2 Copyright (c) 1990-2008 Info-ZIP. All rights reserved.
4 See the accompanying file LICENSE, version 2000-Apr-09 or later
5 (the contents of which are also included in unzip.h) for terms of use.
6 If, for some reason, all these files are missing, the Info-ZIP license
7 also may be found at: ftp://ftp.info-zip.org/pub/infozip/license.html
9 /*---------------------------------------------------------------------------
11 unshrink.c version 1.22 19 Mar 2008
14 NOTE: This code may or may not infringe on the so-called "Welch
15 patent" owned by Unisys. (From reading the patent, it appears
16 that a pure LZW decompressor is *not* covered, but this claim has
17 not been tested in court, and Unisys is reported to believe other-
18 wise.) It is therefore the responsibility of the user to acquire
19 whatever license(s) may be required for legal use of this code.
21 THE INFO-ZIP GROUP DISCLAIMS ALL LIABILITY FOR USE OF THIS CODE
22 IN VIOLATION OF APPLICABLE PATENT LAW.
25 Shrinking is basically a dynamic LZW algorithm with allowed code sizes of
26 up to 13 bits; in addition, there is provision for partial clearing of
27 leaf nodes. PKWARE uses the special code 256 (decimal) to indicate a
28 change in code size or a partial clear of the code tree: 256,1 for the
29 former and 256,2 for the latter. [Note that partial clearing can "orphan"
30 nodes: the parent-to-be can be cleared before its new child is added,
31 but the child is added anyway (as an orphan, as though the parent still
32 existed). When the tree fills up to the point where the parent node is
33 reused, the orphan is effectively "adopted." Versions prior to 1.05 were
34 affected more due to greater use of pointers (to children and siblings
37 This replacement version of unshrink.c was written from scratch. It is
38 based only on the algorithms described in Mark Nelson's _The Data Compres-
39 sion Book_ and in Terry Welch's original paper in the June 1984 issue of
40 IEEE _Computer_; no existing source code, including any in Nelson's book,
43 Memory requirements have been reduced in this version and are now no more
44 than the original Sam Smith code. This is still larger than any of the
45 other algorithms: at a minimum, 8K+8K+16K (stack+values+parents) assuming
46 16-bit short ints, and this does not even include the output buffer (the
47 other algorithms leave the uncompressed data in the work area, typically
48 called slide[]). For machines with a 64KB data space this is a problem,
49 particularly when text conversion is required and line endings have more
50 than one character. UnZip's solution is to use two roughly equal halves
51 of outbuf for the ASCII conversion in such a case; the "unshrink" argument
52 to flush() signals that this is the case.
54 For large-memory machines, a second outbuf is allocated for translations,
55 but only if unshrinking and only if translations are required.
57 | binary mode | text mode
58 ---------------------------------------------------
59 big mem | big outbuf | big outbuf + big outbuf2 <- malloc'd here
60 small mem | small outbuf | half + half small outbuf
62 Copyright 1994, 1995 Greg Roelofs. See the accompanying file "COPYING"
63 in UnZip 5.20 (or later) source or binary distributions.
65 ---------------------------------------------------------------------------*/
68 #define __UNSHRINK_C /* identifies this source module */
69 #define UNZIP_INTERNAL
75 static void partial_clear
OF((__GPRO__
int lastcodeused
));
79 if ((c)<32 || (c)>=127) fprintf(stderr,"\\x%02x",(c)); else putc((c),stderr);
84 /* HSIZE is defined as 2^13 (8192) in unzip.h (resp. unzpriv.h */
86 #define FLAG_BITS parent /* upper bits of parent[] used as flag bits */
87 #define CODE_MASK (HSIZE - 1) /* 0x1fff (lower bits are parent's index) */
88 #define FREE_CODE HSIZE /* 0x2000 (code is unused or was cleared) */
89 #define HAS_CHILD (HSIZE << 1) /* 0x4000 (code has a child--do not clear) */
91 #define parent G.area.shrink.Parent
92 #define Value G.area.shrink.value /* "value" conflicts with Pyramid ioctl.h */
93 #define stack G.area.shrink.Stack
96 /***********************/
97 /* Function unshrink() */
98 /***********************/
103 uch
*stacktop
= stack
+ (HSIZE
- 1);
104 register uch
*newstr
;
106 int codesize
=9, len
, error
;
107 shrint code
, oldcode
, curcode
;
109 unsigned int outbufsiz
;
110 #if (defined(DLL) && !defined(NO_SLIDE_REDIR))
111 /* Normally realbuf and outbuf will be the same. However, if the data
112 * are redirected to a large memory buffer, realbuf will point to the
113 * new location while outbuf will remain pointing to the malloc'd
115 uch
*realbuf
= G
.outbuf
;
117 # define realbuf G.outbuf
121 /*---------------------------------------------------------------------------
122 Initialize various variables.
123 ---------------------------------------------------------------------------*/
125 lastfreecode
= BOGUSCODE
;
127 #ifndef VMS /* VMS uses its own buffer scheme for textmode flush(). */
129 /* non-memory-limited machines: allocate second (large) buffer for
130 * textmode conversion in flush(), but only if needed */
131 if (G
.pInfo
->textmode
&& !G
.outbuf2
&&
132 (G
.outbuf2
= (uch
*)malloc(TRANSBUFSIZ
)) == (uch
*)NULL
)
137 for (code
= 0; code
< BOGUSCODE
; ++code
) {
138 Value
[code
] = (uch
)code
;
139 parent
[code
] = BOGUSCODE
;
141 for (code
= BOGUSCODE
+1; code
< HSIZE
; ++code
)
142 parent
[code
] = FREE_CODE
;
144 #if (defined(DLL) && !defined(NO_SLIDE_REDIR))
145 if (G
.redirect_slide
) { /* use normal outbuf unless we're a DLL routine */
146 realbuf
= G
.redirect_buffer
;
147 outbufsiz
= (unsigned)G
.redirect_size
;
151 if (G
.pInfo
->textmode
&& !G
.redirect_data
)
153 if (G
.pInfo
->textmode
)
155 outbufsiz
= RAWBUFSIZ
;
157 outbufsiz
= OUTBUFSIZ
;
161 /*---------------------------------------------------------------------------
162 Get and output first code, then loop over remaining ones.
163 ---------------------------------------------------------------------------*/
165 READBITS(codesize
, oldcode
)
169 finalval
= (uch
)oldcode
;
171 *G
.outptr
++ = finalval
;
175 READBITS(codesize
, code
)
178 if (code
== BOGUSCODE
) { /* possible to have consecutive escapes? */
179 READBITS(codesize
, code
)
184 Trace((stderr
, " (codesize now %d bits)\n", codesize
));
185 if (codesize
> MAX_BITS
) return PK_ERR
;
186 } else if (code
== 2) {
187 Trace((stderr
, " (partial clear code)\n"));
188 /* clear leafs (nodes with no children) */
189 partial_clear(__G__ lastfreecode
);
190 Trace((stderr
, " (done with partial clear)\n"));
191 lastfreecode
= BOGUSCODE
; /* reset start of free-node search */
196 /*-----------------------------------------------------------------------
197 Translate code: traverse tree from leaf back to root.
198 -----------------------------------------------------------------------*/
203 if (parent
[code
] == FREE_CODE
) {
204 /* or (FLAG_BITS[code] & FREE_CODE)? */
205 Trace((stderr
, " (found a KwKwK code %d; oldcode = %d)\n", code
,
207 *newstr
-- = finalval
;
211 while (code
!= BOGUSCODE
) {
212 if (newstr
< stack
) {
213 /* Bogus compression stream caused buffer underflow! */
214 Trace((stderr
, "unshrink stack overflow!\n"));
217 if (parent
[code
] == FREE_CODE
) {
218 /* or (FLAG_BITS[code] & FREE_CODE)? */
219 Trace((stderr
, " (found a KwKwK code %d; oldcode = %d)\n",
221 *newstr
-- = finalval
;
224 *newstr
-- = Value
[code
];
225 code
= (shrint
)(parent
[code
] & CODE_MASK
);
229 len
= (int)(stacktop
- newstr
++);
232 /*-----------------------------------------------------------------------
233 Write expanded string in reverse order to output buffer.
234 -----------------------------------------------------------------------*/
237 "code %4d; oldcode %4d; char %3d (%c); len %d; string [", curcode
,
238 oldcode
, (int)(*newstr
), (*newstr
<32 || *newstr
>=127)? ' ':*newstr
,
244 for (p
= newstr
; p
< newstr
+len
; ++p
) {
247 if (++G
.outcnt
== outbufsiz
) {
248 Trace((stderr
, "doing flush(), outcnt = %lu\n", G
.outcnt
));
249 if ((error
= flush(__G__ realbuf
, G
.outcnt
, TRUE
)) != 0) {
250 Trace((stderr
, "unshrink: flush() error (%d)\n",
256 Trace((stderr
, "done with flush()\n"));
261 /*-----------------------------------------------------------------------
262 Add new leaf (first character of newstr) to tree as child of oldcode.
263 -----------------------------------------------------------------------*/
265 /* search for freecode */
266 code
= (shrint
)(lastfreecode
+ 1);
267 /* add if-test before loop for speed? */
268 while ((code
< HSIZE
) && (parent
[code
] != FREE_CODE
))
271 Trace((stderr
, "]; newcode %d\n", code
));
273 /* invalid compressed data caused max-code overflow! */
276 Value
[code
] = finalval
;
277 parent
[code
] = oldcode
;
282 /*---------------------------------------------------------------------------
283 Flush any remaining data and return to sender...
284 ---------------------------------------------------------------------------*/
287 Trace((stderr
, "doing final flush(), outcnt = %lu\n", G
.outcnt
));
288 if ((error
= flush(__G__ realbuf
, G
.outcnt
, TRUE
)) != 0) {
289 Trace((stderr
, "unshrink: flush() error (%d)\n", error
));
292 Trace((stderr
, "done with flush()\n"));
297 } /* end function unshrink() */
303 /****************************/
304 /* Function partial_clear() */ /* no longer recursive... */
305 /****************************/
307 static void partial_clear(__G__ lastcodeused
)
311 register shrint code
;
313 /* clear all nodes which have no children (i.e., leaf nodes only) */
315 /* first loop: mark each parent as such */
316 for (code
= BOGUSCODE
+1; code
<= lastcodeused
; ++code
) {
317 register shrint cparent
= (shrint
)(parent
[code
] & CODE_MASK
);
319 if (cparent
> BOGUSCODE
)
320 FLAG_BITS
[cparent
] |= HAS_CHILD
; /* set parent's child-bit */
323 /* second loop: clear all nodes *not* marked as parents; reset flag bits */
324 for (code
= BOGUSCODE
+1; code
<= lastcodeused
; ++code
) {
325 if (FLAG_BITS
[code
] & HAS_CHILD
) /* just clear child-bit */
326 FLAG_BITS
[code
] &= ~HAS_CHILD
;
327 else { /* leaf: lose it */
328 Trace((stderr
, "%d\n", code
));
329 parent
[code
] = FREE_CODE
;
336 #endif /* !LZW_CLEAN */