Rusty's patch to gzip. Here for informational purposes only. Please

don't complain if applying it to gzip causes monkeys to fly out of
your butt.

diff --git a/gzip-rsyncable.diff b/gzip-rsyncable.diff
new file mode 100644 (file)
index 0000000..724c2d5
--- /dev/null
+++ b/gzip-rsyncable.diff
@@ -0,0 +1,303 @@
+In message <20020403094739.GA5434@samba.org> you write:
+> On  3 Apr 2002, Rusty Russell <rusty@rustcorp.com.au> wrote:
+> > In message <20020403020455.GC18851@samba.org> you write:
+> > > Hi,
+> > > 
+> > > I think you said the other day that you had a working --rsyncable
+> > > patch for gzip.  Could I have it please?
+> 
+> Thanks for that.  I'll put it in with our contrib patches and probably
+> merge it soon.
+> 
+> I was actually after the patch to gzip, rather than the fuzzy patch
+> for rsync.  I'd like to try it myself and also distribute it along
+> with rsync.
+
+It's pending for actual gzip inclusion!  The maintainer was dug up and
+everything: it'll even be on by default.
+
+But here's the old patch,
+Rusty.
+--
+  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.
+
+diff -ur gzip-1.2.4.orig/deflate.c gzip-1.2.4-rsync/deflate.c
+--- gzip-1.2.4.orig/deflate.c  Fri Aug 13 22:35:31 1993
++++ gzip-1.2.4-rsync/deflate.c Sat Dec 30 15:33:25 2000
+@@ -121,6 +121,14 @@
+ #endif
+ /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
+ 
++#ifndef RSYNC_WIN
++#  define RSYNC_WIN 4096
++#endif
++/* Size of rsync window, must be < MAX_DIST */
++
++#define RSYNC_SUM_MATCH(sum) ((sum) % RSYNC_WIN == 0)
++/* Whether window sum matches magic value */
++
+ /* ===========================================================================
+  * Local data used by the "longest match" routines.
+  */
+@@ -202,6 +210,8 @@
+ unsigned near good_match;
+ /* Use a faster search when the previous match is longer than this */
+ 
++local ulg rsync_sum;  /* rolling sum of rsync window */
++local ulg rsync_chunk_end; /* next rsync sequence point */
+ 
+ /* 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
+@@ -300,6 +310,10 @@
+ #endif
+     /* prev will be initialized on the fly */
+ 
++    /* rsync params */
++    rsync_chunk_end = 0xFFFFFFFFUL;
++    rsync_sum = 0;
++
+     /* Set the default configuration parameters:
+      */
+     max_lazy_match   = configuration_table[pack_level].max_lazy;
+@@ -536,6 +550,8 @@
+         memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE);
+         match_start -= WSIZE;
+         strstart    -= WSIZE; /* we now have strstart >= MAX_DIST: */
++      if (rsync_chunk_end != 0xFFFFFFFFUL)
++          rsync_chunk_end -= WSIZE;
+ 
+         block_start -= (long) WSIZE;
+ 
+@@ -563,13 +579,46 @@
+     }
+ }
+ 
++local void rsync_roll(start, num)
++    unsigned start;
++    unsigned num;
++{
++    unsigned i;
++
++    if (start < RSYNC_WIN) {
++      /* before window fills. */
++      for (i = start; i < RSYNC_WIN; i++) {
++          if (i == start + num) return;
++          rsync_sum += (ulg)window[i];
++      }
++      num -= (RSYNC_WIN - start);
++      start = RSYNC_WIN;
++    }
++
++    /* buffer after window full */
++    for (i = start; i < start+num; i++) {
++      /* New character in */
++      rsync_sum += (ulg)window[i];
++      /* Old character out */
++      rsync_sum -= (ulg)window[i - RSYNC_WIN];
++      if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum))
++          rsync_chunk_end = i;
++    }
++}
++
++/* ===========================================================================
++ * Set rsync_chunk_end if window sum matches magic value.
++ */
++#define RSYNC_ROLL(s, n) \
++   do { if (rsync) rsync_roll((s), (n)); } 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(eof) \
+    flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \
+-                (char*)NULL, (long)strstart - block_start, (eof))
++                (char*)NULL, (long)strstart - block_start, flush-1, (eof))
+ 
+ /* ===========================================================================
+  * Processes a new input file and return its compressed length. This
+@@ -580,7 +629,7 @@
+ local ulg deflate_fast()
+ {
+     IPos hash_head; /* head of the hash chain */
+-    int flush;      /* set if current block must be flushed */
++    int flush;      /* set if current block must be flushed, 2=>and padded  */
+     unsigned match_length = 0;  /* length of best match */
+ 
+     prev_length = MIN_MATCH-1;
+@@ -609,6 +658,7 @@
+ 
+             lookahead -= match_length;
+ 
++          RSYNC_ROLL(strstart, match_length);
+           /* Insert new strings in the hash table only if the match length
+              * is not too large. This saves time but degrades compression.
+              */
+@@ -637,9 +687,14 @@
+             /* No match, output a literal byte */
+             Tracevv((stderr,"%c",window[strstart]));
+             flush = ct_tally (0, window[strstart]);
++          RSYNC_ROLL(strstart, 1);
+             lookahead--;
+           strstart++; 
+         }
++      if (rsync && strstart > rsync_chunk_end) {
++          rsync_chunk_end = 0xFFFFFFFFUL;
++          flush = 2;
++      } 
+         if (flush) FLUSH_BLOCK(0), block_start = strstart;
+ 
+         /* Make sure that we always have enough lookahead, except
+@@ -715,6 +770,7 @@
+              */
+             lookahead -= prev_length-1;
+             prev_length -= 2;
++          RSYNC_ROLL(strstart, prev_length+1);
+             do {
+                 strstart++;
+                 INSERT_STRING(strstart, hash_head);
+@@ -727,24 +783,39 @@
+             match_available = 0;
+             match_length = MIN_MATCH-1;
+             strstart++;
+-            if (flush) FLUSH_BLOCK(0), block_start = strstart;
+ 
++          if (rsync && strstart > rsync_chunk_end) {
++              rsync_chunk_end = 0xFFFFFFFFUL;
++              flush = 2;
++          }
++            if (flush) FLUSH_BLOCK(0), block_start = strstart;
+         } else if (match_available) {
+             /* If there was no match at the previous position, output a
+              * single literal. If there was a match but the current match
+              * is longer, truncate the previous match to a single literal.
+              */
+             Tracevv((stderr,"%c",window[strstart-1]));
+-            if (ct_tally (0, window[strstart-1])) {
+-                FLUSH_BLOCK(0), block_start = strstart;
+-            }
++          flush = ct_tally (0, window[strstart-1]);
++          if (rsync && strstart > rsync_chunk_end) {
++              rsync_chunk_end = 0xFFFFFFFFUL;
++              flush = 2;
++          }
++            if (flush) FLUSH_BLOCK(0), block_start = strstart;
++          RSYNC_ROLL(strstart, 1);
+             strstart++;
+             lookahead--;
+         } else {
+             /* There is no previous match to compare with, wait for
+              * the next step to decide.
+              */
++          if (rsync && strstart > rsync_chunk_end) {
++              /* Reset huffman tree */
++              rsync_chunk_end = 0xFFFFFFFFUL;
++              flush = 2;
++              FLUSH_BLOCK(0), block_start = strstart;
++          }
+             match_available = 1;
++          RSYNC_ROLL(strstart, 1);
+             strstart++;
+             lookahead--;
+         }
+diff -ur gzip-1.2.4.orig/gzip.c gzip-1.2.4-rsync/gzip.c
+--- gzip-1.2.4.orig/gzip.c     Thu Aug 19 23:39:43 1993
++++ gzip-1.2.4-rsync/gzip.c    Fri Dec 29 21:20:54 2000
+@@ -239,6 +239,7 @@
+ unsigned insize;           /* valid bytes in inbuf */
+ unsigned inptr;            /* index of next byte to be processed in inbuf */
+ unsigned outcnt;           /* bytes in output buffer */
++int rsync = 0;             /* make ryncable chunks */
+ 
+ struct option longopts[] =
+ {
+@@ -268,6 +269,7 @@
+     {"best",       0, 0, '9'}, /* compress better */
+     {"lzw",        0, 0, 'Z'}, /* make output compatible with old compress */
+     {"bits",       1, 0, 'b'}, /* max number of bits per code (implies -Z) */
++    {"rsyncable",  0, 0, 'R'}, /* make rsync-friendly archive */
+     { 0, 0, 0, 0 }
+ };
+ 
+@@ -357,6 +359,7 @@
+  " -Z --lzw         produce output compatible with old compress",
+  " -b --bits maxbits   max number of bits per code (implies -Z)",
+ #endif
++ "    --rsyncable   Make rsync-friendly archive",
+  " file...          files to (de)compress. If none given, use standard input.",
+   0};
+     char **p = help_msg;
+@@ -516,6 +519,9 @@
+ #else
+           recursive = 1; break;
+ #endif
++      case 'R':
++          rsync = 1; break;
++
+       case 'S':
+ #ifdef NO_MULTIPLE_DOTS
+             if (*optarg == '.') optarg++;
+diff -ur gzip-1.2.4.orig/gzip.h gzip-1.2.4-rsync/gzip.h
+--- gzip-1.2.4.orig/gzip.h     Fri Aug 13 22:35:33 1993
++++ gzip-1.2.4-rsync/gzip.h    Sat Dec 30 15:26:56 2000
+@@ -131,6 +131,7 @@
+ extern unsigned insize; /* valid bytes in inbuf */
+ extern unsigned inptr;  /* index of next byte to be processed in inbuf */
+ extern unsigned outcnt; /* bytes in output buffer */
++extern int rsync;  /* deflate into rsyncable chunks */
+ 
+ extern long bytes_in;   /* number of input bytes */
+ extern long bytes_out;  /* number of output bytes */
+@@ -282,7 +283,7 @@
+         /* in trees.c */
+ void ct_init     OF((ush *attr, int *method));
+ int  ct_tally    OF((int dist, int lc));
+-ulg  flush_block OF((char *buf, ulg stored_len, int eof));
++ulg  flush_block OF((char *buf, ulg stored_len, int pad, int eof));
+ 
+         /* in bits.c */
+ void     bi_init    OF((file_t zipfile));
+diff -ur gzip-1.2.4.orig/gzip.texi gzip-1.2.4-rsync/gzip.texi
+--- gzip-1.2.4.orig/gzip.texi  Thu Aug 19 06:42:50 1993
++++ gzip-1.2.4-rsync/gzip.texi Fri Dec 29 21:20:54 2000
+@@ -316,6 +316,14 @@
+ into the directory and compress all the files it finds there (or
+ decompress them in the case of @code{gunzip}).
+ 
++@item --rsyncable
++While compressing, synchronize the output occasionally based on the
++input.  This reduces compression by about 1 percent most cases, but
++means that the @code{rsync} program can take advantage of similarities
++in the uncompressed input when syncronizing two files compressed with
++this flag.  @code{gunzip} cannot tell the difference between a
++compressed file created with this option, and one created without it.
++
+ @item --suffix @var{suf}
+ @itemx -S @var{suf}
+ Use suffix @samp{@var{suf}} instead of @samp{.gz}. Any suffix can be
+diff -ur gzip-1.2.4.orig/trees.c gzip-1.2.4-rsync/trees.c
+--- gzip-1.2.4.orig/trees.c    Wed Aug 18 03:36:32 1993
++++ gzip-1.2.4-rsync/trees.c   Sat Dec 30 15:37:00 2000
+@@ -850,9 +850,10 @@
+  * trees or store, and output the encoded block to the zip file. This function
+  * returns the total compressed length for the file so far.
+  */
+-ulg flush_block(buf, stored_len, eof)
++ulg flush_block(buf, stored_len, pad, eof)
+     char *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 */
+@@ -944,6 +967,10 @@
+         Assert (input_len == isize, "bad input size");
+         bi_windup();
+         compressed_len += 7;  /* align on byte boundary */
++    } else if (pad && (compressed_len % 8) != 0) {
++        send_bits((STORED_BLOCK<<1)+eof, 3);  /* send block type */
++        compressed_len = (compressed_len + 3 + 7) & ~7L;
++        copy_block(buf, 0, 1); /* with header */
+     }
+     Tracev((stderr,"\ncomprlen %lu(%lu) ", compressed_len>>3,
+            compressed_len-7*eof));