home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
BURKS 2
/
BURKS_AUG97.ISO
/
BURKS
/
SOFTWARE
/
SOURCES
/
UNZIP52.ZIP
/
unshrink.c
(
.txt
)
< prev
next >
Wrap
C/C++ Source or Header
|
1996-02-18
|
10KB
|
285 lines
/*---------------------------------------------------------------------------
unshrink.c version 1.21 23 Nov 95
Shrinking is basically a dynamic LZW algorithm with allowed code sizes of
up to 13 bits; in addition, there is provision for partial clearing of
leaf nodes. PKWARE uses the special code 256 (decimal) to indicate a
change in code size or a partial clear of the code tree: 256,1 for the
former and 256,2 for the latter. [Note that partial clearing can "orphan"
nodes: the parent-to-be can be cleared before its new child is added,
but the child is added anyway (as an orphan, as though the parent still
exists). When the tree fills up to the point where the parent node is
reused, the orphan is effectively "adopted." Versions prior to 1.05 were
more affected due to their use of more pointers (to children and siblings
as well as parents).]
This replacement version of unshrink.c was written from scratch. It is
based only on the algorithms described in Mark Nelson's _The Data Compres-
sion Book_ and in Terry Welch's original paper in the June 1984 issue of
IEEE _Computer_; no existing source code, including any in Nelson's book,
was used.
Memory requirements have been reduced in this version and are now no more
than the original Sam Smith code. This is still larger than any of the
other algorithms: at a minimum, 8K+8K+16K (stack+values+parents) assuming
16-bit short ints, and this does not even include the output buffer (the
other algorithms leave the uncompressed data in the work area, typically
called slide[]). For machines with a 64KB data space this is a problem,
particularly when text conversion is required and line endings have more
than one character. UnZip's solution is to use two roughly equal halves
of outbuf for the ASCII conversion in such a case; the "unshrink" argument
to flush() signals that this is the case.
For large-memory machines, a second outbuf is allocated for translations,
but only if unshrinking and only if translations are required.
| binary mode | text mode
---------------------------------------------------
big mem | big outbuf | big outbuf + big outbuf2 <- malloc'd here
small mem | small outbuf | half + half small outbuf
Copyright 1994, 1995 Greg Roelofs. See the accompanying file "COPYING"
in UnZip 5.20 (or later) source or binary distributions.
---------------------------------------------------------------------------*/
#define UNZIP_INTERNAL
#include "unzip.h"
static void partial_clear OF((__GPRO));
#ifdef DEBUG
# define OUTDBG(c) \
if ((c)<32 || (c)>=127) fprintf(stderr,"\\x%02x",(c)); else putc((c),stderr);
#else
# define OUTDBG(c)
#endif
/* HSIZE is defined as 2^13 (8192) in unzip.h */
#define BOGUSCODE 256
#define FLAG_BITS parent /* upper bits of parent[] used as flag bits */
#define CODE_MASK (HSIZE - 1) /* 0x1fff (lower bits are parent's index) */
#define FREE_CODE HSIZE /* 0x2000 (code is unused or was cleared) */
#define HAS_CHILD (HSIZE << 1) /* 0x4000 (code has a child--do not clear) */
#define parent G.area.shrink.Parent
#define Value G.area.shrink.value /* "value" conflicts with Pyramid ioctl.h */
#define stack G.area.shrink.Stack
/***********************/
/* Function unshrink() */
/***********************/
int unshrink(__G)
__GDEF
{
#if 0
static uch *stacktop = NULL;
#else
uch *stacktop = stack + (HSIZE - 1);
#endif
register uch *newstr;
int codesize=9, len, KwKwK, error;
shrint code, oldcode, freecode, curcode;
shrint lastfreecode;
unsigned int outbufsiz;
/*---------------------------------------------------------------------------
Initialize various variables.
---------------------------------------------------------------------------*/
#if 0
/* SPC: no longer needed, stacktop is now an automatic variable that
* is initialized on every unshrink() entry.
*/
/* this is required for MACOS, but performance hit is minuscule */
if (stacktop == NULL)
stacktop = stack + (HSIZE - 1);
#endif
lastfreecode = BOGUSCODE;
#ifndef VMS /* VMS uses its own buffer scheme for textmode flush(). */
#ifndef SMALL_MEM
/* non-memory-limited machines: allocate second (large) buffer for
* textmode conversion in flush(), but only if needed */
if (G.pInfo->textmode && !G.outbuf2 &&
(G.outbuf2 = (uch *)malloc(TRANSBUFSIZ)) == (uch *)NULL)
return PK_MEM3;
#endif
#endif /* !VMS */
for (code = 0; code < BOGUSCODE; ++code) {
Value[code] = (uch)code;
parent[code] = BOGUSCODE;
}
for (code = BOGUSCODE+1; code < HSIZE; ++code)
parent[code] = FREE_CODE;
G.realbuf = G.outbuf; /* use normal outbuf unless we're a DLL routine */
#ifdef DLL
if (G.redirect_data) {
G.realbuf = G.redirect_buffer;
outbufsiz = G.redirect_size;
} else
#endif
if (G.pInfo->textmode)
outbufsiz = RAWBUFSIZ;
else
outbufsiz = OUTBUFSIZ;
G.outptr = G.realbuf;
G.outcnt = 0L;
/*---------------------------------------------------------------------------
Get and output first code, then loop over remaining ones.
---------------------------------------------------------------------------*/
READBITS(codesize, oldcode)
if (!G.zipeof) {
*G.outptr++ = (uch)oldcode;
OUTDBG((uch)oldcode)
++G.outcnt;
}
do {
READBITS(codesize, code)
if (G.zipeof)
break;
if (code == BOGUSCODE) { /* possible to have consecutive escapes? */
READBITS(codesize, code)
if (code == 1) {
++codesize;
Trace((stderr, " (codesize now %d bits)\n", codesize));
} else if (code == 2) {
Trace((stderr, " (partial clear code)\n"));
partial_clear(__G); /* clear leafs (nodes with no children) */
Trace((stderr, " (done with partial clear)\n"));
lastfreecode = BOGUSCODE; /* reset start of free-node search */
}
continue;
}
/*-----------------------------------------------------------------------
Translate code: traverse tree from leaf back to root.
-----------------------------------------------------------------------*/
newstr = stacktop;
curcode = code;
if (parent[curcode] == FREE_CODE) {
/* or (FLAG_BITS[curcode] & FREE_CODE)? */
KwKwK = TRUE;
Trace((stderr, " (found a KwKwK code %d; oldcode = %d)\n", code,
oldcode));
--newstr; /* last character will be same as first character */
curcode = oldcode;
} else
KwKwK = FALSE;
do {
*newstr-- = Value[curcode];
curcode = (shrint)(parent[curcode] & CODE_MASK);
} while (curcode != BOGUSCODE);
len = (int)(stacktop - newstr++);
if (KwKwK)
*stacktop = *newstr;
/*-----------------------------------------------------------------------
Write expanded string in reverse order to output buffer.
-----------------------------------------------------------------------*/
Trace((stderr, "code %4d; oldcode %4d; char %3d (%c); string [", code,
oldcode, (int)(*newstr), (*newstr<32 || *newstr>=127)? ' ':*newstr));
{
register uch *p;
for (p = newstr; p < newstr+len; ++p) {
*G.outptr++ = *p;
OUTDBG(*p)
if (++G.outcnt == outbufsiz) {
Trace((stderr, "doing flush(), outcnt = %lu\n", G.outcnt));
if ((error = flush(__G__ G.realbuf, G.outcnt, TRUE)) != 0)
fprintf(stderr, "unshrink: flush() error (%d)\n",
error);
Trace((stderr, "done with flush()\n"));
G.outptr = G.realbuf;
G.outcnt = 0L;
}
}
}
/*------------------------------------