| 1 | NOTE: this patch is for _gzip_! |
| 2 | |
| 3 | This is pending for actual inclusion in gzip. It is currently being |
| 4 | tried out in the default gzip for Debian Sarge, and may go into the |
| 5 | upstream gzip at somepoint in the not-too-distant future. |
| 6 | |
| 7 | --- gzip-1.3.5/deflate.c 1999-10-07 23:46:28 -0700 |
| 8 | +++ deflate.c 2005-02-05 09:40:33 -0800 |
| 9 | @@ -122,6 +122,14 @@ |
| 10 | #endif |
| 11 | /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ |
| 12 | |
| 13 | +#ifndef RSYNC_WIN |
| 14 | +# define RSYNC_WIN 4096 |
| 15 | +#endif |
| 16 | +/* Size of rsync window, must be < MAX_DIST */ |
| 17 | + |
| 18 | +#define RSYNC_SUM_MATCH(sum) ((sum) % RSYNC_WIN == 0) |
| 19 | +/* Whether window sum matches magic value */ |
| 20 | + |
| 21 | /* =========================================================================== |
| 22 | * Local data used by the "longest match" routines. |
| 23 | */ |
| 24 | @@ -203,6 +211,8 @@ |
| 25 | unsigned near good_match; |
| 26 | /* Use a faster search when the previous match is longer than this */ |
| 27 | |
| 28 | +local ulg rsync_sum; /* rolling sum of rsync window */ |
| 29 | +local ulg rsync_chunk_end; /* next rsync sequence point */ |
| 30 | |
| 31 | /* Values for max_lazy_match, good_match and max_chain_length, depending on |
| 32 | * the desired pack level (0..9). The values given below have been tuned to |
| 33 | @@ -301,6 +311,10 @@ |
| 34 | #endif |
| 35 | /* prev will be initialized on the fly */ |
| 36 | |
| 37 | + /* rsync params */ |
| 38 | + rsync_chunk_end = 0xFFFFFFFFUL; |
| 39 | + rsync_sum = 0; |
| 40 | + |
| 41 | /* Set the default configuration parameters: |
| 42 | */ |
| 43 | max_lazy_match = configuration_table[pack_level].max_lazy; |
| 44 | @@ -537,6 +551,8 @@ |
| 45 | memcpy((char*)window, (char*)window+WSIZE, (unsigned)WSIZE); |
| 46 | match_start -= WSIZE; |
| 47 | strstart -= WSIZE; /* we now have strstart >= MAX_DIST: */ |
| 48 | + if (rsync_chunk_end != 0xFFFFFFFFUL) |
| 49 | + rsync_chunk_end -= WSIZE; |
| 50 | |
| 51 | block_start -= (long) WSIZE; |
| 52 | |
| 53 | @@ -564,13 +580,46 @@ |
| 54 | } |
| 55 | } |
| 56 | |
| 57 | +local void rsync_roll(start, num) |
| 58 | + unsigned start; |
| 59 | + unsigned num; |
| 60 | +{ |
| 61 | + unsigned i; |
| 62 | + |
| 63 | + if (start < RSYNC_WIN) { |
| 64 | + /* before window fills. */ |
| 65 | + for (i = start; i < RSYNC_WIN; i++) { |
| 66 | + if (i == start + num) return; |
| 67 | + rsync_sum += (ulg)window[i]; |
| 68 | + } |
| 69 | + num -= (RSYNC_WIN - start); |
| 70 | + start = RSYNC_WIN; |
| 71 | + } |
| 72 | + |
| 73 | + /* buffer after window full */ |
| 74 | + for (i = start; i < start+num; i++) { |
| 75 | + /* New character in */ |
| 76 | + rsync_sum += (ulg)window[i]; |
| 77 | + /* Old character out */ |
| 78 | + rsync_sum -= (ulg)window[i - RSYNC_WIN]; |
| 79 | + if (rsync_chunk_end == 0xFFFFFFFFUL && RSYNC_SUM_MATCH(rsync_sum)) |
| 80 | + rsync_chunk_end = i; |
| 81 | + } |
| 82 | +} |
| 83 | + |
| 84 | +/* =========================================================================== |
| 85 | + * Set rsync_chunk_end if window sum matches magic value. |
| 86 | + */ |
| 87 | +#define RSYNC_ROLL(s, n) \ |
| 88 | + do { if (rsync) rsync_roll((s), (n)); } while(0) |
| 89 | + |
| 90 | /* =========================================================================== |
| 91 | * Flush the current block, with given end-of-file flag. |
| 92 | * IN assertion: strstart is set to the end of the current match. |
| 93 | */ |
| 94 | #define FLUSH_BLOCK(eof) \ |
| 95 | flush_block(block_start >= 0L ? (char*)&window[(unsigned)block_start] : \ |
| 96 | - (char*)NULL, (long)strstart - block_start, (eof)) |
| 97 | + (char*)NULL, (long)strstart - block_start, flush-1, (eof)) |
| 98 | |
| 99 | /* =========================================================================== |
| 100 | * Processes a new input file and return its compressed length. This |
| 101 | @@ -581,7 +630,7 @@ |
| 102 | local off_t deflate_fast() |
| 103 | { |
| 104 | IPos hash_head; /* head of the hash chain */ |
| 105 | - int flush; /* set if current block must be flushed */ |
| 106 | + int flush; /* set if current block must be flushed, 2=>and padded */ |
| 107 | unsigned match_length = 0; /* length of best match */ |
| 108 | |
| 109 | prev_length = MIN_MATCH-1; |
| 110 | @@ -610,6 +659,7 @@ |
| 111 | |
| 112 | lookahead -= match_length; |
| 113 | |
| 114 | + RSYNC_ROLL(strstart, match_length); |
| 115 | /* Insert new strings in the hash table only if the match length |
| 116 | * is not too large. This saves time but degrades compression. |
| 117 | */ |
| 118 | @@ -638,9 +688,14 @@ |
| 119 | /* No match, output a literal byte */ |
| 120 | Tracevv((stderr,"%c",window[strstart])); |
| 121 | flush = ct_tally (0, window[strstart]); |
| 122 | + RSYNC_ROLL(strstart, 1); |
| 123 | lookahead--; |
| 124 | strstart++; |
| 125 | } |
| 126 | + if (rsync && strstart > rsync_chunk_end) { |
| 127 | + rsync_chunk_end = 0xFFFFFFFFUL; |
| 128 | + flush = 2; |
| 129 | + } |
| 130 | if (flush) FLUSH_BLOCK(0), block_start = strstart; |
| 131 | |
| 132 | /* Make sure that we always have enough lookahead, except |
| 133 | @@ -713,6 +768,7 @@ |
| 134 | */ |
| 135 | lookahead -= prev_length-1; |
| 136 | prev_length -= 2; |
| 137 | + RSYNC_ROLL(strstart, prev_length+1); |
| 138 | do { |
| 139 | strstart++; |
| 140 | INSERT_STRING(strstart, hash_head); |
| 141 | @@ -725,24 +781,39 @@ |
| 142 | match_available = 0; |
| 143 | match_length = MIN_MATCH-1; |
| 144 | strstart++; |
| 145 | - if (flush) FLUSH_BLOCK(0), block_start = strstart; |
| 146 | |
| 147 | + if (rsync && strstart > rsync_chunk_end) { |
| 148 | + rsync_chunk_end = 0xFFFFFFFFUL; |
| 149 | + flush = 2; |
| 150 | + } |
| 151 | + if (flush) FLUSH_BLOCK(0), block_start = strstart; |
| 152 | } else if (match_available) { |
| 153 | /* If there was no match at the previous position, output a |
| 154 | * single literal. If there was a match but the current match |
| 155 | * is longer, truncate the previous match to a single literal. |
| 156 | */ |
| 157 | Tracevv((stderr,"%c",window[strstart-1])); |
| 158 | - if (ct_tally (0, window[strstart-1])) { |
| 159 | - FLUSH_BLOCK(0), block_start = strstart; |
| 160 | - } |
| 161 | + flush = ct_tally (0, window[strstart-1]); |
| 162 | + if (rsync && strstart > rsync_chunk_end) { |
| 163 | + rsync_chunk_end = 0xFFFFFFFFUL; |
| 164 | + flush = 2; |
| 165 | + } |
| 166 | + if (flush) FLUSH_BLOCK(0), block_start = strstart; |
| 167 | + RSYNC_ROLL(strstart, 1); |
| 168 | strstart++; |
| 169 | lookahead--; |
| 170 | } else { |
| 171 | /* There is no previous match to compare with, wait for |
| 172 | * the next step to decide. |
| 173 | */ |
| 174 | + if (rsync && strstart > rsync_chunk_end) { |
| 175 | + /* Reset huffman tree */ |
| 176 | + rsync_chunk_end = 0xFFFFFFFFUL; |
| 177 | + flush = 2; |
| 178 | + FLUSH_BLOCK(0), block_start = strstart; |
| 179 | + } |
| 180 | match_available = 1; |
| 181 | + RSYNC_ROLL(strstart, 1); |
| 182 | strstart++; |
| 183 | lookahead--; |
| 184 | } |
| 185 | --- gzip-1.3.5/gzip.c 2002-09-28 00:38:43 -0700 |
| 186 | +++ gzip.c 2005-02-05 09:40:33 -0800 |
| 187 | @@ -256,6 +256,7 @@ |
| 188 | unsigned insize; /* valid bytes in inbuf */ |
| 189 | unsigned inptr; /* index of next byte to be processed in inbuf */ |
| 190 | unsigned outcnt; /* bytes in output buffer */ |
| 191 | +int rsync = 0; /* make ryncable chunks */ |
| 192 | |
| 193 | struct option longopts[] = |
| 194 | { |
| 195 | @@ -285,6 +286,7 @@ |
| 196 | {"best", 0, 0, '9'}, /* compress better */ |
| 197 | {"lzw", 0, 0, 'Z'}, /* make output compatible with old compress */ |
| 198 | {"bits", 1, 0, 'b'}, /* max number of bits per code (implies -Z) */ |
| 199 | + {"rsyncable", 0, 0, 'R'}, /* make rsync-friendly archive */ |
| 200 | { 0, 0, 0, 0 } |
| 201 | }; |
| 202 | |
| 203 | @@ -365,6 +367,7 @@ |
| 204 | " -Z --lzw produce output compatible with old compress", |
| 205 | " -b --bits maxbits max number of bits per code (implies -Z)", |
| 206 | #endif |
| 207 | + " --rsyncable Make rsync-friendly archive", |
| 208 | " file... files to (de)compress. If none given, use standard input.", |
| 209 | "Report bugs to <bug-gzip@gnu.org>.", |
| 210 | 0}; |
| 211 | @@ -543,6 +546,9 @@ |
| 212 | #else |
| 213 | recursive = 1; break; |
| 214 | #endif |
| 215 | + case 'R': |
| 216 | + rsync = 1; break; |
| 217 | + |
| 218 | case 'S': |
| 219 | #ifdef NO_MULTIPLE_DOTS |
| 220 | if (*optarg == '.') optarg++; |
| 221 | --- gzip-1.3.5/gzip.h 2001-09-30 23:53:41 -0700 |
| 222 | +++ gzip.h 2005-02-05 09:40:33 -0800 |
| 223 | @@ -133,6 +133,7 @@ |
| 224 | extern unsigned insize; /* valid bytes in inbuf */ |
| 225 | extern unsigned inptr; /* index of next byte to be processed in inbuf */ |
| 226 | extern unsigned outcnt; /* bytes in output buffer */ |
| 227 | +extern int rsync; /* deflate into rsyncable chunks */ |
| 228 | |
| 229 | extern off_t bytes_in; /* number of input bytes */ |
| 230 | extern off_t bytes_out; /* number of output bytes */ |
| 231 | @@ -281,7 +282,7 @@ |
| 232 | /* in trees.c */ |
| 233 | void ct_init OF((ush *attr, int *method)); |
| 234 | int ct_tally OF((int dist, int lc)); |
| 235 | -off_t flush_block OF((char *buf, ulg stored_len, int eof)); |
| 236 | +off_t flush_block OF((char *buf, ulg stored_len, int pad, int eof)); |
| 237 | |
| 238 | /* in bits.c */ |
| 239 | void bi_init OF((file_t zipfile)); |
| 240 | --- gzip-1.3.5/gzip.texi 2002-09-29 23:57:29 -0700 |
| 241 | +++ gzip.texi 2005-02-05 09:40:33 -0800 |
| 242 | @@ -334,6 +334,14 @@ |
| 243 | into the directory and compress all the files it finds there (or |
| 244 | decompress them in the case of @code{gunzip}). |
| 245 | |
| 246 | +@item --rsyncable |
| 247 | +While compressing, synchronize the output occasionally based on the |
| 248 | +input. This increases size by less than 1 percent most cases, but |
| 249 | +means that the @code{rsync} program can take advantage of similarities |
| 250 | +in the uncompressed input when syncronizing two files compressed with |
| 251 | +this flag. @code{gunzip} cannot tell the difference between a |
| 252 | +compressed file created with this option, and one created without it. |
| 253 | + |
| 254 | @item --suffix @var{suf} |
| 255 | @itemx -S @var{suf} |
| 256 | Use suffix @samp{@var{suf}} instead of @samp{.gz}. Any suffix can be |
| 257 | --- gzip-1.3.5/trees.c 1999-10-05 22:00:00 -0700 |
| 258 | +++ trees.c 2005-02-05 09:40:33 -0800 |
| 259 | @@ -46,12 +46,13 @@ |
| 260 | * void ct_tally (int dist, int lc); |
| 261 | * Save the match info and tally the frequency counts. |
| 262 | * |
| 263 | - * off_t flush_block (char *buf, ulg stored_len, int eof) |
| 264 | + * off_t flush_block (char *buf, ulg stored_len, int pad, int eof) |
| 265 | * Determine the best encoding for the current block: dynamic trees, |
| 266 | * static trees or store, and output the encoded block to the zip |
| 267 | - * file. Returns the total compressed length for the file so far. |
| 268 | - * |
| 269 | - */ |
| 270 | + * file. If pad is set, pads the block to the next |
| 271 | + * byte. Returns the total compressed length for the file so |
| 272 | + * far. |
| 273 | + * */ |
| 274 | |
| 275 | #include <config.h> |
| 276 | #include <ctype.h> |
| 277 | @@ -847,9 +848,10 @@ |
| 278 | * trees or store, and output the encoded block to the zip file. This function |
| 279 | * returns the total compressed length for the file so far. |
| 280 | */ |
| 281 | -off_t flush_block(buf, stored_len, eof) |
| 282 | +off_t flush_block(buf, stored_len, pad, eof) |
| 283 | char *buf; /* input block, or NULL if too old */ |
| 284 | ulg stored_len; /* length of input block */ |
| 285 | + int pad; /* pad output to byte boundary */ |
| 286 | int eof; /* true if this is the last block for a file */ |
| 287 | { |
| 288 | ulg opt_lenb, static_lenb; /* opt_len and static_len in bytes */ |
| 289 | @@ -941,6 +943,10 @@ |
| 290 | Assert (input_len == bytes_in, "bad input size"); |
| 291 | bi_windup(); |
| 292 | compressed_len += 7; /* align on byte boundary */ |
| 293 | + } else if (pad && (compressed_len % 8) != 0) { |
| 294 | + send_bits((STORED_BLOCK<<1)+eof, 3); /* send block type */ |
| 295 | + compressed_len = (compressed_len + 3 + 7) & ~7L; |
| 296 | + copy_block(buf, 0, 1); /* with header */ |
| 297 | } |
| 298 | |
| 299 | return compressed_len >> 3; |