->writeCount=count;
-		if(len) {
-			gotcount+=bd->outbufPos;
-			memcpy(outbuf,bd->outbuf,len);
-			/* If we got enough data, checkpoint loop state and return */
-			if((len-=bd->outbufPos)<1) {
-				bd->outbufPos-=len;
-				if(bd->outbufPos)
-					memmove(bd->outbuf,bd->outbuf+len,bd->outbufPos);
-				bd->writePos=pos;
-				bd->writeCurrent=current;
-				bd->writeRun=run;
-				return gotcount;
-			}
-		}
 	}
+
+	/* Refill the intermediate buffer by huffman-decoding next block of input */
+	/* (previous is just a convenient unused temp variable here) */
+	previous=get_next_block(bd);
+	if(previous) {
+		bd->writeCount=previous;
+		return (previous!=RETVAL_LAST_BLOCK) ? previous : gotcount;
+	}
+	bd->writeCRC=0xffffffffUL;
+	pos=bd->writePos;
+	current=bd->writeCurrent;
+	goto decode_next_byte;
 }
 
-/* Allocate the structure, read file header.  If !len, src_fd contains
-   filehandle to read from.  Else inbuf contains data. */
-extern int start_bunzip(bunzip_data **bdp, int src_fd, char *inbuf, int len)
+/* Allocate the structure, read file header.  If in_fd==-1, inbuf must contain
+   a complete bunzip file (len bytes long).  If in_fd!=-1, inbuf and len are
+   ignored, and data is read from file handle into temporary buffer. */
+extern int start_bunzip(bunzip_data **bdp, int in_fd, char *inbuf, int len)
 {
 	bunzip_data *bd;
 	unsigned int i,j,c;
+	const unsigned int BZh0=(((unsigned int)'B')<<24)+(((unsigned int)'Z')<<16)
+							+(((unsigned int)'h')<<8)+(unsigned int)'0';
 
 	/* Figure out how much data to allocate */
 	i=sizeof(bunzip_data);
-	if(!len) i+=IOBUF_SIZE;
+	if(in_fd!=-1) i+=IOBUF_SIZE;
 	/* Allocate bunzip_data.  Most fields initialize to zero. */
 	if(!(bd=*bdp=malloc(i))) return RETVAL_OUT_OF_MEMORY;
 	memset(bd,0,sizeof(bunzip_data));
-	if(len) {
+	/* Setup input buffer */
+	if(-1==(bd->in_fd=in_fd)) {
 		bd->inbuf=inbuf;
 		bd->inbufCount=len;
-		bd->in_fd=-1;
-	} else {
-		bd->inbuf=(char *)(bd+1);
-		bd->in_fd=src_fd;
-	}
+	} else bd->inbuf=(unsigned char *)(bd+1);
 	/* Init the CRC32 table (big endian) */
 	for(i=0;i<256;i++) {
 		c=i<<24;
@@ -498,55 +537,60 @@ extern int start_bunzip(bunzip_data **bdp, int src_fd, char *inbuf, int len)
 	/* Setup for I/O error handling via longjmp */
 	i=setjmp(bd->jmpbuf);
 	if(i) return i;
-	/* Ensure that file starts with "BZh" */
-    for(i=0;i<3;i++) if(get_bits(bd,8)!="BZh"[i]) return RETVAL_NOT_BZIP_DATA;
-	/* Next byte ascii '1'-'9', indicates block size in units of 100k of
+
+	/* Ensure that file starts with "BZh['1'-'9']." */
+	i = get_bits(bd,32);
+	if (((unsigned int)(i-BZh0-1)) >= 9) return RETVAL_NOT_BZIP_DATA;
+
+	/* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of
 	   uncompressed data.  Allocate intermediate buffer for block. */
-	i=get_bits(bd,8);
-	if (i<'1' || i>'9') return RETVAL_NOT_BZIP_DATA;
-	bd->dbufSize=100000*(i-'0');
+	bd->dbufSize=100000*(i-BZh0);
+
 	if(!(bd->dbuf=malloc(bd->dbufSize * sizeof(int))))
 		return RETVAL_OUT_OF_MEMORY;
 	return RETVAL_OK;
 }
 
-extern char *uncompressStream(int src_fd, int dst_fd)
+/* Example usage: decompress src_fd to dst_fd.  (Stops at end of bzip data,
+   not end of file.) */
+extern int uncompressStream(int src_fd, int dst_fd)
 {
+	char *outbuf;
 	bunzip_data *bd;
 	int i;
 
+	if(!(outbuf=malloc(IOBUF_SIZE))) return RETVAL_OUT_OF_MEMORY;
 	if(!(i=start_bunzip(&bd,src_fd,0,0))) {
-		i=write_bunzip_data(bd,dst_fd,0,0);
-		if(i==RETVAL_LAST_BLOCK && bd->headerCRC==bd->totalCRC) i=RETVAL_OK;
+		for(;;) {
+			if((i=read_bunzip(bd,outbuf,IOBUF_SIZE)) <= 0) break;
+			if(i!=write(dst_fd,outbuf,i)) {
+				i=RETVAL_UNEXPECTED_OUTPUT_EOF;
+				break;
+			}
+		}
 	}
-	flush_bunzip_outbuf(bd,dst_fd);
+	/* Check CRC and release memory */
+	if(i==RETVAL_LAST_BLOCK && bd->headerCRC==bd->totalCRC) i=RETVAL_OK;
 	if(bd->dbuf) free(bd->dbuf);
 	free(bd);
-	return bunzip_errors[-i];
+	free(outbuf);
+	return i;
 }
 
-/* This new version is not yet properly integrated with tar */
-extern ssize_t read_bz2(int fd, void *buf, size_t count)
-{
-#warning FIXME
-	return(0);
-}
+#ifdef TESTING
 
-extern void BZ2_bzReadOpen(int fd, void *unused, int nUnused)
-{
-#warning FIXME
-	return;
-}
-extern void BZ2_bzReadClose(void)
-{
-#warning FIXME
-}
+static char * const bunzip_errors[]={NULL,"Bad file checksum","Not bzip data",
+		"Unexpected input EOF","Unexpected output EOF","Data error",
+		 "Out of memory","Obsolete (pre 0.9.5) bzip format not supported."};
 
-#if 0
 /* Dumb little test thing, decompress stdin to stdout */
 int main(int argc, char *argv[])
 {
-	char *c=uncompressStream(0,1);
-	fprintf(stderr,"\n%s\n", c ? c : "Completed OK");
+	int i=uncompressStream(0,1);
+	char c;
+
+	if(i) fprintf(stderr,"%s\n", bunzip_errors[-i]);
+    else if(read(0,&c,1)) fprintf(stderr,"Trailing garbage ignored\n");
+	return -i;
 }
 #endif