xref: /illumos-gate/usr/src/contrib/zlib/deflate.c (revision 1fa2a66491e7d8ae0be84e7da4da8e812480c710)
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     do { \
192         s->head[s->hash_size-1] = NIL; \
193         zmemzero((Bytef *)s->head, \
194         (unsigned)(s->hash_size-1)*sizeof(*s->head)); \
195     } while (0)
196 
197 /* ===========================================================================
198  * Slide the hash table when sliding the window down (could be avoided with 32
199  * bit values at the expense of memory usage). We slide even when level == 0 to
200  * keep the hash table consistent if we switch back to level > 0 later.
201  */
202 local void slide_hash(s)
203     deflate_state *s;
204 {
205     unsigned n, m;
206     Posf *p;
207     uInt wsize = s->w_size;
208 
209     n = s->hash_size;
210     p = &s->head[n];
211     do {
212         m = *--p;
213         *p = (Pos)(m >= wsize ? m - wsize : NIL);
214     } while (--n);
215     n = wsize;
216 #ifndef FASTEST
217     p = &s->prev[n];
218     do {
219         m = *--p;
220         *p = (Pos)(m >= wsize ? m - wsize : NIL);
221         /* If n is not on any hash chain, prev[n] is garbage but
222          * its value will never be used.
223          */
224     } while (--n);
225 #endif
226 }
227 
228 /* ========================================================================= */
229 int ZEXPORT deflateInit_(strm, level, version, stream_size)
230     z_streamp strm;
231     int level;
232     const char *version;
233     int stream_size;
234 {
235     return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
236                          Z_DEFAULT_STRATEGY, version, stream_size);
237     /* To do: ignore strm->next_in if we use it as window */
238 }
239 
240 /* ========================================================================= */
241 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
242                   version, stream_size)
243     z_streamp strm;
244     int  level;
245     int  method;
246     int  windowBits;
247     int  memLevel;
248     int  strategy;
249     const char *version;
250     int stream_size;
251 {
252     deflate_state *s;
253     int wrap = 1;
254     static const char my_version[] = ZLIB_VERSION;
255 
256     ushf *overlay;
257     /* We overlay pending_buf and d_buf+l_buf. This works since the average
258      * output size for (length,distance) codes is <= 24 bits.
259      */
260 
261     if (version == Z_NULL || version[0] != my_version[0] ||
262         stream_size != sizeof(z_stream)) {
263         return Z_VERSION_ERROR;
264     }
265     if (strm == Z_NULL) return Z_STREAM_ERROR;
266 
267     strm->msg = Z_NULL;
268     if (strm->zalloc == (alloc_func)0) {
269 #ifdef Z_SOLO
270         return Z_STREAM_ERROR;
271 #else
272         strm->zalloc = zcalloc;
273         strm->opaque = (voidpf)0;
274 #endif
275     }
276     if (strm->zfree == (free_func)0)
277 #ifdef Z_SOLO
278         return Z_STREAM_ERROR;
279 #else
280         strm->zfree = zcfree;
281 #endif
282 
283 #ifdef FASTEST
284     if (level != 0) level = 1;
285 #else
286     if (level == Z_DEFAULT_COMPRESSION) level = 6;
287 #endif
288 
289     if (windowBits < 0) { /* suppress zlib wrapper */
290         wrap = 0;
291         windowBits = -windowBits;
292     }
293 #ifdef GZIP
294     else if (windowBits > 15) {
295         wrap = 2;       /* write gzip wrapper instead */
296         windowBits -= 16;
297     }
298 #endif
299     if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
300         windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
301         strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
302         return Z_STREAM_ERROR;
303     }
304     if (windowBits == 8) windowBits = 9;  /* until 256-byte window bug fixed */
305     s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
306     if (s == Z_NULL) return Z_MEM_ERROR;
307     strm->state = (struct internal_state FAR *)s;
308     s->strm = strm;
309     s->status = INIT_STATE;     /* to pass state test in deflateReset() */
310 
311     s->wrap = wrap;
312     s->gzhead = Z_NULL;
313     s->w_bits = (uInt)windowBits;
314     s->w_size = 1 << s->w_bits;
315     s->w_mask = s->w_size - 1;
316 
317     s->hash_bits = (uInt)memLevel + 7;
318     s->hash_size = 1 << s->hash_bits;
319     s->hash_mask = s->hash_size - 1;
320     s->hash_shift =  ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
321 
322     s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
323     s->prev   = (Posf *)  ZALLOC(strm, s->w_size, sizeof(Pos));
324     s->head   = (Posf *)  ZALLOC(strm, s->hash_size, sizeof(Pos));
325 
326     s->high_water = 0;      /* nothing written to s->window yet */
327 
328     s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
329 
330     overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
331     s->pending_buf = (uchf *) overlay;
332     s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
333 
334     if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
335         s->pending_buf == Z_NULL) {
336         s->status = FINISH_STATE;
337         strm->msg = ERR_MSG(Z_MEM_ERROR);
338         deflateEnd (strm);
339         return Z_MEM_ERROR;
340     }
341     s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
342     s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
343 
344     s->level = level;
345     s->strategy = strategy;
346     s->method = (Byte)method;
347 
348     return deflateReset(strm);
349 }
350 
351 /* =========================================================================
352  * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
353  */
354 local int deflateStateCheck (strm)
355     z_streamp strm;
356 {
357     deflate_state *s;
358     if (strm == Z_NULL ||
359         strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
360         return 1;
361     s = strm->state;
362     if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
363 #ifdef GZIP
364                                            s->status != GZIP_STATE &&
365 #endif
366                                            s->status != EXTRA_STATE &&
367                                            s->status != NAME_STATE &&
368                                            s->status != COMMENT_STATE &&
369                                            s->status != HCRC_STATE &&
370                                            s->status != BUSY_STATE &&
371                                            s->status != FINISH_STATE))
372         return 1;
373     return 0;
374 }
375 
376 /* ========================================================================= */
377 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
378     z_streamp strm;
379     const Bytef *dictionary;
380     uInt  dictLength;
381 {
382     deflate_state *s;
383     uInt str, n;
384     int wrap;
385     unsigned avail;
386     z_const unsigned char *next;
387 
388     if (deflateStateCheck(strm) || dictionary == Z_NULL)
389         return Z_STREAM_ERROR;
390     s = strm->state;
391     wrap = s->wrap;
392     if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
393         return Z_STREAM_ERROR;
394 
395     /* when using zlib wrappers, compute Adler-32 for provided dictionary */
396     if (wrap == 1)
397         strm->adler = adler32(strm->adler, dictionary, dictLength);
398     s->wrap = 0;                    /* avoid computing Adler-32 in read_buf */
399 
400     /* if dictionary would fill window, just replace the history */
401     if (dictLength >= s->w_size) {
402         if (wrap == 0) {            /* already empty otherwise */
403             CLEAR_HASH(s);
404             s->strstart = 0;
405             s->block_start = 0L;
406             s->insert = 0;
407         }
408         dictionary += dictLength - s->w_size;  /* use the tail */
409         dictLength = s->w_size;
410     }
411 
412     /* insert dictionary into window and hash */
413     avail = strm->avail_in;
414     next = strm->next_in;
415     strm->avail_in = dictLength;
416     strm->next_in = (z_const Bytef *)dictionary;
417     fill_window(s);
418     while (s->lookahead >= MIN_MATCH) {
419         str = s->strstart;
420         n = s->lookahead - (MIN_MATCH-1);
421         do {
422             UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
423 #ifndef FASTEST
424             s->prev[str & s->w_mask] = s->head[s->ins_h];
425 #endif
426             s->head[s->ins_h] = (Pos)str;
427             str++;
428         } while (--n);
429         s->strstart = str;
430         s->lookahead = MIN_MATCH-1;
431         fill_window(s);
432     }
433     s->strstart += s->lookahead;
434     s->block_start = (long)s->strstart;
435     s->insert = s->lookahead;
436     s->lookahead = 0;
437     s->match_length = s->prev_length = MIN_MATCH-1;
438     s->match_available = 0;
439     strm->next_in = next;
440     strm->avail_in = avail;
441     s->wrap = wrap;
442     return Z_OK;
443 }
444 
445 /* ========================================================================= */
446 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
447     z_streamp strm;
448     Bytef *dictionary;
449     uInt  *dictLength;
450 {
451     deflate_state *s;
452     uInt len;
453 
454     if (deflateStateCheck(strm))
455         return Z_STREAM_ERROR;
456     s = strm->state;
457     len = s->strstart + s->lookahead;
458     if (len > s->w_size)
459         len = s->w_size;
460     if (dictionary != Z_NULL && len)
461         zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
462     if (dictLength != Z_NULL)
463         *dictLength = len;
464     return Z_OK;
465 }
466 
467 /* ========================================================================= */
468 int ZEXPORT deflateResetKeep (strm)
469     z_streamp strm;
470 {
471     deflate_state *s;
472 
473     if (deflateStateCheck(strm)) {
474         return Z_STREAM_ERROR;
475     }
476 
477     strm->total_in = strm->total_out = 0;
478     strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
479     strm->data_type = Z_UNKNOWN;
480 
481     s = (deflate_state *)strm->state;
482     s->pending = 0;
483     s->pending_out = s->pending_buf;
484 
485     if (s->wrap < 0) {
486         s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
487     }
488     s->status =
489 #ifdef GZIP
490         s->wrap == 2 ? GZIP_STATE :
491 #endif
492         s->wrap ? INIT_STATE : BUSY_STATE;
493     strm->adler =
494 #ifdef GZIP
495         s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
496 #endif
497         adler32(0L, Z_NULL, 0);
498     s->last_flush = Z_NO_FLUSH;
499 
500     _tr_init(s);
501 
502     return Z_OK;
503 }
504 
505 /* ========================================================================= */
506 int ZEXPORT deflateReset (strm)
507     z_streamp strm;
508 {
509     int ret;
510 
511     ret = deflateResetKeep(strm);
512     if (ret == Z_OK)
513         lm_init(strm->state);
514     return ret;
515 }
516 
517 /* ========================================================================= */
518 int ZEXPORT deflateSetHeader (strm, head)
519     z_streamp strm;
520     gz_headerp head;
521 {
522     if (deflateStateCheck(strm) || strm->state->wrap != 2)
523         return Z_STREAM_ERROR;
524     strm->state->gzhead = head;
525     return Z_OK;
526 }
527 
528 /* ========================================================================= */
529 int ZEXPORT deflatePending (strm, pending, bits)
530     unsigned *pending;
531     int *bits;
532     z_streamp strm;
533 {
534     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
535     if (pending != Z_NULL)
536         *pending = strm->state->pending;
537     if (bits != Z_NULL)
538         *bits = strm->state->bi_valid;
539     return Z_OK;
540 }
541 
542 /* ========================================================================= */
543 int ZEXPORT deflatePrime (strm, bits, value)
544     z_streamp strm;
545     int bits;
546     int value;
547 {
548     deflate_state *s;
549     int put;
550 
551     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
552     s = strm->state;
553     if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
554         return Z_BUF_ERROR;
555     do {
556         put = Buf_size - s->bi_valid;
557         if (put > bits)
558             put = bits;
559         s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
560         s->bi_valid += put;
561         _tr_flush_bits(s);
562         value >>= put;
563         bits -= put;
564     } while (bits);
565     return Z_OK;
566 }
567 
568 /* ========================================================================= */
569 int ZEXPORT deflateParams(strm, level, strategy)
570     z_streamp strm;
571     int level;
572     int strategy;
573 {
574     deflate_state *s;
575     compress_func func;
576 
577     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
578     s = strm->state;
579 
580 #ifdef FASTEST
581     if (level != 0) level = 1;
582 #else
583     if (level == Z_DEFAULT_COMPRESSION) level = 6;
584 #endif
585     if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
586         return Z_STREAM_ERROR;
587     }
588     func = configuration_table[s->level].func;
589 
590     if ((strategy != s->strategy || func != configuration_table[level].func) &&
591         s->high_water) {
592         /* Flush the last buffer: */
593         int err = deflate(strm, Z_BLOCK);
594         if (err == Z_STREAM_ERROR)
595             return err;
596         if (strm->avail_out == 0)
597             return Z_BUF_ERROR;
598     }
599     if (s->level != level) {
600         if (s->level == 0 && s->matches != 0) {
601             if (s->matches == 1)
602                 slide_hash(s);
603             else
604                 CLEAR_HASH(s);
605             s->matches = 0;
606         }
607         s->level = level;
608         s->max_lazy_match   = configuration_table[level].max_lazy;
609         s->good_match       = configuration_table[level].good_length;
610         s->nice_match       = configuration_table[level].nice_length;
611         s->max_chain_length = configuration_table[level].max_chain;
612     }
613     s->strategy = strategy;
614     return Z_OK;
615 }
616 
617 /* ========================================================================= */
618 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
619     z_streamp strm;
620     int good_length;
621     int max_lazy;
622     int nice_length;
623     int max_chain;
624 {
625     deflate_state *s;
626 
627     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
628     s = strm->state;
629     s->good_match = (uInt)good_length;
630     s->max_lazy_match = (uInt)max_lazy;
631     s->nice_match = nice_length;
632     s->max_chain_length = (uInt)max_chain;
633     return Z_OK;
634 }
635 
636 /* =========================================================================
637  * For the default windowBits of 15 and memLevel of 8, this function returns
638  * a close to exact, as well as small, upper bound on the compressed size.
639  * They are coded as constants here for a reason--if the #define's are
640  * changed, then this function needs to be changed as well.  The return
641  * value for 15 and 8 only works for those exact settings.
642  *
643  * For any setting other than those defaults for windowBits and memLevel,
644  * the value returned is a conservative worst case for the maximum expansion
645  * resulting from using fixed blocks instead of stored blocks, which deflate
646  * can emit on compressed data for some combinations of the parameters.
647  *
648  * This function could be more sophisticated to provide closer upper bounds for
649  * every combination of windowBits and memLevel.  But even the conservative
650  * upper bound of about 14% expansion does not seem onerous for output buffer
651  * allocation.
652  */
653 uLong ZEXPORT deflateBound(strm, sourceLen)
654     z_streamp strm;
655     uLong sourceLen;
656 {
657     deflate_state *s;
658     uLong complen, wraplen;
659 
660     /* conservative upper bound for compressed data */
661     complen = sourceLen +
662               ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
663 
664     /* if can't get parameters, return conservative bound plus zlib wrapper */
665     if (deflateStateCheck(strm))
666         return complen + 6;
667 
668     /* compute wrapper length */
669     s = strm->state;
670     switch (s->wrap) {
671     case 0:                                 /* raw deflate */
672         wraplen = 0;
673         break;
674     case 1:                                 /* zlib wrapper */
675         wraplen = 6 + (s->strstart ? 4 : 0);
676         break;
677 #ifdef GZIP
678     case 2:                                 /* gzip wrapper */
679         wraplen = 18;
680         if (s->gzhead != Z_NULL) {          /* user-supplied gzip header */
681             Bytef *str;
682             if (s->gzhead->extra != Z_NULL)
683                 wraplen += 2 + s->gzhead->extra_len;
684             str = s->gzhead->name;
685             if (str != Z_NULL)
686                 do {
687                     wraplen++;
688                 } while (*str++);
689             str = s->gzhead->comment;
690             if (str != Z_NULL)
691                 do {
692                     wraplen++;
693                 } while (*str++);
694             if (s->gzhead->hcrc)
695                 wraplen += 2;
696         }
697         break;
698 #endif
699     default:                                /* for compiler happiness */
700         wraplen = 6;
701     }
702 
703     /* if not default parameters, return conservative bound */
704     if (s->w_bits != 15 || s->hash_bits != 8 + 7)
705         return complen + wraplen;
706 
707     /* default settings: return tight bound for that case */
708     return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
709            (sourceLen >> 25) + 13 - 6 + wraplen;
710 }
711 
712 /* =========================================================================
713  * Put a short in the pending buffer. The 16-bit value is put in MSB order.
714  * IN assertion: the stream state is correct and there is enough room in
715  * pending_buf.
716  */
717 local void putShortMSB (s, b)
718     deflate_state *s;
719     uInt b;
720 {
721     put_byte(s, (Byte)(b >> 8));
722     put_byte(s, (Byte)(b & 0xff));
723 }
724 
725 /* =========================================================================
726  * Flush as much pending output as possible. All deflate() output, except for
727  * some deflate_stored() output, goes through this function so some
728  * applications may wish to modify it to avoid allocating a large
729  * strm->next_out buffer and copying into it. (See also read_buf()).
730  */
731 local void flush_pending(strm)
732     z_streamp strm;
733 {
734     unsigned len;
735     deflate_state *s = strm->state;
736 
737     _tr_flush_bits(s);
738     len = s->pending;
739     if (len > strm->avail_out) len = strm->avail_out;
740     if (len == 0) return;
741 
742     zmemcpy(strm->next_out, s->pending_out, len);
743     strm->next_out  += len;
744     s->pending_out  += len;
745     strm->total_out += len;
746     strm->avail_out -= len;
747     s->pending      -= len;
748     if (s->pending == 0) {
749         s->pending_out = s->pending_buf;
750     }
751 }
752 
753 /* ===========================================================================
754  * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
755  */
756 #define HCRC_UPDATE(beg) \
757     do { \
758         if (s->gzhead->hcrc && s->pending > (beg)) \
759             strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
760                                 s->pending - (beg)); \
761     } while (0)
762 
763 /* ========================================================================= */
764 int ZEXPORT deflate (strm, flush)
765     z_streamp strm;
766     int flush;
767 {
768     int old_flush; /* value of flush param for previous deflate call */
769     deflate_state *s;
770 
771     if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
772         return Z_STREAM_ERROR;
773     }
774     s = strm->state;
775 
776     if (strm->next_out == Z_NULL ||
777         (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
778         (s->status == FINISH_STATE && flush != Z_FINISH)) {
779         ERR_RETURN(strm, Z_STREAM_ERROR);
780     }
781     if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
782 
783     old_flush = s->last_flush;
784     s->last_flush = flush;
785 
786     /* Flush as much pending output as possible */
787     if (s->pending != 0) {
788         flush_pending(strm);
789         if (strm->avail_out == 0) {
790             /* Since avail_out is 0, deflate will be called again with
791              * more output space, but possibly with both pending and
792              * avail_in equal to zero. There won't be anything to do,
793              * but this is not an error situation so make sure we
794              * return OK instead of BUF_ERROR at next call of deflate:
795              */
796             s->last_flush = -1;
797             return Z_OK;
798         }
799 
800     /* Make sure there is something to do and avoid duplicate consecutive
801      * flushes. For repeated and useless calls with Z_FINISH, we keep
802      * returning Z_STREAM_END instead of Z_BUF_ERROR.
803      */
804     } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
805                flush != Z_FINISH) {
806         ERR_RETURN(strm, Z_BUF_ERROR);
807     }
808 
809     /* User must not provide more input after the first FINISH: */
810     if (s->status == FINISH_STATE && strm->avail_in != 0) {
811         ERR_RETURN(strm, Z_BUF_ERROR);
812     }
813 
814     /* Write the header */
815     if (s->status == INIT_STATE) {
816         /* zlib header */
817         uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
818         uInt level_flags;
819 
820         if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
821             level_flags = 0;
822         else if (s->level < 6)
823             level_flags = 1;
824         else if (s->level == 6)
825             level_flags = 2;
826         else
827             level_flags = 3;
828         header |= (level_flags << 6);
829         if (s->strstart != 0) header |= PRESET_DICT;
830         header += 31 - (header % 31);
831 
832         putShortMSB(s, header);
833 
834         /* Save the adler32 of the preset dictionary: */
835         if (s->strstart != 0) {
836             putShortMSB(s, (uInt)(strm->adler >> 16));
837             putShortMSB(s, (uInt)(strm->adler & 0xffff));
838         }
839         strm->adler = adler32(0L, Z_NULL, 0);
840         s->status = BUSY_STATE;
841 
842         /* Compression must start with an empty pending buffer */
843         flush_pending(strm);
844         if (s->pending != 0) {
845             s->last_flush = -1;
846             return Z_OK;
847         }
848     }
849 #ifdef GZIP
850     if (s->status == GZIP_STATE) {
851         /* gzip header */
852         strm->adler = crc32(0L, Z_NULL, 0);
853         put_byte(s, 31);
854         put_byte(s, 139);
855         put_byte(s, 8);
856         if (s->gzhead == Z_NULL) {
857             put_byte(s, 0);
858             put_byte(s, 0);
859             put_byte(s, 0);
860             put_byte(s, 0);
861             put_byte(s, 0);
862             put_byte(s, s->level == 9 ? 2 :
863                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
864                       4 : 0));
865             put_byte(s, OS_CODE);
866             s->status = BUSY_STATE;
867 
868             /* Compression must start with an empty pending buffer */
869             flush_pending(strm);
870             if (s->pending != 0) {
871                 s->last_flush = -1;
872                 return Z_OK;
873             }
874         }
875         else {
876             put_byte(s, (s->gzhead->text ? 1 : 0) +
877                      (s->gzhead->hcrc ? 2 : 0) +
878                      (s->gzhead->extra == Z_NULL ? 0 : 4) +
879                      (s->gzhead->name == Z_NULL ? 0 : 8) +
880                      (s->gzhead->comment == Z_NULL ? 0 : 16)
881                      );
882             put_byte(s, (Byte)(s->gzhead->time & 0xff));
883             put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
884             put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
885             put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
886             put_byte(s, s->level == 9 ? 2 :
887                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
888                       4 : 0));
889             put_byte(s, s->gzhead->os & 0xff);
890             if (s->gzhead->extra != Z_NULL) {
891                 put_byte(s, s->gzhead->extra_len & 0xff);
892                 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
893             }
894             if (s->gzhead->hcrc)
895                 strm->adler = crc32(strm->adler, s->pending_buf,
896                                     s->pending);
897             s->gzindex = 0;
898             s->status = EXTRA_STATE;
899         }
900     }
901     if (s->status == EXTRA_STATE) {
902         if (s->gzhead->extra != Z_NULL) {
903             ulg beg = s->pending;   /* start of bytes to update crc */
904             uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
905             while (s->pending + left > s->pending_buf_size) {
906                 uInt copy = s->pending_buf_size - s->pending;
907                 zmemcpy(s->pending_buf + s->pending,
908                         s->gzhead->extra + s->gzindex, copy);
909                 s->pending = s->pending_buf_size;
910                 HCRC_UPDATE(beg);
911                 s->gzindex += copy;
912                 flush_pending(strm);
913                 if (s->pending != 0) {
914                     s->last_flush = -1;
915                     return Z_OK;
916                 }
917                 beg = 0;
918                 left -= copy;
919             }
920             zmemcpy(s->pending_buf + s->pending,
921                     s->gzhead->extra + s->gzindex, left);
922             s->pending += left;
923             HCRC_UPDATE(beg);
924             s->gzindex = 0;
925         }
926         s->status = NAME_STATE;
927     }
928     if (s->status == NAME_STATE) {
929         if (s->gzhead->name != Z_NULL) {
930             ulg beg = s->pending;   /* start of bytes to update crc */
931             int val;
932             do {
933                 if (s->pending == s->pending_buf_size) {
934                     HCRC_UPDATE(beg);
935                     flush_pending(strm);
936                     if (s->pending != 0) {
937                         s->last_flush = -1;
938                         return Z_OK;
939                     }
940                     beg = 0;
941                 }
942                 val = s->gzhead->name[s->gzindex++];
943                 put_byte(s, val);
944             } while (val != 0);
945             HCRC_UPDATE(beg);
946             s->gzindex = 0;
947         }
948         s->status = COMMENT_STATE;
949     }
950     if (s->status == COMMENT_STATE) {
951         if (s->gzhead->comment != Z_NULL) {
952             ulg beg = s->pending;   /* start of bytes to update crc */
953             int val;
954             do {
955                 if (s->pending == s->pending_buf_size) {
956                     HCRC_UPDATE(beg);
957                     flush_pending(strm);
958                     if (s->pending != 0) {
959                         s->last_flush = -1;
960                         return Z_OK;
961                     }
962                     beg = 0;
963                 }
964                 val = s->gzhead->comment[s->gzindex++];
965                 put_byte(s, val);
966             } while (val != 0);
967             HCRC_UPDATE(beg);
968         }
969         s->status = HCRC_STATE;
970     }
971     if (s->status == HCRC_STATE) {
972         if (s->gzhead->hcrc) {
973             if (s->pending + 2 > s->pending_buf_size) {
974                 flush_pending(strm);
975                 if (s->pending != 0) {
976                     s->last_flush = -1;
977                     return Z_OK;
978                 }
979             }
980             put_byte(s, (Byte)(strm->adler & 0xff));
981             put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
982             strm->adler = crc32(0L, Z_NULL, 0);
983         }
984         s->status = BUSY_STATE;
985 
986         /* Compression must start with an empty pending buffer */
987         flush_pending(strm);
988         if (s->pending != 0) {
989             s->last_flush = -1;
990             return Z_OK;
991         }
992     }
993 #endif
994 
995     /* Start a new block or continue the current one.
996      */
997     if (strm->avail_in != 0 || s->lookahead != 0 ||
998         (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
999         block_state bstate;
1000 
1001         bstate = s->level == 0 ? deflate_stored(s, flush) :
1002                  s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1003                  s->strategy == Z_RLE ? deflate_rle(s, flush) :
1004                  (*(configuration_table[s->level].func))(s, flush);
1005 
1006         if (bstate == finish_started || bstate == finish_done) {
1007             s->status = FINISH_STATE;
1008         }
1009         if (bstate == need_more || bstate == finish_started) {
1010             if (strm->avail_out == 0) {
1011                 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1012             }
1013             return Z_OK;
1014             /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1015              * of deflate should use the same flush parameter to make sure
1016              * that the flush is complete. So we don't have to output an
1017              * empty block here, this will be done at next call. This also
1018              * ensures that for a very small output buffer, we emit at most
1019              * one empty block.
1020              */
1021         }
1022         if (bstate == block_done) {
1023             if (flush == Z_PARTIAL_FLUSH) {
1024                 _tr_align(s);
1025             } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1026                 _tr_stored_block(s, (char*)0, 0L, 0);
1027                 /* For a full flush, this empty block will be recognized
1028                  * as a special marker by inflate_sync().
1029                  */
1030                 if (flush == Z_FULL_FLUSH) {
1031                     CLEAR_HASH(s);             /* forget history */
1032                     if (s->lookahead == 0) {
1033                         s->strstart = 0;
1034                         s->block_start = 0L;
1035                         s->insert = 0;
1036                     }
1037                 }
1038             }
1039             flush_pending(strm);
1040             if (strm->avail_out == 0) {
1041               s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1042               return Z_OK;
1043             }
1044         }
1045     }
1046 
1047     if (flush != Z_FINISH) return Z_OK;
1048     if (s->wrap <= 0) return Z_STREAM_END;
1049 
1050     /* Write the trailer */
1051 #ifdef GZIP
1052     if (s->wrap == 2) {
1053         put_byte(s, (Byte)(strm->adler & 0xff));
1054         put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1055         put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1056         put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1057         put_byte(s, (Byte)(strm->total_in & 0xff));
1058         put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1059         put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1060         put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1061     }
1062     else
1063 #endif
1064     {
1065         putShortMSB(s, (uInt)(strm->adler >> 16));
1066         putShortMSB(s, (uInt)(strm->adler & 0xffff));
1067     }
1068     flush_pending(strm);
1069     /* If avail_out is zero, the application will call deflate again
1070      * to flush the rest.
1071      */
1072     if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1073     return s->pending != 0 ? Z_OK : Z_STREAM_END;
1074 }
1075 
1076 /* ========================================================================= */
1077 int ZEXPORT deflateEnd (strm)
1078     z_streamp strm;
1079 {
1080     int status;
1081 
1082     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1083 
1084     status = strm->state->status;
1085 
1086     /* Deallocate in reverse order of allocations: */
1087     TRY_FREE(strm, strm->state->pending_buf);
1088     TRY_FREE(strm, strm->state->head);
1089     TRY_FREE(strm, strm->state->prev);
1090     TRY_FREE(strm, strm->state->window);
1091 
1092     ZFREE(strm, strm->state);
1093     strm->state = Z_NULL;
1094 
1095     return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1096 }
1097 
1098 /* =========================================================================
1099  * Copy the source state to the destination state.
1100  * To simplify the source, this is not supported for 16-bit MSDOS (which
1101  * doesn't have enough memory anyway to duplicate compression states).
1102  */
1103 int ZEXPORT deflateCopy (dest, source)
1104     z_streamp dest;
1105     z_streamp source;
1106 {
1107 #ifdef MAXSEG_64K
1108     return Z_STREAM_ERROR;
1109 #else
1110     deflate_state *ds;
1111     deflate_state *ss;
1112     ushf *overlay;
1113 
1114 
1115     if (deflateStateCheck(source) || dest == Z_NULL) {
1116         return Z_STREAM_ERROR;
1117     }
1118 
1119     ss = source->state;
1120 
1121     zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1122 
1123     ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1124     if (ds == Z_NULL) return Z_MEM_ERROR;
1125     dest->state = (struct internal_state FAR *) ds;
1126     zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1127     ds->strm = dest;
1128 
1129     ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1130     ds->prev   = (Posf *)  ZALLOC(dest, ds->w_size, sizeof(Pos));
1131     ds->head   = (Posf *)  ZALLOC(dest, ds->hash_size, sizeof(Pos));
1132     overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1133     ds->pending_buf = (uchf *) overlay;
1134 
1135     if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1136         ds->pending_buf == Z_NULL) {
1137         deflateEnd (dest);
1138         return Z_MEM_ERROR;
1139     }
1140     /* following zmemcpy do not work for 16-bit MSDOS */
1141     zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1142     zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1143     zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1144     zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1145 
1146     ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1147     ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1148     ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1149 
1150     ds->l_desc.dyn_tree = ds->dyn_ltree;
1151     ds->d_desc.dyn_tree = ds->dyn_dtree;
1152     ds->bl_desc.dyn_tree = ds->bl_tree;
1153 
1154     return Z_OK;
1155 #endif /* MAXSEG_64K */
1156 }
1157 
1158 /* ===========================================================================
1159  * Read a new buffer from the current input stream, update the adler32
1160  * and total number of bytes read.  All deflate() input goes through
1161  * this function so some applications may wish to modify it to avoid
1162  * allocating a large strm->next_in buffer and copying from it.
1163  * (See also flush_pending()).
1164  */
1165 local unsigned read_buf(strm, buf, size)
1166     z_streamp strm;
1167     Bytef *buf;
1168     unsigned size;
1169 {
1170     unsigned len = strm->avail_in;
1171 
1172     if (len > size) len = size;
1173     if (len == 0) return 0;
1174 
1175     strm->avail_in  -= len;
1176 
1177     zmemcpy(buf, strm->next_in, len);
1178     if (strm->state->wrap == 1) {
1179         strm->adler = adler32(strm->adler, buf, len);
1180     }
1181 #ifdef GZIP
1182     else if (strm->state->wrap == 2) {
1183         strm->adler = crc32(strm->adler, buf, len);
1184     }
1185 #endif
1186     strm->next_in  += len;
1187     strm->total_in += len;
1188 
1189     return len;
1190 }
1191 
1192 /* ===========================================================================
1193  * Initialize the "longest match" routines for a new zlib stream
1194  */
1195 local void lm_init (s)
1196     deflate_state *s;
1197 {
1198     s->window_size = (ulg)2L*s->w_size;
1199 
1200     CLEAR_HASH(s);
1201 
1202     /* Set the default configuration parameters:
1203      */
1204     s->max_lazy_match   = configuration_table[s->level].max_lazy;
1205     s->good_match       = configuration_table[s->level].good_length;
1206     s->nice_match       = configuration_table[s->level].nice_length;
1207     s->max_chain_length = configuration_table[s->level].max_chain;
1208 
1209     s->strstart = 0;
1210     s->block_start = 0L;
1211     s->lookahead = 0;
1212     s->insert = 0;
1213     s->match_length = s->prev_length = MIN_MATCH-1;
1214     s->match_available = 0;
1215     s->ins_h = 0;
1216 #ifndef FASTEST
1217 #ifdef ASMV
1218     match_init(); /* initialize the asm code */
1219 #endif
1220 #endif
1221 }
1222 
1223 #ifndef FASTEST
1224 /* ===========================================================================
1225  * Set match_start to the longest match starting at the given string and
1226  * return its length. Matches shorter or equal to prev_length are discarded,
1227  * in which case the result is equal to prev_length and match_start is
1228  * garbage.
1229  * IN assertions: cur_match is the head of the hash chain for the current
1230  *   string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1231  * OUT assertion: the match length is not greater than s->lookahead.
1232  */
1233 #ifndef ASMV
1234 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1235  * match.S. The code will be functionally equivalent.
1236  */
1237 local uInt longest_match(s, cur_match)
1238     deflate_state *s;
1239     IPos cur_match;                             /* current match */
1240 {
1241     unsigned chain_length = s->max_chain_length;/* max hash chain length */
1242     register Bytef *scan = s->window + s->strstart; /* current string */
1243     register Bytef *match;                      /* matched string */
1244     register int len;                           /* length of current match */
1245     int best_len = (int)s->prev_length;         /* best match length so far */
1246     int nice_match = s->nice_match;             /* stop if match long enough */
1247     IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1248         s->strstart - (IPos)MAX_DIST(s) : NIL;
1249     /* Stop when cur_match becomes <= limit. To simplify the code,
1250      * we prevent matches with the string of window index 0.
1251      */
1252     Posf *prev = s->prev;
1253     uInt wmask = s->w_mask;
1254 
1255 #ifdef UNALIGNED_OK
1256     /* Compare two bytes at a time. Note: this is not always beneficial.
1257      * Try with and without -DUNALIGNED_OK to check.
1258      */
1259     register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1260     register ush scan_start = *(ushf*)scan;
1261     register ush scan_end   = *(ushf*)(scan+best_len-1);
1262 #else
1263     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1264     register Byte scan_end1  = scan[best_len-1];
1265     register Byte scan_end   = scan[best_len];
1266 #endif
1267 
1268     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1269      * It is easy to get rid of this optimization if necessary.
1270      */
1271     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1272 
1273     /* Do not waste too much time if we already have a good match: */
1274     if (s->prev_length >= s->good_match) {
1275         chain_length >>= 2;
1276     }
1277     /* Do not look for matches beyond the end of the input. This is necessary
1278      * to make deflate deterministic.
1279      */
1280     if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1281 
1282     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1283 
1284     do {
1285         Assert(cur_match < s->strstart, "no future");
1286         match = s->window + cur_match;
1287 
1288         /* Skip to next match if the match length cannot increase
1289          * or if the match length is less than 2.  Note that the checks below
1290          * for insufficient lookahead only occur occasionally for performance
1291          * reasons.  Therefore uninitialized memory will be accessed, and
1292          * conditional jumps will be made that depend on those values.
1293          * However the length of the match is limited to the lookahead, so
1294          * the output of deflate is not affected by the uninitialized values.
1295          */
1296 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1297         /* This code assumes sizeof(unsigned short) == 2. Do not use
1298          * UNALIGNED_OK if your compiler uses a different size.
1299          */
1300         if (*(ushf*)(match+best_len-1) != scan_end ||
1301             *(ushf*)match != scan_start) continue;
1302 
1303         /* It is not necessary to compare scan[2] and match[2] since they are
1304          * always equal when the other bytes match, given that the hash keys
1305          * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1306          * strstart+3, +5, ... up to strstart+257. We check for insufficient
1307          * lookahead only every 4th comparison; the 128th check will be made
1308          * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1309          * necessary to put more guard bytes at the end of the window, or
1310          * to check more often for insufficient lookahead.
1311          */
1312         Assert(scan[2] == match[2], "scan[2]?");
1313         scan++, match++;
1314         do {
1315         } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1316                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1317                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1318                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1319                  scan < strend);
1320         /* The funny "do {}" generates better code on most compilers */
1321 
1322         /* Here, scan <= window+strstart+257 */
1323         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1324         if (*scan == *match) scan++;
1325 
1326         len = (MAX_MATCH - 1) - (int)(strend-scan);
1327         scan = strend - (MAX_MATCH-1);
1328 
1329 #else /* UNALIGNED_OK */
1330 
1331         if (match[best_len]   != scan_end  ||
1332             match[best_len-1] != scan_end1 ||
1333             *match            != *scan     ||
1334             *++match          != scan[1])      continue;
1335 
1336         /* The check at best_len-1 can be removed because it will be made
1337          * again later. (This heuristic is not always a win.)
1338          * It is not necessary to compare scan[2] and match[2] since they
1339          * are always equal when the other bytes match, given that
1340          * the hash keys are equal and that HASH_BITS >= 8.
1341          */
1342         scan += 2, match++;
1343         Assert(*scan == *match, "match[2]?");
1344 
1345         /* We check for insufficient lookahead only every 8th comparison;
1346          * the 256th check will be made at strstart+258.
1347          */
1348         do {
1349         } while (*++scan == *++match && *++scan == *++match &&
1350                  *++scan == *++match && *++scan == *++match &&
1351                  *++scan == *++match && *++scan == *++match &&
1352                  *++scan == *++match && *++scan == *++match &&
1353                  scan < strend);
1354 
1355         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1356 
1357         len = MAX_MATCH - (int)(strend - scan);
1358         scan = strend - MAX_MATCH;
1359 
1360 #endif /* UNALIGNED_OK */
1361 
1362         if (len > best_len) {
1363             s->match_start = cur_match;
1364             best_len = len;
1365             if (len >= nice_match) break;
1366 #ifdef UNALIGNED_OK
1367             scan_end = *(ushf*)(scan+best_len-1);
1368 #else
1369             scan_end1  = scan[best_len-1];
1370             scan_end   = scan[best_len];
1371 #endif
1372         }
1373     } while ((cur_match = prev[cur_match & wmask]) > limit
1374              && --chain_length != 0);
1375 
1376     if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1377     return s->lookahead;
1378 }
1379 #endif /* ASMV */
1380 
1381 #else /* FASTEST */
1382 
1383 /* ---------------------------------------------------------------------------
1384  * Optimized version for FASTEST only
1385  */
1386 local uInt longest_match(s, cur_match)
1387     deflate_state *s;
1388     IPos cur_match;                             /* current match */
1389 {
1390     register Bytef *scan = s->window + s->strstart; /* current string */
1391     register Bytef *match;                       /* matched string */
1392     register int len;                           /* length of current match */
1393     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1394 
1395     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1396      * It is easy to get rid of this optimization if necessary.
1397      */
1398     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1399 
1400     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1401 
1402     Assert(cur_match < s->strstart, "no future");
1403 
1404     match = s->window + cur_match;
1405 
1406     /* Return failure if the match length is less than 2:
1407      */
1408     if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1409 
1410     /* The check at best_len-1 can be removed because it will be made
1411      * again later. (This heuristic is not always a win.)
1412      * It is not necessary to compare scan[2] and match[2] since they
1413      * are always equal when the other bytes match, given that
1414      * the hash keys are equal and that HASH_BITS >= 8.
1415      */
1416     scan += 2, match += 2;
1417     Assert(*scan == *match, "match[2]?");
1418 
1419     /* We check for insufficient lookahead only every 8th comparison;
1420      * the 256th check will be made at strstart+258.
1421      */
1422     do {
1423     } while (*++scan == *++match && *++scan == *++match &&
1424              *++scan == *++match && *++scan == *++match &&
1425              *++scan == *++match && *++scan == *++match &&
1426              *++scan == *++match && *++scan == *++match &&
1427              scan < strend);
1428 
1429     Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1430 
1431     len = MAX_MATCH - (int)(strend - scan);
1432 
1433     if (len < MIN_MATCH) return MIN_MATCH - 1;
1434 
1435     s->match_start = cur_match;
1436     return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1437 }
1438 
1439 #endif /* FASTEST */
1440 
1441 #ifdef ZLIB_DEBUG
1442 
1443 #define EQUAL 0
1444 /* result of memcmp for equal strings */
1445 
1446 /* ===========================================================================
1447  * Check that the match at match_start is indeed a match.
1448  */
1449 local void check_match(s, start, match, length)
1450     deflate_state *s;
1451     IPos start, match;
1452     int length;
1453 {
1454     /* check that the match is indeed a match */
1455     if (zmemcmp(s->window + match,
1456                 s->window + start, length) != EQUAL) {
1457         fprintf(stderr, " start %u, match %u, length %d\n",
1458                 start, match, length);
1459         do {
1460             fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1461         } while (--length != 0);
1462         z_error("invalid match");
1463     }
1464     if (z_verbose > 1) {
1465         fprintf(stderr,"\\[%d,%d]", start-match, length);
1466         do { putc(s->window[start++], stderr); } while (--length != 0);
1467     }
1468 }
1469 #else
1470 #  define check_match(s, start, match, length)
1471 #endif /* ZLIB_DEBUG */
1472 
1473 /* ===========================================================================
1474  * Fill the window when the lookahead becomes insufficient.
1475  * Updates strstart and lookahead.
1476  *
1477  * IN assertion: lookahead < MIN_LOOKAHEAD
1478  * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1479  *    At least one byte has been read, or avail_in == 0; reads are
1480  *    performed for at least two bytes (required for the zip translate_eol
1481  *    option -- not supported here).
1482  */
1483 local void fill_window(s)
1484     deflate_state *s;
1485 {
1486     unsigned n;
1487     unsigned more;    /* Amount of free space at the end of the window. */
1488     uInt wsize = s->w_size;
1489 
1490     Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1491 
1492     do {
1493         more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1494 
1495         /* Deal with !@#$% 64K limit: */
1496         if (sizeof(int) <= 2) {
1497             if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1498                 more = wsize;
1499 
1500             } else if (more == (unsigned)(-1)) {
1501                 /* Very unlikely, but possible on 16 bit machine if
1502                  * strstart == 0 && lookahead == 1 (input done a byte at time)
1503                  */
1504                 more--;
1505             }
1506         }
1507 
1508         /* If the window is almost full and there is insufficient lookahead,
1509          * move the upper half to the lower one to make room in the upper half.
1510          */
1511         if (s->strstart >= wsize+MAX_DIST(s)) {
1512 
1513             zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1514             s->match_start -= wsize;
1515             s->strstart    -= wsize; /* we now have strstart >= MAX_DIST */
1516             s->block_start -= (long) wsize;
1517             slide_hash(s);
1518             more += wsize;
1519         }
1520         if (s->strm->avail_in == 0) break;
1521 
1522         /* If there was no sliding:
1523          *    strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1524          *    more == window_size - lookahead - strstart
1525          * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1526          * => more >= window_size - 2*WSIZE + 2
1527          * In the BIG_MEM or MMAP case (not yet supported),
1528          *   window_size == input_size + MIN_LOOKAHEAD  &&
1529          *   strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1530          * Otherwise, window_size == 2*WSIZE so more >= 2.
1531          * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1532          */
1533         Assert(more >= 2, "more < 2");
1534 
1535         n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1536         s->lookahead += n;
1537 
1538         /* Initialize the hash value now that we have some input: */
1539         if (s->lookahead + s->insert >= MIN_MATCH) {
1540             uInt str = s->strstart - s->insert;
1541             s->ins_h = s->window[str];
1542             UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1543 #if MIN_MATCH != 3
1544             Call UPDATE_HASH() MIN_MATCH-3 more times
1545 #endif
1546             while (s->insert) {
1547                 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1548 #ifndef FASTEST
1549                 s->prev[str & s->w_mask] = s->head[s->ins_h];
1550 #endif
1551                 s->head[s->ins_h] = (Pos)str;
1552                 str++;
1553                 s->insert--;
1554                 if (s->lookahead + s->insert < MIN_MATCH)
1555                     break;
1556             }
1557         }
1558         /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1559          * but this is not important since only literal bytes will be emitted.
1560          */
1561 
1562     } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1563 
1564     /* If the WIN_INIT bytes after the end of the current data have never been
1565      * written, then zero those bytes in order to avoid memory check reports of
1566      * the use of uninitialized (or uninitialised as Julian writes) bytes by
1567      * the longest match routines.  Update the high water mark for the next
1568      * time through here.  WIN_INIT is set to MAX_MATCH since the longest match
1569      * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1570      */
1571     if (s->high_water < s->window_size) {
1572         ulg curr = s->strstart + (ulg)(s->lookahead);
1573         ulg init;
1574 
1575         if (s->high_water < curr) {
1576             /* Previous high water mark below current data -- zero WIN_INIT
1577              * bytes or up to end of window, whichever is less.
1578              */
1579             init = s->window_size - curr;
1580             if (init > WIN_INIT)
1581                 init = WIN_INIT;
1582             zmemzero(s->window + curr, (unsigned)init);
1583             s->high_water = curr + init;
1584         }
1585         else if (s->high_water < (ulg)curr + WIN_INIT) {
1586             /* High water mark at or above current data, but below current data
1587              * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1588              * to end of window, whichever is less.
1589              */
1590             init = (ulg)curr + WIN_INIT - s->high_water;
1591             if (init > s->window_size - s->high_water)
1592                 init = s->window_size - s->high_water;
1593             zmemzero(s->window + s->high_water, (unsigned)init);
1594             s->high_water += init;
1595         }
1596     }
1597 
1598     Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1599            "not enough room for search");
1600 }
1601 
1602 /* ===========================================================================
1603  * Flush the current block, with given end-of-file flag.
1604  * IN assertion: strstart is set to the end of the current match.
1605  */
1606 #define FLUSH_BLOCK_ONLY(s, last) { \
1607    _tr_flush_block(s, (s->block_start >= 0L ? \
1608                    (charf *)&s->window[(unsigned)s->block_start] : \
1609                    (charf *)Z_NULL), \
1610                 (ulg)((long)s->strstart - s->block_start), \
1611                 (last)); \
1612    s->block_start = s->strstart; \
1613    flush_pending(s->strm); \
1614    Tracev((stderr,"[FLUSH]")); \
1615 }
1616 
1617 /* Same but force premature exit if necessary. */
1618 #define FLUSH_BLOCK(s, last) { \
1619    FLUSH_BLOCK_ONLY(s, last); \
1620    if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1621 }
1622 
1623 /* Maximum stored block length in deflate format (not including header). */
1624 #define MAX_STORED 65535
1625 
1626 /* Minimum of a and b. */
1627 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1628 
1629 /* ===========================================================================
1630  * Copy without compression as much as possible from the input stream, return
1631  * the current block state.
1632  *
1633  * In case deflateParams() is used to later switch to a non-zero compression
1634  * level, s->matches (otherwise unused when storing) keeps track of the number
1635  * of hash table slides to perform. If s->matches is 1, then one hash table
1636  * slide will be done when switching. If s->matches is 2, the maximum value
1637  * allowed here, then the hash table will be cleared, since two or more slides
1638  * is the same as a clear.
1639  *
1640  * deflate_stored() is written to minimize the number of times an input byte is
1641  * copied. It is most efficient with large input and output buffers, which
1642  * maximizes the opportunites to have a single copy from next_in to next_out.
1643  */
1644 local block_state deflate_stored(s, flush)
1645     deflate_state *s;
1646     int flush;
1647 {
1648     /* Smallest worthy block size when not flushing or finishing. By default
1649      * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1650      * large input and output buffers, the stored block size will be larger.
1651      */
1652     unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1653 
1654     /* Copy as many min_block or larger stored blocks directly to next_out as
1655      * possible. If flushing, copy the remaining available input to next_out as
1656      * stored blocks, if there is enough space.
1657      */
1658     unsigned len, left, have, last = 0;
1659     unsigned used = s->strm->avail_in;
1660     do {
1661         /* Set len to the maximum size block that we can copy directly with the
1662          * available input data and output space. Set left to how much of that
1663          * would be copied from what's left in the window.
1664          */
1665         len = MAX_STORED;       /* maximum deflate stored block length */
1666         have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1667         if (s->strm->avail_out < have)          /* need room for header */
1668             break;
1669             /* maximum stored block length that will fit in avail_out: */
1670         have = s->strm->avail_out - have;
1671         left = s->strstart - s->block_start;    /* bytes left in window */
1672         if (len > (ulg)left + s->strm->avail_in)
1673             len = left + s->strm->avail_in;     /* limit len to the input */
1674         if (len > have)
1675             len = have;                         /* limit len to the output */
1676 
1677         /* If the stored block would be less than min_block in length, or if
1678          * unable to copy all of the available input when flushing, then try
1679          * copying to the window and the pending buffer instead. Also don't
1680          * write an empty block when flushing -- deflate() does that.
1681          */
1682         if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1683                                 flush == Z_NO_FLUSH ||
1684                                 len != left + s->strm->avail_in))
1685             break;
1686 
1687         /* Make a dummy stored block in pending to get the header bytes,
1688          * including any pending bits. This also updates the debugging counts.
1689          */
1690         last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1691         _tr_stored_block(s, (char *)0, 0L, last);
1692 
1693         /* Replace the lengths in the dummy stored block with len. */
1694         s->pending_buf[s->pending - 4] = len;
1695         s->pending_buf[s->pending - 3] = len >> 8;
1696         s->pending_buf[s->pending - 2] = ~len;
1697         s->pending_buf[s->pending - 1] = ~len >> 8;
1698 
1699         /* Write the stored block header bytes. */
1700         flush_pending(s->strm);
1701 
1702 #ifdef ZLIB_DEBUG
1703         /* Update debugging counts for the data about to be copied. */
1704         s->compressed_len += len << 3;
1705         s->bits_sent += len << 3;
1706 #endif
1707 
1708         /* Copy uncompressed bytes from the window to next_out. */
1709         if (left) {
1710             if (left > len)
1711                 left = len;
1712             zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1713             s->strm->next_out += left;
1714             s->strm->avail_out -= left;
1715             s->strm->total_out += left;
1716             s->block_start += left;
1717             len -= left;
1718         }
1719 
1720         /* Copy uncompressed bytes directly from next_in to next_out, updating
1721          * the check value.
1722          */
1723         if (len) {
1724             read_buf(s->strm, s->strm->next_out, len);
1725             s->strm->next_out += len;
1726             s->strm->avail_out -= len;
1727             s->strm->total_out += len;
1728         }
1729     } while (last == 0);
1730 
1731     /* Update the sliding window with the last s->w_size bytes of the copied
1732      * data, or append all of the copied data to the existing window if less
1733      * than s->w_size bytes were copied. Also update the number of bytes to
1734      * insert in the hash tables, in the event that deflateParams() switches to
1735      * a non-zero compression level.
1736      */
1737     used -= s->strm->avail_in;      /* number of input bytes directly copied */
1738     if (used) {
1739         /* If any input was used, then no unused input remains in the window,
1740          * therefore s->block_start == s->strstart.
1741          */
1742         if (used >= s->w_size) {    /* supplant the previous history */
1743             s->matches = 2;         /* clear hash */
1744             zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1745             s->strstart = s->w_size;
1746         }
1747         else {
1748             if (s->window_size - s->strstart <= used) {
1749                 /* Slide the window down. */
1750                 s->strstart -= s->w_size;
1751                 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1752                 if (s->matches < 2)
1753                     s->matches++;   /* add a pending slide_hash() */
1754             }
1755             zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1756             s->strstart += used;
1757         }
1758         s->block_start = s->strstart;
1759         s->insert += MIN(used, s->w_size - s->insert);
1760     }
1761     if (s->high_water < s->strstart)
1762         s->high_water = s->strstart;
1763 
1764     /* If the last block was written to next_out, then done. */
1765     if (last)
1766         return finish_done;
1767 
1768     /* If flushing and all input has been consumed, then done. */
1769     if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1770         s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1771         return block_done;
1772 
1773     /* Fill the window with any remaining input. */
1774     have = s->window_size - s->strstart - 1;
1775     if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1776         /* Slide the window down. */
1777         s->block_start -= s->w_size;
1778         s->strstart -= s->w_size;
1779         zmemcpy(s->window, s->window + s->w_size, s->strstart);
1780         if (s->matches < 2)
1781             s->matches++;           /* add a pending slide_hash() */
1782         have += s->w_size;          /* more space now */
1783     }
1784     if (have > s->strm->avail_in)
1785         have = s->strm->avail_in;
1786     if (have) {
1787         read_buf(s->strm, s->window + s->strstart, have);
1788         s->strstart += have;
1789     }
1790     if (s->high_water < s->strstart)
1791         s->high_water = s->strstart;
1792 
1793     /* There was not enough avail_out to write a complete worthy or flushed
1794      * stored block to next_out. Write a stored block to pending instead, if we
1795      * have enough input for a worthy block, or if flushing and there is enough
1796      * room for the remaining input as a stored block in the pending buffer.
1797      */
1798     have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1799         /* maximum stored block length that will fit in pending: */
1800     have = MIN(s->pending_buf_size - have, MAX_STORED);
1801     min_block = MIN(have, s->w_size);
1802     left = s->strstart - s->block_start;
1803     if (left >= min_block ||
1804         ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1805          s->strm->avail_in == 0 && left <= have)) {
1806         len = MIN(left, have);
1807         last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1808                len == left ? 1 : 0;
1809         _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1810         s->block_start += len;
1811         flush_pending(s->strm);
1812     }
1813 
1814     /* We've done all we can with the available input and output. */
1815     return last ? finish_started : need_more;
1816 }
1817 
1818 /* ===========================================================================
1819  * Compress as much as possible from the input stream, return the current
1820  * block state.
1821  * This function does not perform lazy evaluation of matches and inserts
1822  * new strings in the dictionary only for unmatched strings or for short
1823  * matches. It is used only for the fast compression options.
1824  */
1825 local block_state deflate_fast(s, flush)
1826     deflate_state *s;
1827     int flush;
1828 {
1829     IPos hash_head;       /* head of the hash chain */
1830     int bflush;           /* set if current block must be flushed */
1831 
1832     for (;;) {
1833         /* Make sure that we always have enough lookahead, except
1834          * at the end of the input file. We need MAX_MATCH bytes
1835          * for the next match, plus MIN_MATCH bytes to insert the
1836          * string following the next match.
1837          */
1838         if (s->lookahead < MIN_LOOKAHEAD) {
1839             fill_window(s);
1840             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1841                 return need_more;
1842             }
1843             if (s->lookahead == 0) break; /* flush the current block */
1844         }
1845 
1846         /* Insert the string window[strstart .. strstart+2] in the
1847          * dictionary, and set hash_head to the head of the hash chain:
1848          */
1849         hash_head = NIL;
1850         if (s->lookahead >= MIN_MATCH) {
1851             INSERT_STRING(s, s->strstart, hash_head);
1852         }
1853 
1854         /* Find the longest match, discarding those <= prev_length.
1855          * At this point we have always match_length < MIN_MATCH
1856          */
1857         if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1858             /* To simplify the code, we prevent matches with the string
1859              * of window index 0 (in particular we have to avoid a match
1860              * of the string with itself at the start of the input file).
1861              */
1862             s->match_length = longest_match (s, hash_head);
1863             /* longest_match() sets match_start */
1864         }
1865         if (s->match_length >= MIN_MATCH) {
1866             check_match(s, s->strstart, s->match_start, s->match_length);
1867 
1868             _tr_tally_dist(s, s->strstart - s->match_start,
1869                            s->match_length - MIN_MATCH, bflush);
1870 
1871             s->lookahead -= s->match_length;
1872 
1873             /* Insert new strings in the hash table only if the match length
1874              * is not too large. This saves time but degrades compression.
1875              */
1876 #ifndef FASTEST
1877             if (s->match_length <= s->max_insert_length &&
1878                 s->lookahead >= MIN_MATCH) {
1879                 s->match_length--; /* string at strstart already in table */
1880                 do {
1881                     s->strstart++;
1882                     INSERT_STRING(s, s->strstart, hash_head);
1883                     /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1884                      * always MIN_MATCH bytes ahead.
1885                      */
1886                 } while (--s->match_length != 0);
1887                 s->strstart++;
1888             } else
1889 #endif
1890             {
1891                 s->strstart += s->match_length;
1892                 s->match_length = 0;
1893                 s->ins_h = s->window[s->strstart];
1894                 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1895 #if MIN_MATCH != 3
1896                 Call UPDATE_HASH() MIN_MATCH-3 more times
1897 #endif
1898                 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1899                  * matter since it will be recomputed at next deflate call.
1900                  */
1901             }
1902         } else {
1903             /* No match, output a literal byte */
1904             Tracevv((stderr,"%c", s->window[s->strstart]));
1905             _tr_tally_lit (s, s->window[s->strstart], bflush);
1906             s->lookahead--;
1907             s->strstart++;
1908         }
1909         if (bflush) FLUSH_BLOCK(s, 0);
1910     }
1911     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1912     if (flush == Z_FINISH) {
1913         FLUSH_BLOCK(s, 1);
1914         return finish_done;
1915     }
1916     if (s->last_lit)
1917         FLUSH_BLOCK(s, 0);
1918     return block_done;
1919 }
1920 
1921 #ifndef FASTEST
1922 /* ===========================================================================
1923  * Same as above, but achieves better compression. We use a lazy
1924  * evaluation for matches: a match is finally adopted only if there is
1925  * no better match at the next window position.
1926  */
1927 local block_state deflate_slow(s, flush)
1928     deflate_state *s;
1929     int flush;
1930 {
1931     IPos hash_head;          /* head of hash chain */
1932     int bflush;              /* set if current block must be flushed */
1933 
1934     /* Process the input block. */
1935     for (;;) {
1936         /* Make sure that we always have enough lookahead, except
1937          * at the end of the input file. We need MAX_MATCH bytes
1938          * for the next match, plus MIN_MATCH bytes to insert the
1939          * string following the next match.
1940          */
1941         if (s->lookahead < MIN_LOOKAHEAD) {
1942             fill_window(s);
1943             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1944                 return need_more;
1945             }
1946             if (s->lookahead == 0) break; /* flush the current block */
1947         }
1948 
1949         /* Insert the string window[strstart .. strstart+2] in the
1950          * dictionary, and set hash_head to the head of the hash chain:
1951          */
1952         hash_head = NIL;
1953         if (s->lookahead >= MIN_MATCH) {
1954             INSERT_STRING(s, s->strstart, hash_head);
1955         }
1956 
1957         /* Find the longest match, discarding those <= prev_length.
1958          */
1959         s->prev_length = s->match_length, s->prev_match = s->match_start;
1960         s->match_length = MIN_MATCH-1;
1961 
1962         if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1963             s->strstart - hash_head <= MAX_DIST(s)) {
1964             /* To simplify the code, we prevent matches with the string
1965              * of window index 0 (in particular we have to avoid a match
1966              * of the string with itself at the start of the input file).
1967              */
1968             s->match_length = longest_match (s, hash_head);
1969             /* longest_match() sets match_start */
1970 
1971             if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1972 #if TOO_FAR <= 32767
1973                 || (s->match_length == MIN_MATCH &&
1974                     s->strstart - s->match_start > TOO_FAR)
1975 #endif
1976                 )) {
1977 
1978                 /* If prev_match is also MIN_MATCH, match_start is garbage
1979                  * but we will ignore the current match anyway.
1980                  */
1981                 s->match_length = MIN_MATCH-1;
1982             }
1983         }
1984         /* If there was a match at the previous step and the current
1985          * match is not better, output the previous match:
1986          */
1987         if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1988             uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1989             /* Do not insert strings in hash table beyond this. */
1990 
1991             check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1992 
1993             _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1994                            s->prev_length - MIN_MATCH, bflush);
1995 
1996             /* Insert in hash table all strings up to the end of the match.
1997              * strstart-1 and strstart are already inserted. If there is not
1998              * enough lookahead, the last two strings are not inserted in
1999              * the hash table.
2000              */
2001             s->lookahead -= s->prev_length-1;
2002             s->prev_length -= 2;
2003             do {
2004                 if (++s->strstart <= max_insert) {
2005                     INSERT_STRING(s, s->strstart, hash_head);
2006                 }
2007             } while (--s->prev_length != 0);
2008             s->match_available = 0;
2009             s->match_length = MIN_MATCH-1;
2010             s->strstart++;
2011 
2012             if (bflush) FLUSH_BLOCK(s, 0);
2013 
2014         } else if (s->match_available) {
2015             /* If there was no match at the previous position, output a
2016              * single literal. If there was a match but the current match
2017              * is longer, truncate the previous match to a single literal.
2018              */
2019             Tracevv((stderr,"%c", s->window[s->strstart-1]));
2020             _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2021             if (bflush) {
2022                 FLUSH_BLOCK_ONLY(s, 0);
2023             }
2024             s->strstart++;
2025             s->lookahead--;
2026             if (s->strm->avail_out == 0) return need_more;
2027         } else {
2028             /* There is no previous match to compare with, wait for
2029              * the next step to decide.
2030              */
2031             s->match_available = 1;
2032             s->strstart++;
2033             s->lookahead--;
2034         }
2035     }
2036     Assert (flush != Z_NO_FLUSH, "no flush?");
2037     if (s->match_available) {
2038         Tracevv((stderr,"%c", s->window[s->strstart-1]));
2039         _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2040         s->match_available = 0;
2041     }
2042     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2043     if (flush == Z_FINISH) {
2044         FLUSH_BLOCK(s, 1);
2045         return finish_done;
2046     }
2047     if (s->last_lit)
2048         FLUSH_BLOCK(s, 0);
2049     return block_done;
2050 }
2051 #endif /* FASTEST */
2052 
2053 /* ===========================================================================
2054  * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2055  * one.  Do not maintain a hash table.  (It will be regenerated if this run of
2056  * deflate switches away from Z_RLE.)
2057  */
2058 local block_state deflate_rle(s, flush)
2059     deflate_state *s;
2060     int flush;
2061 {
2062     int bflush;             /* set if current block must be flushed */
2063     uInt prev;              /* byte at distance one to match */
2064     Bytef *scan, *strend;   /* scan goes up to strend for length of run */
2065 
2066     for (;;) {
2067         /* Make sure that we always have enough lookahead, except
2068          * at the end of the input file. We need MAX_MATCH bytes
2069          * for the longest run, plus one for the unrolled loop.
2070          */
2071         if (s->lookahead <= MAX_MATCH) {
2072             fill_window(s);
2073             if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2074                 return need_more;
2075             }
2076             if (s->lookahead == 0) break; /* flush the current block */
2077         }
2078 
2079         /* See how many times the previous byte repeats */
2080         s->match_length = 0;
2081         if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2082             scan = s->window + s->strstart - 1;
2083             prev = *scan;
2084             if (prev == *++scan && prev == *++scan && prev == *++scan) {
2085                 strend = s->window + s->strstart + MAX_MATCH;
2086                 do {
2087                 } while (prev == *++scan && prev == *++scan &&
2088                          prev == *++scan && prev == *++scan &&
2089                          prev == *++scan && prev == *++scan &&
2090                          prev == *++scan && prev == *++scan &&
2091                          scan < strend);
2092                 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2093                 if (s->match_length > s->lookahead)
2094                     s->match_length = s->lookahead;
2095             }
2096             Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2097         }
2098 
2099         /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2100         if (s->match_length >= MIN_MATCH) {
2101             check_match(s, s->strstart, s->strstart - 1, s->match_length);
2102 
2103             _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2104 
2105             s->lookahead -= s->match_length;
2106             s->strstart += s->match_length;
2107             s->match_length = 0;
2108         } else {
2109             /* No match, output a literal byte */
2110             Tracevv((stderr,"%c", s->window[s->strstart]));
2111             _tr_tally_lit (s, s->window[s->strstart], bflush);
2112             s->lookahead--;
2113             s->strstart++;
2114         }
2115         if (bflush) FLUSH_BLOCK(s, 0);
2116     }
2117     s->insert = 0;
2118     if (flush == Z_FINISH) {
2119         FLUSH_BLOCK(s, 1);
2120         return finish_done;
2121     }
2122     if (s->last_lit)
2123         FLUSH_BLOCK(s, 0);
2124     return block_done;
2125 }
2126 
2127 /* ===========================================================================
2128  * For Z_HUFFMAN_ONLY, do not look for matches.  Do not maintain a hash table.
2129  * (It will be regenerated if this run of deflate switches away from Huffman.)
2130  */
2131 local block_state deflate_huff(s, flush)
2132     deflate_state *s;
2133     int flush;
2134 {
2135     int bflush;             /* set if current block must be flushed */
2136 
2137     for (;;) {
2138         /* Make sure that we have a literal to write. */
2139         if (s->lookahead == 0) {
2140             fill_window(s);
2141             if (s->lookahead == 0) {
2142                 if (flush == Z_NO_FLUSH)
2143                     return need_more;
2144                 break;      /* flush the current block */
2145             }
2146         }
2147 
2148         /* Output a literal byte */
2149         s->match_length = 0;
2150         Tracevv((stderr,"%c", s->window[s->strstart]));
2151         _tr_tally_lit (s, s->window[s->strstart], bflush);
2152         s->lookahead--;
2153         s->strstart++;
2154         if (bflush) FLUSH_BLOCK(s, 0);
2155     }
2156     s->insert = 0;
2157     if (flush == Z_FINISH) {
2158         FLUSH_BLOCK(s, 1);
2159         return finish_done;
2160     }
2161     if (s->last_lit)
2162         FLUSH_BLOCK(s, 0);
2163     return block_done;
2164 }
2165