gzip-rsyncable-checksum.diff   [plain text]


NOTE: this patch is for _gzip_!

Kevin Day's version of the gzip-rsyncable patch that uses the rsync
checksum algorithm.

--- gzip-1.2.2/deflate.c	2004-09-15 10:28:14.000000000 -0700
+++ rsyncable/deflate.c	2005-02-17 14:37:14.660957200 -0700
@@ -98,6 +98,10 @@
                             int length));
 #endif
 
+local void rsync_roll(deflate_state *s, unsigned start, unsigned num);
+local void rsync_roll_noop(deflate_state *s, unsigned start, unsigned num);
+local void rsync_roll2(deflate_state *s, unsigned start, unsigned num);
+
 /* ===========================================================================
  * Local data
  */
@@ -115,6 +119,39 @@
  * See deflate.c for comments about the MIN_MATCH+1.
  */
 
+
+
+/* 
+	Valid values for RSYNC_DEFAULT_CHECKSUM_TYPE are:
+
+	Z_RSYNCABLE_OFF
+	Z_RSYNCABLE_SIMPLESUM
+	Z_RSYNCABLE_RSSUM
+*/
+
+#ifndef RSYNC_DEFAULT_CHECKSUM_TYPE
+#	define RSYNC_DEFAULT_CHECKSUM_TYPE Z_RSYNCABLE_RSSUM
+#endif
+
+#ifndef RSYNC_DEFAULT_WINDOW_SIZE
+#	define RSYNC_DEFAULT_WINDOW_SIZE 30
+#endif
+
+#ifndef RSYNC_DEFAULT_RESET_BLOCK_SIZE
+#	define RSYNC_DEFAULT_RESET_BLOCK_SIZE 4096 
+#endif
+
+#ifndef RSYNC_RESET_MAGIC_VALUE
+#	define RSYNC_RESET_MAGIC_VALUE 0
+#endif
+
+#define RSYNC_SUM_MATCH(s) ((s)->rsync_sum % (s)->rsync_reset_block_size == RSYNC_RESET_MAGIC_VALUE)
+/* Whether window sum matches magic value */
+
+/* Global rsync mode control variable */
+int zlib_rsync = 1 ;
+
+
 /* Values for max_lazy_match, good_match and max_chain_length, depending on
  * the desired pack level (0..9). The values given below have been tuned to
  * exclude worst case performance for pathological files. Better values may be
@@ -212,6 +249,36 @@
     /* To do: ignore strm->next_in if we use it as window */
 }
 
+int ZEXPORT deflateSetRsyncParameters_(strm, checksum_type, window_size, reset_block_size)
+    z_streamp strm;
+    int  checksum_type;
+	ulg window_size;
+	ulg reset_block_size;
+{
+	deflate_state *s = strm->state;
+
+	switch(checksum_type){
+		case Z_RSYNCABLE_SIMPLESUM: 
+			s->rsync_rollfunction = rsync_roll; 
+			break;
+		case Z_RSYNCABLE_RSSUM: 
+			s->rsync_rollfunction = rsync_roll2; 
+			break;
+		default: 
+			s->rsync_rollfunction = rsync_roll_noop;
+	}
+
+	s->rsync_window_size = window_size != 0 ? window_size : RSYNC_DEFAULT_WINDOW_SIZE;
+	s->rsync_reset_block_size = reset_block_size != 0 ? reset_block_size : s->rsync_window_size;
+
+    s->rsync_chunk_end = 0xFFFFFFFFUL;
+    s->rsync_sum = 0;
+	s->rsync_s1 = 0;
+	s->rsync_s2 = 0;
+
+	return Z_OK;
+}
+
 /* ========================================================================= */
 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
                   version, stream_size)
@@ -307,9 +374,13 @@
     s->strategy = strategy;
     s->method = (Byte)method;
 
+	deflateSetRsyncParameters_(strm, RSYNC_DEFAULT_CHECKSUM_TYPE, RSYNC_DEFAULT_WINDOW_SIZE, RSYNC_DEFAULT_RESET_BLOCK_SIZE);
+
     return deflateReset(strm);
 }
 
+
+
 /* ========================================================================= */
 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
     z_streamp strm;
@@ -841,6 +912,13 @@
 #ifdef ASMV
     match_init(); /* initialize the asm code */
 #endif
+
+    /* rsync params */
+    s->rsync_chunk_end = 0xFFFFFFFFUL;
+    s->rsync_sum = 0;
+	s->rsync_s1 = 0;
+	s->rsync_s2 = 0;
+
 }
 
 #ifndef FASTEST
@@ -1123,6 +1201,8 @@
             zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
             s->match_start -= wsize;
             s->strstart    -= wsize; /* we now have strstart >= MAX_DIST */
+            if (s->rsync_chunk_end != 0xFFFFFFFFUL)
+                s->rsync_chunk_end -= wsize;
             s->block_start -= (long) wsize;
 
             /* Slide the hash table (could be avoided with 32 bit values
@@ -1184,15 +1264,98 @@
     } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
 }
 
+local void rsync_roll(s, start, num)
+    deflate_state *s;
+    unsigned start;
+    unsigned num;
+{
+    unsigned i;
+
+    if (start < s->rsync_window_size) {
+	/* before window fills. */
+	for (i = start; i < s->rsync_window_size; i++) {
+	    if (i == start + num) return;
+	    s->rsync_sum += (ulg)s->window[i];
+	}
+	num -= (s->rsync_window_size - start);
+	start = s->rsync_window_size;
+    }
+
+    /* buffer after window full */
+    for (i = start; i < start+num; i++) {
+	/* New character in */
+	s->rsync_sum += (ulg)s->window[i];
+	/* Old character out */
+	s->rsync_sum -= (ulg)s->window[i - s->rsync_window_size];
+	if (s->rsync_chunk_end == 0xFFFFFFFFUL
+            && RSYNC_SUM_MATCH(s))
+	    s->rsync_chunk_end = i;
+    }
+}
+
+local void rsync_roll_noop(s, start, num)
+    deflate_state *s;
+    unsigned start;
+    unsigned num;
+{
+}
+
+/*
+ Implements the 2 part rsync checksum, instead of a simple summation checksum.
+*/
+local void rsync_roll2(deflate_state *s, unsigned start, unsigned num)
+{
+    unsigned i;
+
+    if (start < s->rsync_window_size) {
+		/* before window fills. */
+		for (i = start; i < s->rsync_window_size; i++) {
+			if (i == start + num) return;
+			s->rsync_s1 = (s->rsync_s1 + (ulg)s->window[i]) & 0xffff;
+			s->rsync_s2 = (s->rsync_s2 + s->rsync_s1) & 0xffff;
+		}
+		num -= (s->rsync_window_size - start);
+		start = s->rsync_window_size;
+    }
+
+    /* buffer after window full */
+    for (i = start; i < start+num; i++) {
+		/* Old character out */
+
+		s->rsync_s1 = (s->rsync_s1 - (ulg)s->window[i - s->rsync_window_size]) & 0xffff;
+		s->rsync_s2 = (s->rsync_s2 - s->rsync_window_size * (ulg)s->window[i - s->rsync_window_size]) & 0xffff;
+
+		/* New character in */
+		s->rsync_s1 = (s->rsync_s1 + (ulg)s->window[i]) & 0xffff;
+		s->rsync_s2 = (s->rsync_s2 + s->rsync_s1) & 0xffff;
+
+		// add the two together for the match calculation
+		s->rsync_sum = s->rsync_s1 + s->rsync_s2;
+
+
+		if (s->rsync_chunk_end == 0xFFFFFFFFUL
+			&& RSYNC_SUM_MATCH(s)){
+			s->rsync_chunk_end = i;
+		}
+    }
+}
+
+/* ===========================================================================
+ * Set rsync_chunk_end if window sum matches magic value.
+ */
+#define RSYNC_ROLL(s, start, num) \
+   do { if (zlib_rsync) (s)->rsync_rollfunction((s), (start), (num)); } while(0)
+
 /* ===========================================================================
  * Flush the current block, with given end-of-file flag.
  * IN assertion: strstart is set to the end of the current match.
  */
-#define FLUSH_BLOCK_ONLY(s, eof) { \
+#define FLUSH_BLOCK_ONLY(s, eof, pad) { \
    _tr_flush_block(s, (s->block_start >= 0L ? \
                    (charf *)&s->window[(unsigned)s->block_start] : \
                    (charf *)Z_NULL), \
                 (ulg)((long)s->strstart - s->block_start), \
+                (pad), \
                 (eof)); \
    s->block_start = s->strstart; \
    flush_pending(s->strm); \
@@ -1200,8 +1363,8 @@
 }
 
 /* Same but force premature exit if necessary. */
-#define FLUSH_BLOCK(s, eof) { \
-   FLUSH_BLOCK_ONLY(s, eof); \
+#define FLUSH_BLOCK(s, eof, pad) { \
+   FLUSH_BLOCK_ONLY(s, eof, pad); \
    if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
 }
 
@@ -1252,16 +1415,16 @@
             /* strstart == 0 is possible when wraparound on 16-bit machine */
             s->lookahead = (uInt)(s->strstart - max_start);
             s->strstart = (uInt)max_start;
-            FLUSH_BLOCK(s, 0);
+            FLUSH_BLOCK(s, 0, 0);
         }
         /* Flush if we may have to slide, otherwise block_start may become
          * negative and the data will be gone:
          */
         if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
-            FLUSH_BLOCK(s, 0);
+            FLUSH_BLOCK(s, 0, 0);
         }
     }
-    FLUSH_BLOCK(s, flush == Z_FINISH);
+    FLUSH_BLOCK(s, flush == Z_FINISH, 0);
     return flush == Z_FINISH ? finish_done : block_done;
 }
 
@@ -1330,6 +1493,7 @@
 
             s->lookahead -= s->match_length;
 
+            RSYNC_ROLL(s, s->strstart, s->match_length);
             /* Insert new strings in the hash table only if the match length
              * is not too large. This saves time but degrades compression.
              */
@@ -1363,12 +1527,17 @@
             /* No match, output a literal byte */
             Tracevv((stderr,"%c", s->window[s->strstart]));
             _tr_tally_lit (s, s->window[s->strstart], bflush);
+            RSYNC_ROLL(s, s->strstart, 1);
             s->lookahead--;
             s->strstart++;
         }
-        if (bflush) FLUSH_BLOCK(s, 0);
+	if (zlib_rsync && s->strstart > s->rsync_chunk_end) {
+	    s->rsync_chunk_end = 0xFFFFFFFFUL;
+	    bflush = 2;
+	}
+        if (bflush) FLUSH_BLOCK(s, 0, bflush-1);
     }
-    FLUSH_BLOCK(s, flush == Z_FINISH);
+    FLUSH_BLOCK(s, flush == Z_FINISH, bflush-1);
     return flush == Z_FINISH ? finish_done : block_done;
 }
 
@@ -1457,6 +1626,7 @@
              */
             s->lookahead -= s->prev_length-1;
             s->prev_length -= 2;
+            RSYNC_ROLL(s, s->strstart, s->prev_length+1);
             do {
                 if (++s->strstart <= max_insert) {
                     INSERT_STRING(s, s->strstart, hash_head);
@@ -1466,7 +1636,11 @@
             s->match_length = MIN_MATCH-1;
             s->strstart++;
 
-            if (bflush) FLUSH_BLOCK(s, 0);
+            if (zlib_rsync && s->strstart > s->rsync_chunk_end) {
+                s->rsync_chunk_end = 0xFFFFFFFFUL;
+                bflush = 2;
+            } 
+            if (bflush) FLUSH_BLOCK(s, 0, bflush-1);
 
         } else if (s->match_available) {
             /* If there was no match at the previous position, output a
@@ -1475,9 +1649,14 @@
              */
             Tracevv((stderr,"%c", s->window[s->strstart-1]));
             _tr_tally_lit(s, s->window[s->strstart-1], bflush);
+            if (zlib_rsync && s->strstart > s->rsync_chunk_end) {
+                s->rsync_chunk_end = 0xFFFFFFFFUL;
+		bflush = 2;
+            } 
             if (bflush) {
-                FLUSH_BLOCK_ONLY(s, 0);
+                FLUSH_BLOCK_ONLY(s, 0, bflush-1);
             }
+            RSYNC_ROLL(s, s->strstart, 1);
             s->strstart++;
             s->lookahead--;
             if (s->strm->avail_out == 0) return need_more;
@@ -1485,7 +1664,14 @@
             /* There is no previous match to compare with, wait for
              * the next step to decide.
              */
+            if (zlib_rsync && s->strstart > s->rsync_chunk_end) {
+                /* Reset huffman tree */
+                s->rsync_chunk_end = 0xFFFFFFFFUL;
+                bflush = 2;
+                FLUSH_BLOCK(s, 0, bflush-1);
+            } 
             s->match_available = 1;
+            RSYNC_ROLL(s, s->strstart, 1);
             s->strstart++;
             s->lookahead--;
         }
@@ -1496,7 +1682,7 @@
         _tr_tally_lit(s, s->window[s->strstart-1], bflush);
         s->match_available = 0;
     }
-    FLUSH_BLOCK(s, flush == Z_FINISH);
+    FLUSH_BLOCK(s, flush == Z_FINISH, bflush-1);
     return flush == Z_FINISH ? finish_done : block_done;
 }
 #endif /* FASTEST */
--- gzip-1.2.2/deflate.h	2004-02-24 07:38:44.000000000 -0700
+++ rsyncable/deflate.h	2005-02-17 13:46:12.056551200 -0700
@@ -254,6 +254,17 @@
      * are always zero.
      */
 
+    ulg rsync_sum;      /* rolling sum of rsync window */
+    ulg rsync_chunk_end; /* next rsync sequence point */
+	ulg rsync_window_size; /* the number of bytes used in computing the rolling checksum */
+	ulg rsync_reset_block_size; /* the compressed stream will be reset approximately every 'rsync_reset_block_size' bytes */
+	ulg rsync_s1; /* part 1 of the checksum for use with checksum type Z_RSYNCABLE_RSSUM*/
+	ulg rsync_s2; /* part 2 of the checksum for use with checksum type Z_RSYNCABLE_RSSUM*/
+
+	/* the function that should be called for performing the rsyncable checksum roll */
+	void (*rsync_rollfunction)(struct internal_state*s , unsigned start, unsigned num);
+
+
 } FAR deflate_state;
 
 /* Output a byte on the stream.
@@ -276,7 +287,7 @@
 void _tr_init         OF((deflate_state *s));
 int  _tr_tally        OF((deflate_state *s, unsigned dist, unsigned lc));
 void _tr_flush_block  OF((deflate_state *s, charf *buf, ulg stored_len,
-                          int eof));
+                          int pad, int eof));
 void _tr_align        OF((deflate_state *s));
 void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,
                           int eof));
--- gzip-1.2.2/minigzip.c	2003-11-04 18:19:26.000000000 -0700
+++ rsyncable/minigzip.c	2005-02-17 13:11:35.472851600 -0700
@@ -215,7 +215,7 @@
     }
     gz_compress(in, out);
 
-    unlink(file);
+    //unlink(file);
 }
 
 
@@ -236,7 +236,10 @@
     if (len > SUFFIX_LEN && strcmp(file+len-SUFFIX_LEN, GZ_SUFFIX) == 0) {
         infile = file;
         outfile = buf;
-        outfile[len-3] = '\0';
+        outfile[len-3] = '.';
+        outfile[len-2] = 'u';
+        outfile[len-1] = 'z';
+        outfile[len-0] = '\0';
     } else {
         outfile = file;
         infile = buf;
@@ -255,7 +258,7 @@
 
     gz_uncompress(in, out);
 
-    unlink(infile);
+    //unlink(infile);
 }
 
 
--- gzip-1.2.2/trees.c	2004-02-24 07:36:38.000000000 -0700
+++ rsyncable/trees.c	2005-02-17 13:09:38.768435100 -0700
@@ -918,10 +918,11 @@
  * Determine the best encoding for the current block: dynamic trees, static
  * trees or store, and output the encoded block to the zip file.
  */
-void _tr_flush_block(s, buf, stored_len, eof)
+void _tr_flush_block(s, buf, stored_len, pad, eof)
     deflate_state *s;
     charf *buf;       /* input block, or NULL if too old */
     ulg stored_len;   /* length of input block */
+    int pad;          /* pad output to byte boundary */
     int eof;          /* true if this is the last block for a file */
 {
     ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */
@@ -1009,6 +1010,12 @@
 #ifdef DEBUG
         s->compressed_len += 7;  /* align on byte boundary */
 #endif
+#ifdef DEBUG
+    } else if (pad && (s->compressed_len % 8) != 0) {
+#else
+    } else if (pad) {
+#endif
+        _tr_stored_block(s, buf, 0, eof);
     }
     Tracev((stderr,"\ncomprlen %lu(%lu) ", s->compressed_len>>3,
            s->compressed_len-7*eof));
--- gzip-1.2.2/zlib.def	1969-12-31 17:00:00.000000000 -0700
+++ rsyncable/zlib.def	2005-02-17 14:01:48.972258000 -0700
@@ -0,0 +1,61 @@
+LIBRARY
+; zlib data compression library
+
+EXPORTS
+; basic functions
+    zlibVersion
+    deflate
+    deflateEnd
+    inflate
+    inflateEnd
+; advanced functions
+    deflateSetDictionary
+    deflateCopy
+    deflateReset
+    deflateParams
+    deflateBound
+    deflatePrime
+    inflateSetDictionary
+    inflateSync
+    inflateCopy
+    inflateReset
+    inflateBack
+    inflateBackEnd
+    zlibCompileFlags
+; utility functions
+    compress
+    compress2
+    compressBound
+    uncompress
+    gzopen
+    gzdopen
+    gzsetparams
+    gzread
+    gzwrite
+    gzprintf
+    gzputs
+    gzgets
+    gzputc
+    gzgetc
+    gzungetc
+    gzflush
+    gzseek
+    gzrewind
+    gztell
+    gzeof
+    gzclose
+    gzerror
+    gzclearerr
+; checksum functions
+    adler32
+    crc32
+; various hacks, don't look :)
+    deflateInit_
+    deflateInit2_
+    inflateInit_
+    inflateInit2_
+    inflateBackInit_
+    inflateSyncPoint
+    get_crc_table
+    zError
+	deflateSetRsyncParameters_
\ No newline at end of file
--- gzip-1.2.2/zlib.h	2004-10-03 22:57:26.000000000 -0700
+++ rsyncable/zlib.h	2005-02-17 14:02:11.753362200 -0700
@@ -179,6 +179,13 @@
 
 #define Z_NULL  0  /* for initializing zalloc, zfree, opaque */
 
+
+/* Constants used for selecting Rsyncable checksum type */
+#define Z_RSYNCABLE_OFF			0
+#define Z_RSYNCABLE_SIMPLESUM	1
+#define Z_RSYNCABLE_RSSUM		2
+
+
 #define zlib_version zlibVersion()
 /* for compatibility with versions < 1.0.2 */
 
@@ -1185,6 +1192,17 @@
         ZLIB_VERSION, sizeof(z_stream))
 
 
+
+/* deflateSetRsyncParameters allows for setting rsyncable parameters on a stream.
+	These parameters MUST be set immediately after the stream is created, and before
+	any data is written to the stream.
+ */
+ZEXTERN int ZEXPORT deflateSetRsyncParameters_ OF((z_stream FAR *strm, int checksum_type, unsigned long window_size, unsigned long reset_block_size));
+
+#define deflateSetRsyncParameters(strm, checksum_type, window_size, reset_block_size) \
+        deflateSetRsyncParameters_((strm), (checksum_type), (window_size), (reset_block_size))
+
+
 #if !defined(ZUTIL_H) && !defined(NO_DUMMY_DECL)
     struct internal_state {int dummy;}; /* hack for buggy compilers */
 #endif
@@ -1193,6 +1211,10 @@
 ZEXTERN int            ZEXPORT inflateSyncPoint OF((z_streamp z));
 ZEXTERN const uLongf * ZEXPORT get_crc_table    OF((void));
 
+/* Global rsync mode control variable */
+extern int zlib_rsync;
+
+
 #ifdef __cplusplus
 }
 #endif