xref: /titanic_52/usr/src/contrib/zlib/deflate.c (revision 110570ceb08d1961475cc4922fdf73e6ffc65924)
1 /* deflate.c -- compress data using the deflation algorithm
2  * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
3  * For conditions of distribution and use, see copyright notice in zlib.h
4  */
5 
6 /*
7  *  ALGORITHM
8  *
9  *      The "deflation" process depends on being able to identify portions
10  *      of the input text which are identical to earlier input (within a
11  *      sliding window trailing behind the input currently being processed).
12  *
13  *      The most straightforward technique turns out to be the fastest for
14  *      most input files: try all possible matches and select the longest.
15  *      The key feature of this algorithm is that insertions into the string
16  *      dictionary are very simple and thus fast, and deletions are avoided
17  *      completely. Insertions are performed at each input character, whereas
18  *      string matches are performed only when the previous match ends. So it
19  *      is preferable to spend more time in matches to allow very fast string
20  *      insertions and avoid deletions. The matching algorithm for small
21  *      strings is inspired from that of Rabin & Karp. A brute force approach
22  *      is used to find longer strings when a small match has been found.
23  *      A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24  *      (by Leonid Broukhis).
25  *         A previous version of this file used a more sophisticated algorithm
26  *      (by Fiala and Greene) which is guaranteed to run in linear amortized
27  *      time, but has a larger average cost, uses more memory and is patented.
28  *      However the F&G algorithm may be faster for some highly redundant
29  *      files if the parameter max_chain_length (described below) is too large.
30  *
31  *  ACKNOWLEDGEMENTS
32  *
33  *      The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34  *      I found it in 'freeze' written by Leonid Broukhis.
35  *      Thanks to many people for bug reports and testing.
36  *
37  *  REFERENCES
38  *
39  *      Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40  *      Available in http://tools.ietf.org/html/rfc1951
41  *
42  *      A description of the Rabin and Karp algorithm is given in the book
43  *         "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44  *
45  *      Fiala,E.R., and Greene,D.H.
46  *         Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47  *
48  */
49 
50 #include "deflate.h"
51 
52 const char deflate_copyright[] =
53    " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler ";
54 /*
55   If you use the zlib library in a product, an acknowledgment is welcome
56   in the documentation of your product. If for some reason you cannot
57   include such an acknowledgment, I would appreciate that you keep this
58   copyright string in the executable of your product.
59  */
60 
61 /* ===========================================================================
62  *  Function prototypes.
63  */
64 typedef enum {
65     need_more,      /* block not completed, need more input or more output */
66     block_done,     /* block flush performed */
67     finish_started, /* finish started, need only more output at next deflate */
68     finish_done     /* finish done, accept no more input or output */
69 } block_state;
70 
71 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
72 /* Compression function. Returns the block state after the call. */
73 
74 local int deflateStateCheck      OF((z_streamp strm));
75 local void slide_hash     OF((deflate_state *s));
76 local void fill_window    OF((deflate_state *s));
77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 local block_state deflate_fast   OF((deflate_state *s, int flush));
79 #ifndef FASTEST
80 local block_state deflate_slow   OF((deflate_state *s, int flush));
81 #endif
82 local block_state deflate_rle    OF((deflate_state *s, int flush));
83 local block_state deflate_huff   OF((deflate_state *s, int flush));
84 local void lm_init        OF((deflate_state *s));
85 local void putShortMSB    OF((deflate_state *s, uInt b));
86 local void flush_pending  OF((z_streamp strm));
87 local unsigned read_buf   OF((z_streamp strm, Bytef *buf, unsigned size));
88 #ifdef ASMV
89 #  pragma message("Assembler code may have bugs -- use at your own risk")
90       void match_init OF((void)); /* asm code initialization */
91       uInt longest_match  OF((deflate_state *s, IPos cur_match));
92 #else
93 local uInt longest_match  OF((deflate_state *s, IPos cur_match));
94 #endif
95 
96 #ifdef ZLIB_DEBUG
97 local  void check_match OF((deflate_state *s, IPos start, IPos match,
98                             int length));
99 #endif
100 
101 /* ===========================================================================
102  * Local data
103  */
104 
105 #define NIL 0
106 /* Tail of hash chains */
107 
108 #ifndef TOO_FAR
109 #  define TOO_FAR 4096
110 #endif
111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112 
113 /* Values for max_lazy_match, good_match and max_chain_length, depending on
114  * the desired pack level (0..9). The values given below have been tuned to
115  * exclude worst case performance for pathological files. Better values may be
116  * found for specific files.
117  */
118 typedef struct config_s {
119    ush good_length; /* reduce lazy search above this match length */
120    ush max_lazy;    /* do not perform lazy search above this match length */
121    ush nice_length; /* quit search above this match length */
122    ush max_chain;
123    compress_func func;
124 } config;
125 
126 #ifdef FASTEST
127 local const config configuration_table[2] = {
128 /*      good lazy nice chain */
129 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
130 /* 1 */ {4,    4,  8,    4, deflate_fast}}; /* max speed, no lazy matches */
131 #else
132 local const config configuration_table[10] = {
133 /*      good lazy nice chain */
134 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
135 /* 1 */ {4,    4,  8,    4, deflate_fast}, /* max speed, no lazy matches */
136 /* 2 */ {4,    5, 16,    8, deflate_fast},
137 /* 3 */ {4,    6, 32,   32, deflate_fast},
138 
139 /* 4 */ {4,    4, 16,   16, deflate_slow},  /* lazy matches */
140 /* 5 */ {8,   16, 32,   32, deflate_slow},
141 /* 6 */ {8,   16, 128, 128, deflate_slow},
142 /* 7 */ {8,   32, 128, 256, deflate_slow},
143 /* 8 */ {32, 128, 258, 1024, deflate_slow},
144 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
145 #endif
146 
147 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
148  * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
149  * meaning.
150  */
151 
152 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
153 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
154 
155 /* ===========================================================================
156  * Update a hash value with the given input byte
157  * IN  assertion: all calls to UPDATE_HASH are made with consecutive input
158  *    characters, so that a running hash key can be computed from the previous
159  *    key instead of complete recalculation each time.
160  */
161 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
162 
163 
164 /* ===========================================================================
165  * Insert string str in the dictionary and set match_head to the previous head
166  * of the hash chain (the most recent string with same hash key). Return
167  * the previous length of the hash chain.
168  * If this file is compiled with -DFASTEST, the compression level is forced
169  * to 1, and no hash chains are maintained.
170  * IN  assertion: all calls to INSERT_STRING are made with consecutive input
171  *    characters and the first MIN_MATCH bytes of str are valid (except for
172  *    the last MIN_MATCH-1 bytes of the input file).
173  */
174 #ifdef FASTEST
175 #define INSERT_STRING(s, str, match_head) \
176    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
177     match_head = s->head[s->ins_h], \
178     s->head[s->ins_h] = (Pos)(str))
179 #else
180 #define INSERT_STRING(s, str, match_head) \
181    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
182     match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
183     s->head[s->ins_h] = (Pos)(str))
184 #endif
185 
186 /* ===========================================================================
187  * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
188  * prev[] will be initialized on the fly.
189  */
190 #define CLEAR_HASH(s) \
191     s->head[s->hash_size-1] = NIL; \
192     zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
193 
194 /* ===========================================================================
195  * Slide the hash table when sliding the window down (could be avoided with 32
196  * bit values at the expense of memory usage). We slide even when level == 0 to
197  * keep the hash table consistent if we switch back to level > 0 later.
198  */
199 local void slide_hash(s)
200     deflate_state *s;
201 {
202     unsigned n, m;
203     Posf *p;
204     uInt wsize = s->w_size;
205 
206     n = s->hash_size;
207     p = &s->head[n];
208     do {
209         m = *--p;
210         *p = (Pos)(m >= wsize ? m - wsize : NIL);
211     } while (--n);
212     n = wsize;
213 #ifndef FASTEST
214     p = &s->prev[n];
215     do {
216         m = *--p;
217         *p = (Pos)(m >= wsize ? m - wsize : NIL);
218         /* If n is not on any hash chain, prev[n] is garbage but
219          * its value will never be used.
220          */
221     } while (--n);
222 #endif
223 }
224 
225 /* ========================================================================= */
226 int ZEXPORT deflateInit_(strm, level, version, stream_size)
227     z_streamp strm;
228     int level;
229     const char *version;
230     int stream_size;
231 {
232     return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
233                          Z_DEFAULT_STRATEGY, version, stream_size);
234     /* To do: ignore strm->next_in if we use it as window */
235 }
236 
237 /* ========================================================================= */
238 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
239                   version, stream_size)
240     z_streamp strm;
241     int  level;
242     int  method;
243     int  windowBits;
244     int  memLevel;
245     int  strategy;
246     const char *version;
247     int stream_size;
248 {
249     deflate_state *s;
250     int wrap = 1;
251     static const char my_version[] = ZLIB_VERSION;
252 
253     ushf *overlay;
254     /* We overlay pending_buf and d_buf+l_buf. This works since the average
255      * output size for (length,distance) codes is <= 24 bits.
256      */
257 
258     if (version == Z_NULL || version[0] != my_version[0] ||
259         stream_size != sizeof(z_stream)) {
260         return Z_VERSION_ERROR;
261     }
262     if (strm == Z_NULL) return Z_STREAM_ERROR;
263 
264     strm->msg = Z_NULL;
265     if (strm->zalloc == (alloc_func)0) {
266 #ifdef Z_SOLO
267         return Z_STREAM_ERROR;
268 #else
269         strm->zalloc = zcalloc;
270         strm->opaque = (voidpf)0;
271 #endif
272     }
273     if (strm->zfree == (free_func)0)
274 #ifdef Z_SOLO
275         return Z_STREAM_ERROR;
276 #else
277         strm->zfree = zcfree;
278 #endif
279 
280 #ifdef FASTEST
281     if (level != 0) level = 1;
282 #else
283     if (level == Z_DEFAULT_COMPRESSION) level = 6;
284 #endif
285 
286     if (windowBits < 0) { /* suppress zlib wrapper */
287         wrap = 0;
288         windowBits = -windowBits;
289     }
290 #ifdef GZIP
291     else if (windowBits > 15) {
292         wrap = 2;       /* write gzip wrapper instead */
293         windowBits -= 16;
294     }
295 #endif
296     if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
297         windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
298         strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
299         return Z_STREAM_ERROR;
300     }
301     if (windowBits == 8) windowBits = 9;  /* until 256-byte window bug fixed */
302     s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
303     if (s == Z_NULL) return Z_MEM_ERROR;
304     strm->state = (struct internal_state FAR *)s;
305     s->strm = strm;
306     s->status = INIT_STATE;     /* to pass state test in deflateReset() */
307 
308     s->wrap = wrap;
309     s->gzhead = Z_NULL;
310     s->w_bits = (uInt)windowBits;
311     s->w_size = 1 << s->w_bits;
312     s->w_mask = s->w_size - 1;
313 
314     s->hash_bits = (uInt)memLevel + 7;
315     s->hash_size = 1 << s->hash_bits;
316     s->hash_mask = s->hash_size - 1;
317     s->hash_shift =  ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
318 
319     s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
320     s->prev   = (Posf *)  ZALLOC(strm, s->w_size, sizeof(Pos));
321     s->head   = (Posf *)  ZALLOC(strm, s->hash_size, sizeof(Pos));
322 
323     s->high_water = 0;      /* nothing written to s->window yet */
324 
325     s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
326 
327     overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
328     s->pending_buf = (uchf *) overlay;
329     s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
330 
331     if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
332         s->pending_buf == Z_NULL) {
333         s->status = FINISH_STATE;
334         strm->msg = ERR_MSG(Z_MEM_ERROR);
335         deflateEnd (strm);
336         return Z_MEM_ERROR;
337     }
338     s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
339     s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
340 
341     s->level = level;
342     s->strategy = strategy;
343     s->method = (Byte)method;
344 
345     return deflateReset(strm);
346 }
347 
348 /* =========================================================================
349  * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
350  */
351 local int deflateStateCheck (strm)
352     z_streamp strm;
353 {
354     deflate_state *s;
355     if (strm == Z_NULL ||
356         strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
357         return 1;
358     s = strm->state;
359     if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
360 #ifdef GZIP
361                                            s->status != GZIP_STATE &&
362 #endif
363                                            s->status != EXTRA_STATE &&
364                                            s->status != NAME_STATE &&
365                                            s->status != COMMENT_STATE &&
366                                            s->status != HCRC_STATE &&
367                                            s->status != BUSY_STATE &&
368                                            s->status != FINISH_STATE))
369         return 1;
370     return 0;
371 }
372 
373 /* ========================================================================= */
374 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
375     z_streamp strm;
376     const Bytef *dictionary;
377     uInt  dictLength;
378 {
379     deflate_state *s;
380     uInt str, n;
381     int wrap;
382     unsigned avail;
383     z_const unsigned char *next;
384 
385     if (deflateStateCheck(strm) || dictionary == Z_NULL)
386         return Z_STREAM_ERROR;
387     s = strm->state;
388     wrap = s->wrap;
389     if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
390         return Z_STREAM_ERROR;
391 
392     /* when using zlib wrappers, compute Adler-32 for provided dictionary */
393     if (wrap == 1)
394         strm->adler = adler32(strm->adler, dictionary, dictLength);
395     s->wrap = 0;                    /* avoid computing Adler-32 in read_buf */
396 
397     /* if dictionary would fill window, just replace the history */
398     if (dictLength >= s->w_size) {
399         if (wrap == 0) {            /* already empty otherwise */
400             CLEAR_HASH(s);
401             s->strstart = 0;
402             s->block_start = 0L;
403             s->insert = 0;
404         }
405         dictionary += dictLength - s->w_size;  /* use the tail */
406         dictLength = s->w_size;
407     }
408 
409     /* insert dictionary into window and hash */
410     avail = strm->avail_in;
411     next = strm->next_in;
412     strm->avail_in = dictLength;
413     strm->next_in = (z_const Bytef *)dictionary;
414     fill_window(s);
415     while (s->lookahead >= MIN_MATCH) {
416         str = s->strstart;
417         n = s->lookahead - (MIN_MATCH-1);
418         do {
419             UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
420 #ifndef FASTEST
421             s->prev[str & s->w_mask] = s->head[s->ins_h];
422 #endif
423             s->head[s->ins_h] = (Pos)str;
424             str++;
425         } while (--n);
426         s->strstart = str;
427         s->lookahead = MIN_MATCH-1;
428         fill_window(s);
429     }
430     s->strstart += s->lookahead;
431     s->block_start = (long)s->strstart;
432     s->insert = s->lookahead;
433     s->lookahead = 0;
434     s->match_length = s->prev_length = MIN_MATCH-1;
435     s->match_available = 0;
436     strm->next_in = next;
437     strm->avail_in = avail;
438     s->wrap = wrap;
439     return Z_OK;
440 }
441 
442 /* ========================================================================= */
443 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
444     z_streamp strm;
445     Bytef *dictionary;
446     uInt  *dictLength;
447 {
448     deflate_state *s;
449     uInt len;
450 
451     if (deflateStateCheck(strm))
452         return Z_STREAM_ERROR;
453     s = strm->state;
454     len = s->strstart + s->lookahead;
455     if (len > s->w_size)
456         len = s->w_size;
457     if (dictionary != Z_NULL && len)
458         zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
459     if (dictLength != Z_NULL)
460         *dictLength = len;
461     return Z_OK;
462 }
463 
464 /* ========================================================================= */
465 int ZEXPORT deflateResetKeep (strm)
466     z_streamp strm;
467 {
468     deflate_state *s;
469 
470     if (deflateStateCheck(strm)) {
471         return Z_STREAM_ERROR;
472     }
473 
474     strm->total_in = strm->total_out = 0;
475     strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
476     strm->data_type = Z_UNKNOWN;
477 
478     s = (deflate_state *)strm->state;
479     s->pending = 0;
480     s->pending_out = s->pending_buf;
481 
482     if (s->wrap < 0) {
483         s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
484     }
485     s->status =
486 #ifdef GZIP
487         s->wrap == 2 ? GZIP_STATE :
488 #endif
489         s->wrap ? INIT_STATE : BUSY_STATE;
490     strm->adler =
491 #ifdef GZIP
492         s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
493 #endif
494         adler32(0L, Z_NULL, 0);
495     s->last_flush = Z_NO_FLUSH;
496 
497     _tr_init(s);
498 
499     return Z_OK;
500 }
501 
502 /* ========================================================================= */
503 int ZEXPORT deflateReset (strm)
504     z_streamp strm;
505 {
506     int ret;
507 
508     ret = deflateResetKeep(strm);
509     if (ret == Z_OK)
510         lm_init(strm->state);
511     return ret;
512 }
513 
514 /* ========================================================================= */
515 int ZEXPORT deflateSetHeader (strm, head)
516     z_streamp strm;
517     gz_headerp head;
518 {
519     if (deflateStateCheck(strm) || strm->state->wrap != 2)
520         return Z_STREAM_ERROR;
521     strm->state->gzhead = head;
522     return Z_OK;
523 }
524 
525 /* ========================================================================= */
526 int ZEXPORT deflatePending (strm, pending, bits)
527     unsigned *pending;
528     int *bits;
529     z_streamp strm;
530 {
531     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
532     if (pending != Z_NULL)
533         *pending = strm->state->pending;
534     if (bits != Z_NULL)
535         *bits = strm->state->bi_valid;
536     return Z_OK;
537 }
538 
539 /* ========================================================================= */
540 int ZEXPORT deflatePrime (strm, bits, value)
541     z_streamp strm;
542     int bits;
543     int value;
544 {
545     deflate_state *s;
546     int put;
547 
548     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
549     s = strm->state;
550     if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
551         return Z_BUF_ERROR;
552     do {
553         put = Buf_size - s->bi_valid;
554         if (put > bits)
555             put = bits;
556         s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
557         s->bi_valid += put;
558         _tr_flush_bits(s);
559         value >>= put;
560         bits -= put;
561     } while (bits);
562     return Z_OK;
563 }
564 
565 /* ========================================================================= */
566 int ZEXPORT deflateParams(strm, level, strategy)
567     z_streamp strm;
568     int level;
569     int strategy;
570 {
571     deflate_state *s;
572     compress_func func;
573 
574     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
575     s = strm->state;
576 
577 #ifdef FASTEST
578     if (level != 0) level = 1;
579 #else
580     if (level == Z_DEFAULT_COMPRESSION) level = 6;
581 #endif
582     if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
583         return Z_STREAM_ERROR;
584     }
585     func = configuration_table[s->level].func;
586 
587     if ((strategy != s->strategy || func != configuration_table[level].func) &&
588         s->high_water) {
589         /* Flush the last buffer: */
590         int err = deflate(strm, Z_BLOCK);
591         if (err == Z_STREAM_ERROR)
592             return err;
593         if (strm->avail_out == 0)
594             return Z_BUF_ERROR;
595     }
596     if (s->level != level) {
597         if (s->level == 0 && s->matches != 0) {
598             if (s->matches == 1)
599                 slide_hash(s);
600             else
601                 CLEAR_HASH(s);
602             s->matches = 0;
603         }
604         s->level = level;
605         s->max_lazy_match   = configuration_table[level].max_lazy;
606         s->good_match       = configuration_table[level].good_length;
607         s->nice_match       = configuration_table[level].nice_length;
608         s->max_chain_length = configuration_table[level].max_chain;
609     }
610     s->strategy = strategy;
611     return Z_OK;
612 }
613 
614 /* ========================================================================= */
615 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
616     z_streamp strm;
617     int good_length;
618     int max_lazy;
619     int nice_length;
620     int max_chain;
621 {
622     deflate_state *s;
623 
624     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
625     s = strm->state;
626     s->good_match = (uInt)good_length;
627     s->max_lazy_match = (uInt)max_lazy;
628     s->nice_match = nice_length;
629     s->max_chain_length = (uInt)max_chain;
630     return Z_OK;
631 }
632 
633 /* =========================================================================
634  * For the default windowBits of 15 and memLevel of 8, this function returns
635  * a close to exact, as well as small, upper bound on the compressed size.
636  * They are coded as constants here for a reason--if the #define's are
637  * changed, then this function needs to be changed as well.  The return
638  * value for 15 and 8 only works for those exact settings.
639  *
640  * For any setting other than those defaults for windowBits and memLevel,
641  * the value returned is a conservative worst case for the maximum expansion
642  * resulting from using fixed blocks instead of stored blocks, which deflate
643  * can emit on compressed data for some combinations of the parameters.
644  *
645  * This function could be more sophisticated to provide closer upper bounds for
646  * every combination of windowBits and memLevel.  But even the conservative
647  * upper bound of about 14% expansion does not seem onerous for output buffer
648  * allocation.
649  */
650 uLong ZEXPORT deflateBound(strm, sourceLen)
651     z_streamp strm;
652     uLong sourceLen;
653 {
654     deflate_state *s;
655     uLong complen, wraplen;
656 
657     /* conservative upper bound for compressed data */
658     complen = sourceLen +
659               ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
660 
661     /* if can't get parameters, return conservative bound plus zlib wrapper */
662     if (deflateStateCheck(strm))
663         return complen + 6;
664 
665     /* compute wrapper length */
666     s = strm->state;
667     switch (s->wrap) {
668     case 0:                                 /* raw deflate */
669         wraplen = 0;
670         break;
671     case 1:                                 /* zlib wrapper */
672         wraplen = 6 + (s->strstart ? 4 : 0);
673         break;
674 #ifdef GZIP
675     case 2:                                 /* gzip wrapper */
676         wraplen = 18;
677         if (s->gzhead != Z_NULL) {          /* user-supplied gzip header */
678             Bytef *str;
679             if (s->gzhead->extra != Z_NULL)
680                 wraplen += 2 + s->gzhead->extra_len;
681             str = s->gzhead->name;
682             if (str != Z_NULL)
683                 do {
684                     wraplen++;
685                 } while (*str++);
686             str = s->gzhead->comment;
687             if (str != Z_NULL)
688                 do {
689                     wraplen++;
690                 } while (*str++);
691             if (s->gzhead->hcrc)
692                 wraplen += 2;
693         }
694         break;
695 #endif
696     default:                                /* for compiler happiness */
697         wraplen = 6;
698     }
699 
700     /* if not default parameters, return conservative bound */
701     if (s->w_bits != 15 || s->hash_bits != 8 + 7)
702         return complen + wraplen;
703 
704     /* default settings: return tight bound for that case */
705     return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
706            (sourceLen >> 25) + 13 - 6 + wraplen;
707 }
708 
709 /* =========================================================================
710  * Put a short in the pending buffer. The 16-bit value is put in MSB order.
711  * IN assertion: the stream state is correct and there is enough room in
712  * pending_buf.
713  */
714 local void putShortMSB (s, b)
715     deflate_state *s;
716     uInt b;
717 {
718     put_byte(s, (Byte)(b >> 8));
719     put_byte(s, (Byte)(b & 0xff));
720 }
721 
722 /* =========================================================================
723  * Flush as much pending output as possible. All deflate() output, except for
724  * some deflate_stored() output, goes through this function so some
725  * applications may wish to modify it to avoid allocating a large
726  * strm->next_out buffer and copying into it. (See also read_buf()).
727  */
728 local void flush_pending(strm)
729     z_streamp strm;
730 {
731     unsigned len;
732     deflate_state *s = strm->state;
733 
734     _tr_flush_bits(s);
735     len = s->pending;
736     if (len > strm->avail_out) len = strm->avail_out;
737     if (len == 0) return;
738 
739     zmemcpy(strm->next_out, s->pending_out, len);
740     strm->next_out  += len;
741     s->pending_out  += len;
742     strm->total_out += len;
743     strm->avail_out -= len;
744     s->pending      -= len;
745     if (s->pending == 0) {
746         s->pending_out = s->pending_buf;
747     }
748 }
749 
750 /* ===========================================================================
751  * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
752  */
753 #define HCRC_UPDATE(beg) \
754     do { \
755         if (s->gzhead->hcrc && s->pending > (beg)) \
756             strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
757                                 s->pending - (beg)); \
758     } while (0)
759 
760 /* ========================================================================= */
761 int ZEXPORT deflate (strm, flush)
762     z_streamp strm;
763     int flush;
764 {
765     int old_flush; /* value of flush param for previous deflate call */
766     deflate_state *s;
767 
768     if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
769         return Z_STREAM_ERROR;
770     }
771     s = strm->state;
772 
773     if (strm->next_out == Z_NULL ||
774         (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
775         (s->status == FINISH_STATE && flush != Z_FINISH)) {
776         ERR_RETURN(strm, Z_STREAM_ERROR);
777     }
778     if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
779 
780     old_flush = s->last_flush;
781     s->last_flush = flush;
782 
783     /* Flush as much pending output as possible */
784     if (s->pending != 0) {
785         flush_pending(strm);
786         if (strm->avail_out == 0) {
787             /* Since avail_out is 0, deflate will be called again with
788              * more output space, but possibly with both pending and
789              * avail_in equal to zero. There won't be anything to do,
790              * but this is not an error situation so make sure we
791              * return OK instead of BUF_ERROR at next call of deflate:
792              */
793             s->last_flush = -1;
794             return Z_OK;
795         }
796 
797     /* Make sure there is something to do and avoid duplicate consecutive
798      * flushes. For repeated and useless calls with Z_FINISH, we keep
799      * returning Z_STREAM_END instead of Z_BUF_ERROR.
800      */
801     } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
802                flush != Z_FINISH) {
803         ERR_RETURN(strm, Z_BUF_ERROR);
804     }
805 
806     /* User must not provide more input after the first FINISH: */
807     if (s->status == FINISH_STATE && strm->avail_in != 0) {
808         ERR_RETURN(strm, Z_BUF_ERROR);
809     }
810 
811     /* Write the header */
812     if (s->status == INIT_STATE) {
813         /* zlib header */
814         uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
815         uInt level_flags;
816 
817         if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
818             level_flags = 0;
819         else if (s->level < 6)
820             level_flags = 1;
821         else if (s->level == 6)
822             level_flags = 2;
823         else
824             level_flags = 3;
825         header |= (level_flags << 6);
826         if (s->strstart != 0) header |= PRESET_DICT;
827         header += 31 - (header % 31);
828 
829         putShortMSB(s, header);
830 
831         /* Save the adler32 of the preset dictionary: */
832         if (s->strstart != 0) {
833             putShortMSB(s, (uInt)(strm->adler >> 16));
834             putShortMSB(s, (uInt)(strm->adler & 0xffff));
835         }
836         strm->adler = adler32(0L, Z_NULL, 0);
837         s->status = BUSY_STATE;
838 
839         /* Compression must start with an empty pending buffer */
840         flush_pending(strm);
841         if (s->pending != 0) {
842             s->last_flush = -1;
843             return Z_OK;
844         }
845     }
846 #ifdef GZIP
847     if (s->status == GZIP_STATE) {
848         /* gzip header */
849         strm->adler = crc32(0L, Z_NULL, 0);
850         put_byte(s, 31);
851         put_byte(s, 139);
852         put_byte(s, 8);
853         if (s->gzhead == Z_NULL) {
854             put_byte(s, 0);
855             put_byte(s, 0);
856             put_byte(s, 0);
857             put_byte(s, 0);
858             put_byte(s, 0);
859             put_byte(s, s->level == 9 ? 2 :
860                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
861                       4 : 0));
862             put_byte(s, OS_CODE);
863             s->status = BUSY_STATE;
864 
865             /* Compression must start with an empty pending buffer */
866             flush_pending(strm);
867             if (s->pending != 0) {
868                 s->last_flush = -1;
869                 return Z_OK;
870             }
871         }
872         else {
873             put_byte(s, (s->gzhead->text ? 1 : 0) +
874                      (s->gzhead->hcrc ? 2 : 0) +
875                      (s->gzhead->extra == Z_NULL ? 0 : 4) +
876                      (s->gzhead->name == Z_NULL ? 0 : 8) +
877                      (s->gzhead->comment == Z_NULL ? 0 : 16)
878                      );
879             put_byte(s, (Byte)(s->gzhead->time & 0xff));
880             put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
881             put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
882             put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
883             put_byte(s, s->level == 9 ? 2 :
884                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
885                       4 : 0));
886             put_byte(s, s->gzhead->os & 0xff);
887             if (s->gzhead->extra != Z_NULL) {
888                 put_byte(s, s->gzhead->extra_len & 0xff);
889                 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
890             }
891             if (s->gzhead->hcrc)
892                 strm->adler = crc32(strm->adler, s->pending_buf,
893                                     s->pending);
894             s->gzindex = 0;
895             s->status = EXTRA_STATE;
896         }
897     }
898     if (s->status == EXTRA_STATE) {
899         if (s->gzhead->extra != Z_NULL) {
900             ulg beg = s->pending;   /* start of bytes to update crc */
901             uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
902             while (s->pending + left > s->pending_buf_size) {
903                 uInt copy = s->pending_buf_size - s->pending;
904                 zmemcpy(s->pending_buf + s->pending,
905                         s->gzhead->extra + s->gzindex, copy);
906                 s->pending = s->pending_buf_size;
907                 HCRC_UPDATE(beg);
908                 s->gzindex += copy;
909                 flush_pending(strm);
910                 if (s->pending != 0) {
911                     s->last_flush = -1;
912                     return Z_OK;
913                 }
914                 beg = 0;
915                 left -= copy;
916             }
917             zmemcpy(s->pending_buf + s->pending,
918                     s->gzhead->extra + s->gzindex, left);
919             s->pending += left;
920             HCRC_UPDATE(beg);
921             s->gzindex = 0;
922         }
923         s->status = NAME_STATE;
924     }
925     if (s->status == NAME_STATE) {
926         if (s->gzhead->name != Z_NULL) {
927             ulg beg = s->pending;   /* start of bytes to update crc */
928             int val;
929             do {
930                 if (s->pending == s->pending_buf_size) {
931                     HCRC_UPDATE(beg);
932                     flush_pending(strm);
933                     if (s->pending != 0) {
934                         s->last_flush = -1;
935                         return Z_OK;
936                     }
937                     beg = 0;
938                 }
939                 val = s->gzhead->name[s->gzindex++];
940                 put_byte(s, val);
941             } while (val != 0);
942             HCRC_UPDATE(beg);
943             s->gzindex = 0;
944         }
945         s->status = COMMENT_STATE;
946     }
947     if (s->status == COMMENT_STATE) {
948         if (s->gzhead->comment != Z_NULL) {
949             ulg beg = s->pending;   /* start of bytes to update crc */
950             int val;
951             do {
952                 if (s->pending == s->pending_buf_size) {
953                     HCRC_UPDATE(beg);
954                     flush_pending(strm);
955                     if (s->pending != 0) {
956                         s->last_flush = -1;
957                         return Z_OK;
958                     }
959                     beg = 0;
960                 }
961                 val = s->gzhead->comment[s->gzindex++];
962                 put_byte(s, val);
963             } while (val != 0);
964             HCRC_UPDATE(beg);
965         }
966         s->status = HCRC_STATE;
967     }
968     if (s->status == HCRC_STATE) {
969         if (s->gzhead->hcrc) {
970             if (s->pending + 2 > s->pending_buf_size) {
971                 flush_pending(strm);
972                 if (s->pending != 0) {
973                     s->last_flush = -1;
974                     return Z_OK;
975                 }
976             }
977             put_byte(s, (Byte)(strm->adler & 0xff));
978             put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
979             strm->adler = crc32(0L, Z_NULL, 0);
980         }
981         s->status = BUSY_STATE;
982 
983         /* Compression must start with an empty pending buffer */
984         flush_pending(strm);
985         if (s->pending != 0) {
986             s->last_flush = -1;
987             return Z_OK;
988         }
989     }
990 #endif
991 
992     /* Start a new block or continue the current one.
993      */
994     if (strm->avail_in != 0 || s->lookahead != 0 ||
995         (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
996         block_state bstate;
997 
998         bstate = s->level == 0 ? deflate_stored(s, flush) :
999                  s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1000                  s->strategy == Z_RLE ? deflate_rle(s, flush) :
1001                  (*(configuration_table[s->level].func))(s, flush);
1002 
1003         if (bstate == finish_started || bstate == finish_done) {
1004             s->status = FINISH_STATE;
1005         }
1006         if (bstate == need_more || bstate == finish_started) {
1007             if (strm->avail_out == 0) {
1008                 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1009             }
1010             return Z_OK;
1011             /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1012              * of deflate should use the same flush parameter to make sure
1013              * that the flush is complete. So we don't have to output an
1014              * empty block here, this will be done at next call. This also
1015              * ensures that for a very small output buffer, we emit at most
1016              * one empty block.
1017              */
1018         }
1019         if (bstate == block_done) {
1020             if (flush == Z_PARTIAL_FLUSH) {
1021                 _tr_align(s);
1022             } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1023                 _tr_stored_block(s, (char*)0, 0L, 0);
1024                 /* For a full flush, this empty block will be recognized
1025                  * as a special marker by inflate_sync().
1026                  */
1027                 if (flush == Z_FULL_FLUSH) {
1028                     CLEAR_HASH(s);             /* forget history */
1029                     if (s->lookahead == 0) {
1030                         s->strstart = 0;
1031                         s->block_start = 0L;
1032                         s->insert = 0;
1033                     }
1034                 }
1035             }
1036             flush_pending(strm);
1037             if (strm->avail_out == 0) {
1038               s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1039               return Z_OK;
1040             }
1041         }
1042     }
1043 
1044     if (flush != Z_FINISH) return Z_OK;
1045     if (s->wrap <= 0) return Z_STREAM_END;
1046 
1047     /* Write the trailer */
1048 #ifdef GZIP
1049     if (s->wrap == 2) {
1050         put_byte(s, (Byte)(strm->adler & 0xff));
1051         put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1052         put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1053         put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1054         put_byte(s, (Byte)(strm->total_in & 0xff));
1055         put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1056         put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1057         put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1058     }
1059     else
1060 #endif
1061     {
1062         putShortMSB(s, (uInt)(strm->adler >> 16));
1063         putShortMSB(s, (uInt)(strm->adler & 0xffff));
1064     }
1065     flush_pending(strm);
1066     /* If avail_out is zero, the application will call deflate again
1067      * to flush the rest.
1068      */
1069     if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1070     return s->pending != 0 ? Z_OK : Z_STREAM_END;
1071 }
1072 
1073 /* ========================================================================= */
1074 int ZEXPORT deflateEnd (strm)
1075     z_streamp strm;
1076 {
1077     int status;
1078 
1079     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1080 
1081     status = strm->state->status;
1082 
1083     /* Deallocate in reverse order of allocations: */
1084     TRY_FREE(strm, strm->state->pending_buf);
1085     TRY_FREE(strm, strm->state->head);
1086     TRY_FREE(strm, strm->state->prev);
1087     TRY_FREE(strm, strm->state->window);
1088 
1089     ZFREE(strm, strm->state);
1090     strm->state = Z_NULL;
1091 
1092     return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1093 }
1094 
1095 /* =========================================================================
1096  * Copy the source state to the destination state.
1097  * To simplify the source, this is not supported for 16-bit MSDOS (which
1098  * doesn't have enough memory anyway to duplicate compression states).
1099  */
1100 int ZEXPORT deflateCopy (dest, source)
1101     z_streamp dest;
1102     z_streamp source;
1103 {
1104 #ifdef MAXSEG_64K
1105     return Z_STREAM_ERROR;
1106 #else
1107     deflate_state *ds;
1108     deflate_state *ss;
1109     ushf *overlay;
1110 
1111 
1112     if (deflateStateCheck(source) || dest == Z_NULL) {
1113         return Z_STREAM_ERROR;
1114     }
1115 
1116     ss = source->state;
1117 
1118     zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1119 
1120     ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1121     if (ds == Z_NULL) return Z_MEM_ERROR;
1122     dest->state = (struct internal_state FAR *) ds;
1123     zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1124     ds->strm = dest;
1125 
1126     ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1127     ds->prev   = (Posf *)  ZALLOC(dest, ds->w_size, sizeof(Pos));
1128     ds->head   = (Posf *)  ZALLOC(dest, ds->hash_size, sizeof(Pos));
1129     overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1130     ds->pending_buf = (uchf *) overlay;
1131 
1132     if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1133         ds->pending_buf == Z_NULL) {
1134         deflateEnd (dest);
1135         return Z_MEM_ERROR;
1136     }
1137     /* following zmemcpy do not work for 16-bit MSDOS */
1138     zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1139     zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1140     zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1141     zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1142 
1143     ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1144     ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1145     ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1146 
1147     ds->l_desc.dyn_tree = ds->dyn_ltree;
1148     ds->d_desc.dyn_tree = ds->dyn_dtree;
1149     ds->bl_desc.dyn_tree = ds->bl_tree;
1150 
1151     return Z_OK;
1152 #endif /* MAXSEG_64K */
1153 }
1154 
1155 /* ===========================================================================
1156  * Read a new buffer from the current input stream, update the adler32
1157  * and total number of bytes read.  All deflate() input goes through
1158  * this function so some applications may wish to modify it to avoid
1159  * allocating a large strm->next_in buffer and copying from it.
1160  * (See also flush_pending()).
1161  */
1162 local unsigned read_buf(strm, buf, size)
1163     z_streamp strm;
1164     Bytef *buf;
1165     unsigned size;
1166 {
1167     unsigned len = strm->avail_in;
1168 
1169     if (len > size) len = size;
1170     if (len == 0) return 0;
1171 
1172     strm->avail_in  -= len;
1173 
1174     zmemcpy(buf, strm->next_in, len);
1175     if (strm->state->wrap == 1) {
1176         strm->adler = adler32(strm->adler, buf, len);
1177     }
1178 #ifdef GZIP
1179     else if (strm->state->wrap == 2) {
1180         strm->adler = crc32(strm->adler, buf, len);
1181     }
1182 #endif
1183     strm->next_in  += len;
1184     strm->total_in += len;
1185 
1186     return len;
1187 }
1188 
1189 /* ===========================================================================
1190  * Initialize the "longest match" routines for a new zlib stream
1191  */
1192 local void lm_init (s)
1193     deflate_state *s;
1194 {
1195     s->window_size = (ulg)2L*s->w_size;
1196 
1197     CLEAR_HASH(s);
1198 
1199     /* Set the default configuration parameters:
1200      */
1201     s->max_lazy_match   = configuration_table[s->level].max_lazy;
1202     s->good_match       = configuration_table[s->level].good_length;
1203     s->nice_match       = configuration_table[s->level].nice_length;
1204     s->max_chain_length = configuration_table[s->level].max_chain;
1205 
1206     s->strstart = 0;
1207     s->block_start = 0L;
1208     s->lookahead = 0;
1209     s->insert = 0;
1210     s->match_length = s->prev_length = MIN_MATCH-1;
1211     s->match_available = 0;
1212     s->ins_h = 0;
1213 #ifndef FASTEST
1214 #ifdef ASMV
1215     match_init(); /* initialize the asm code */
1216 #endif
1217 #endif
1218 }
1219 
1220 #ifndef FASTEST
1221 /* ===========================================================================
1222  * Set match_start to the longest match starting at the given string and
1223  * return its length. Matches shorter or equal to prev_length are discarded,
1224  * in which case the result is equal to prev_length and match_start is
1225  * garbage.
1226  * IN assertions: cur_match is the head of the hash chain for the current
1227  *   string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1228  * OUT assertion: the match length is not greater than s->lookahead.
1229  */
1230 #ifndef ASMV
1231 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1232  * match.S. The code will be functionally equivalent.
1233  */
1234 local uInt longest_match(s, cur_match)
1235     deflate_state *s;
1236     IPos cur_match;                             /* current match */
1237 {
1238     unsigned chain_length = s->max_chain_length;/* max hash chain length */
1239     register Bytef *scan = s->window + s->strstart; /* current string */
1240     register Bytef *match;                      /* matched string */
1241     register int len;                           /* length of current match */
1242     int best_len = (int)s->prev_length;         /* best match length so far */
1243     int nice_match = s->nice_match;             /* stop if match long enough */
1244     IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1245         s->strstart - (IPos)MAX_DIST(s) : NIL;
1246     /* Stop when cur_match becomes <= limit. To simplify the code,
1247      * we prevent matches with the string of window index 0.
1248      */
1249     Posf *prev = s->prev;
1250     uInt wmask = s->w_mask;
1251 
1252 #ifdef UNALIGNED_OK
1253     /* Compare two bytes at a time. Note: this is not always beneficial.
1254      * Try with and without -DUNALIGNED_OK to check.
1255      */
1256     register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1257     register ush scan_start = *(ushf*)scan;
1258     register ush scan_end   = *(ushf*)(scan+best_len-1);
1259 #else
1260     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1261     register Byte scan_end1  = scan[best_len-1];
1262     register Byte scan_end   = scan[best_len];
1263 #endif
1264 
1265     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1266      * It is easy to get rid of this optimization if necessary.
1267      */
1268     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1269 
1270     /* Do not waste too much time if we already have a good match: */
1271     if (s->prev_length >= s->good_match) {
1272         chain_length >>= 2;
1273     }
1274     /* Do not look for matches beyond the end of the input. This is necessary
1275      * to make deflate deterministic.
1276      */
1277     if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1278 
1279     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1280 
1281     do {
1282         Assert(cur_match < s->strstart, "no future");
1283         match = s->window + cur_match;
1284 
1285         /* Skip to next match if the match length cannot increase
1286          * or if the match length is less than 2.  Note that the checks below
1287          * for insufficient lookahead only occur occasionally for performance
1288          * reasons.  Therefore uninitialized memory will be accessed, and
1289          * conditional jumps will be made that depend on those values.
1290          * However the length of the match is limited to the lookahead, so
1291          * the output of deflate is not affected by the uninitialized values.
1292          */
1293 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1294         /* This code assumes sizeof(unsigned short) == 2. Do not use
1295          * UNALIGNED_OK if your compiler uses a different size.
1296          */
1297         if (*(ushf*)(match+best_len-1) != scan_end ||
1298             *(ushf*)match != scan_start) continue;
1299 
1300         /* It is not necessary to compare scan[2] and match[2] since they are
1301          * always equal when the other bytes match, given that the hash keys
1302          * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1303          * strstart+3, +5, ... up to strstart+257. We check for insufficient
1304          * lookahead only every 4th comparison; the 128th check will be made
1305          * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1306          * necessary to put more guard bytes at the end of the window, or
1307          * to check more often for insufficient lookahead.
1308          */
1309         Assert(scan[2] == match[2], "scan[2]?");
1310         scan++, match++;
1311         do {
1312         } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1313                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1314                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1315                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1316                  scan < strend);
1317         /* The funny "do {}" generates better code on most compilers */
1318 
1319         /* Here, scan <= window+strstart+257 */
1320         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1321         if (*scan == *match) scan++;
1322 
1323         len = (MAX_MATCH - 1) - (int)(strend-scan);
1324         scan = strend - (MAX_MATCH-1);
1325 
1326 #else /* UNALIGNED_OK */
1327 
1328         if (match[best_len]   != scan_end  ||
1329             match[best_len-1] != scan_end1 ||
1330             *match            != *scan     ||
1331             *++match          != scan[1])      continue;
1332 
1333         /* The check at best_len-1 can be removed because it will be made
1334          * again later. (This heuristic is not always a win.)
1335          * It is not necessary to compare scan[2] and match[2] since they
1336          * are always equal when the other bytes match, given that
1337          * the hash keys are equal and that HASH_BITS >= 8.
1338          */
1339         scan += 2, match++;
1340         Assert(*scan == *match, "match[2]?");
1341 
1342         /* We check for insufficient lookahead only every 8th comparison;
1343          * the 256th check will be made at strstart+258.
1344          */
1345         do {
1346         } while (*++scan == *++match && *++scan == *++match &&
1347                  *++scan == *++match && *++scan == *++match &&
1348                  *++scan == *++match && *++scan == *++match &&
1349                  *++scan == *++match && *++scan == *++match &&
1350                  scan < strend);
1351 
1352         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1353 
1354         len = MAX_MATCH - (int)(strend - scan);
1355         scan = strend - MAX_MATCH;
1356 
1357 #endif /* UNALIGNED_OK */
1358 
1359         if (len > best_len) {
1360             s->match_start = cur_match;
1361             best_len = len;
1362             if (len >= nice_match) break;
1363 #ifdef UNALIGNED_OK
1364             scan_end = *(ushf*)(scan+best_len-1);
1365 #else
1366             scan_end1  = scan[best_len-1];
1367             scan_end   = scan[best_len];
1368 #endif
1369         }
1370     } while ((cur_match = prev[cur_match & wmask]) > limit
1371              && --chain_length != 0);
1372 
1373     if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1374     return s->lookahead;
1375 }
1376 #endif /* ASMV */
1377 
1378 #else /* FASTEST */
1379 
1380 /* ---------------------------------------------------------------------------
1381  * Optimized version for FASTEST only
1382  */
1383 local uInt longest_match(s, cur_match)
1384     deflate_state *s;
1385     IPos cur_match;                             /* current match */
1386 {
1387     register Bytef *scan = s->window + s->strstart; /* current string */
1388     register Bytef *match;                       /* matched string */
1389     register int len;                           /* length of current match */
1390     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1391 
1392     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1393      * It is easy to get rid of this optimization if necessary.
1394      */
1395     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1396 
1397     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1398 
1399     Assert(cur_match < s->strstart, "no future");
1400 
1401     match = s->window + cur_match;
1402 
1403     /* Return failure if the match length is less than 2:
1404      */
1405     if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1406 
1407     /* The check at best_len-1 can be removed because it will be made
1408      * again later. (This heuristic is not always a win.)
1409      * It is not necessary to compare scan[2] and match[2] since they
1410      * are always equal when the other bytes match, given that
1411      * the hash keys are equal and that HASH_BITS >= 8.
1412      */
1413     scan += 2, match += 2;
1414     Assert(*scan == *match, "match[2]?");
1415 
1416     /* We check for insufficient lookahead only every 8th comparison;
1417      * the 256th check will be made at strstart+258.
1418      */
1419     do {
1420     } while (*++scan == *++match && *++scan == *++match &&
1421              *++scan == *++match && *++scan == *++match &&
1422              *++scan == *++match && *++scan == *++match &&
1423              *++scan == *++match && *++scan == *++match &&
1424              scan < strend);
1425 
1426     Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1427 
1428     len = MAX_MATCH - (int)(strend - scan);
1429 
1430     if (len < MIN_MATCH) return MIN_MATCH - 1;
1431 
1432     s->match_start = cur_match;
1433     return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1434 }
1435 
1436 #endif /* FASTEST */
1437 
1438 #ifdef ZLIB_DEBUG
1439 
1440 #define EQUAL 0
1441 /* result of memcmp for equal strings */
1442 
1443 /* ===========================================================================
1444  * Check that the match at match_start is indeed a match.
1445  */
1446 local void check_match(s, start, match, length)
1447     deflate_state *s;
1448     IPos start, match;
1449     int length;
1450 {
1451     /* check that the match is indeed a match */
1452     if (zmemcmp(s->window + match,
1453                 s->window + start, length) != EQUAL) {
1454         fprintf(stderr, " start %u, match %u, length %d\n",
1455                 start, match, length);
1456         do {
1457             fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1458         } while (--length != 0);
1459         z_error("invalid match");
1460     }
1461     if (z_verbose > 1) {
1462         fprintf(stderr,"\\[%d,%d]", start-match, length);
1463         do { putc(s->window[start++], stderr); } while (--length != 0);
1464     }
1465 }
1466 #else
1467 #  define check_match(s, start, match, length)
1468 #endif /* ZLIB_DEBUG */
1469 
1470 /* ===========================================================================
1471  * Fill the window when the lookahead becomes insufficient.
1472  * Updates strstart and lookahead.
1473  *
1474  * IN assertion: lookahead < MIN_LOOKAHEAD
1475  * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1476  *    At least one byte has been read, or avail_in == 0; reads are
1477  *    performed for at least two bytes (required for the zip translate_eol
1478  *    option -- not supported here).
1479  */
1480 local void fill_window(s)
1481     deflate_state *s;
1482 {
1483     unsigned n;
1484     unsigned more;    /* Amount of free space at the end of the window. */
1485     uInt wsize = s->w_size;
1486 
1487     Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1488 
1489     do {
1490         more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1491 
1492         /* Deal with !@#$% 64K limit: */
1493         if (sizeof(int) <= 2) {
1494             if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1495                 more = wsize;
1496 
1497             } else if (more == (unsigned)(-1)) {
1498                 /* Very unlikely, but possible on 16 bit machine if
1499                  * strstart == 0 && lookahead == 1 (input done a byte at time)
1500                  */
1501                 more--;
1502             }
1503         }
1504 
1505         /* If the window is almost full and there is insufficient lookahead,
1506          * move the upper half to the lower one to make room in the upper half.
1507          */
1508         if (s->strstart >= wsize+MAX_DIST(s)) {
1509 
1510             zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1511             s->match_start -= wsize;
1512             s->strstart    -= wsize; /* we now have strstart >= MAX_DIST */
1513             s->block_start -= (long) wsize;
1514             slide_hash(s);
1515             more += wsize;
1516         }
1517         if (s->strm->avail_in == 0) break;
1518 
1519         /* If there was no sliding:
1520          *    strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1521          *    more == window_size - lookahead - strstart
1522          * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1523          * => more >= window_size - 2*WSIZE + 2
1524          * In the BIG_MEM or MMAP case (not yet supported),
1525          *   window_size == input_size + MIN_LOOKAHEAD  &&
1526          *   strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1527          * Otherwise, window_size == 2*WSIZE so more >= 2.
1528          * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1529          */
1530         Assert(more >= 2, "more < 2");
1531 
1532         n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1533         s->lookahead += n;
1534 
1535         /* Initialize the hash value now that we have some input: */
1536         if (s->lookahead + s->insert >= MIN_MATCH) {
1537             uInt str = s->strstart - s->insert;
1538             s->ins_h = s->window[str];
1539             UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1540 #if MIN_MATCH != 3
1541             Call UPDATE_HASH() MIN_MATCH-3 more times
1542 #endif
1543             while (s->insert) {
1544                 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1545 #ifndef FASTEST
1546                 s->prev[str & s->w_mask] = s->head[s->ins_h];
1547 #endif
1548                 s->head[s->ins_h] = (Pos)str;
1549                 str++;
1550                 s->insert--;
1551                 if (s->lookahead + s->insert < MIN_MATCH)
1552                     break;
1553             }
1554         }
1555         /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1556          * but this is not important since only literal bytes will be emitted.
1557          */
1558 
1559     } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1560 
1561     /* If the WIN_INIT bytes after the end of the current data have never been
1562      * written, then zero those bytes in order to avoid memory check reports of
1563      * the use of uninitialized (or uninitialised as Julian writes) bytes by
1564      * the longest match routines.  Update the high water mark for the next
1565      * time through here.  WIN_INIT is set to MAX_MATCH since the longest match
1566      * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1567      */
1568     if (s->high_water < s->window_size) {
1569         ulg curr = s->strstart + (ulg)(s->lookahead);
1570         ulg init;
1571 
1572         if (s->high_water < curr) {
1573             /* Previous high water mark below current data -- zero WIN_INIT
1574              * bytes or up to end of window, whichever is less.
1575              */
1576             init = s->window_size - curr;
1577             if (init > WIN_INIT)
1578                 init = WIN_INIT;
1579             zmemzero(s->window + curr, (unsigned)init);
1580             s->high_water = curr + init;
1581         }
1582         else if (s->high_water < (ulg)curr + WIN_INIT) {
1583             /* High water mark at or above current data, but below current data
1584              * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1585              * to end of window, whichever is less.
1586              */
1587             init = (ulg)curr + WIN_INIT - s->high_water;
1588             if (init > s->window_size - s->high_water)
1589                 init = s->window_size - s->high_water;
1590             zmemzero(s->window + s->high_water, (unsigned)init);
1591             s->high_water += init;
1592         }
1593     }
1594 
1595     Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1596            "not enough room for search");
1597 }
1598 
1599 /* ===========================================================================
1600  * Flush the current block, with given end-of-file flag.
1601  * IN assertion: strstart is set to the end of the current match.
1602  */
1603 #define FLUSH_BLOCK_ONLY(s, last) { \
1604    _tr_flush_block(s, (s->block_start >= 0L ? \
1605                    (charf *)&s->window[(unsigned)s->block_start] : \
1606                    (charf *)Z_NULL), \
1607                 (ulg)((long)s->strstart - s->block_start), \
1608                 (last)); \
1609    s->block_start = s->strstart; \
1610    flush_pending(s->strm); \
1611    Tracev((stderr,"[FLUSH]")); \
1612 }
1613 
1614 /* Same but force premature exit if necessary. */
1615 #define FLUSH_BLOCK(s, last) { \
1616    FLUSH_BLOCK_ONLY(s, last); \
1617    if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1618 }
1619 
1620 /* Maximum stored block length in deflate format (not including header). */
1621 #define MAX_STORED 65535
1622 
1623 /* Minimum of a and b. */
1624 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1625 
1626 /* ===========================================================================
1627  * Copy without compression as much as possible from the input stream, return
1628  * the current block state.
1629  *
1630  * In case deflateParams() is used to later switch to a non-zero compression
1631  * level, s->matches (otherwise unused when storing) keeps track of the number
1632  * of hash table slides to perform. If s->matches is 1, then one hash table
1633  * slide will be done when switching. If s->matches is 2, the maximum value
1634  * allowed here, then the hash table will be cleared, since two or more slides
1635  * is the same as a clear.
1636  *
1637  * deflate_stored() is written to minimize the number of times an input byte is
1638  * copied. It is most efficient with large input and output buffers, which
1639  * maximizes the opportunites to have a single copy from next_in to next_out.
1640  */
1641 local block_state deflate_stored(s, flush)
1642     deflate_state *s;
1643     int flush;
1644 {
1645     /* Smallest worthy block size when not flushing or finishing. By default
1646      * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1647      * large input and output buffers, the stored block size will be larger.
1648      */
1649     unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1650 
1651     /* Copy as many min_block or larger stored blocks directly to next_out as
1652      * possible. If flushing, copy the remaining available input to next_out as
1653      * stored blocks, if there is enough space.
1654      */
1655     unsigned len, left, have, last = 0;
1656     unsigned used = s->strm->avail_in;
1657     do {
1658         /* Set len to the maximum size block that we can copy directly with the
1659          * available input data and output space. Set left to how much of that
1660          * would be copied from what's left in the window.
1661          */
1662         len = MAX_STORED;       /* maximum deflate stored block length */
1663         have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1664         if (s->strm->avail_out < have)          /* need room for header */
1665             break;
1666             /* maximum stored block length that will fit in avail_out: */
1667         have = s->strm->avail_out - have;
1668         left = s->strstart - s->block_start;    /* bytes left in window */
1669         if (len > (ulg)left + s->strm->avail_in)
1670             len = left + s->strm->avail_in;     /* limit len to the input */
1671         if (len > have)
1672             len = have;                         /* limit len to the output */
1673 
1674         /* If the stored block would be less than min_block in length, or if
1675          * unable to copy all of the available input when flushing, then try
1676          * copying to the window and the pending buffer instead. Also don't
1677          * write an empty block when flushing -- deflate() does that.
1678          */
1679         if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1680                                 flush == Z_NO_FLUSH ||
1681                                 len != left + s->strm->avail_in))
1682             break;
1683 
1684         /* Make a dummy stored block in pending to get the header bytes,
1685          * including any pending bits. This also updates the debugging counts.
1686          */
1687         last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1688         _tr_stored_block(s, (char *)0, 0L, last);
1689 
1690         /* Replace the lengths in the dummy stored block with len. */
1691         s->pending_buf[s->pending - 4] = len;
1692         s->pending_buf[s->pending - 3] = len >> 8;
1693         s->pending_buf[s->pending - 2] = ~len;
1694         s->pending_buf[s->pending - 1] = ~len >> 8;
1695 
1696         /* Write the stored block header bytes. */
1697         flush_pending(s->strm);
1698 
1699 #ifdef ZLIB_DEBUG
1700         /* Update debugging counts for the data about to be copied. */
1701         s->compressed_len += len << 3;
1702         s->bits_sent += len << 3;
1703 #endif
1704 
1705         /* Copy uncompressed bytes from the window to next_out. */
1706         if (left) {
1707             if (left > len)
1708                 left = len;
1709             zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1710             s->strm->next_out += left;
1711             s->strm->avail_out -= left;
1712             s->strm->total_out += left;
1713             s->block_start += left;
1714             len -= left;
1715         }
1716 
1717         /* Copy uncompressed bytes directly from next_in to next_out, updating
1718          * the check value.
1719          */
1720         if (len) {
1721             read_buf(s->strm, s->strm->next_out, len);
1722             s->strm->next_out += len;
1723             s->strm->avail_out -= len;
1724             s->strm->total_out += len;
1725         }
1726     } while (last == 0);
1727 
1728     /* Update the sliding window with the last s->w_size bytes of the copied
1729      * data, or append all of the copied data to the existing window if less
1730      * than s->w_size bytes were copied. Also update the number of bytes to
1731      * insert in the hash tables, in the event that deflateParams() switches to
1732      * a non-zero compression level.
1733      */
1734     used -= s->strm->avail_in;      /* number of input bytes directly copied */
1735     if (used) {
1736         /* If any input was used, then no unused input remains in the window,
1737          * therefore s->block_start == s->strstart.
1738          */
1739         if (used >= s->w_size) {    /* supplant the previous history */
1740             s->matches = 2;         /* clear hash */
1741             zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1742             s->strstart = s->w_size;
1743         }
1744         else {
1745             if (s->window_size - s->strstart <= used) {
1746                 /* Slide the window down. */
1747                 s->strstart -= s->w_size;
1748                 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1749                 if (s->matches < 2)
1750                     s->matches++;   /* add a pending slide_hash() */
1751             }
1752             zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1753             s->strstart += used;
1754         }
1755         s->block_start = s->strstart;
1756         s->insert += MIN(used, s->w_size - s->insert);
1757     }
1758     if (s->high_water < s->strstart)
1759         s->high_water = s->strstart;
1760 
1761     /* If the last block was written to next_out, then done. */
1762     if (last)
1763         return finish_done;
1764 
1765     /* If flushing and all input has been consumed, then done. */
1766     if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1767         s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1768         return block_done;
1769 
1770     /* Fill the window with any remaining input. */
1771     have = s->window_size - s->strstart - 1;
1772     if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1773         /* Slide the window down. */
1774         s->block_start -= s->w_size;
1775         s->strstart -= s->w_size;
1776         zmemcpy(s->window, s->window + s->w_size, s->strstart);
1777         if (s->matches < 2)
1778             s->matches++;           /* add a pending slide_hash() */
1779         have += s->w_size;          /* more space now */
1780     }
1781     if (have > s->strm->avail_in)
1782         have = s->strm->avail_in;
1783     if (have) {
1784         read_buf(s->strm, s->window + s->strstart, have);
1785         s->strstart += have;
1786     }
1787     if (s->high_water < s->strstart)
1788         s->high_water = s->strstart;
1789 
1790     /* There was not enough avail_out to write a complete worthy or flushed
1791      * stored block to next_out. Write a stored block to pending instead, if we
1792      * have enough input for a worthy block, or if flushing and there is enough
1793      * room for the remaining input as a stored block in the pending buffer.
1794      */
1795     have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1796         /* maximum stored block length that will fit in pending: */
1797     have = MIN(s->pending_buf_size - have, MAX_STORED);
1798     min_block = MIN(have, s->w_size);
1799     left = s->strstart - s->block_start;
1800     if (left >= min_block ||
1801         ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1802          s->strm->avail_in == 0 && left <= have)) {
1803         len = MIN(left, have);
1804         last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1805                len == left ? 1 : 0;
1806         _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1807         s->block_start += len;
1808         flush_pending(s->strm);
1809     }
1810 
1811     /* We've done all we can with the available input and output. */
1812     return last ? finish_started : need_more;
1813 }
1814 
1815 /* ===========================================================================
1816  * Compress as much as possible from the input stream, return the current
1817  * block state.
1818  * This function does not perform lazy evaluation of matches and inserts
1819  * new strings in the dictionary only for unmatched strings or for short
1820  * matches. It is used only for the fast compression options.
1821  */
1822 local block_state deflate_fast(s, flush)
1823     deflate_state *s;
1824     int flush;
1825 {
1826     IPos hash_head;       /* head of the hash chain */
1827     int bflush;           /* set if current block must be flushed */
1828 
1829     for (;;) {
1830         /* Make sure that we always have enough lookahead, except
1831          * at the end of the input file. We need MAX_MATCH bytes
1832          * for the next match, plus MIN_MATCH bytes to insert the
1833          * string following the next match.
1834          */
1835         if (s->lookahead < MIN_LOOKAHEAD) {
1836             fill_window(s);
1837             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1838                 return need_more;
1839             }
1840             if (s->lookahead == 0) break; /* flush the current block */
1841         }
1842 
1843         /* Insert the string window[strstart .. strstart+2] in the
1844          * dictionary, and set hash_head to the head of the hash chain:
1845          */
1846         hash_head = NIL;
1847         if (s->lookahead >= MIN_MATCH) {
1848             INSERT_STRING(s, s->strstart, hash_head);
1849         }
1850 
1851         /* Find the longest match, discarding those <= prev_length.
1852          * At this point we have always match_length < MIN_MATCH
1853          */
1854         if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1855             /* To simplify the code, we prevent matches with the string
1856              * of window index 0 (in particular we have to avoid a match
1857              * of the string with itself at the start of the input file).
1858              */
1859             s->match_length = longest_match (s, hash_head);
1860             /* longest_match() sets match_start */
1861         }
1862         if (s->match_length >= MIN_MATCH) {
1863             check_match(s, s->strstart, s->match_start, s->match_length);
1864 
1865             _tr_tally_dist(s, s->strstart - s->match_start,
1866                            s->match_length - MIN_MATCH, bflush);
1867 
1868             s->lookahead -= s->match_length;
1869 
1870             /* Insert new strings in the hash table only if the match length
1871              * is not too large. This saves time but degrades compression.
1872              */
1873 #ifndef FASTEST
1874             if (s->match_length <= s->max_insert_length &&
1875                 s->lookahead >= MIN_MATCH) {
1876                 s->match_length--; /* string at strstart already in table */
1877                 do {
1878                     s->strstart++;
1879                     INSERT_STRING(s, s->strstart, hash_head);
1880                     /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1881                      * always MIN_MATCH bytes ahead.
1882                      */
1883                 } while (--s->match_length != 0);
1884                 s->strstart++;
1885             } else
1886 #endif
1887             {
1888                 s->strstart += s->match_length;
1889                 s->match_length = 0;
1890                 s->ins_h = s->window[s->strstart];
1891                 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1892 #if MIN_MATCH != 3
1893                 Call UPDATE_HASH() MIN_MATCH-3 more times
1894 #endif
1895                 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1896                  * matter since it will be recomputed at next deflate call.
1897                  */
1898             }
1899         } else {
1900             /* No match, output a literal byte */
1901             Tracevv((stderr,"%c", s->window[s->strstart]));
1902             _tr_tally_lit (s, s->window[s->strstart], bflush);
1903             s->lookahead--;
1904             s->strstart++;
1905         }
1906         if (bflush) FLUSH_BLOCK(s, 0);
1907     }
1908     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1909     if (flush == Z_FINISH) {
1910         FLUSH_BLOCK(s, 1);
1911         return finish_done;
1912     }
1913     if (s->last_lit)
1914         FLUSH_BLOCK(s, 0);
1915     return block_done;
1916 }
1917 
1918 #ifndef FASTEST
1919 /* ===========================================================================
1920  * Same as above, but achieves better compression. We use a lazy
1921  * evaluation for matches: a match is finally adopted only if there is
1922  * no better match at the next window position.
1923  */
1924 local block_state deflate_slow(s, flush)
1925     deflate_state *s;
1926     int flush;
1927 {
1928     IPos hash_head;          /* head of hash chain */
1929     int bflush;              /* set if current block must be flushed */
1930 
1931     /* Process the input block. */
1932     for (;;) {
1933         /* Make sure that we always have enough lookahead, except
1934          * at the end of the input file. We need MAX_MATCH bytes
1935          * for the next match, plus MIN_MATCH bytes to insert the
1936          * string following the next match.
1937          */
1938         if (s->lookahead < MIN_LOOKAHEAD) {
1939             fill_window(s);
1940             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1941                 return need_more;
1942             }
1943             if (s->lookahead == 0) break; /* flush the current block */
1944         }
1945 
1946         /* Insert the string window[strstart .. strstart+2] in the
1947          * dictionary, and set hash_head to the head of the hash chain:
1948          */
1949         hash_head = NIL;
1950         if (s->lookahead >= MIN_MATCH) {
1951             INSERT_STRING(s, s->strstart, hash_head);
1952         }
1953 
1954         /* Find the longest match, discarding those <= prev_length.
1955          */
1956         s->prev_length = s->match_length, s->prev_match = s->match_start;
1957         s->match_length = MIN_MATCH-1;
1958 
1959         if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1960             s->strstart - hash_head <= MAX_DIST(s)) {
1961             /* To simplify the code, we prevent matches with the string
1962              * of window index 0 (in particular we have to avoid a match
1963              * of the string with itself at the start of the input file).
1964              */
1965             s->match_length = longest_match (s, hash_head);
1966             /* longest_match() sets match_start */
1967 
1968             if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1969 #if TOO_FAR <= 32767
1970                 || (s->match_length == MIN_MATCH &&
1971                     s->strstart - s->match_start > TOO_FAR)
1972 #endif
1973                 )) {
1974 
1975                 /* If prev_match is also MIN_MATCH, match_start is garbage
1976                  * but we will ignore the current match anyway.
1977                  */
1978                 s->match_length = MIN_MATCH-1;
1979             }
1980         }
1981         /* If there was a match at the previous step and the current
1982          * match is not better, output the previous match:
1983          */
1984         if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1985             uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1986             /* Do not insert strings in hash table beyond this. */
1987 
1988             check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1989 
1990             _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1991                            s->prev_length - MIN_MATCH, bflush);
1992 
1993             /* Insert in hash table all strings up to the end of the match.
1994              * strstart-1 and strstart are already inserted. If there is not
1995              * enough lookahead, the last two strings are not inserted in
1996              * the hash table.
1997              */
1998             s->lookahead -= s->prev_length-1;
1999             s->prev_length -= 2;
2000             do {
2001                 if (++s->strstart <= max_insert) {
2002                     INSERT_STRING(s, s->strstart, hash_head);
2003                 }
2004             } while (--s->prev_length != 0);
2005             s->match_available = 0;
2006             s->match_length = MIN_MATCH-1;
2007             s->strstart++;
2008 
2009             if (bflush) FLUSH_BLOCK(s, 0);
2010 
2011         } else if (s->match_available) {
2012             /* If there was no match at the previous position, output a
2013              * single literal. If there was a match but the current match
2014              * is longer, truncate the previous match to a single literal.
2015              */
2016             Tracevv((stderr,"%c", s->window[s->strstart-1]));
2017             _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2018             if (bflush) {
2019                 FLUSH_BLOCK_ONLY(s, 0);
2020             }
2021             s->strstart++;
2022             s->lookahead--;
2023             if (s->strm->avail_out == 0) return need_more;
2024         } else {
2025             /* There is no previous match to compare with, wait for
2026              * the next step to decide.
2027              */
2028             s->match_available = 1;
2029             s->strstart++;
2030             s->lookahead--;
2031         }
2032     }
2033     Assert (flush != Z_NO_FLUSH, "no flush?");
2034     if (s->match_available) {
2035         Tracevv((stderr,"%c", s->window[s->strstart-1]));
2036         _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2037         s->match_available = 0;
2038     }
2039     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2040     if (flush == Z_FINISH) {
2041         FLUSH_BLOCK(s, 1);
2042         return finish_done;
2043     }
2044     if (s->last_lit)
2045         FLUSH_BLOCK(s, 0);
2046     return block_done;
2047 }
2048 #endif /* FASTEST */
2049 
2050 /* ===========================================================================
2051  * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2052  * one.  Do not maintain a hash table.  (It will be regenerated if this run of
2053  * deflate switches away from Z_RLE.)
2054  */
2055 local block_state deflate_rle(s, flush)
2056     deflate_state *s;
2057     int flush;
2058 {
2059     int bflush;             /* set if current block must be flushed */
2060     uInt prev;              /* byte at distance one to match */
2061     Bytef *scan, *strend;   /* scan goes up to strend for length of run */
2062 
2063     for (;;) {
2064         /* Make sure that we always have enough lookahead, except
2065          * at the end of the input file. We need MAX_MATCH bytes
2066          * for the longest run, plus one for the unrolled loop.
2067          */
2068         if (s->lookahead <= MAX_MATCH) {
2069             fill_window(s);
2070             if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2071                 return need_more;
2072             }
2073             if (s->lookahead == 0) break; /* flush the current block */
2074         }
2075 
2076         /* See how many times the previous byte repeats */
2077         s->match_length = 0;
2078         if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2079             scan = s->window + s->strstart - 1;
2080             prev = *scan;
2081             if (prev == *++scan && prev == *++scan && prev == *++scan) {
2082                 strend = s->window + s->strstart + MAX_MATCH;
2083                 do {
2084                 } while (prev == *++scan && prev == *++scan &&
2085                          prev == *++scan && prev == *++scan &&
2086                          prev == *++scan && prev == *++scan &&
2087                          prev == *++scan && prev == *++scan &&
2088                          scan < strend);
2089                 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2090                 if (s->match_length > s->lookahead)
2091                     s->match_length = s->lookahead;
2092             }
2093             Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2094         }
2095 
2096         /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2097         if (s->match_length >= MIN_MATCH) {
2098             check_match(s, s->strstart, s->strstart - 1, s->match_length);
2099 
2100             _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2101 
2102             s->lookahead -= s->match_length;
2103             s->strstart += s->match_length;
2104             s->match_length = 0;
2105         } else {
2106             /* No match, output a literal byte */
2107             Tracevv((stderr,"%c", s->window[s->strstart]));
2108             _tr_tally_lit (s, s->window[s->strstart], bflush);
2109             s->lookahead--;
2110             s->strstart++;
2111         }
2112         if (bflush) FLUSH_BLOCK(s, 0);
2113     }
2114     s->insert = 0;
2115     if (flush == Z_FINISH) {
2116         FLUSH_BLOCK(s, 1);
2117         return finish_done;
2118     }
2119     if (s->last_lit)
2120         FLUSH_BLOCK(s, 0);
2121     return block_done;
2122 }
2123 
2124 /* ===========================================================================
2125  * For Z_HUFFMAN_ONLY, do not look for matches.  Do not maintain a hash table.
2126  * (It will be regenerated if this run of deflate switches away from Huffman.)
2127  */
2128 local block_state deflate_huff(s, flush)
2129     deflate_state *s;
2130     int flush;
2131 {
2132     int bflush;             /* set if current block must be flushed */
2133 
2134     for (;;) {
2135         /* Make sure that we have a literal to write. */
2136         if (s->lookahead == 0) {
2137             fill_window(s);
2138             if (s->lookahead == 0) {
2139                 if (flush == Z_NO_FLUSH)
2140                     return need_more;
2141                 break;      /* flush the current block */
2142             }
2143         }
2144 
2145         /* Output a literal byte */
2146         s->match_length = 0;
2147         Tracevv((stderr,"%c", s->window[s->strstart]));
2148         _tr_tally_lit (s, s->window[s->strstart], bflush);
2149         s->lookahead--;
2150         s->strstart++;
2151         if (bflush) FLUSH_BLOCK(s, 0);
2152     }
2153     s->insert = 0;
2154     if (flush == Z_FINISH) {
2155         FLUSH_BLOCK(s, 1);
2156         return finish_done;
2157     }
2158     if (s->last_lit)
2159         FLUSH_BLOCK(s, 0);
2160     return block_done;
2161 }
2162