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 /* @(#) $Id$ */
51 
52 #include "deflate.h"
53 
54 const char deflate_copyright[] =
55    " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler ";
56 /*
57   If you use the zlib library in a product, an acknowledgment is welcome
58   in the documentation of your product. If for some reason you cannot
59   include such an acknowledgment, I would appreciate that you keep this
60   copyright string in the executable of your product.
61  */
62 
63 /* ===========================================================================
64  *  Function prototypes.
65  */
66 typedef enum {
67     need_more,      /* block not completed, need more input or more output */
68     block_done,     /* block flush performed */
69     finish_started, /* finish started, need only more output at next deflate */
70     finish_done     /* finish done, accept no more input or output */
71 } block_state;
72 
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74 /* Compression function. Returns the block state after the call. */
75 
76 local int deflateStateCheck      OF((z_streamp strm));
77 local void slide_hash     OF((deflate_state *s));
78 local void fill_window    OF((deflate_state *s));
79 local block_state deflate_stored OF((deflate_state *s, int flush));
80 local block_state deflate_fast   OF((deflate_state *s, int flush));
81 #ifndef FASTEST
82 local block_state deflate_slow   OF((deflate_state *s, int flush));
83 #endif
84 local block_state deflate_rle    OF((deflate_state *s, int flush));
85 local block_state deflate_huff   OF((deflate_state *s, int flush));
86 local void lm_init        OF((deflate_state *s));
87 local void putShortMSB    OF((deflate_state *s, uInt b));
88 local void flush_pending  OF((z_streamp strm));
89 local unsigned read_buf   OF((z_streamp strm, Bytef *buf, unsigned size));
90 #ifdef ASMV
91 #  pragma message("Assembler code may have bugs -- use at your own risk")
92       void match_init OF((void)); /* asm code initialization */
93       uInt longest_match  OF((deflate_state *s, IPos cur_match));
94 #else
95 local uInt longest_match  OF((deflate_state *s, IPos cur_match));
96 #endif
97 
98 #ifdef ZLIB_DEBUG
99 local  void check_match OF((deflate_state *s, IPos start, IPos match,
100                             int length));
101 #endif
102 
103 /* ===========================================================================
104  * Local data
105  */
106 
107 #define NIL 0
108 /* Tail of hash chains */
109 
110 #ifndef TOO_FAR
111 #  define TOO_FAR 4096
112 #endif
113 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
114 
115 /* Values for max_lazy_match, good_match and max_chain_length, depending on
116  * the desired pack level (0..9). The values given below have been tuned to
117  * exclude worst case performance for pathological files. Better values may be
118  * found for specific files.
119  */
120 typedef struct config_s {
121    ush good_length; /* reduce lazy search above this match length */
122    ush max_lazy;    /* do not perform lazy search above this match length */
123    ush nice_length; /* quit search above this match length */
124    ush max_chain;
125    compress_func func;
126 } config;
127 
128 #ifdef FASTEST
129 local const config configuration_table[2] = {
130 /*      good lazy nice chain */
131 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
132 /* 1 */ {4,    4,  8,    4, deflate_fast}}; /* max speed, no lazy matches */
133 #else
134 local const config configuration_table[10] = {
135 /*      good lazy nice chain */
136 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
137 /* 1 */ {4,    4,  8,    4, deflate_fast}, /* max speed, no lazy matches */
138 /* 2 */ {4,    5, 16,    8, deflate_fast},
139 /* 3 */ {4,    6, 32,   32, deflate_fast},
140 
141 /* 4 */ {4,    4, 16,   16, deflate_slow},  /* lazy matches */
142 /* 5 */ {8,   16, 32,   32, deflate_slow},
143 /* 6 */ {8,   16, 128, 128, deflate_slow},
144 /* 7 */ {8,   32, 128, 256, deflate_slow},
145 /* 8 */ {32, 128, 258, 1024, deflate_slow},
146 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
147 #endif
148 
149 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
150  * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
151  * meaning.
152  */
153 
154 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
155 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
156 
157 /* ===========================================================================
158  * Update a hash value with the given input byte
159  * IN  assertion: all calls to UPDATE_HASH are made with consecutive input
160  *    characters, so that a running hash key can be computed from the previous
161  *    key instead of complete recalculation each time.
162  */
163 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
164 
165 
166 /* ===========================================================================
167  * Insert string str in the dictionary and set match_head to the previous head
168  * of the hash chain (the most recent string with same hash key). Return
169  * the previous length of the hash chain.
170  * If this file is compiled with -DFASTEST, the compression level is forced
171  * to 1, and no hash chains are maintained.
172  * IN  assertion: all calls to INSERT_STRING are made with consecutive input
173  *    characters and the first MIN_MATCH bytes of str are valid (except for
174  *    the last MIN_MATCH-1 bytes of the input file).
175  */
176 #ifdef FASTEST
177 #define INSERT_STRING(s, str, match_head) \
178    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
179     match_head = s->head[s->ins_h], \
180     s->head[s->ins_h] = (Pos)(str))
181 #else
182 #define INSERT_STRING(s, str, match_head) \
183    (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
184     match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
185     s->head[s->ins_h] = (Pos)(str))
186 #endif
187 
188 /* ===========================================================================
189  * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
190  * prev[] will be initialized on the fly.
191  */
192 #define CLEAR_HASH(s) \
193     s->head[s->hash_size-1] = NIL; \
194     zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
195 
196 /* ===========================================================================
197  * Slide the hash table when sliding the window down (could be avoided with 32
198  * bit values at the expense of memory usage). We slide even when level == 0 to
199  * keep the hash table consistent if we switch back to level > 0 later.
200  */
201 local void slide_hash(
202     deflate_state *s)
203 {
204     unsigned n, m;
205     Posf *p;
206     uInt wsize = s->w_size;
207 
208     n = s->hash_size;
209     p = &s->head[n];
210     do {
211         m = *--p;
212         *p = (Pos)(m >= wsize ? m - wsize : NIL);
213     } while (--n);
214     n = wsize;
215 #ifndef FASTEST
216     p = &s->prev[n];
217     do {
218         m = *--p;
219         *p = (Pos)(m >= wsize ? m - wsize : NIL);
220         /* If n is not on any hash chain, prev[n] is garbage but
221          * its value will never be used.
222          */
223     } while (--n);
224 #endif
225 }
226 
227 /* ========================================================================= */
228 int ZEXPORT deflateInit_(
229     z_streamp strm,
230     int level,
231     const char *version,
232     int stream_size)
233 {
234     return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
235                          Z_DEFAULT_STRATEGY, version, stream_size);
236     /* To do: ignore strm->next_in if we use it as window */
237 }
238 
239 /* ========================================================================= */
240 int ZEXPORT deflateInit2_(
241     z_streamp strm,
242     int  level,
243     int  method,
244     int  windowBits,
245     int  memLevel,
246     int  strategy,
247     const char *version,
248     int stream_size)
249 {
250     deflate_state *s;
251     int wrap = 1;
252     static const char my_version[] = ZLIB_VERSION;
253 
254     ushf *overlay;
255     /* We overlay pending_buf and d_buf+l_buf. This works since the average
256      * output size for (length,distance) codes is <= 24 bits.
257      */
258 
259     if (version == Z_NULL || version[0] != my_version[0] ||
260         stream_size != sizeof(z_stream)) {
261         return Z_VERSION_ERROR;
262     }
263     if (strm == Z_NULL) return Z_STREAM_ERROR;
264 
265     strm->msg = Z_NULL;
266     if (strm->zalloc == (alloc_func)0) {
267 #ifdef Z_SOLO
268         return Z_STREAM_ERROR;
269 #else
270         strm->zalloc = zcalloc;
271         strm->opaque = (voidpf)0;
272 #endif
273     }
274     if (strm->zfree == (free_func)0)
275 #ifdef Z_SOLO
276         return Z_STREAM_ERROR;
277 #else
278         strm->zfree = zcfree;
279 #endif
280 
281 #ifdef FASTEST
282     if (level != 0) level = 1;
283 #else
284     if (level == Z_DEFAULT_COMPRESSION) level = 6;
285 #endif
286 
287     if (windowBits < 0) { /* suppress zlib wrapper */
288         wrap = 0;
289         windowBits = -windowBits;
290     }
291 #ifdef GZIP
292     else if (windowBits > 15) {
293         wrap = 2;       /* write gzip wrapper instead */
294         windowBits -= 16;
295     }
296 #endif
297     if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
298         windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
299         strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
300         return Z_STREAM_ERROR;
301     }
302     if (windowBits == 8) windowBits = 9;  /* until 256-byte window bug fixed */
303     s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
304     if (s == Z_NULL) return Z_MEM_ERROR;
305     strm->state = (struct internal_state FAR *)s;
306     s->strm = strm;
307     s->status = INIT_STATE;     /* to pass state test in deflateReset() */
308 
309     s->wrap = wrap;
310     s->gzhead = Z_NULL;
311     s->w_bits = (uInt)windowBits;
312     s->w_size = 1 << s->w_bits;
313     s->w_mask = s->w_size - 1;
314 
315     s->hash_bits = (uInt)memLevel + 7;
316     s->hash_size = 1 << s->hash_bits;
317     s->hash_mask = s->hash_size - 1;
318     s->hash_shift =  ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
319 
320     s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
321     s->prev   = (Posf *)  ZALLOC(strm, s->w_size, sizeof(Pos));
322     s->head   = (Posf *)  ZALLOC(strm, s->hash_size, sizeof(Pos));
323 
324     s->high_water = 0;      /* nothing written to s->window yet */
325 
326     s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
327 
328     overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
329     s->pending_buf = (uchf *) overlay;
330     s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
331 
332     if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
333         s->pending_buf == Z_NULL) {
334         s->status = FINISH_STATE;
335         strm->msg = ERR_MSG(Z_MEM_ERROR);
336         deflateEnd (strm);
337         return Z_MEM_ERROR;
338     }
339     s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
340     s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
341 
342     s->level = level;
343     s->strategy = strategy;
344     s->method = (Byte)method;
345 
346     return deflateReset(strm);
347 }
348 
349 /* =========================================================================
350  * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
351  */
352 local int deflateStateCheck (
353     z_streamp strm)
354 {
355     deflate_state *s;
356     if (strm == Z_NULL ||
357         strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
358         return 1;
359     s = strm->state;
360     if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
361 #ifdef GZIP
362                                            s->status != GZIP_STATE &&
363 #endif
364                                            s->status != EXTRA_STATE &&
365                                            s->status != NAME_STATE &&
366                                            s->status != COMMENT_STATE &&
367                                            s->status != HCRC_STATE &&
368                                            s->status != BUSY_STATE &&
369                                            s->status != FINISH_STATE))
370         return 1;
371     return 0;
372 }
373 
374 /* ========================================================================= */
375 int ZEXPORT deflateSetDictionary (
376     z_streamp strm,
377     const Bytef *dictionary,
378     uInt  dictLength)
379 {
380     deflate_state *s;
381     uInt str, n;
382     int wrap;
383     unsigned avail;
384     z_const unsigned char *next;
385 
386     if (deflateStateCheck(strm) || dictionary == Z_NULL)
387         return Z_STREAM_ERROR;
388     s = strm->state;
389     wrap = s->wrap;
390     if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
391         return Z_STREAM_ERROR;
392 
393     /* when using zlib wrappers, compute Adler-32 for provided dictionary */
394     if (wrap == 1)
395         strm->adler = adler32(strm->adler, dictionary, dictLength);
396     s->wrap = 0;                    /* avoid computing Adler-32 in read_buf */
397 
398     /* if dictionary would fill window, just replace the history */
399     if (dictLength >= s->w_size) {
400         if (wrap == 0) {            /* already empty otherwise */
401             CLEAR_HASH(s);
402             s->strstart = 0;
403             s->block_start = 0L;
404             s->insert = 0;
405         }
406         dictionary += dictLength - s->w_size;  /* use the tail */
407         dictLength = s->w_size;
408     }
409 
410     /* insert dictionary into window and hash */
411     avail = strm->avail_in;
412     next = strm->next_in;
413     strm->avail_in = dictLength;
414     strm->next_in = (z_const Bytef *)dictionary;
415     fill_window(s);
416     while (s->lookahead >= MIN_MATCH) {
417         str = s->strstart;
418         n = s->lookahead - (MIN_MATCH-1);
419         do {
420             UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
421 #ifndef FASTEST
422             s->prev[str & s->w_mask] = s->head[s->ins_h];
423 #endif
424             s->head[s->ins_h] = (Pos)str;
425             str++;
426         } while (--n);
427         s->strstart = str;
428         s->lookahead = MIN_MATCH-1;
429         fill_window(s);
430     }
431     s->strstart += s->lookahead;
432     s->block_start = (long)s->strstart;
433     s->insert = s->lookahead;
434     s->lookahead = 0;
435     s->match_length = s->prev_length = MIN_MATCH-1;
436     s->match_available = 0;
437     strm->next_in = next;
438     strm->avail_in = avail;
439     s->wrap = wrap;
440     return Z_OK;
441 }
442 
443 /* ========================================================================= */
444 int ZEXPORT deflateGetDictionary (
445     z_streamp strm,
446     Bytef *dictionary,
447     uInt  *dictLength)
448 {
449     deflate_state *s;
450     uInt len;
451 
452     if (deflateStateCheck(strm))
453         return Z_STREAM_ERROR;
454     s = strm->state;
455     len = s->strstart + s->lookahead;
456     if (len > s->w_size)
457         len = s->w_size;
458     if (dictionary != Z_NULL && len)
459         zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
460     if (dictLength != Z_NULL)
461         *dictLength = len;
462     return Z_OK;
463 }
464 
465 /* ========================================================================= */
466 int ZEXPORT deflateResetKeep (
467     z_streamp strm)
468 {
469     deflate_state *s;
470 
471     if (deflateStateCheck(strm)) {
472         return Z_STREAM_ERROR;
473     }
474 
475     strm->total_in = strm->total_out = 0;
476     strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
477     strm->data_type = Z_UNKNOWN;
478 
479     s = (deflate_state *)strm->state;
480     s->pending = 0;
481     s->pending_out = s->pending_buf;
482 
483     if (s->wrap < 0) {
484         s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
485     }
486     s->status =
487 #ifdef GZIP
488         s->wrap == 2 ? GZIP_STATE :
489 #endif
490         s->wrap ? INIT_STATE : BUSY_STATE;
491     strm->adler =
492 #ifdef GZIP
493         s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
494 #endif
495         adler32(0L, Z_NULL, 0);
496     s->last_flush = Z_NO_FLUSH;
497 
498     _tr_init(s);
499 
500     return Z_OK;
501 }
502 
503 /* ========================================================================= */
504 int ZEXPORT deflateReset (
505     z_streamp strm)
506 {
507     int ret;
508 
509     ret = deflateResetKeep(strm);
510     if (ret == Z_OK)
511         lm_init(strm->state);
512     return ret;
513 }
514 
515 /* ========================================================================= */
516 int ZEXPORT deflateSetHeader (
517     z_streamp strm,
518     gz_headerp head)
519 {
520     if (deflateStateCheck(strm) || strm->state->wrap != 2)
521         return Z_STREAM_ERROR;
522     strm->state->gzhead = head;
523     return Z_OK;
524 }
525 
526 /* ========================================================================= */
527 int ZEXPORT deflatePending (
528     z_streamp strm,
529     unsigned *pending,
530     int *bits)
531 {
532     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
533     if (pending != Z_NULL)
534         *pending = strm->state->pending;
535     if (bits != Z_NULL)
536         *bits = strm->state->bi_valid;
537     return Z_OK;
538 }
539 
540 /* ========================================================================= */
541 int ZEXPORT deflatePrime (
542     z_streamp strm,
543     int bits,
544     int value)
545 {
546     deflate_state *s;
547     int put;
548 
549     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
550     s = strm->state;
551     if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
552         return Z_BUF_ERROR;
553     do {
554         put = Buf_size - s->bi_valid;
555         if (put > bits)
556             put = bits;
557         s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
558         s->bi_valid += put;
559         _tr_flush_bits(s);
560         value >>= put;
561         bits -= put;
562     } while (bits);
563     return Z_OK;
564 }
565 
566 /* ========================================================================= */
567 int ZEXPORT deflateParams(
568     z_streamp strm,
569     int level,
570     int strategy)
571 {
572     deflate_state *s;
573     compress_func func;
574 
575     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
576     s = strm->state;
577 
578 #ifdef FASTEST
579     if (level != 0) level = 1;
580 #else
581     if (level == Z_DEFAULT_COMPRESSION) level = 6;
582 #endif
583     if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
584         return Z_STREAM_ERROR;
585     }
586     func = configuration_table[s->level].func;
587 
588     if ((strategy != s->strategy || func != configuration_table[level].func) &&
589         s->high_water) {
590         /* Flush the last buffer: */
591         int err = deflate(strm, Z_BLOCK);
592         if (err == Z_STREAM_ERROR)
593             return err;
594         if (strm->avail_out == 0)
595             return Z_BUF_ERROR;
596     }
597     if (s->level != level) {
598         if (s->level == 0 && s->matches != 0) {
599             if (s->matches == 1)
600                 slide_hash(s);
601             else
602                 CLEAR_HASH(s);
603             s->matches = 0;
604         }
605         s->level = level;
606         s->max_lazy_match   = configuration_table[level].max_lazy;
607         s->good_match       = configuration_table[level].good_length;
608         s->nice_match       = configuration_table[level].nice_length;
609         s->max_chain_length = configuration_table[level].max_chain;
610     }
611     s->strategy = strategy;
612     return Z_OK;
613 }
614 
615 /* ========================================================================= */
616 int ZEXPORT deflateTune(
617     z_streamp strm,
618     int good_length,
619     int max_lazy,
620     int nice_length,
621     int max_chain)
622 {
623     deflate_state *s;
624 
625     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
626     s = strm->state;
627     s->good_match = (uInt)good_length;
628     s->max_lazy_match = (uInt)max_lazy;
629     s->nice_match = nice_length;
630     s->max_chain_length = (uInt)max_chain;
631     return Z_OK;
632 }
633 
634 /* =========================================================================
635  * For the default windowBits of 15 and memLevel of 8, this function returns
636  * a close to exact, as well as small, upper bound on the compressed size.
637  * They are coded as constants here for a reason--if the #define's are
638  * changed, then this function needs to be changed as well.  The return
639  * value for 15 and 8 only works for those exact settings.
640  *
641  * For any setting other than those defaults for windowBits and memLevel,
642  * the value returned is a conservative worst case for the maximum expansion
643  * resulting from using fixed blocks instead of stored blocks, which deflate
644  * can emit on compressed data for some combinations of the parameters.
645  *
646  * This function could be more sophisticated to provide closer upper bounds for
647  * every combination of windowBits and memLevel.  But even the conservative
648  * upper bound of about 14% expansion does not seem onerous for output buffer
649  * allocation.
650  */
651 uLong ZEXPORT deflateBound(
652     z_streamp strm,
653     uLong sourceLen)
654 {
655     deflate_state *s;
656     uLong complen, wraplen;
657 
658     /* conservative upper bound for compressed data */
659     complen = sourceLen +
660               ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
661 
662     /* if can't get parameters, return conservative bound plus zlib wrapper */
663     if (deflateStateCheck(strm))
664         return complen + 6;
665 
666     /* compute wrapper length */
667     s = strm->state;
668     switch (s->wrap) {
669     case 0:                                 /* raw deflate */
670         wraplen = 0;
671         break;
672     case 1:                                 /* zlib wrapper */
673         wraplen = 6 + (s->strstart ? 4 : 0);
674         break;
675 #ifdef GZIP
676     case 2:                                 /* gzip wrapper */
677         wraplen = 18;
678         if (s->gzhead != Z_NULL) {          /* user-supplied gzip header */
679             Bytef *str;
680             if (s->gzhead->extra != Z_NULL)
681                 wraplen += 2 + s->gzhead->extra_len;
682             str = s->gzhead->name;
683             if (str != Z_NULL)
684                 do {
685                     wraplen++;
686                 } while (*str++);
687             str = s->gzhead->comment;
688             if (str != Z_NULL)
689                 do {
690                     wraplen++;
691                 } while (*str++);
692             if (s->gzhead->hcrc)
693                 wraplen += 2;
694         }
695         break;
696 #endif
697     default:                                /* for compiler happiness */
698         wraplen = 6;
699     }
700 
701     /* if not default parameters, return conservative bound */
702     if (s->w_bits != 15 || s->hash_bits != 8 + 7)
703         return complen + wraplen;
704 
705     /* default settings: return tight bound for that case */
706     return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
707            (sourceLen >> 25) + 13 - 6 + wraplen;
708 }
709 
710 /* =========================================================================
711  * Put a short in the pending buffer. The 16-bit value is put in MSB order.
712  * IN assertion: the stream state is correct and there is enough room in
713  * pending_buf.
714  */
715 local void putShortMSB (
716     deflate_state *s,
717     uInt b)
718 {
719     put_byte(s, (Byte)(b >> 8));
720     put_byte(s, (Byte)(b & 0xff));
721 }
722 
723 /* =========================================================================
724  * Flush as much pending output as possible. All deflate() output, except for
725  * some deflate_stored() output, goes through this function so some
726  * applications may wish to modify it to avoid allocating a large
727  * strm->next_out buffer and copying into it. (See also read_buf()).
728  */
729 local void flush_pending(
730     z_streamp strm)
731 {
732     unsigned len;
733     deflate_state *s = strm->state;
734 
735     _tr_flush_bits(s);
736     len = s->pending;
737     if (len > strm->avail_out) len = strm->avail_out;
738     if (len == 0) return;
739 
740     zmemcpy(strm->next_out, s->pending_out, len);
741     strm->next_out  += len;
742     s->pending_out  += len;
743     strm->total_out += len;
744     strm->avail_out -= len;
745     s->pending      -= len;
746     if (s->pending == 0) {
747         s->pending_out = s->pending_buf;
748     }
749 }
750 
751 /* ===========================================================================
752  * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
753  */
754 #define HCRC_UPDATE(beg) \
755     do { \
756         if (s->gzhead->hcrc && s->pending > (beg)) \
757             strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
758                                 s->pending - (beg)); \
759     } while (0)
760 
761 /* ========================================================================= */
762 int ZEXPORT deflate (
763     z_streamp strm,
764     int flush)
765 {
766     int old_flush; /* value of flush param for previous deflate call */
767     deflate_state *s;
768 
769     if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
770         return Z_STREAM_ERROR;
771     }
772     s = strm->state;
773 
774     if (strm->next_out == Z_NULL ||
775         (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
776         (s->status == FINISH_STATE && flush != Z_FINISH)) {
777         ERR_RETURN(strm, Z_STREAM_ERROR);
778     }
779     if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
780 
781     old_flush = s->last_flush;
782     s->last_flush = flush;
783 
784     /* Flush as much pending output as possible */
785     if (s->pending != 0) {
786         flush_pending(strm);
787         if (strm->avail_out == 0) {
788             /* Since avail_out is 0, deflate will be called again with
789              * more output space, but possibly with both pending and
790              * avail_in equal to zero. There won't be anything to do,
791              * but this is not an error situation so make sure we
792              * return OK instead of BUF_ERROR at next call of deflate:
793              */
794             s->last_flush = -1;
795             return Z_OK;
796         }
797 
798     /* Make sure there is something to do and avoid duplicate consecutive
799      * flushes. For repeated and useless calls with Z_FINISH, we keep
800      * returning Z_STREAM_END instead of Z_BUF_ERROR.
801      */
802     } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
803                flush != Z_FINISH) {
804         ERR_RETURN(strm, Z_BUF_ERROR);
805     }
806 
807     /* User must not provide more input after the first FINISH: */
808     if (s->status == FINISH_STATE && strm->avail_in != 0) {
809         ERR_RETURN(strm, Z_BUF_ERROR);
810     }
811 
812     /* Write the header */
813     if (s->status == INIT_STATE) {
814         /* zlib header */
815         uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
816         uInt level_flags;
817 
818         if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
819             level_flags = 0;
820         else if (s->level < 6)
821             level_flags = 1;
822         else if (s->level == 6)
823             level_flags = 2;
824         else
825             level_flags = 3;
826         header |= (level_flags << 6);
827         if (s->strstart != 0) header |= PRESET_DICT;
828         header += 31 - (header % 31);
829 
830         putShortMSB(s, header);
831 
832         /* Save the adler32 of the preset dictionary: */
833         if (s->strstart != 0) {
834             putShortMSB(s, (uInt)(strm->adler >> 16));
835             putShortMSB(s, (uInt)(strm->adler & 0xffff));
836         }
837         strm->adler = adler32(0L, Z_NULL, 0);
838         s->status = BUSY_STATE;
839 
840         /* Compression must start with an empty pending buffer */
841         flush_pending(strm);
842         if (s->pending != 0) {
843             s->last_flush = -1;
844             return Z_OK;
845         }
846     }
847 #ifdef GZIP
848     if (s->status == GZIP_STATE) {
849         /* gzip header */
850         strm->adler = crc32(0L, Z_NULL, 0);
851         put_byte(s, 31);
852         put_byte(s, 139);
853         put_byte(s, 8);
854         if (s->gzhead == Z_NULL) {
855             put_byte(s, 0);
856             put_byte(s, 0);
857             put_byte(s, 0);
858             put_byte(s, 0);
859             put_byte(s, 0);
860             put_byte(s, s->level == 9 ? 2 :
861                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
862                       4 : 0));
863             put_byte(s, OS_CODE);
864             s->status = BUSY_STATE;
865 
866             /* Compression must start with an empty pending buffer */
867             flush_pending(strm);
868             if (s->pending != 0) {
869                 s->last_flush = -1;
870                 return Z_OK;
871             }
872         }
873         else {
874             put_byte(s, (s->gzhead->text ? 1 : 0) +
875                      (s->gzhead->hcrc ? 2 : 0) +
876                      (s->gzhead->extra == Z_NULL ? 0 : 4) +
877                      (s->gzhead->name == Z_NULL ? 0 : 8) +
878                      (s->gzhead->comment == Z_NULL ? 0 : 16)
879                      );
880             put_byte(s, (Byte)(s->gzhead->time & 0xff));
881             put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
882             put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
883             put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
884             put_byte(s, s->level == 9 ? 2 :
885                      (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
886                       4 : 0));
887             put_byte(s, s->gzhead->os & 0xff);
888             if (s->gzhead->extra != Z_NULL) {
889                 put_byte(s, s->gzhead->extra_len & 0xff);
890                 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
891             }
892             if (s->gzhead->hcrc)
893                 strm->adler = crc32(strm->adler, s->pending_buf,
894                                     s->pending);
895             s->gzindex = 0;
896             s->status = EXTRA_STATE;
897         }
898     }
899     if (s->status == EXTRA_STATE) {
900         if (s->gzhead->extra != Z_NULL) {
901             ulg beg = s->pending;   /* start of bytes to update crc */
902             uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
903             while (s->pending + left > s->pending_buf_size) {
904                 uInt copy = s->pending_buf_size - s->pending;
905                 zmemcpy(s->pending_buf + s->pending,
906                         s->gzhead->extra + s->gzindex, copy);
907                 s->pending = s->pending_buf_size;
908                 HCRC_UPDATE(beg);
909                 s->gzindex += copy;
910                 flush_pending(strm);
911                 if (s->pending != 0) {
912                     s->last_flush = -1;
913                     return Z_OK;
914                 }
915                 beg = 0;
916                 left -= copy;
917             }
918             zmemcpy(s->pending_buf + s->pending,
919                     s->gzhead->extra + s->gzindex, left);
920             s->pending += left;
921             HCRC_UPDATE(beg);
922             s->gzindex = 0;
923         }
924         s->status = NAME_STATE;
925     }
926     if (s->status == NAME_STATE) {
927         if (s->gzhead->name != Z_NULL) {
928             ulg beg = s->pending;   /* start of bytes to update crc */
929             int val;
930             do {
931                 if (s->pending == s->pending_buf_size) {
932                     HCRC_UPDATE(beg);
933                     flush_pending(strm);
934                     if (s->pending != 0) {
935                         s->last_flush = -1;
936                         return Z_OK;
937                     }
938                     beg = 0;
939                 }
940                 val = s->gzhead->name[s->gzindex++];
941                 put_byte(s, val);
942             } while (val != 0);
943             HCRC_UPDATE(beg);
944             s->gzindex = 0;
945         }
946         s->status = COMMENT_STATE;
947     }
948     if (s->status == COMMENT_STATE) {
949         if (s->gzhead->comment != Z_NULL) {
950             ulg beg = s->pending;   /* start of bytes to update crc */
951             int val;
952             do {
953                 if (s->pending == s->pending_buf_size) {
954                     HCRC_UPDATE(beg);
955                     flush_pending(strm);
956                     if (s->pending != 0) {
957                         s->last_flush = -1;
958                         return Z_OK;
959                     }
960                     beg = 0;
961                 }
962                 val = s->gzhead->comment[s->gzindex++];
963                 put_byte(s, val);
964             } while (val != 0);
965             HCRC_UPDATE(beg);
966         }
967         s->status = HCRC_STATE;
968     }
969     if (s->status == HCRC_STATE) {
970         if (s->gzhead->hcrc) {
971             if (s->pending + 2 > s->pending_buf_size) {
972                 flush_pending(strm);
973                 if (s->pending != 0) {
974                     s->last_flush = -1;
975                     return Z_OK;
976                 }
977             }
978             put_byte(s, (Byte)(strm->adler & 0xff));
979             put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
980             strm->adler = crc32(0L, Z_NULL, 0);
981         }
982         s->status = BUSY_STATE;
983 
984         /* Compression must start with an empty pending buffer */
985         flush_pending(strm);
986         if (s->pending != 0) {
987             s->last_flush = -1;
988             return Z_OK;
989         }
990     }
991 #endif
992 
993     /* Start a new block or continue the current one.
994      */
995     if (strm->avail_in != 0 || s->lookahead != 0 ||
996         (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
997         block_state bstate;
998 
999         bstate = s->level == 0 ? deflate_stored(s, flush) :
1000                  s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1001                  s->strategy == Z_RLE ? deflate_rle(s, flush) :
1002                  (*(configuration_table[s->level].func))(s, flush);
1003 
1004         if (bstate == finish_started || bstate == finish_done) {
1005             s->status = FINISH_STATE;
1006         }
1007         if (bstate == need_more || bstate == finish_started) {
1008             if (strm->avail_out == 0) {
1009                 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1010             }
1011             return Z_OK;
1012             /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1013              * of deflate should use the same flush parameter to make sure
1014              * that the flush is complete. So we don't have to output an
1015              * empty block here, this will be done at next call. This also
1016              * ensures that for a very small output buffer, we emit at most
1017              * one empty block.
1018              */
1019         }
1020         if (bstate == block_done) {
1021             if (flush == Z_PARTIAL_FLUSH) {
1022                 _tr_align(s);
1023             } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1024                 _tr_stored_block(s, (char*)0, 0L, 0);
1025                 /* For a full flush, this empty block will be recognized
1026                  * as a special marker by inflate_sync().
1027                  */
1028                 if (flush == Z_FULL_FLUSH) {
1029                     CLEAR_HASH(s);             /* forget history */
1030                     if (s->lookahead == 0) {
1031                         s->strstart = 0;
1032                         s->block_start = 0L;
1033                         s->insert = 0;
1034                     }
1035                 }
1036             }
1037             flush_pending(strm);
1038             if (strm->avail_out == 0) {
1039               s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1040               return Z_OK;
1041             }
1042         }
1043     }
1044 
1045     if (flush != Z_FINISH) return Z_OK;
1046     if (s->wrap <= 0) return Z_STREAM_END;
1047 
1048     /* Write the trailer */
1049 #ifdef GZIP
1050     if (s->wrap == 2) {
1051         put_byte(s, (Byte)(strm->adler & 0xff));
1052         put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1053         put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1054         put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1055         put_byte(s, (Byte)(strm->total_in & 0xff));
1056         put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1057         put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1058         put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1059     }
1060     else
1061 #endif
1062     {
1063         putShortMSB(s, (uInt)(strm->adler >> 16));
1064         putShortMSB(s, (uInt)(strm->adler & 0xffff));
1065     }
1066     flush_pending(strm);
1067     /* If avail_out is zero, the application will call deflate again
1068      * to flush the rest.
1069      */
1070     if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1071     return s->pending != 0 ? Z_OK : Z_STREAM_END;
1072 }
1073 
1074 /* ========================================================================= */
1075 int ZEXPORT deflateEnd (
1076     z_streamp strm)
1077 {
1078     int status;
1079 
1080     if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1081 
1082     status = strm->state->status;
1083 
1084     /* Deallocate in reverse order of allocations: */
1085     TRY_FREE(strm, strm->state->pending_buf);
1086     TRY_FREE(strm, strm->state->head);
1087     TRY_FREE(strm, strm->state->prev);
1088     TRY_FREE(strm, strm->state->window);
1089 
1090     ZFREE(strm, strm->state);
1091     strm->state = Z_NULL;
1092 
1093     return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1094 }
1095 
1096 /* =========================================================================
1097  * Copy the source state to the destination state.
1098  * To simplify the source, this is not supported for 16-bit MSDOS (which
1099  * doesn't have enough memory anyway to duplicate compression states).
1100  */
1101 int ZEXPORT deflateCopy (
1102     z_streamp dest,
1103     z_streamp source)
1104 {
1105 #ifdef MAXSEG_64K
1106     return Z_STREAM_ERROR;
1107 #else
1108     deflate_state *ds;
1109     deflate_state *ss;
1110     ushf *overlay;
1111 
1112 
1113     if (deflateStateCheck(source) || dest == Z_NULL) {
1114         return Z_STREAM_ERROR;
1115     }
1116 
1117     ss = source->state;
1118 
1119     zmemcpy((Bytef*)dest, (Bytef*)source, sizeof(z_stream));
1120 
1121     ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1122     if (ds == Z_NULL) return Z_MEM_ERROR;
1123     dest->state = (struct internal_state FAR *) ds;
1124     zmemcpy((Bytef*)ds, (Bytef*)ss, sizeof(deflate_state));
1125     ds->strm = dest;
1126 
1127     ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1128     ds->prev   = (Posf *)  ZALLOC(dest, ds->w_size, sizeof(Pos));
1129     ds->head   = (Posf *)  ZALLOC(dest, ds->hash_size, sizeof(Pos));
1130     overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1131     ds->pending_buf = (uchf *) overlay;
1132 
1133     if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1134         ds->pending_buf == Z_NULL) {
1135         deflateEnd (dest);
1136         return Z_MEM_ERROR;
1137     }
1138     /* following zmemcpy do not work for 16-bit MSDOS */
1139     zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1140     zmemcpy((Bytef*)ds->prev, (Bytef*)ss->prev, ds->w_size * sizeof(Pos));
1141     zmemcpy((Bytef*)ds->head, (Bytef*)ss->head, ds->hash_size * sizeof(Pos));
1142     zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1143 
1144     ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1145     ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1146     ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1147 
1148     ds->l_desc.dyn_tree = ds->dyn_ltree;
1149     ds->d_desc.dyn_tree = ds->dyn_dtree;
1150     ds->bl_desc.dyn_tree = ds->bl_tree;
1151 
1152     return Z_OK;
1153 #endif /* MAXSEG_64K */
1154 }
1155 
1156 /* ===========================================================================
1157  * Read a new buffer from the current input stream, update the adler32
1158  * and total number of bytes read.  All deflate() input goes through
1159  * this function so some applications may wish to modify it to avoid
1160  * allocating a large strm->next_in buffer and copying from it.
1161  * (See also flush_pending()).
1162  */
1163 local unsigned read_buf(
1164     z_streamp strm,
1165     Bytef *buf,
1166     unsigned size)
1167 {
1168     unsigned len = strm->avail_in;
1169 
1170     if (len > size) len = size;
1171     if (len == 0) return 0;
1172 
1173     strm->avail_in  -= len;
1174 
1175     zmemcpy(buf, strm->next_in, len);
1176     if (strm->state->wrap == 1) {
1177         strm->adler = adler32(strm->adler, buf, len);
1178     }
1179 #ifdef GZIP
1180     else if (strm->state->wrap == 2) {
1181         strm->adler = crc32(strm->adler, buf, len);
1182     }
1183 #endif
1184     strm->next_in  += len;
1185     strm->total_in += len;
1186 
1187     return len;
1188 }
1189 
1190 /* ===========================================================================
1191  * Initialize the "longest match" routines for a new zlib stream
1192  */
1193 local void lm_init (
1194     deflate_state *s)
1195 {
1196     s->window_size = (ulg)2L*s->w_size;
1197 
1198     CLEAR_HASH(s);
1199 
1200     /* Set the default configuration parameters:
1201      */
1202     s->max_lazy_match   = configuration_table[s->level].max_lazy;
1203     s->good_match       = configuration_table[s->level].good_length;
1204     s->nice_match       = configuration_table[s->level].nice_length;
1205     s->max_chain_length = configuration_table[s->level].max_chain;
1206 
1207     s->strstart = 0;
1208     s->block_start = 0L;
1209     s->lookahead = 0;
1210     s->insert = 0;
1211     s->match_length = s->prev_length = MIN_MATCH-1;
1212     s->match_available = 0;
1213     s->ins_h = 0;
1214 #ifndef FASTEST
1215 #ifdef ASMV
1216     match_init(); /* initialize the asm code */
1217 #endif
1218 #endif
1219 }
1220 
1221 #ifndef FASTEST
1222 /* ===========================================================================
1223  * Set match_start to the longest match starting at the given string and
1224  * return its length. Matches shorter or equal to prev_length are discarded,
1225  * in which case the result is equal to prev_length and match_start is
1226  * garbage.
1227  * IN assertions: cur_match is the head of the hash chain for the current
1228  *   string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1229  * OUT assertion: the match length is not greater than s->lookahead.
1230  */
1231 #ifndef ASMV
1232 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1233  * match.S. The code will be functionally equivalent.
1234  */
1235 local uInt longest_match(
1236     deflate_state *s,
1237     IPos cur_match)
1238 {
1239     unsigned chain_length = s->max_chain_length;/* max hash chain length */
1240     register Bytef *scan = s->window + s->strstart; /* current string */
1241     register Bytef *match;                      /* matched string */
1242     register int len;                           /* length of current match */
1243     int best_len = (int)s->prev_length;         /* best match length so far */
1244     int nice_match = s->nice_match;             /* stop if match long enough */
1245     IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1246         s->strstart - (IPos)MAX_DIST(s) : NIL;
1247     /* Stop when cur_match becomes <= limit. To simplify the code,
1248      * we prevent matches with the string of window index 0.
1249      */
1250     Posf *prev = s->prev;
1251     uInt wmask = s->w_mask;
1252 
1253 #ifdef UNALIGNED_OK
1254     /* Compare two bytes at a time. Note: this is not always beneficial.
1255      * Try with and without -DUNALIGNED_OK to check.
1256      */
1257     register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1258     register ush scan_start = *(ushf*)scan;
1259     register ush scan_end   = *(ushf*)(scan+best_len-1);
1260 #else
1261     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1262     register Byte scan_end1  = scan[best_len-1];
1263     register Byte scan_end   = scan[best_len];
1264 #endif
1265 
1266     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1267      * It is easy to get rid of this optimization if necessary.
1268      */
1269     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1270 
1271     /* Do not waste too much time if we already have a good match: */
1272     if (s->prev_length >= s->good_match) {
1273         chain_length >>= 2;
1274     }
1275     /* Do not look for matches beyond the end of the input. This is necessary
1276      * to make deflate deterministic.
1277      */
1278     if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1279 
1280     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1281 
1282     do {
1283         Assert(cur_match < s->strstart, "no future");
1284         match = s->window + cur_match;
1285 
1286         /* Skip to next match if the match length cannot increase
1287          * or if the match length is less than 2.  Note that the checks below
1288          * for insufficient lookahead only occur occasionally for performance
1289          * reasons.  Therefore uninitialized memory will be accessed, and
1290          * conditional jumps will be made that depend on those values.
1291          * However the length of the match is limited to the lookahead, so
1292          * the output of deflate is not affected by the uninitialized values.
1293          */
1294 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1295         /* This code assumes sizeof(unsigned short) == 2. Do not use
1296          * UNALIGNED_OK if your compiler uses a different size.
1297          */
1298         if (*(ushf*)(match+best_len-1) != scan_end ||
1299             *(ushf*)match != scan_start) continue;
1300 
1301         /* It is not necessary to compare scan[2] and match[2] since they are
1302          * always equal when the other bytes match, given that the hash keys
1303          * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1304          * strstart+3, +5, ... up to strstart+257. We check for insufficient
1305          * lookahead only every 4th comparison; the 128th check will be made
1306          * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1307          * necessary to put more guard bytes at the end of the window, or
1308          * to check more often for insufficient lookahead.
1309          */
1310         Assert(scan[2] == match[2], "scan[2]?");
1311         scan++, match++;
1312         do {
1313         } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1314                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1315                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1316                  *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1317                  scan < strend);
1318         /* The funny "do {}" generates better code on most compilers */
1319 
1320         /* Here, scan <= window+strstart+257 */
1321         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1322         if (*scan == *match) scan++;
1323 
1324         len = (MAX_MATCH - 1) - (int)(strend-scan);
1325         scan = strend - (MAX_MATCH-1);
1326 
1327 #else /* UNALIGNED_OK */
1328 
1329         if (match[best_len]   != scan_end  ||
1330             match[best_len-1] != scan_end1 ||
1331             *match            != *scan     ||
1332             *++match          != scan[1])      continue;
1333 
1334         /* The check at best_len-1 can be removed because it will be made
1335          * again later. (This heuristic is not always a win.)
1336          * It is not necessary to compare scan[2] and match[2] since they
1337          * are always equal when the other bytes match, given that
1338          * the hash keys are equal and that HASH_BITS >= 8.
1339          */
1340         scan += 2, match++;
1341         Assert(*scan == *match, "match[2]?");
1342 
1343         /* We check for insufficient lookahead only every 8th comparison;
1344          * the 256th check will be made at strstart+258.
1345          */
1346         do {
1347         } while (*++scan == *++match && *++scan == *++match &&
1348                  *++scan == *++match && *++scan == *++match &&
1349                  *++scan == *++match && *++scan == *++match &&
1350                  *++scan == *++match && *++scan == *++match &&
1351                  scan < strend);
1352 
1353         Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1354 
1355         len = MAX_MATCH - (int)(strend - scan);
1356         scan = strend - MAX_MATCH;
1357 
1358 #endif /* UNALIGNED_OK */
1359 
1360         if (len > best_len) {
1361             s->match_start = cur_match;
1362             best_len = len;
1363             if (len >= nice_match) break;
1364 #ifdef UNALIGNED_OK
1365             scan_end = *(ushf*)(scan+best_len-1);
1366 #else
1367             scan_end1  = scan[best_len-1];
1368             scan_end   = scan[best_len];
1369 #endif
1370         }
1371     } while ((cur_match = prev[cur_match & wmask]) > limit
1372              && --chain_length != 0);
1373 
1374     if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1375     return s->lookahead;
1376 }
1377 #endif /* ASMV */
1378 
1379 #else /* FASTEST */
1380 
1381 /* ---------------------------------------------------------------------------
1382  * Optimized version for FASTEST only
1383  */
1384 local uInt longest_match(
1385     deflate_state *s,
1386     IPos cur_match)
1387 {
1388     register Bytef *scan = s->window + s->strstart; /* current string */
1389     register Bytef *match;                       /* matched string */
1390     register int len;                           /* length of current match */
1391     register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1392 
1393     /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1394      * It is easy to get rid of this optimization if necessary.
1395      */
1396     Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1397 
1398     Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1399 
1400     Assert(cur_match < s->strstart, "no future");
1401 
1402     match = s->window + cur_match;
1403 
1404     /* Return failure if the match length is less than 2:
1405      */
1406     if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1407 
1408     /* The check at best_len-1 can be removed because it will be made
1409      * again later. (This heuristic is not always a win.)
1410      * It is not necessary to compare scan[2] and match[2] since they
1411      * are always equal when the other bytes match, given that
1412      * the hash keys are equal and that HASH_BITS >= 8.
1413      */
1414     scan += 2, match += 2;
1415     Assert(*scan == *match, "match[2]?");
1416 
1417     /* We check for insufficient lookahead only every 8th comparison;
1418      * the 256th check will be made at strstart+258.
1419      */
1420     do {
1421     } while (*++scan == *++match && *++scan == *++match &&
1422              *++scan == *++match && *++scan == *++match &&
1423              *++scan == *++match && *++scan == *++match &&
1424              *++scan == *++match && *++scan == *++match &&
1425              scan < strend);
1426 
1427     Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1428 
1429     len = MAX_MATCH - (int)(strend - scan);
1430 
1431     if (len < MIN_MATCH) return MIN_MATCH - 1;
1432 
1433     s->match_start = cur_match;
1434     return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1435 }
1436 
1437 #endif /* FASTEST */
1438 
1439 #ifdef ZLIB_DEBUG
1440 
1441 #define EQUAL 0
1442 /* result of memcmp for equal strings */
1443 
1444 /* ===========================================================================
1445  * Check that the match at match_start is indeed a match.
1446  */
1447 local void check_match(
1448     deflate_state *s,
1449     IPos start,
1450     IPos match,
1451     int length)
1452 {
1453     /* check that the match is indeed a match */
1454     if (zmemcmp(s->window + match,
1455                 s->window + start, length) != EQUAL) {
1456         fprintf(stderr, " start %u, match %u, length %d\n",
1457                 start, match, length);
1458         do {
1459             fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1460         } while (--length != 0);
1461         z_error("invalid match");
1462     }
1463     if (z_verbose > 1) {
1464         fprintf(stderr,"\\[%d,%d]", start-match, length);
1465         do { putc(s->window[start++], stderr); } while (--length != 0);
1466     }
1467 }
1468 #else
1469 #  define check_match(s, start, match, length)
1470 #endif /* ZLIB_DEBUG */
1471 
1472 /* ===========================================================================
1473  * Fill the window when the lookahead becomes insufficient.
1474  * Updates strstart and lookahead.
1475  *
1476  * IN assertion: lookahead < MIN_LOOKAHEAD
1477  * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1478  *    At least one byte has been read, or avail_in == 0; reads are
1479  *    performed for at least two bytes (required for the zip translate_eol
1480  *    option -- not supported here).
1481  */
1482 local void fill_window(
1483     deflate_state *s)
1484 {
1485     unsigned n;
1486     unsigned more;    /* Amount of free space at the end of the window. */
1487     uInt wsize = s->w_size;
1488 
1489     Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1490 
1491     do {
1492         more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1493 
1494         /* Deal with !@#$% 64K limit: */
1495         if (sizeof(int) <= 2) {
1496             if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1497                 more = wsize;
1498 
1499             } else if (more == (unsigned)(-1)) {
1500                 /* Very unlikely, but possible on 16 bit machine if
1501                  * strstart == 0 && lookahead == 1 (input done a byte at time)
1502                  */
1503                 more--;
1504             }
1505         }
1506 
1507         /* If the window is almost full and there is insufficient lookahead,
1508          * move the upper half to the lower one to make room in the upper half.
1509          */
1510         if (s->strstart >= wsize+MAX_DIST(s)) {
1511 
1512             zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1513             s->match_start -= wsize;
1514             s->strstart    -= wsize; /* we now have strstart >= MAX_DIST */
1515             s->block_start -= (long) wsize;
1516             slide_hash(s);
1517             more += wsize;
1518         }
1519         if (s->strm->avail_in == 0) break;
1520 
1521         /* If there was no sliding:
1522          *    strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1523          *    more == window_size - lookahead - strstart
1524          * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1525          * => more >= window_size - 2*WSIZE + 2
1526          * In the BIG_MEM or MMAP case (not yet supported),
1527          *   window_size == input_size + MIN_LOOKAHEAD  &&
1528          *   strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1529          * Otherwise, window_size == 2*WSIZE so more >= 2.
1530          * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1531          */
1532         Assert(more >= 2, "more < 2");
1533 
1534         n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1535         s->lookahead += n;
1536 
1537         /* Initialize the hash value now that we have some input: */
1538         if (s->lookahead + s->insert >= MIN_MATCH) {
1539             uInt str = s->strstart - s->insert;
1540             s->ins_h = s->window[str];
1541             UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1542 #if MIN_MATCH != 3
1543             Call UPDATE_HASH() MIN_MATCH-3 more times
1544 #endif
1545             while (s->insert) {
1546                 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1547 #ifndef FASTEST
1548                 s->prev[str & s->w_mask] = s->head[s->ins_h];
1549 #endif
1550                 s->head[s->ins_h] = (Pos)str;
1551                 str++;
1552                 s->insert--;
1553                 if (s->lookahead + s->insert < MIN_MATCH)
1554                     break;
1555             }
1556         }
1557         /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1558          * but this is not important since only literal bytes will be emitted.
1559          */
1560 
1561     } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1562 
1563     /* If the WIN_INIT bytes after the end of the current data have never been
1564      * written, then zero those bytes in order to avoid memory check reports of
1565      * the use of uninitialized (or uninitialised as Julian writes) bytes by
1566      * the longest match routines.  Update the high water mark for the next
1567      * time through here.  WIN_INIT is set to MAX_MATCH since the longest match
1568      * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1569      */
1570     if (s->high_water < s->window_size) {
1571         ulg curr = s->strstart + (ulg)(s->lookahead);
1572         ulg init;
1573 
1574         if (s->high_water < curr) {
1575             /* Previous high water mark below current data -- zero WIN_INIT
1576              * bytes or up to end of window, whichever is less.
1577              */
1578             init = s->window_size - curr;
1579             if (init > WIN_INIT)
1580                 init = WIN_INIT;
1581             zmemzero(s->window + curr, (unsigned)init);
1582             s->high_water = curr + init;
1583         }
1584         else if (s->high_water < (ulg)curr + WIN_INIT) {
1585             /* High water mark at or above current data, but below current data
1586              * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1587              * to end of window, whichever is less.
1588              */
1589             init = (ulg)curr + WIN_INIT - s->high_water;
1590             if (init > s->window_size - s->high_water)
1591                 init = s->window_size - s->high_water;
1592             zmemzero(s->window + s->high_water, (unsigned)init);
1593             s->high_water += init;
1594         }
1595     }
1596 
1597     Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1598            "not enough room for search");
1599 }
1600 
1601 /* ===========================================================================
1602  * Flush the current block, with given end-of-file flag.
1603  * IN assertion: strstart is set to the end of the current match.
1604  */
1605 #define FLUSH_BLOCK_ONLY(s, last) { \
1606    _tr_flush_block(s, (s->block_start >= 0L ? \
1607                    (charf *)&s->window[(unsigned)s->block_start] : \
1608                    (charf *)Z_NULL), \
1609                 (ulg)((long)s->strstart - s->block_start), \
1610                 (last)); \
1611    s->block_start = s->strstart; \
1612    flush_pending(s->strm); \
1613    Tracev((stderr,"[FLUSH]")); \
1614 }
1615 
1616 /* Same but force premature exit if necessary. */
1617 #define FLUSH_BLOCK(s, last) { \
1618    FLUSH_BLOCK_ONLY(s, last); \
1619    if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1620 }
1621 
1622 /* Maximum stored block length in deflate format (not including header). */
1623 #define MAX_STORED 65535
1624 
1625 /* Minimum of a and b. */
1626 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1627 
1628 /* ===========================================================================
1629  * Copy without compression as much as possible from the input stream, return
1630  * the current block state.
1631  *
1632  * In case deflateParams() is used to later switch to a non-zero compression
1633  * level, s->matches (otherwise unused when storing) keeps track of the number
1634  * of hash table slides to perform. If s->matches is 1, then one hash table
1635  * slide will be done when switching. If s->matches is 2, the maximum value
1636  * allowed here, then the hash table will be cleared, since two or more slides
1637  * is the same as a clear.
1638  *
1639  * deflate_stored() is written to minimize the number of times an input byte is
1640  * copied. It is most efficient with large input and output buffers, which
1641  * maximizes the opportunites to have a single copy from next_in to next_out.
1642  */
1643 local block_state deflate_stored(
1644     deflate_state *s,
1645     int flush)
1646 {
1647     /* Smallest worthy block size when not flushing or finishing. By default
1648      * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1649      * large input and output buffers, the stored block size will be larger.
1650      */
1651     unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1652 
1653     /* Copy as many min_block or larger stored blocks directly to next_out as
1654      * possible. If flushing, copy the remaining available input to next_out as
1655      * stored blocks, if there is enough space.
1656      */
1657     unsigned len, left, have, last = 0;
1658     unsigned used = s->strm->avail_in;
1659     do {
1660         /* Set len to the maximum size block that we can copy directly with the
1661          * available input data and output space. Set left to how much of that
1662          * would be copied from what's left in the window.
1663          */
1664         len = MAX_STORED;       /* maximum deflate stored block length */
1665         have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1666         if (s->strm->avail_out < have)          /* need room for header */
1667             break;
1668             /* maximum stored block length that will fit in avail_out: */
1669         have = s->strm->avail_out - have;
1670         left = s->strstart - s->block_start;    /* bytes left in window */
1671         if (len > (ulg)left + s->strm->avail_in)
1672             len = left + s->strm->avail_in;     /* limit len to the input */
1673         if (len > have)
1674             len = have;                         /* limit len to the output */
1675 
1676         /* If the stored block would be less than min_block in length, or if
1677          * unable to copy all of the available input when flushing, then try
1678          * copying to the window and the pending buffer instead. Also don't
1679          * write an empty block when flushing -- deflate() does that.
1680          */
1681         if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1682                                 flush == Z_NO_FLUSH ||
1683                                 len != left + s->strm->avail_in))
1684             break;
1685 
1686         /* Make a dummy stored block in pending to get the header bytes,
1687          * including any pending bits. This also updates the debugging counts.
1688          */
1689         last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1690         _tr_stored_block(s, (char *)0, 0L, last);
1691 
1692         /* Replace the lengths in the dummy stored block with len. */
1693         s->pending_buf[s->pending - 4] = len;
1694         s->pending_buf[s->pending - 3] = len >> 8;
1695         s->pending_buf[s->pending - 2] = ~len;
1696         s->pending_buf[s->pending - 1] = ~len >> 8;
1697 
1698         /* Write the stored block header bytes. */
1699         flush_pending(s->strm);
1700 
1701 #ifdef ZLIB_DEBUG
1702         /* Update debugging counts for the data about to be copied. */
1703         s->compressed_len += len << 3;
1704         s->bits_sent += len << 3;
1705 #endif
1706 
1707         /* Copy uncompressed bytes from the window to next_out. */
1708         if (left) {
1709             if (left > len)
1710                 left = len;
1711             zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1712             s->strm->next_out += left;
1713             s->strm->avail_out -= left;
1714             s->strm->total_out += left;
1715             s->block_start += left;
1716             len -= left;
1717         }
1718 
1719         /* Copy uncompressed bytes directly from next_in to next_out, updating
1720          * the check value.
1721          */
1722         if (len) {
1723             read_buf(s->strm, s->strm->next_out, len);
1724             s->strm->next_out += len;
1725             s->strm->avail_out -= len;
1726             s->strm->total_out += len;
1727         }
1728     } while (last == 0);
1729 
1730     /* Update the sliding window with the last s->w_size bytes of the copied
1731      * data, or append all of the copied data to the existing window if less
1732      * than s->w_size bytes were copied. Also update the number of bytes to
1733      * insert in the hash tables, in the event that deflateParams() switches to
1734      * a non-zero compression level.
1735      */
1736     used -= s->strm->avail_in;      /* number of input bytes directly copied */
1737     if (used) {
1738         /* If any input was used, then no unused input remains in the window,
1739          * therefore s->block_start == s->strstart.
1740          */
1741         if (used >= s->w_size) {    /* supplant the previous history */
1742             s->matches = 2;         /* clear hash */
1743             zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1744             s->strstart = s->w_size;
1745         }
1746         else {
1747             if (s->window_size - s->strstart <= used) {
1748                 /* Slide the window down. */
1749                 s->strstart -= s->w_size;
1750                 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1751                 if (s->matches < 2)
1752                     s->matches++;   /* add a pending slide_hash() */
1753             }
1754             zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1755             s->strstart += used;
1756         }
1757         s->block_start = s->strstart;
1758         s->insert += MIN(used, s->w_size - s->insert);
1759     }
1760     if (s->high_water < s->strstart)
1761         s->high_water = s->strstart;
1762 
1763     /* If the last block was written to next_out, then done. */
1764     if (last)
1765         return finish_done;
1766 
1767     /* If flushing and all input has been consumed, then done. */
1768     if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1769         s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1770         return block_done;
1771 
1772     /* Fill the window with any remaining input. */
1773     have = s->window_size - s->strstart - 1;
1774     if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1775         /* Slide the window down. */
1776         s->block_start -= s->w_size;
1777         s->strstart -= s->w_size;
1778         zmemcpy(s->window, s->window + s->w_size, s->strstart);
1779         if (s->matches < 2)
1780             s->matches++;           /* add a pending slide_hash() */
1781         have += s->w_size;          /* more space now */
1782     }
1783     if (have > s->strm->avail_in)
1784         have = s->strm->avail_in;
1785     if (have) {
1786         read_buf(s->strm, s->window + s->strstart, have);
1787         s->strstart += have;
1788     }
1789     if (s->high_water < s->strstart)
1790         s->high_water = s->strstart;
1791 
1792     /* There was not enough avail_out to write a complete worthy or flushed
1793      * stored block to next_out. Write a stored block to pending instead, if we
1794      * have enough input for a worthy block, or if flushing and there is enough
1795      * room for the remaining input as a stored block in the pending buffer.
1796      */
1797     have = (s->bi_valid + 42) >> 3;         /* number of header bytes */
1798         /* maximum stored block length that will fit in pending: */
1799     have = MIN(s->pending_buf_size - have, MAX_STORED);
1800     min_block = MIN(have, s->w_size);
1801     left = s->strstart - s->block_start;
1802     if (left >= min_block ||
1803         ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1804          s->strm->avail_in == 0 && left <= have)) {
1805         len = MIN(left, have);
1806         last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1807                len == left ? 1 : 0;
1808         _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1809         s->block_start += len;
1810         flush_pending(s->strm);
1811     }
1812 
1813     /* We've done all we can with the available input and output. */
1814     return last ? finish_started : need_more;
1815 }
1816 
1817 /* ===========================================================================
1818  * Compress as much as possible from the input stream, return the current
1819  * block state.
1820  * This function does not perform lazy evaluation of matches and inserts
1821  * new strings in the dictionary only for unmatched strings or for short
1822  * matches. It is used only for the fast compression options.
1823  */
1824 local block_state deflate_fast(
1825     deflate_state *s,
1826     int flush)
1827 {
1828     IPos hash_head;       /* head of the hash chain */
1829     int bflush;           /* set if current block must be flushed */
1830 
1831     for (;;) {
1832         /* Make sure that we always have enough lookahead, except
1833          * at the end of the input file. We need MAX_MATCH bytes
1834          * for the next match, plus MIN_MATCH bytes to insert the
1835          * string following the next match.
1836          */
1837         if (s->lookahead < MIN_LOOKAHEAD) {
1838             fill_window(s);
1839             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1840                 return need_more;
1841             }
1842             if (s->lookahead == 0) break; /* flush the current block */
1843         }
1844 
1845         /* Insert the string window[strstart .. strstart+2] in the
1846          * dictionary, and set hash_head to the head of the hash chain:
1847          */
1848         hash_head = NIL;
1849         if (s->lookahead >= MIN_MATCH) {
1850             INSERT_STRING(s, s->strstart, hash_head);
1851         }
1852 
1853         /* Find the longest match, discarding those <= prev_length.
1854          * At this point we have always match_length < MIN_MATCH
1855          */
1856         if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1857             /* To simplify the code, we prevent matches with the string
1858              * of window index 0 (in particular we have to avoid a match
1859              * of the string with itself at the start of the input file).
1860              */
1861             s->match_length = longest_match (s, hash_head);
1862             /* longest_match() sets match_start */
1863         }
1864         if (s->match_length >= MIN_MATCH) {
1865             check_match(s, s->strstart, s->match_start, s->match_length);
1866 
1867             _tr_tally_dist(s, s->strstart - s->match_start,
1868                            s->match_length - MIN_MATCH, bflush);
1869 
1870             s->lookahead -= s->match_length;
1871 
1872             /* Insert new strings in the hash table only if the match length
1873              * is not too large. This saves time but degrades compression.
1874              */
1875 #ifndef FASTEST
1876             if (s->match_length <= s->max_insert_length &&
1877                 s->lookahead >= MIN_MATCH) {
1878                 s->match_length--; /* string at strstart already in table */
1879                 do {
1880                     s->strstart++;
1881                     INSERT_STRING(s, s->strstart, hash_head);
1882                     /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1883                      * always MIN_MATCH bytes ahead.
1884                      */
1885                 } while (--s->match_length != 0);
1886                 s->strstart++;
1887             } else
1888 #endif
1889             {
1890                 s->strstart += s->match_length;
1891                 s->match_length = 0;
1892                 s->ins_h = s->window[s->strstart];
1893                 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1894 #if MIN_MATCH != 3
1895                 Call UPDATE_HASH() MIN_MATCH-3 more times
1896 #endif
1897                 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1898                  * matter since it will be recomputed at next deflate call.
1899                  */
1900             }
1901         } else {
1902             /* No match, output a literal byte */
1903             Tracevv((stderr,"%c", s->window[s->strstart]));
1904             _tr_tally_lit (s, s->window[s->strstart], bflush);
1905             s->lookahead--;
1906             s->strstart++;
1907         }
1908         if (bflush) FLUSH_BLOCK(s, 0);
1909     }
1910     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1911     if (flush == Z_FINISH) {
1912         FLUSH_BLOCK(s, 1);
1913         return finish_done;
1914     }
1915     if (s->last_lit)
1916         FLUSH_BLOCK(s, 0);
1917     return block_done;
1918 }
1919 
1920 #ifndef FASTEST
1921 /* ===========================================================================
1922  * Same as above, but achieves better compression. We use a lazy
1923  * evaluation for matches: a match is finally adopted only if there is
1924  * no better match at the next window position.
1925  */
1926 local block_state deflate_slow(
1927     deflate_state *s,
1928     int flush)
1929 {
1930     IPos hash_head;          /* head of hash chain */
1931     int bflush;              /* set if current block must be flushed */
1932 
1933     /* Process the input block. */
1934     for (;;) {
1935         /* Make sure that we always have enough lookahead, except
1936          * at the end of the input file. We need MAX_MATCH bytes
1937          * for the next match, plus MIN_MATCH bytes to insert the
1938          * string following the next match.
1939          */
1940         if (s->lookahead < MIN_LOOKAHEAD) {
1941             fill_window(s);
1942             if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1943                 return need_more;
1944             }
1945             if (s->lookahead == 0) break; /* flush the current block */
1946         }
1947 
1948         /* Insert the string window[strstart .. strstart+2] in the
1949          * dictionary, and set hash_head to the head of the hash chain:
1950          */
1951         hash_head = NIL;
1952         if (s->lookahead >= MIN_MATCH) {
1953             INSERT_STRING(s, s->strstart, hash_head);
1954         }
1955 
1956         /* Find the longest match, discarding those <= prev_length.
1957          */
1958         s->prev_length = s->match_length, s->prev_match = s->match_start;
1959         s->match_length = MIN_MATCH-1;
1960 
1961         if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1962             s->strstart - hash_head <= MAX_DIST(s)) {
1963             /* To simplify the code, we prevent matches with the string
1964              * of window index 0 (in particular we have to avoid a match
1965              * of the string with itself at the start of the input file).
1966              */
1967             s->match_length = longest_match (s, hash_head);
1968             /* longest_match() sets match_start */
1969 
1970             if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1971 #if TOO_FAR <= 32767
1972                 || (s->match_length == MIN_MATCH &&
1973                     s->strstart - s->match_start > TOO_FAR)
1974 #endif
1975                 )) {
1976 
1977                 /* If prev_match is also MIN_MATCH, match_start is garbage
1978                  * but we will ignore the current match anyway.
1979                  */
1980                 s->match_length = MIN_MATCH-1;
1981             }
1982         }
1983         /* If there was a match at the previous step and the current
1984          * match is not better, output the previous match:
1985          */
1986         if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1987             uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1988             /* Do not insert strings in hash table beyond this. */
1989 
1990             check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1991 
1992             _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1993                            s->prev_length - MIN_MATCH, bflush);
1994 
1995             /* Insert in hash table all strings up to the end of the match.
1996              * strstart-1 and strstart are already inserted. If there is not
1997              * enough lookahead, the last two strings are not inserted in
1998              * the hash table.
1999              */
2000             s->lookahead -= s->prev_length-1;
2001             s->prev_length -= 2;
2002             do {
2003                 if (++s->strstart <= max_insert) {
2004                     INSERT_STRING(s, s->strstart, hash_head);
2005                 }
2006             } while (--s->prev_length != 0);
2007             s->match_available = 0;
2008             s->match_length = MIN_MATCH-1;
2009             s->strstart++;
2010 
2011             if (bflush) FLUSH_BLOCK(s, 0);
2012 
2013         } else if (s->match_available) {
2014             /* If there was no match at the previous position, output a
2015              * single literal. If there was a match but the current match
2016              * is longer, truncate the previous match to a single literal.
2017              */
2018             Tracevv((stderr,"%c", s->window[s->strstart-1]));
2019             _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2020             if (bflush) {
2021                 FLUSH_BLOCK_ONLY(s, 0);
2022             }
2023             s->strstart++;
2024             s->lookahead--;
2025             if (s->strm->avail_out == 0) return need_more;
2026         } else {
2027             /* There is no previous match to compare with, wait for
2028              * the next step to decide.
2029              */
2030             s->match_available = 1;
2031             s->strstart++;
2032             s->lookahead--;
2033         }
2034     }
2035     Assert (flush != Z_NO_FLUSH, "no flush?");
2036     if (s->match_available) {
2037         Tracevv((stderr,"%c", s->window[s->strstart-1]));
2038         _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2039         s->match_available = 0;
2040     }
2041     s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2042     if (flush == Z_FINISH) {
2043         FLUSH_BLOCK(s, 1);
2044         return finish_done;
2045     }
2046     if (s->last_lit)
2047         FLUSH_BLOCK(s, 0);
2048     return block_done;
2049 }
2050 #endif /* FASTEST */
2051 
2052 /* ===========================================================================
2053  * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2054  * one.  Do not maintain a hash table.  (It will be regenerated if this run of
2055  * deflate switches away from Z_RLE.)
2056  */
2057 local block_state deflate_rle(
2058     deflate_state *s,
2059     int flush)
2060 {
2061     int bflush;             /* set if current block must be flushed */
2062     uInt prev;              /* byte at distance one to match */
2063     Bytef *scan, *strend;   /* scan goes up to strend for length of run */
2064 
2065     for (;;) {
2066         /* Make sure that we always have enough lookahead, except
2067          * at the end of the input file. We need MAX_MATCH bytes
2068          * for the longest run, plus one for the unrolled loop.
2069          */
2070         if (s->lookahead <= MAX_MATCH) {
2071             fill_window(s);
2072             if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2073                 return need_more;
2074             }
2075             if (s->lookahead == 0) break; /* flush the current block */
2076         }
2077 
2078         /* See how many times the previous byte repeats */
2079         s->match_length = 0;
2080         if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2081             scan = s->window + s->strstart - 1;
2082             prev = *scan;
2083             if (prev == *++scan && prev == *++scan && prev == *++scan) {
2084                 strend = s->window + s->strstart + MAX_MATCH;
2085                 do {
2086                 } while (prev == *++scan && prev == *++scan &&
2087                          prev == *++scan && prev == *++scan &&
2088                          prev == *++scan && prev == *++scan &&
2089                          prev == *++scan && prev == *++scan &&
2090                          scan < strend);
2091                 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2092                 if (s->match_length > s->lookahead)
2093                     s->match_length = s->lookahead;
2094             }
2095             Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2096         }
2097 
2098         /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2099         if (s->match_length >= MIN_MATCH) {
2100             check_match(s, s->strstart, s->strstart - 1, s->match_length);
2101 
2102             _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2103 
2104             s->lookahead -= s->match_length;
2105             s->strstart += s->match_length;
2106             s->match_length = 0;
2107         } else {
2108             /* No match, output a literal byte */
2109             Tracevv((stderr,"%c", s->window[s->strstart]));
2110             _tr_tally_lit (s, s->window[s->strstart], bflush);
2111             s->lookahead--;
2112             s->strstart++;
2113         }
2114         if (bflush) FLUSH_BLOCK(s, 0);
2115     }
2116     s->insert = 0;
2117     if (flush == Z_FINISH) {
2118         FLUSH_BLOCK(s, 1);
2119         return finish_done;
2120     }
2121     if (s->last_lit)
2122         FLUSH_BLOCK(s, 0);
2123     return block_done;
2124 }
2125 
2126 /* ===========================================================================
2127  * For Z_HUFFMAN_ONLY, do not look for matches.  Do not maintain a hash table.
2128  * (It will be regenerated if this run of deflate switches away from Huffman.)
2129  */
2130 local block_state deflate_huff(
2131     deflate_state *s,
2132     int flush)
2133 {
2134     int bflush;             /* set if current block must be flushed */
2135 
2136     for (;;) {
2137         /* Make sure that we have a literal to write. */
2138         if (s->lookahead == 0) {
2139             fill_window(s);
2140             if (s->lookahead == 0) {
2141                 if (flush == Z_NO_FLUSH)
2142                     return need_more;
2143                 break;      /* flush the current block */
2144             }
2145         }
2146 
2147         /* Output a literal byte */
2148         s->match_length = 0;
2149         Tracevv((stderr,"%c", s->window[s->strstart]));
2150         _tr_tally_lit (s, s->window[s->strstart], bflush);
2151         s->lookahead--;
2152         s->strstart++;
2153         if (bflush) FLUSH_BLOCK(s, 0);
2154     }
2155     s->insert = 0;
2156     if (flush == Z_FINISH) {
2157         FLUSH_BLOCK(s, 1);
2158         return finish_done;
2159     }
2160     if (s->last_lit)
2161         FLUSH_BLOCK(s, 0);
2162     return block_done;
2163 }
2164