xref: /openbsd-src/sys/lib/libz/deflate.c (revision 4e1ee0786f11cc571bd0be17d38e46f635c719fc)
1 /*	$OpenBSD: deflate.c,v 1.6 2021/07/28 07:36:06 tb Exp $ */
2 /* deflate.c -- compress data using the deflation algorithm
3  * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
4  * For conditions of distribution and use, see copyright notice in zlib.h
5  */
6 
7 /*
8  *  ALGORITHM
9  *
10  *      The "deflation" process depends on being able to identify portions
11  *      of the input text which are identical to earlier input (within a
12  *      sliding window trailing behind the input currently being processed).
13  *
14  *      The most straightforward technique turns out to be the fastest for
15  *      most input files: try all possible matches and select the longest.
16  *      The key feature of this algorithm is that insertions into the string
17  *      dictionary are very simple and thus fast, and deletions are avoided
18  *      completely. Insertions are performed at each input character, whereas
19  *      string matches are performed only when the previous match ends. So it
20  *      is preferable to spend more time in matches to allow very fast string
21  *      insertions and avoid deletions. The matching algorithm for small
22  *      strings is inspired from that of Rabin & Karp. A brute force approach
23  *      is used to find longer strings when a small match has been found.
24  *      A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
25  *      (by Leonid Broukhis).
26  *         A previous version of this file used a more sophisticated algorithm
27  *      (by Fiala and Greene) which is guaranteed to run in linear amortized
28  *      time, but has a larger average cost, uses more memory and is patented.
29  *      However the F&G algorithm may be faster for some highly redundant
30  *      files if the parameter max_chain_length (described below) is too large.
31  *
32  *  ACKNOWLEDGEMENTS
33  *
34  *      The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
35  *      I found it in 'freeze' written by Leonid Broukhis.
36  *      Thanks to many people for bug reports and testing.
37  *
38  *  REFERENCES
39  *
40  *      Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
41  *      Available in http://tools.ietf.org/html/rfc1951
42  *
43  *      A description of the Rabin and Karp algorithm is given in the book
44  *         "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45  *
46  *      Fiala,E.R., and Greene,D.H.
47  *         Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
48  *
49  */
50 
51 #include "deflate.h"
52 
53 /*
54   If you use the zlib library in a product, an acknowledgment is welcome
55   in the documentation of your product. If for some reason you cannot
56   include such an acknowledgment, I would appreciate that you keep this
57   copyright string in the executable of your product.
58  */
59 
60 /* ===========================================================================
61  *  Function prototypes.
62  */
63 typedef enum {
64     need_more,      /* block not completed, need more input or more output */
65     block_done,     /* block flush performed */
66     finish_started, /* finish started, need only more output at next deflate */
67     finish_done     /* finish done, accept no more input or output */
68 } block_state;
69 
70 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
71 /* Compression function. Returns the block state after the call. */
72 
73 local int deflateStateCheck      OF((z_streamp strm));
74 local void slide_hash     OF((deflate_state *s));
75 local void fill_window    OF((deflate_state *s));
76 local block_state deflate_stored OF((deflate_state *s, int flush));
77 local block_state deflate_fast   OF((deflate_state *s, int flush));
78 #ifndef FASTEST
79 local block_state deflate_slow   OF((deflate_state *s, int flush));
80 #endif
81 local block_state deflate_rle    OF((deflate_state *s, int flush));
82 local block_state deflate_huff   OF((deflate_state *s, int flush));
83 local void lm_init        OF((deflate_state *s));
84 local void putShortMSB    OF((deflate_state *s, uInt b));
85 local void flush_pending  OF((z_streamp strm));
86 local unsigned read_buf   OF((z_streamp strm, Bytef *buf, unsigned size));
87 #ifdef ASMV
88 #  pragma message("Assembler code may have bugs -- use at your own risk")
89       void match_init OF((void)); /* asm code initialization */
90       uInt longest_match  OF((deflate_state *s, IPos cur_match));
91 #else
92 local uInt longest_match  OF((deflate_state *s, IPos cur_match));
93 #endif
94 
95 #ifdef ZLIB_DEBUG
96 local  void check_match OF((deflate_state *s, IPos start, IPos match,
97                             int length));
98 #endif
99 
100 /* ===========================================================================
101  * Local data
102  */
103 
104 #define NIL 0
105 /* Tail of hash chains */
106 
107 #ifndef TOO_FAR
108 #  define TOO_FAR 4096
109 #endif
110 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
111 
112 /* Values for max_lazy_match, good_match and max_chain_length, depending on
113  * the desired pack level (0..9). The values given below have been tuned to
114  * exclude worst case performance for pathological files. Better values may be
115  * found for specific files.
116  */
117 typedef struct config_s {
118    ush good_length; /* reduce lazy search above this match length */
119    ush max_lazy;    /* do not perform lazy search above this match length */
120    ush nice_length; /* quit search above this match length */
121    ush max_chain;
122    compress_func func;
123 } config;
124 
125 #ifdef FASTEST
126 local const config configuration_table[2] = {
127 /*      good lazy nice chain */
128 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
129 /* 1 */ {4,    4,  8,    4, deflate_fast}}; /* max speed, no lazy matches */
130 #else
131 local const config configuration_table[10] = {
132 /*      good lazy nice chain */
133 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
134 /* 1 */ {4,    4,  8,    4, deflate_fast}, /* max speed, no lazy matches */
135 /* 2 */ {4,    5, 16,    8, deflate_fast},
136 /* 3 */ {4,    6, 32,   32, deflate_fast},
137 
138 /* 4 */ {4,    4, 16,   16, deflate_slow},  /* lazy matches */
139 /* 5 */ {8,   16, 32,   32, deflate_slow},
140 /* 6 */ {8,   16, 128, 128, deflate_slow},
141 /* 7 */ {8,   32, 128, 256, deflate_slow},
142 /* 8 */ {32, 128, 258, 1024, deflate_slow},
143 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
144 #endif
145 
146 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147  * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
148  * meaning.
149  */
150 
151 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
152 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
153 
154 /* ===========================================================================
155  * Update a hash value with the given input byte
156  * IN  assertion: all calls to UPDATE_HASH are made with consecutive input
157  *    characters, so that a running hash key can be computed from the previous
158  *    key instead of complete recalculation each time.
159  */
160 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
161 
162 
163 /* ===========================================================================
164  * Insert string str in the dictionary and set match_head to the previous head
165  * of the hash chain (the most recent string with same hash key). Return
166  * the previous length of the hash chain.
167  * If this file is compiled with -DFASTEST, the compression level is forced
168  * to 1, and no hash chains are maintained.
169  * IN  assertion: all calls to INSERT_STRING are made with consecutive input
170  *    characters and the first MIN_MATCH bytes of str are valid (except for
171  *    the last MIN_MATCH-1 bytes of the input file).
172  */
173 #ifdef FASTEST
174 #define INSERT_STRING(s, str, match_head) \
175    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
176     match_head = s->head[s->ins_h], \
177     s->head[s->ins_h] = (Pos)(str))
178 #else
179 #define INSERT_STRING(s, str, match_head) \
180    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
181     match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
182     s->head[s->ins_h] = (Pos)(str))
183 #endif
184 
185 /* ===========================================================================
186  * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
187  * prev[] will be initialized on the fly.
188  */
189 #define CLEAR_HASH(s) \
190     s->head[s->hash_size-1] = NIL; \
191     zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
192 
193 /* ===========================================================================
194  * Slide the hash table when sliding the window down (could be avoided with 32
195  * bit values at the expense of memory usage). We slide even when level == 0 to
196  * keep the hash table consistent if we switch back to level > 0 later.
197  */
198 local void slide_hash(s)
199     deflate_state *s;
200 {
201     unsigned n, m;
202     Posf *p;
203     uInt wsize = s->w_size;
204 
205     n = s->hash_size;
206     p = &s->head[n];
207     do {
208         m = *--p;
209         *p = (Pos)(m >= wsize ? m - wsize : NIL);
210     } while (--n);
211     n = wsize;
212 #ifndef FASTEST
213     p = &s->prev[n];
214     do {
215         m = *--p;
216         *p = (Pos)(m >= wsize ? m - wsize : NIL);
217         /* If n is not on any hash chain, prev[n] is garbage but
218          * its value will never be used.
219          */
220     } while (--n);
221 #endif
222 }
223 
224 /* ========================================================================= */
225 int ZEXPORT deflateInit_(strm, level, version, stream_size)
226     z_streamp strm;
227     int level;
228     const char *version;
229     int stream_size;
230 {
231     return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
232                          Z_DEFAULT_STRATEGY, version, stream_size);
233     /* To do: ignore strm->next_in if we use it as window */
234 }
235 
236 /* ========================================================================= */
237 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
238                   version, stream_size)
239     z_streamp strm;
240     int  level;
241     int  method;
242     int  windowBits;
243     int  memLevel;
244     int  strategy;
245     const char *version;
246     int stream_size;
247 {
248     deflate_state *s;
249     int wrap = 1;
250     static const char my_version[] = ZLIB_VERSION;
251 
252     ushf *overlay;
253     /* We overlay pending_buf and d_buf+l_buf. This works since the average
254      * output size for (length,distance) codes is <= 24 bits.
255      */
256 
257     if (version == Z_NULL || version[0] != my_version[0] ||
258         stream_size != sizeof(z_stream)) {
259         return Z_VERSION_ERROR;
260     }
261     if (strm == Z_NULL) return Z_STREAM_ERROR;
262 
263     strm->msg = Z_NULL;
264     if (strm->zalloc == (alloc_func)0) {
265 #ifdef Z_SOLO
266         return Z_STREAM_ERROR;
267 #else
268         strm->zalloc = zcalloc;
269         strm->opaque = (voidpf)0;
270 #endif
271     }
272     if (strm->zfree == (free_func)0)
273 #ifdef Z_SOLO
274         return Z_STREAM_ERROR;
275 #else
276         strm->zfree = zcfree;
277 #endif
278 
279 #ifdef FASTEST
280     if (level != 0) level = 1;
281 #else
282     if (level == Z_DEFAULT_COMPRESSION) level = 6;
283 #endif
284 
285     if (windowBits < 0) { /* suppress zlib wrapper */
286         wrap = 0;
287         windowBits = -windowBits;
288     }
289 #ifdef GZIP
290     else if (windowBits > 15) {
291         wrap = 2;       /* write gzip wrapper instead */
292         windowBits -= 16;
293     }
294 #endif
295     if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
296         windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
297         strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
298         return Z_STREAM_ERROR;
299     }
300     if (windowBits == 8) windowBits = 9;  /* until 256-byte window bug fixed */
301     s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
302     if (s == Z_NULL) return Z_MEM_ERROR;
303     strm->state = (struct internal_state FAR *)s;
304     s->strm = strm;
305     s->status = INIT_STATE;     /* to pass state test in deflateReset() */
306 
307     s->wrap = wrap;
308     s->gzhead = Z_NULL;
309     s->w_bits = (uInt)windowBits;
310     s->w_size = 1 << s->w_bits;
311     s->w_mask = s->w_size - 1;
312 
313     s->hash_bits = (uInt)memLevel + 7;
314     s->hash_size = 1 << s->hash_bits;
315     s->hash_mask = s->hash_size - 1;
316     s->hash_shift =  ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
317 
318     s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
319     s->prev   = (Posf *)  ZALLOC(strm, s->w_size, sizeof(Pos));
320     s->head   = (Posf *)  ZALLOC(strm, s->hash_size, sizeof(Pos));
321 
322     s->high_water = 0;      /* nothing written to s->window yet */
323 
324     s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
325 
326     overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
327     s->pending_buf = (uchf *) overlay;
328     s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
329 
330     if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
331         s->pending_buf == Z_NULL) {
332         s->status = FINISH_STATE;
333         strm->msg = ERR_MSG(Z_MEM_ERROR);
334         deflateEnd (strm);
335         return Z_MEM_ERROR;
336     }
337     s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
338     s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
339 
340     s->level = level;
341     s->strategy = strategy;
342     s->method = (Byte)method;
343 
344     return deflateReset(strm);
345 }
346 
347 /* =========================================================================
348  * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
349  */
350 local int deflateStateCheck (strm)
351     z_streamp strm;
352 {
353     deflate_state *s;
354     if (strm == Z_NULL ||
355         strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
356         return 1;
357     s = strm->state;
358     if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
359 #ifdef GZIP
360                                            s->status != GZIP_STATE &&
361 #endif
362                                            s->status != EXTRA_STATE &&
363                                            s->status != NAME_STATE &&
364                                            s->status != COMMENT_STATE &&
365                                            s->status != HCRC_STATE &&
366                                            s->status != BUSY_STATE &&
367                                            s->status != FINISH_STATE))
368         return 1;
369     return 0;
370 }
371 
372 /* ========================================================================= */
373 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
374     z_streamp strm;
375     const Bytef *dictionary;
376     uInt  dictLength;
377 {
378     deflate_state *s;
379     uInt str, n;
380     int wrap;
381     unsigned avail;
382     z_const unsigned char *next;
383 
384     if (deflateStateCheck(strm) || dictionary == Z_NULL)
385         return Z_STREAM_ERROR;
386     s = strm->state;
387     wrap = s->wrap;
388     if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
389         return Z_STREAM_ERROR;
390 
391     /* when using zlib wrappers, compute Adler-32 for provided dictionary */
392     if (wrap == 1)
393         strm->adler = adler32(strm->adler, dictionary, dictLength);
394     s->wrap = 0;                    /* avoid computing Adler-32 in read_buf */
395 
396     /* if dictionary would fill window, just replace the history */
397     if (dictLength >= s->w_size) {
398         if (wrap == 0) {            /* already empty otherwise */
399             CLEAR_HASH(s);
400             s->strstart = 0;
401             s->block_start = 0L;
402             s->insert = 0;
403         }
404         dictionary += dictLength - s->w_size;  /* use the tail */
405         dictLength = s->w_size;
406     }
407 
408     /* insert dictionary into window and hash */
409     avail = strm->avail_in;
410     next = strm->next_in;
411     strm->avail_in = dictLength;
412     strm->next_in = (z_const Bytef *)dictionary;
413     fill_window(s);
414     while (s->lookahead >= MIN_MATCH) {
415         str = s->strstart;
416         n = s->lookahead - (MIN_MATCH-1);
417         do {
418             UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
419 #ifndef FASTEST
420             s->prev[str & s->w_mask] = s->head[s->ins_h];
421 #endif
422             s->head[s->ins_h] = (Pos)str;
423             str++;
424         } while (--n);
425         s->strstart = str;
426         s->lookahead = MIN_MATCH-1;
427         fill_window(s);
428     }
429     s->strstart += s->lookahead;
430     s->block_start = (long)s->strstart;
431     s->insert = s->lookahead;
432     s->lookahead = 0;
433     s->match_length = s->prev_length = MIN_MATCH-1;
434     s->match_available = 0;
435     strm->next_in = next;
436     strm->avail_in = avail;
437     s->wrap = wrap;
438     return Z_OK;
439 }
440 
441 /* ========================================================================= */
442 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
443     z_streamp strm;
444     Bytef *dictionary;
445     uInt  *dictLength;
446 {
447     deflate_state *s;
448     uInt len;
449 
450     if (deflateStateCheck(strm))
451         return Z_STREAM_ERROR;
452     s = strm->state;
453     len = s->strstart + s->lookahead;
454     if (len > s->w_size)
455         len = s->w_size;
456     if (dictionary != Z_NULL && len)
457         zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
458     if (dictLength != Z_NULL)
459         *dictLength = len;
460     return Z_OK;
461 }
462 
463 /* ========================================================================= */
464 int ZEXPORT deflateResetKeep (strm)
465     z_streamp strm;
466 {
467     deflate_state *s;
468 
469     if (deflateStateCheck(strm)) {
470         return Z_STREAM_ERROR;
471     }
472 
473     strm->total_in = strm->total_out = 0;
474     strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
475     strm->data_type = Z_UNKNOWN;
476 
477     s = (deflate_state *)strm->state;
478     s->pending = 0;
479     s->pending_out = s->pending_buf;
480 
481     if (s->wrap < 0) {
482         s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
483     }
484     s->status =
485 #ifdef GZIP
486         s->wrap == 2 ? GZIP_STATE :
487 #endif
488         s->wrap ? INIT_STATE : BUSY_STATE;
489     strm->adler =
490 #ifdef GZIP
491         s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
492 #endif
493         adler32(0L, Z_NULL, 0);
494     s->last_flush = Z_NO_FLUSH;
495 
496     _tr_init(s);
497 
498     return Z_OK;
499 }
500 
501 /* ========================================================================= */
502 int ZEXPORT deflateReset (strm)
503     z_streamp strm;
504 {
505     int ret;
506 
507     ret = deflateResetKeep(strm);
508     if (ret == Z_OK)
509         lm_init(strm->state);
510     return ret;
511 }
512 
513 /* ========================================================================= */
514 int ZEXPORT deflateSetHeader (strm, head)
515     z_streamp strm;
516     gz_headerp head;
517 {
518     if (deflateStateCheck(strm) || strm->state->wrap != 2)
519         return Z_STREAM_ERROR;
520     strm->state->gzhead = head;
521     return Z_OK;
522 }
523 
524 /* ========================================================================= */
525 int ZEXPORT deflatePending (strm, pending, bits)
526     unsigned *pending;
527     int *bits;
528     z_streamp strm;
529 {
530     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
531     if (pending != Z_NULL)
532         *pending = strm->state->pending;
533     if (bits != Z_NULL)
534         *bits = strm->state->bi_valid;
535     return Z_OK;
536 }
537 
538 /* ========================================================================= */
539 int ZEXPORT deflatePrime (strm, bits, value)
540     z_streamp strm;
541     int bits;
542     int value;
543 {
544     deflate_state *s;
545     int put;
546 
547     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
548     s = strm->state;
549     if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
550         return Z_BUF_ERROR;
551     do {
552         put = Buf_size - s->bi_valid;
553         if (put > bits)
554             put = bits;
555         s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
556         s->bi_valid += put;
557         _tr_flush_bits(s);
558         value >>= put;
559         bits -= put;
560     } while (bits);
561     return Z_OK;
562 }
563 
564 /* ========================================================================= */
565 int ZEXPORT deflateParams(strm, level, strategy)
566     z_streamp strm;
567     int level;
568     int strategy;
569 {
570     deflate_state *s;
571     compress_func func;
572 
573     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
574     s = strm->state;
575 
576 #ifdef FASTEST
577     if (level != 0) level = 1;
578 #else
579     if (level == Z_DEFAULT_COMPRESSION) level = 6;
580 #endif
581     if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
582         return Z_STREAM_ERROR;
583     }
584     func = configuration_table[s->level].func;
585 
586     if ((strategy != s->strategy || func != configuration_table[level].func) &&
587         s->high_water) {
588         /* Flush the last buffer: */
589         int err = deflate(strm, Z_BLOCK);
590         if (err == Z_STREAM_ERROR)
591             return err;
592         if (strm->avail_out == 0)
593             return Z_BUF_ERROR;
594     }
595     if (s->level != level) {
596         if (s->level == 0 && s->matches != 0) {
597             if (s->matches == 1)
598                 slide_hash(s);
599             else
600                 CLEAR_HASH(s);
601             s->matches = 0;
602         }
603         s->level = level;
604         s->max_lazy_match   = configuration_table[level].max_lazy;
605         s->good_match       = configuration_table[level].good_length;
606         s->nice_match       = configuration_table[level].nice_length;
607         s->max_chain_length = configuration_table[level].max_chain;
608     }
609     s->strategy = strategy;
610     return Z_OK;
611 }
612 
613 /* ========================================================================= */
614 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
615     z_streamp strm;
616     int good_length;
617     int max_lazy;
618     int nice_length;
619     int max_chain;
620 {
621     deflate_state *s;
622 
623     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
624     s = strm->state;
625     s->good_match = (uInt)good_length;
626     s->max_lazy_match = (uInt)max_lazy;
627     s->nice_match = nice_length;
628     s->max_chain_length = (uInt)max_chain;
629     return Z_OK;
630 }
631 
632 /* =========================================================================
633  * For the default windowBits of 15 and memLevel of 8, this function returns
634  * a close to exact, as well as small, upper bound on the compressed size.
635  * They are coded as constants here for a reason--if the #define's are
636  * changed, then this function needs to be changed as well.  The return
637  * value for 15 and 8 only works for those exact settings.
638  *
639  * For any setting other than those defaults for windowBits and memLevel,
640  * the value returned is a conservative worst case for the maximum expansion
641  * resulting from using fixed blocks instead of stored blocks, which deflate
642  * can emit on compressed data for some combinations of the parameters.
643  *
644  * This function could be more sophisticated to provide closer upper bounds for
645  * every combination of windowBits and memLevel.  But even the conservative
646  * upper bound of about 14% expansion does not seem onerous for output buffer
647  * allocation.
648  */
649 uLong ZEXPORT deflateBound(strm, sourceLen)
650     z_streamp strm;
651     uLong sourceLen;
652 {
653     deflate_state *s;
654     uLong complen, wraplen;
655 
656     /* conservative upper bound for compressed data */
657     complen = sourceLen +
658               ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
659 
660     /* if can't get parameters, return conservative bound plus zlib wrapper */
661     if (deflateStateCheck(strm))
662         return complen + 6;
663 
664     /* compute wrapper length */
665     s = strm->state;
666     switch (s->wrap) {
667     case 0:                                 /* raw deflate */
668         wraplen = 0;
669         break;
670     case 1:                                 /* zlib wrapper */
671         wraplen = 6 + (s->strstart ? 4 : 0);
672         break;
673 #ifdef GZIP
674     case 2:                                 /* gzip wrapper */
675         wraplen = 18;
676         if (s->gzhead != Z_NULL) {          /* user-supplied gzip header */
677             Bytef *str;
678             if (s->gzhead->extra != Z_NULL)
679                 wraplen += 2 + s->gzhead->extra_len;
680             str = s->gzhead->name;
681             if (str != Z_NULL)
682                 do {
683                     wraplen++;
684                 } while (*str++);
685             str = s->gzhead->comment;
686             if (str != Z_NULL)
687                 do {
688                     wraplen++;
689                 } while (*str++);
690             if (s->gzhead->hcrc)
691                 wraplen += 2;
692         }
693         break;
694 #endif
695     default:                                /* for compiler happiness */
696         wraplen = 6;
697     }
698 
699     /* if not default parameters, return conservative bound */
700     if (s->w_bits != 15 || s->hash_bits != 8 + 7)
701         return complen + wraplen;
702 
703     /* default settings: return tight bound for that case */
704     return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
705            (sourceLen >> 25) + 13 - 6 + wraplen;
706 }
707 
708 /* =========================================================================
709  * Put a short in the pending buffer. The 16-bit value is put in MSB order.
710  * IN assertion: the stream state is correct and there is enough room in
711  * pending_buf.
712  */
713 local void putShortMSB (s, b)
714     deflate_state *s;
715     uInt b;
716 {
717     put_byte(s, (Byte)(b >> 8));
718     put_byte(s, (Byte)(b & 0xff));
719 }
720 
721 /* =========================================================================
722  * Flush as much pending output as possible. All deflate() output, except for
723  * some deflate_stored() output, goes through this function so some
724  * applications may wish to modify it to avoid allocating a large
725  * strm->next_out buffer and copying into it. (See also read_buf()).
726  */
727 local void flush_pending(strm)
728     z_streamp strm;
729 {
730     unsigned len;
731     deflate_state *s = strm->state;
732 
733     _tr_flush_bits(s);
734     len = s->pending;
735     if (len > strm->avail_out) len = strm->avail_out;
736     if (len == 0) return;
737 
738     zmemcpy(strm->next_out, s->pending_out, len);
739     strm->next_out  += len;
740     s->pending_out  += len;
741     strm->total_out += len;
742     strm->avail_out -= len;
743     s->pending      -= len;
744     if (s->pending == 0) {
745         s->pending_out = s->pending_buf;
746     }
747 }
748 
749 /* ===========================================================================
750  * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
751  */
752 #define HCRC_UPDATE(beg) \
753     do { \
754         if (s->gzhead->hcrc && s->pending > (beg)) \
755             strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
756                                 s->pending - (beg)); \
757     } while (0)
758 
759 /* ========================================================================= */
760 int ZEXPORT deflate (strm, flush)
761     z_streamp strm;
762     int flush;
763 {
764     int old_flush; /* value of flush param for previous deflate call */
765     deflate_state *s;
766 
767     if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
768         return Z_STREAM_ERROR;
769     }
770     s = strm->state;
771 
772     if (strm->next_out == Z_NULL ||
773         (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
774         (s->status == FINISH_STATE && flush != Z_FINISH)) {
775         ERR_RETURN(strm, Z_STREAM_ERROR);
776     }
777     if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
778 
779     old_flush = s->last_flush;
780     s->last_flush = flush;
781 
782     /* Flush as much pending output as possible */
783     if (s->pending != 0) {
784         flush_pending(strm);
785         if (strm->avail_out == 0) {
786             /* Since avail_out is 0, deflate will be called again with
787              * more output space, but possibly with both pending and
788              * avail_in equal to zero. There won't be anything to do,
789              * but this is not an error situation so make sure we
790              * return OK instead of BUF_ERROR at next call of deflate:
791              */
792             s->last_flush = -1;
793             return Z_OK;
794         }
795 
796     /* Make sure there is something to do and avoid duplicate consecutive
797      * flushes. For repeated and useless calls with Z_FINISH, we keep
798      * returning Z_STREAM_END instead of Z_BUF_ERROR.
799      */
800     } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
801                flush != Z_FINISH) {
802         ERR_RETURN(strm, Z_BUF_ERROR);
803     }
804 
805     /* User must not provide more input after the first FINISH: */
806     if (s->status == FINISH_STATE && strm->avail_in != 0) {
807         ERR_RETURN(strm, Z_BUF_ERROR);
808     }
809 
810     /* Write the header */
811     if (s->status == INIT_STATE) {
812         /* zlib header */
813         uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
814         uInt level_flags;
815 
816         if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
817             level_flags = 0;
818         else if (s->level < 6)
819             level_flags = 1;
820         else if (s->level == 6)
821             level_flags = 2;
822         else
823             level_flags = 3;
824         header |= (level_flags << 6);
825         if (s->strstart != 0) header |= PRESET_DICT;
826         header += 31 - (header % 31);
827 
828         putShortMSB(s, header);
829 
830         /* Save the adler32 of the preset dictionary: */
831         if (s->strstart != 0) {
832             putShortMSB(s, (uInt)(strm->adler >> 16));
833             putShortMSB(s, (uInt)(strm->adler & 0xffff));
834         }
835         strm->adler = adler32(0L, Z_NULL, 0);
836         s->status = BUSY_STATE;
837 
838         /* Compression must start with an empty pending buffer */
839         flush_pending(strm);
840         if (s->pending != 0) {
841             s->last_flush = -1;
842             return Z_OK;
843         }
844     }
845 #ifdef GZIP
846     if (s->status == GZIP_STATE) {
847         /* gzip header */
848         strm->adler = crc32(0L, Z_NULL, 0);
849         put_byte(s, 31);
850         put_byte(s, 139);
851         put_byte(s, 8);
852         if (s->gzhead == Z_NULL) {
853             put_byte(s, 0);
854             put_byte(s, 0);
855             put_byte(s, 0);
856             put_byte(s, 0);
857             put_byte(s, 0);
858             put_byte(s, s->level == 9 ? 2 :
859                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
860                       4 : 0));
861             put_byte(s, OS_CODE);
862             s->status = BUSY_STATE;
863 
864             /* Compression must start with an empty pending buffer */
865             flush_pending(strm);
866             if (s->pending != 0) {
867                 s->last_flush = -1;
868                 return Z_OK;
869             }
870         }
871         else {
872             put_byte(s, (s->gzhead->text ? 1 : 0) +
873                      (s->gzhead->hcrc ? 2 : 0) +
874                      (s->gzhead->extra == Z_NULL ? 0 : 4) +
875                      (s->gzhead->name == Z_NULL ? 0 : 8) +
876                      (s->gzhead->comment == Z_NULL ? 0 : 16)
877                      );
878             put_byte(s, (Byte)(s->gzhead->time & 0xff));
879             put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
880             put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
881             put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
882             put_byte(s, s->level == 9 ? 2 :
883                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
884                       4 : 0));
885             put_byte(s, s->gzhead->os & 0xff);
886             if (s->gzhead->extra != Z_NULL) {
887                 put_byte(s, s->gzhead->extra_len & 0xff);
888                 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
889             }
890             if (s->gzhead->hcrc)
891                 strm->adler = crc32(strm->adler, s->pending_buf,
892                                     s->pending);
893             s->gzindex = 0;
894             s->status = EXTRA_STATE;
895         }
896     }
897     if (s->status == EXTRA_STATE) {
898         if (s->gzhead->extra != Z_NULL) {
899             ulg beg = s->pending;   /* start of bytes to update crc */
900             uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
901             while (s->pending + left > s->pending_buf_size) {
902                 uInt copy = s->pending_buf_size - s->pending;
903                 zmemcpy(s->pending_buf + s->pending,
904                         s->gzhead->extra + s->gzindex, copy);
905                 s->pending = s->pending_buf_size;
906                 HCRC_UPDATE(beg);
907                 s->gzindex += copy;
908                 flush_pending(strm);
909                 if (s->pending != 0) {
910                     s->last_flush = -1;
911                     return Z_OK;
912                 }
913                 beg = 0;
914                 left -= copy;
915             }
916             zmemcpy(s->pending_buf + s->pending,
917                     s->gzhead->extra + s->gzindex, left);
918             s->pending += left;
919             HCRC_UPDATE(beg);
920             s->gzindex = 0;
921         }
922         s->status = NAME_STATE;
923     }
924     if (s->status == NAME_STATE) {
925         if (s->gzhead->name != Z_NULL) {
926             ulg beg = s->pending;   /* start of bytes to update crc */
927             int val;
928             do {
929                 if (s->pending == s->pending_buf_size) {
930                     HCRC_UPDATE(beg);
931                     flush_pending(strm);
932                     if (s->pending != 0) {
933                         s->last_flush = -1;
934                         return Z_OK;
935                     }
936                     beg = 0;
937                 }
938                 val = s->gzhead->name[s->gzindex++];
939                 put_byte(s, val);
940             } while (val != 0);
941             HCRC_UPDATE(beg);
942             s->gzindex = 0;
943         }
944         s->status = COMMENT_STATE;
945     }
946     if (s->status == COMMENT_STATE) {
947         if (s->gzhead->comment != Z_NULL) {
948             ulg beg = s->pending;   /* start of bytes to update crc */
949             int val;
950             do {
951                 if (s->pending == s->pending_buf_size) {
952                     HCRC_UPDATE(beg);
953                     flush_pending(strm);
954                     if (s->pending != 0) {
955                         s->last_flush = -1;
956                         return Z_OK;
957                     }
958                     beg = 0;
959                 }
960                 val = s->gzhead->comment[s->gzindex++];
961                 put_byte(s, val);
962             } while (val != 0);
963             HCRC_UPDATE(beg);
964         }
965         s->status = HCRC_STATE;
966     }
967     if (s->status == HCRC_STATE) {
968         if (s->gzhead->hcrc) {
969             if (s->pending + 2 > s->pending_buf_size) {
970                 flush_pending(strm);
971                 if (s->pending != 0) {
972                     s->last_flush = -1;
973                     return Z_OK;
974                 }
975             }
976             put_byte(s, (Byte)(strm->adler & 0xff));
977             put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
978             strm->adler = crc32(0L, Z_NULL, 0);
979         }
980         s->status = BUSY_STATE;
981 
982         /* Compression must start with an empty pending buffer */
983         flush_pending(strm);
984         if (s->pending != 0) {
985             s->last_flush = -1;
986             return Z_OK;
987         }
988     }
989 #endif
990 
991     /* Start a new block or continue the current one.
992      */
993     if (strm->avail_in != 0 || s->lookahead != 0 ||
994         (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
995         block_state bstate;
996 
997         bstate = s->level == 0 ? deflate_stored(s, flush) :
998                  s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
999                  s->strategy == Z_RLE ? deflate_rle(s, flush) :
1000                  (*(configuration_table[s->level].func))(s, flush);
1001 
1002         if (bstate == finish_started || bstate == finish_done) {
1003             s->status = FINISH_STATE;
1004         }
1005         if (bstate == need_more || bstate == finish_started) {
1006             if (strm->avail_out == 0) {
1007                 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1008             }
1009             return Z_OK;
1010             /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1011              * of deflate should use the same flush parameter to make sure
1012              * that the flush is complete. So we don't have to output an
1013              * empty block here, this will be done at next call. This also
1014              * ensures that for a very small output buffer, we emit at most
1015              * one empty block.
1016              */
1017         }
1018         if (bstate == block_done) {
1019             if (flush == Z_PARTIAL_FLUSH) {
1020                 _tr_align(s);
1021             } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1022                 _tr_stored_block(s, (char*)0, 0L, 0);
1023                 /* For a full flush, this empty block will be recognized
1024                  * as a special marker by inflate_sync().
1025                  */
1026                 if (flush == Z_FULL_FLUSH) {
1027                     CLEAR_HASH(s);             /* forget history */
1028                     if (s->lookahead == 0) {
1029                         s->strstart = 0;
1030                         s->block_start = 0L;
1031                         s->insert = 0;
1032                     }
1033                 }
1034             }
1035             flush_pending(strm);
1036             if (strm->avail_out == 0) {
1037               s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1038               return Z_OK;
1039             }
1040         }
1041     }
1042 
1043     if (flush != Z_FINISH) return Z_OK;
1044     if (s->wrap <= 0) return Z_STREAM_END;
1045 
1046     /* Write the trailer */
1047 #ifdef GZIP
1048     if (s->wrap == 2) {
1049         put_byte(s, (Byte)(strm->adler & 0xff));
1050         put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1051         put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1052         put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1053         put_byte(s, (Byte)(strm->total_in & 0xff));
1054         put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1055         put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1056         put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1057     }
1058     else
1059 #endif
1060     {
1061         putShortMSB(s, (uInt)(strm->adler >> 16));
1062         putShortMSB(s, (uInt)(strm->adler & 0xffff));
1063     }
1064     flush_pending(strm);
1065     /* If avail_out is zero, the application will call deflate again
1066      * to flush the rest.
1067      */
1068     if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1069     return s->pending != 0 ? Z_OK : Z_STREAM_END;
1070 }
1071 
1072 /* ========================================================================= */
1073 int ZEXPORT deflateEnd (strm)
1074     z_streamp strm;
1075 {
1076     int status;
1077 
1078     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1079 
1080     status = strm->state->status;
1081 
1082     /* Deallocate in reverse order of allocations: */
1083     TRY_FREE(strm, strm->state->pending_buf, strm->state->pending_buf_size);
1084     TRY_FREE(strm, strm->state->head, strm->state->hash_size * sizeof(Pos));
1085     TRY_FREE(strm, strm->state->prev, strm->state->w_size * sizeof(Pos));
1086     TRY_FREE(strm, strm->state->window, strm->state->w_size * 2 * sizeof(Byte));
1087 
1088     ZFREE(strm, strm->state, sizeof(deflate_state));
1089     strm->state = Z_NULL;
1090 
1091     return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1092 }
1093 
1094 /* =========================================================================
1095  * Copy the source state to the destination state.
1096  * To simplify the source, this is not supported for 16-bit MSDOS (which
1097  * doesn't have enough memory anyway to duplicate compression states).
1098  */
1099 int ZEXPORT deflateCopy (dest, source)
1100     z_streamp dest;
1101     z_streamp source;
1102 {
1103 #ifdef MAXSEG_64K
1104     return Z_STREAM_ERROR;
1105 #else
1106     deflate_state *ds;
1107     deflate_state *ss;
1108     ushf *overlay;
1109 
1110 
1111     if (deflateStateCheck(source) || dest == Z_NULL) {
1112         return Z_STREAM_ERROR;
1113     }
1114 
1115     ss = source->state;
1116 
1117     zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1118 
1119     ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1120     if (ds == Z_NULL) return Z_MEM_ERROR;
1121     dest->state = (struct internal_state FAR *) ds;
1122     zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1123     ds->strm = dest;
1124 
1125     ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1126     ds->prev   = (Posf *)  ZALLOC(dest, ds->w_size, sizeof(Pos));
1127     ds->head   = (Posf *)  ZALLOC(dest, ds->hash_size, sizeof(Pos));
1128     overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1129     ds->pending_buf = (uchf *) overlay;
1130 
1131     if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1132         ds->pending_buf == Z_NULL) {
1133         deflateEnd (dest);
1134         return Z_MEM_ERROR;
1135     }
1136     /* following zmemcpy do not work for 16-bit MSDOS */
1137     zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1138     zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1139     zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1140     zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1141 
1142     ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1143     ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1144     ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1145 
1146     ds->l_desc.dyn_tree = ds->dyn_ltree;
1147     ds->d_desc.dyn_tree = ds->dyn_dtree;
1148     ds->bl_desc.dyn_tree = ds->bl_tree;
1149 
1150     return Z_OK;
1151 #endif /* MAXSEG_64K */
1152 }
1153 
1154 /* ===========================================================================
1155  * Read a new buffer from the current input stream, update the adler32
1156  * and total number of bytes read.  All deflate() input goes through
1157  * this function so some applications may wish to modify it to avoid
1158  * allocating a large strm->next_in buffer and copying from it.
1159  * (See also flush_pending()).
1160  */
1161 local unsigned read_buf(strm, buf, size)
1162     z_streamp strm;
1163     Bytef *buf;
1164     unsigned size;
1165 {
1166     unsigned len = strm->avail_in;
1167 
1168     if (len > size) len = size;
1169     if (len == 0) return 0;
1170 
1171     strm->avail_in  -= len;
1172 
1173     zmemcpy(buf, strm->next_in, len);
1174     if (strm->state->wrap == 1) {
1175         strm->adler = adler32(strm->adler, buf, len);
1176     }
1177 #ifdef GZIP
1178     else if (strm->state->wrap == 2) {
1179         strm->adler = crc32(strm->adler, buf, len);
1180     }
1181 #endif
1182     strm->next_in  += len;
1183     strm->total_in += len;
1184 
1185     return len;
1186 }
1187 
1188 /* ===========================================================================
1189  * Initialize the "longest match" routines for a new zlib stream
1190  */
1191 local void lm_init (s)
1192     deflate_state *s;
1193 {
1194     s->window_size = (ulg)2L*s->w_size;
1195 
1196     CLEAR_HASH(s);
1197 
1198     /* Set the default configuration parameters:
1199      */
1200     s->max_lazy_match   = configuration_table[s->level].max_lazy;
1201     s->good_match       = configuration_table[s->level].good_length;
1202     s->nice_match       = configuration_table[s->level].nice_length;
1203     s->max_chain_length = configuration_table[s->level].max_chain;
1204 
1205     s->strstart = 0;
1206     s->block_start = 0L;
1207     s->lookahead = 0;
1208     s->insert = 0;
1209     s->match_length = s->prev_length = MIN_MATCH-1;
1210     s->match_available = 0;
1211     s->ins_h = 0;
1212 #ifndef FASTEST
1213 #ifdef ASMV
1214     match_init(); /* initialize the asm code */
1215 #endif
1216 #endif
1217 }
1218 
1219 #ifndef FASTEST
1220 /* ===========================================================================
1221  * Set match_start to the longest match starting at the given string and
1222  * return its length. Matches shorter or equal to prev_length are discarded,
1223  * in which case the result is equal to prev_length and match_start is
1224  * garbage.
1225  * IN assertions: cur_match is the head of the hash chain for the current
1226  *   string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1227  * OUT assertion: the match length is not greater than s->lookahead.
1228  */
1229 #ifndef ASMV
1230 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1231  * match.S. The code will be functionally equivalent.
1232  */
1233 local uInt longest_match(s, cur_match)
1234     deflate_state *s;
1235     IPos cur_match;                             /* current match */
1236 {
1237     unsigned chain_length = s->max_chain_length;/* max hash chain length */
1238     register Bytef *scan = s->window + s->strstart; /* current string */
1239     register Bytef *match;                      /* matched string */
1240     register int len;                           /* length of current match */
1241     int best_len = (int)s->prev_length;         /* best match length so far */
1242     int nice_match = s->nice_match;             /* stop if match long enough */
1243     IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1244         s->strstart - (IPos)MAX_DIST(s) : NIL;
1245     /* Stop when cur_match becomes <= limit. To simplify the code,
1246      * we prevent matches with the string of window index 0.
1247      */
1248     Posf *prev = s->prev;
1249     uInt wmask = s->w_mask;
1250 
1251 #ifdef UNALIGNED_OK
1252     /* Compare two bytes at a time. Note: this is not always beneficial.
1253      * Try with and without -DUNALIGNED_OK to check.
1254      */
1255     register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1256     register ush scan_start = *(ushf*)scan;
1257     register ush scan_end   = *(ushf*)(scan+best_len-1);
1258 #else
1259     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1260     register Byte scan_end1  = scan[best_len-1];
1261     register Byte scan_end   = scan[best_len];
1262 #endif
1263 
1264     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1265      * It is easy to get rid of this optimization if necessary.
1266      */
1267     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1268 
1269     /* Do not waste too much time if we already have a good match: */
1270     if (s->prev_length >= s->good_match) {
1271         chain_length >>= 2;
1272     }
1273     /* Do not look for matches beyond the end of the input. This is necessary
1274      * to make deflate deterministic.
1275      */
1276     if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1277 
1278     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1279 
1280     do {
1281         Assert(cur_match < s->strstart, "no future");
1282         match = s->window + cur_match;
1283 
1284         /* Skip to next match if the match length cannot increase
1285          * or if the match length is less than 2.  Note that the checks below
1286          * for insufficient lookahead only occur occasionally for performance
1287          * reasons.  Therefore uninitialized memory will be accessed, and
1288          * conditional jumps will be made that depend on those values.
1289          * However the length of the match is limited to the lookahead, so
1290          * the output of deflate is not affected by the uninitialized values.
1291          */
1292 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1293         /* This code assumes sizeof(unsigned short) == 2. Do not use
1294          * UNALIGNED_OK if your compiler uses a different size.
1295          */
1296         if (*(ushf*)(match+best_len-1) != scan_end ||
1297             *(ushf*)match != scan_start) continue;
1298 
1299         /* It is not necessary to compare scan[2] and match[2] since they are
1300          * always equal when the other bytes match, given that the hash keys
1301          * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1302          * strstart+3, +5, ... up to strstart+257. We check for insufficient
1303          * lookahead only every 4th comparison; the 128th check will be made
1304          * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1305          * necessary to put more guard bytes at the end of the window, or
1306          * to check more often for insufficient lookahead.
1307          */
1308         Assert(scan[2] == match[2], "scan[2]?");
1309         scan++, match++;
1310         do {
1311         } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1312                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1313                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1314                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1315                  scan < strend);
1316         /* The funny "do {}" generates better code on most compilers */
1317 
1318         /* Here, scan <= window+strstart+257 */
1319         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1320         if (*scan == *match) scan++;
1321 
1322         len = (MAX_MATCH - 1) - (int)(strend-scan);
1323         scan = strend - (MAX_MATCH-1);
1324 
1325 #else /* UNALIGNED_OK */
1326 
1327         if (match[best_len]   != scan_end  ||
1328             match[best_len-1] != scan_end1 ||
1329             *match            != *scan     ||
1330             *++match          != scan[1])      continue;
1331 
1332         /* The check at best_len-1 can be removed because it will be made
1333          * again later. (This heuristic is not always a win.)
1334          * It is not necessary to compare scan[2] and match[2] since they
1335          * are always equal when the other bytes match, given that
1336          * the hash keys are equal and that HASH_BITS >= 8.
1337          */
1338         scan += 2, match++;
1339         Assert(*scan == *match, "match[2]?");
1340 
1341         /* We check for insufficient lookahead only every 8th comparison;
1342          * the 256th check will be made at strstart+258.
1343          */
1344         do {
1345         } while (*++scan == *++match && *++scan == *++match &&
1346                  *++scan == *++match && *++scan == *++match &&
1347                  *++scan == *++match && *++scan == *++match &&
1348                  *++scan == *++match && *++scan == *++match &&
1349                  scan < strend);
1350 
1351         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1352 
1353         len = MAX_MATCH - (int)(strend - scan);
1354         scan = strend - MAX_MATCH;
1355 
1356 #endif /* UNALIGNED_OK */
1357 
1358         if (len > best_len) {
1359             s->match_start = cur_match;
1360             best_len = len;
1361             if (len >= nice_match) break;
1362 #ifdef UNALIGNED_OK
1363             scan_end = *(ushf*)(scan+best_len-1);
1364 #else
1365             scan_end1  = scan[best_len-1];
1366             scan_end   = scan[best_len];
1367 #endif
1368         }
1369     } while ((cur_match = prev[cur_match & wmask]) > limit
1370              && --chain_length != 0);
1371 
1372     if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1373     return s->lookahead;
1374 }
1375 #endif /* ASMV */
1376 
1377 #else /* FASTEST */
1378 
1379 /* ---------------------------------------------------------------------------
1380  * Optimized version for FASTEST only
1381  */
1382 local uInt longest_match(s, cur_match)
1383     deflate_state *s;
1384     IPos cur_match;                             /* current match */
1385 {
1386     register Bytef *scan = s->window + s->strstart; /* current string */
1387     register Bytef *match;                       /* matched string */
1388     register int len;                           /* length of current match */
1389     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1390 
1391     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1392      * It is easy to get rid of this optimization if necessary.
1393      */
1394     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1395 
1396     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1397 
1398     Assert(cur_match < s->strstart, "no future");
1399 
1400     match = s->window + cur_match;
1401 
1402     /* Return failure if the match length is less than 2:
1403      */
1404     if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1405 
1406     /* The check at best_len-1 can be removed because it will be made
1407      * again later. (This heuristic is not always a win.)
1408      * It is not necessary to compare scan[2] and match[2] since they
1409      * are always equal when the other bytes match, given that
1410      * the hash keys are equal and that HASH_BITS >= 8.
1411      */
1412     scan += 2, match += 2;
1413     Assert(*scan == *match, "match[2]?");
1414 
1415     /* We check for insufficient lookahead only every 8th comparison;
1416      * the 256th check will be made at strstart+258.
1417      */
1418     do {
1419     } while (*++scan == *++match && *++scan == *++match &&
1420              *++scan == *++match && *++scan == *++match &&
1421              *++scan == *++match && *++scan == *++match &&
1422              *++scan == *++match && *++scan == *++match &&
1423              scan < strend);
1424 
1425     Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1426 
1427     len = MAX_MATCH - (int)(strend - scan);
1428 
1429     if (len < MIN_MATCH) return MIN_MATCH - 1;
1430 
1431     s->match_start = cur_match;
1432     return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1433 }
1434 
1435 #endif /* FASTEST */
1436 
1437 #ifdef ZLIB_DEBUG
1438 
1439 #define EQUAL 0
1440 /* result of memcmp for equal strings */
1441 
1442 /* ===========================================================================
1443  * Check that the match at match_start is indeed a match.
1444  */
1445 local void check_match(s, start, match, length)
1446     deflate_state *s;
1447     IPos start, match;
1448     int length;
1449 {
1450     /* check that the match is indeed a match */
1451     if (zmemcmp(s->window + match,
1452                 s->window + start, length) != EQUAL) {
1453         fprintf(stderr, " start %u, match %u, length %d\n",
1454                 start, match, length);
1455         do {
1456             fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1457         } while (--length != 0);
1458         z_error("invalid match");
1459     }
1460     if (z_verbose > 1) {
1461         fprintf(stderr,"\\[%d,%d]", start-match, length);
1462         do { putc(s->window[start++], stderr); } while (--length != 0);
1463     }
1464 }
1465 #else
1466 #  define check_match(s, start, match, length)
1467 #endif /* ZLIB_DEBUG */
1468 
1469 /* ===========================================================================
1470  * Fill the window when the lookahead becomes insufficient.
1471  * Updates strstart and lookahead.
1472  *
1473  * IN assertion: lookahead < MIN_LOOKAHEAD
1474  * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1475  *    At least one byte has been read, or avail_in == 0; reads are
1476  *    performed for at least two bytes (required for the zip translate_eol
1477  *    option -- not supported here).
1478  */
1479 local void fill_window(s)
1480     deflate_state *s;
1481 {
1482     unsigned n;
1483     unsigned more;    /* Amount of free space at the end of the window. */
1484     uInt wsize = s->w_size;
1485 
1486     Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1487 
1488     do {
1489         more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1490 
1491         /* Deal with !@#$% 64K limit: */
1492         if (sizeof(int) <= 2) {
1493             if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1494                 more = wsize;
1495 
1496             } else if (more == (unsigned)(-1)) {
1497                 /* Very unlikely, but possible on 16 bit machine if
1498                  * strstart == 0 && lookahead == 1 (input done a byte at time)
1499                  */
1500                 more--;
1501             }
1502         }
1503 
1504         /* If the window is almost full and there is insufficient lookahead,
1505          * move the upper half to the lower one to make room in the upper half.
1506          */
1507         if (s->strstart >= wsize+MAX_DIST(s)) {
1508 
1509             zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1510             s->match_start -= wsize;
1511             s->strstart    -= wsize; /* we now have strstart >= MAX_DIST */
1512             s->block_start -= (long) wsize;
1513             slide_hash(s);
1514             more += wsize;
1515         }
1516         if (s->strm->avail_in == 0) break;
1517 
1518         /* If there was no sliding:
1519          *    strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1520          *    more == window_size - lookahead - strstart
1521          * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1522          * => more >= window_size - 2*WSIZE + 2
1523          * In the BIG_MEM or MMAP case (not yet supported),
1524          *   window_size == input_size + MIN_LOOKAHEAD  &&
1525          *   strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1526          * Otherwise, window_size == 2*WSIZE so more >= 2.
1527          * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1528          */
1529         Assert(more >= 2, "more < 2");
1530 
1531         n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1532         s->lookahead += n;
1533 
1534         /* Initialize the hash value now that we have some input: */
1535         if (s->lookahead + s->insert >= MIN_MATCH) {
1536             uInt str = s->strstart - s->insert;
1537             s->ins_h = s->window[str];
1538             UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1539 #if MIN_MATCH != 3
1540             Call UPDATE_HASH() MIN_MATCH-3 more times
1541 #endif
1542             while (s->insert) {
1543                 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1544 #ifndef FASTEST
1545                 s->prev[str & s->w_mask] = s->head[s->ins_h];
1546 #endif
1547                 s->head[s->ins_h] = (Pos)str;
1548                 str++;
1549                 s->insert--;
1550                 if (s->lookahead + s->insert < MIN_MATCH)
1551                     break;
1552             }
1553         }
1554         /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1555          * but this is not important since only literal bytes will be emitted.
1556          */
1557 
1558     } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1559 
1560     /* If the WIN_INIT bytes after the end of the current data have never been
1561      * written, then zero those bytes in order to avoid memory check reports of
1562      * the use of uninitialized (or uninitialised as Julian writes) bytes by
1563      * the longest match routines.  Update the high water mark for the next
1564      * time through here.  WIN_INIT is set to MAX_MATCH since the longest match
1565      * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1566      */
1567     if (s->high_water < s->window_size) {
1568         ulg curr = s->strstart + (ulg)(s->lookahead);
1569         ulg init;
1570 
1571         if (s->high_water < curr) {
1572             /* Previous high water mark below current data -- zero WIN_INIT
1573              * bytes or up to end of window, whichever is less.
1574              */
1575             init = s->window_size - curr;
1576             if (init > WIN_INIT)
1577                 init = WIN_INIT;
1578             zmemzero(s->window + curr, (unsigned)init);
1579             s->high_water = curr + init;
1580         }
1581         else if (s->high_water < (ulg)curr + WIN_INIT) {
1582             /* High water mark at or above current data, but below current data
1583              * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1584              * to end of window, whichever is less.
1585              */
1586             init = (ulg)curr + WIN_INIT - s->high_water;
1587             if (init > s->window_size - s->high_water)
1588                 init = s->window_size - s->high_water;
1589             zmemzero(s->window + s->high_water, (unsigned)init);
1590             s->high_water += init;
1591         }
1592     }
1593 
1594     Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1595            "not enough room for search");
1596 }
1597 
1598 /* ===========================================================================
1599  * Flush the current block, with given end-of-file flag.
1600  * IN assertion: strstart is set to the end of the current match.
1601  */
1602 #define FLUSH_BLOCK_ONLY(s, last) { \
1603    _tr_flush_block(s, (s->block_start >= 0L ? \
1604                    (charf *)&s->window[(unsigned)s->block_start] : \
1605                    (charf *)Z_NULL), \
1606                 (ulg)((long)s->strstart - s->block_start), \
1607                 (last)); \
1608    s->block_start = s->strstart; \
1609    flush_pending(s->strm); \
1610    Tracev((stderr,"[FLUSH]")); \
1611 }
1612 
1613 /* Same but force premature exit if necessary. */
1614 #define FLUSH_BLOCK(s, last) { \
1615    FLUSH_BLOCK_ONLY(s, last); \
1616    if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1617 }
1618 
1619 /* Maximum stored block length in deflate format (not including header). */
1620 #define MAX_STORED 65535
1621 
1622 #ifndef MIN
1623 /* Minimum of a and b. */
1624 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1625 #endif
1626 
1627 /* ===========================================================================
1628  * Copy without compression as much as possible from the input stream, return
1629  * the current block state.
1630  *
1631  * In case deflateParams() is used to later switch to a non-zero compression
1632  * level, s->matches (otherwise unused when storing) keeps track of the number
1633  * of hash table slides to perform. If s->matches is 1, then one hash table
1634  * slide will be done when switching. If s->matches is 2, the maximum value
1635  * allowed here, then the hash table will be cleared, since two or more slides
1636  * is the same as a clear.
1637  *
1638  * deflate_stored() is written to minimize the number of times an input byte is
1639  * copied. It is most efficient with large input and output buffers, which
1640  * maximizes the opportunites to have a single copy from next_in to next_out.
1641  */
1642 local block_state deflate_stored(s, flush)
1643     deflate_state *s;
1644     int flush;
1645 {
1646     /* Smallest worthy block size when not flushing or finishing. By default
1647      * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1648      * large input and output buffers, the stored block size will be larger.
1649      */
1650     unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1651 
1652     /* Copy as many min_block or larger stored blocks directly to next_out as
1653      * possible. If flushing, copy the remaining available input to next_out as
1654      * stored blocks, if there is enough space.
1655      */
1656     unsigned len, left, have, last = 0;
1657     unsigned used = s->strm->avail_in;
1658     do {
1659         /* Set len to the maximum size block that we can copy directly with the
1660          * available input data and output space. Set left to how much of that
1661          * would be copied from what's left in the window.
1662          */
1663         len = MAX_STORED;       /* maximum deflate stored block length */
1664         have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1665         if (s->strm->avail_out < have)          /* need room for header */
1666             break;
1667             /* maximum stored block length that will fit in avail_out: */
1668         have = s->strm->avail_out - have;
1669         left = s->strstart - s->block_start;    /* bytes left in window */
1670         if (len > (ulg)left + s->strm->avail_in)
1671             len = left + s->strm->avail_in;     /* limit len to the input */
1672         if (len > have)
1673             len = have;                         /* limit len to the output */
1674 
1675         /* If the stored block would be less than min_block in length, or if
1676          * unable to copy all of the available input when flushing, then try
1677          * copying to the window and the pending buffer instead. Also don't
1678          * write an empty block when flushing -- deflate() does that.
1679          */
1680         if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1681                                 flush == Z_NO_FLUSH ||
1682                                 len != left + s->strm->avail_in))
1683             break;
1684 
1685         /* Make a dummy stored block in pending to get the header bytes,
1686          * including any pending bits. This also updates the debugging counts.
1687          */
1688         last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1689         _tr_stored_block(s, (char *)0, 0L, last);
1690 
1691         /* Replace the lengths in the dummy stored block with len. */
1692         s->pending_buf[s->pending - 4] = len;
1693         s->pending_buf[s->pending - 3] = len >> 8;
1694         s->pending_buf[s->pending - 2] = ~len;
1695         s->pending_buf[s->pending - 1] = ~len >> 8;
1696 
1697         /* Write the stored block header bytes. */
1698         flush_pending(s->strm);
1699 
1700 #ifdef ZLIB_DEBUG
1701         /* Update debugging counts for the data about to be copied. */
1702         s->compressed_len += len << 3;
1703         s->bits_sent += len << 3;
1704 #endif
1705 
1706         /* Copy uncompressed bytes from the window to next_out. */
1707         if (left) {
1708             if (left > len)
1709                 left = len;
1710             zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1711             s->strm->next_out += left;
1712             s->strm->avail_out -= left;
1713             s->strm->total_out += left;
1714             s->block_start += left;
1715             len -= left;
1716         }
1717 
1718         /* Copy uncompressed bytes directly from next_in to next_out, updating
1719          * the check value.
1720          */
1721         if (len) {
1722             read_buf(s->strm, s->strm->next_out, len);
1723             s->strm->next_out += len;
1724             s->strm->avail_out -= len;
1725             s->strm->total_out += len;
1726         }
1727     } while (last == 0);
1728 
1729     /* Update the sliding window with the last s->w_size bytes of the copied
1730      * data, or append all of the copied data to the existing window if less
1731      * than s->w_size bytes were copied. Also update the number of bytes to
1732      * insert in the hash tables, in the event that deflateParams() switches to
1733      * a non-zero compression level.
1734      */
1735     used -= s->strm->avail_in;      /* number of input bytes directly copied */
1736     if (used) {
1737         /* If any input was used, then no unused input remains in the window,
1738          * therefore s->block_start == s->strstart.
1739          */
1740         if (used >= s->w_size) {    /* supplant the previous history */
1741             s->matches = 2;         /* clear hash */
1742             zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1743             s->strstart = s->w_size;
1744         }
1745         else {
1746             if (s->window_size - s->strstart <= used) {
1747                 /* Slide the window down. */
1748                 s->strstart -= s->w_size;
1749                 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1750                 if (s->matches < 2)
1751                     s->matches++;   /* add a pending slide_hash() */
1752             }
1753             zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1754             s->strstart += used;
1755         }
1756         s->block_start = s->strstart;
1757         s->insert += MIN(used, s->w_size - s->insert);
1758     }
1759     if (s->high_water < s->strstart)
1760         s->high_water = s->strstart;
1761 
1762     /* If the last block was written to next_out, then done. */
1763     if (last)
1764         return finish_done;
1765 
1766     /* If flushing and all input has been consumed, then done. */
1767     if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1768         s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1769         return block_done;
1770 
1771     /* Fill the window with any remaining input. */
1772     have = s->window_size - s->strstart - 1;
1773     if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1774         /* Slide the window down. */
1775         s->block_start -= s->w_size;
1776         s->strstart -= s->w_size;
1777         zmemcpy(s->window, s->window + s->w_size, s->strstart);
1778         if (s->matches < 2)
1779             s->matches++;           /* add a pending slide_hash() */
1780         have += s->w_size;          /* more space now */
1781     }
1782     if (have > s->strm->avail_in)
1783         have = s->strm->avail_in;
1784     if (have) {
1785         read_buf(s->strm, s->window + s->strstart, have);
1786         s->strstart += have;
1787     }
1788     if (s->high_water < s->strstart)
1789         s->high_water = s->strstart;
1790 
1791     /* There was not enough avail_out to write a complete worthy or flushed
1792      * stored block to next_out. Write a stored block to pending instead, if we
1793      * have enough input for a worthy block, or if flushing and there is enough
1794      * room for the remaining input as a stored block in the pending buffer.
1795      */
1796     have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1797         /* maximum stored block length that will fit in pending: */
1798     have = MIN(s->pending_buf_size - have, MAX_STORED);
1799     min_block = MIN(have, s->w_size);
1800     left = s->strstart - s->block_start;
1801     if (left >= min_block ||
1802         ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1803          s->strm->avail_in == 0 && left <= have)) {
1804         len = MIN(left, have);
1805         last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1806                len == left ? 1 : 0;
1807         _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1808         s->block_start += len;
1809         flush_pending(s->strm);
1810     }
1811 
1812     /* We've done all we can with the available input and output. */
1813     return last ? finish_started : need_more;
1814 }
1815 
1816 /* ===========================================================================
1817  * Compress as much as possible from the input stream, return the current
1818  * block state.
1819  * This function does not perform lazy evaluation of matches and inserts
1820  * new strings in the dictionary only for unmatched strings or for short
1821  * matches. It is used only for the fast compression options.
1822  */
1823 local block_state deflate_fast(s, flush)
1824     deflate_state *s;
1825     int flush;
1826 {
1827     IPos hash_head;       /* head of the hash chain */
1828     int bflush;           /* set if current block must be flushed */
1829 
1830     for (;;) {
1831         /* Make sure that we always have enough lookahead, except
1832          * at the end of the input file. We need MAX_MATCH bytes
1833          * for the next match, plus MIN_MATCH bytes to insert the
1834          * string following the next match.
1835          */
1836         if (s->lookahead < MIN_LOOKAHEAD) {
1837             fill_window(s);
1838             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1839                 return need_more;
1840             }
1841             if (s->lookahead == 0) break; /* flush the current block */
1842         }
1843 
1844         /* Insert the string window[strstart .. strstart+2] in the
1845          * dictionary, and set hash_head to the head of the hash chain:
1846          */
1847         hash_head = NIL;
1848         if (s->lookahead >= MIN_MATCH) {
1849             INSERT_STRING(s, s->strstart, hash_head);
1850         }
1851 
1852         /* Find the longest match, discarding those <= prev_length.
1853          * At this point we have always match_length < MIN_MATCH
1854          */
1855         if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1856             /* To simplify the code, we prevent matches with the string
1857              * of window index 0 (in particular we have to avoid a match
1858              * of the string with itself at the start of the input file).
1859              */
1860             s->match_length = longest_match (s, hash_head);
1861             /* longest_match() sets match_start */
1862         }
1863         if (s->match_length >= MIN_MATCH) {
1864             check_match(s, s->strstart, s->match_start, s->match_length);
1865 
1866             _tr_tally_dist(s, s->strstart - s->match_start,
1867                            s->match_length - MIN_MATCH, bflush);
1868 
1869             s->lookahead -= s->match_length;
1870 
1871             /* Insert new strings in the hash table only if the match length
1872              * is not too large. This saves time but degrades compression.
1873              */
1874 #ifndef FASTEST
1875             if (s->match_length <= s->max_insert_length &&
1876                 s->lookahead >= MIN_MATCH) {
1877                 s->match_length--; /* string at strstart already in table */
1878                 do {
1879                     s->strstart++;
1880                     INSERT_STRING(s, s->strstart, hash_head);
1881                     /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1882                      * always MIN_MATCH bytes ahead.
1883                      */
1884                 } while (--s->match_length != 0);
1885                 s->strstart++;
1886             } else
1887 #endif
1888             {
1889                 s->strstart += s->match_length;
1890                 s->match_length = 0;
1891                 s->ins_h = s->window[s->strstart];
1892                 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1893 #if MIN_MATCH != 3
1894                 Call UPDATE_HASH() MIN_MATCH-3 more times
1895 #endif
1896                 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1897                  * matter since it will be recomputed at next deflate call.
1898                  */
1899             }
1900         } else {
1901             /* No match, output a literal byte */
1902             Tracevv((stderr,"%c", s->window[s->strstart]));
1903             _tr_tally_lit (s, s->window[s->strstart], bflush);
1904             s->lookahead--;
1905             s->strstart++;
1906         }
1907         if (bflush) FLUSH_BLOCK(s, 0);
1908     }
1909     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1910     if (flush == Z_FINISH) {
1911         FLUSH_BLOCK(s, 1);
1912         return finish_done;
1913     }
1914     if (s->last_lit)
1915         FLUSH_BLOCK(s, 0);
1916     return block_done;
1917 }
1918 
1919 #ifndef FASTEST
1920 /* ===========================================================================
1921  * Same as above, but achieves better compression. We use a lazy
1922  * evaluation for matches: a match is finally adopted only if there is
1923  * no better match at the next window position.
1924  */
1925 local block_state deflate_slow(s, flush)
1926     deflate_state *s;
1927     int flush;
1928 {
1929     IPos hash_head;          /* head of hash chain */
1930     int bflush;              /* set if current block must be flushed */
1931 
1932     /* Process the input block. */
1933     for (;;) {
1934         /* Make sure that we always have enough lookahead, except
1935          * at the end of the input file. We need MAX_MATCH bytes
1936          * for the next match, plus MIN_MATCH bytes to insert the
1937          * string following the next match.
1938          */
1939         if (s->lookahead < MIN_LOOKAHEAD) {
1940             fill_window(s);
1941             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1942                 return need_more;
1943             }
1944             if (s->lookahead == 0) break; /* flush the current block */
1945         }
1946 
1947         /* Insert the string window[strstart .. strstart+2] in the
1948          * dictionary, and set hash_head to the head of the hash chain:
1949          */
1950         hash_head = NIL;
1951         if (s->lookahead >= MIN_MATCH) {
1952             INSERT_STRING(s, s->strstart, hash_head);
1953         }
1954 
1955         /* Find the longest match, discarding those <= prev_length.
1956          */
1957         s->prev_length = s->match_length, s->prev_match = s->match_start;
1958         s->match_length = MIN_MATCH-1;
1959 
1960         if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1961             s->strstart - hash_head <= MAX_DIST(s)) {
1962             /* To simplify the code, we prevent matches with the string
1963              * of window index 0 (in particular we have to avoid a match
1964              * of the string with itself at the start of the input file).
1965              */
1966             s->match_length = longest_match (s, hash_head);
1967             /* longest_match() sets match_start */
1968 
1969             if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1970 #if TOO_FAR <= 32767
1971                 || (s->match_length == MIN_MATCH &&
1972                     s->strstart - s->match_start > TOO_FAR)
1973 #endif
1974                 )) {
1975 
1976                 /* If prev_match is also MIN_MATCH, match_start is garbage
1977                  * but we will ignore the current match anyway.
1978                  */
1979                 s->match_length = MIN_MATCH-1;
1980             }
1981         }
1982         /* If there was a match at the previous step and the current
1983          * match is not better, output the previous match:
1984          */
1985         if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1986             uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1987             /* Do not insert strings in hash table beyond this. */
1988 
1989             check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1990 
1991             _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1992                            s->prev_length - MIN_MATCH, bflush);
1993 
1994             /* Insert in hash table all strings up to the end of the match.
1995              * strstart-1 and strstart are already inserted. If there is not
1996              * enough lookahead, the last two strings are not inserted in
1997              * the hash table.
1998              */
1999             s->lookahead -= s->prev_length-1;
2000             s->prev_length -= 2;
2001             do {
2002                 if (++s->strstart <= max_insert) {
2003                     INSERT_STRING(s, s->strstart, hash_head);
2004                 }
2005             } while (--s->prev_length != 0);
2006             s->match_available = 0;
2007             s->match_length = MIN_MATCH-1;
2008             s->strstart++;
2009 
2010             if (bflush) FLUSH_BLOCK(s, 0);
2011 
2012         } else if (s->match_available) {
2013             /* If there was no match at the previous position, output a
2014              * single literal. If there was a match but the current match
2015              * is longer, truncate the previous match to a single literal.
2016              */
2017             Tracevv((stderr,"%c", s->window[s->strstart-1]));
2018             _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2019             if (bflush) {
2020                 FLUSH_BLOCK_ONLY(s, 0);
2021             }
2022             s->strstart++;
2023             s->lookahead--;
2024             if (s->strm->avail_out == 0) return need_more;
2025         } else {
2026             /* There is no previous match to compare with, wait for
2027              * the next step to decide.
2028              */
2029             s->match_available = 1;
2030             s->strstart++;
2031             s->lookahead--;
2032         }
2033     }
2034     Assert (flush != Z_NO_FLUSH, "no flush?");
2035     if (s->match_available) {
2036         Tracevv((stderr,"%c", s->window[s->strstart-1]));
2037         _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2038         s->match_available = 0;
2039     }
2040     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2041     if (flush == Z_FINISH) {
2042         FLUSH_BLOCK(s, 1);
2043         return finish_done;
2044     }
2045     if (s->last_lit)
2046         FLUSH_BLOCK(s, 0);
2047     return block_done;
2048 }
2049 #endif /* FASTEST */
2050 
2051 /* ===========================================================================
2052  * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2053  * one.  Do not maintain a hash table.  (It will be regenerated if this run of
2054  * deflate switches away from Z_RLE.)
2055  */
2056 local block_state deflate_rle(s, flush)
2057     deflate_state *s;
2058     int flush;
2059 {
2060     int bflush;             /* set if current block must be flushed */
2061     uInt prev;              /* byte at distance one to match */
2062     Bytef *scan, *strend;   /* scan goes up to strend for length of run */
2063 
2064     for (;;) {
2065         /* Make sure that we always have enough lookahead, except
2066          * at the end of the input file. We need MAX_MATCH bytes
2067          * for the longest run, plus one for the unrolled loop.
2068          */
2069         if (s->lookahead <= MAX_MATCH) {
2070             fill_window(s);
2071             if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2072                 return need_more;
2073             }
2074             if (s->lookahead == 0) break; /* flush the current block */
2075         }
2076 
2077         /* See how many times the previous byte repeats */
2078         s->match_length = 0;
2079         if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2080             scan = s->window + s->strstart - 1;
2081             prev = *scan;
2082             if (prev == *++scan && prev == *++scan && prev == *++scan) {
2083                 strend = s->window + s->strstart + MAX_MATCH;
2084                 do {
2085                 } while (prev == *++scan && prev == *++scan &&
2086                          prev == *++scan && prev == *++scan &&
2087                          prev == *++scan && prev == *++scan &&
2088                          prev == *++scan && prev == *++scan &&
2089                          scan < strend);
2090                 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2091                 if (s->match_length > s->lookahead)
2092                     s->match_length = s->lookahead;
2093             }
2094             Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2095         }
2096 
2097         /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2098         if (s->match_length >= MIN_MATCH) {
2099             check_match(s, s->strstart, s->strstart - 1, s->match_length);
2100 
2101             _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2102 
2103             s->lookahead -= s->match_length;
2104             s->strstart += s->match_length;
2105             s->match_length = 0;
2106         } else {
2107             /* No match, output a literal byte */
2108             Tracevv((stderr,"%c", s->window[s->strstart]));
2109             _tr_tally_lit (s, s->window[s->strstart], bflush);
2110             s->lookahead--;
2111             s->strstart++;
2112         }
2113         if (bflush) FLUSH_BLOCK(s, 0);
2114     }
2115     s->insert = 0;
2116     if (flush == Z_FINISH) {
2117         FLUSH_BLOCK(s, 1);
2118         return finish_done;
2119     }
2120     if (s->last_lit)
2121         FLUSH_BLOCK(s, 0);
2122     return block_done;
2123 }
2124 
2125 /* ===========================================================================
2126  * For Z_HUFFMAN_ONLY, do not look for matches.  Do not maintain a hash table.
2127  * (It will be regenerated if this run of deflate switches away from Huffman.)
2128  */
2129 local block_state deflate_huff(s, flush)
2130     deflate_state *s;
2131     int flush;
2132 {
2133     int bflush;             /* set if current block must be flushed */
2134 
2135     for (;;) {
2136         /* Make sure that we have a literal to write. */
2137         if (s->lookahead == 0) {
2138             fill_window(s);
2139             if (s->lookahead == 0) {
2140                 if (flush == Z_NO_FLUSH)
2141                     return need_more;
2142                 break;      /* flush the current block */
2143             }
2144         }
2145 
2146         /* Output a literal byte */
2147         s->match_length = 0;
2148         Tracevv((stderr,"%c", s->window[s->strstart]));
2149         _tr_tally_lit (s, s->window[s->strstart], bflush);
2150         s->lookahead--;
2151         s->strstart++;
2152         if (bflush) FLUSH_BLOCK(s, 0);
2153     }
2154     s->insert = 0;
2155     if (flush == Z_FINISH) {
2156         FLUSH_BLOCK(s, 1);
2157         return finish_done;
2158     }
2159     if (s->last_lit)
2160         FLUSH_BLOCK(s, 0);
2161     return block_done;
2162 }
2163