Oliver Schneider avatar Oliver Schneider committed 3c38e3c

- Adding bzip2 source files relevant for the library (v1.0.6)

Comments (0)

Files changed (11)

+ ------------------------------------------------------------------
+ This file is part of bzip2/libbzip2, a program and library for
+ lossless, block-sorting data compression.
+
+ bzip2/libbzip2 version 1.0.6 of 6 September 2010
+ Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
+
+ Please read the WARNING, DISCLAIMER and PATENTS sections in the 
+ README file.
+
+ This program is released under the terms of the license contained
+ in the file LICENSE.
+ ------------------------------------------------------------------
+
+
+0.9.0
+~~~~~
+First version.
+
+
+0.9.0a
+~~~~~~
+Removed 'ranlib' from Makefile, since most modern Unix-es 
+don't need it, or even know about it.
+
+
+0.9.0b
+~~~~~~
+Fixed a problem with error reporting in bzip2.c.  This does not effect
+the library in any way.  Problem is: versions 0.9.0 and 0.9.0a (of the
+program proper) compress and decompress correctly, but give misleading
+error messages (internal panics) when an I/O error occurs, instead of
+reporting the problem correctly.  This shouldn't give any data loss
+(as far as I can see), but is confusing.
+
+Made the inline declarations disappear for non-GCC compilers.
+
+
+0.9.0c
+~~~~~~
+Fixed some problems in the library pertaining to some boundary cases.
+This makes the library behave more correctly in those situations.  The
+fixes apply only to features (calls and parameters) not used by
+bzip2.c, so the non-fixedness of them in previous versions has no
+effect on reliability of bzip2.c.
+
+In bzlib.c:
+   * made zero-length BZ_FLUSH work correctly in bzCompress().
+   * fixed bzWrite/bzRead to ignore zero-length requests.
+   * fixed bzread to correctly handle read requests after EOF.
+   * wrong parameter order in call to bzDecompressInit in
+     bzBuffToBuffDecompress.  Fixed.
+
+In compress.c:
+   * changed setting of nGroups in sendMTFValues() so as to 
+     do a bit better on small files.  This _does_ effect
+     bzip2.c.
+
+
+0.9.5a
+~~~~~~
+Major change: add a fallback sorting algorithm (blocksort.c)
+to give reasonable behaviour even for very repetitive inputs.
+Nuked --repetitive-best and --repetitive-fast since they are
+no longer useful.
+
+Minor changes: mostly a whole bunch of small changes/
+bugfixes in the driver (bzip2.c).  Changes pertaining to the
+user interface are:
+
+   allow decompression of symlink'd files to stdout
+   decompress/test files even without .bz2 extension
+   give more accurate error messages for I/O errors
+   when compressing/decompressing to stdout, don't catch control-C
+   read flags from BZIP2 and BZIP environment variables
+   decline to break hard links to a file unless forced with -f
+   allow -c flag even with no filenames
+   preserve file ownerships as far as possible
+   make -s -1 give the expected block size (100k)
+   add a flag -q --quiet to suppress nonessential warnings
+   stop decoding flags after --, so files beginning in - can be handled
+   resolved inconsistent naming: bzcat or bz2cat ?
+   bzip2 --help now returns 0
+
+Programming-level changes are:
+
+   fixed syntax error in GET_LL4 for Borland C++ 5.02
+   let bzBuffToBuffDecompress return BZ_DATA_ERROR{_MAGIC}
+   fix overshoot of mode-string end in bzopen_or_bzdopen
+   wrapped bzlib.h in #ifdef __cplusplus ... extern "C" { ... }
+   close file handles under all error conditions
+   added minor mods so it compiles with DJGPP out of the box
+   fixed Makefile so it doesn't give problems with BSD make
+   fix uninitialised memory reads in dlltest.c
+
+0.9.5b
+~~~~~~
+Open stdin/stdout in binary mode for DJGPP.
+
+0.9.5c
+~~~~~~
+Changed BZ_N_OVERSHOOT to be ... + 2 instead of ... + 1.  The + 1
+version could cause the sorted order to be wrong in some extremely
+obscure cases.  Also changed setting of quadrant in blocksort.c.
+
+0.9.5d
+~~~~~~
+The only functional change is to make bzlibVersion() in the library
+return the correct string.  This has no effect whatsoever on the
+functioning of the bzip2 program or library.  Added a couple of casts
+so the library compiles without warnings at level 3 in MS Visual
+Studio 6.0.  Included a Y2K statement in the file Y2K_INFO.  All other
+changes are minor documentation changes.
+
+1.0
+~~~
+Several minor bugfixes and enhancements:
+
+* Large file support.  The library uses 64-bit counters to
+  count the volume of data passing through it.  bzip2.c 
+  is now compiled with -D_FILE_OFFSET_BITS=64 to get large
+  file support from the C library.  -v correctly prints out
+  file sizes greater than 4 gigabytes.  All these changes have
+  been made without assuming a 64-bit platform or a C compiler
+  which supports 64-bit ints, so, except for the C library
+  aspect, they are fully portable.
+
+* Decompression robustness.  The library/program should be
+  robust to any corruption of compressed data, detecting and
+  handling _all_ corruption, instead of merely relying on
+  the CRCs.  What this means is that the program should 
+  never crash, given corrupted data, and the library should
+  always return BZ_DATA_ERROR.
+
+* Fixed an obscure race-condition bug only ever observed on
+  Solaris, in which, if you were very unlucky and issued
+  control-C at exactly the wrong time, both input and output
+  files would be deleted.
+
+* Don't run out of file handles on test/decompression when
+  large numbers of files have invalid magic numbers.
+
+* Avoid library namespace pollution.  Prefix all exported 
+  symbols with BZ2_.
+
+* Minor sorting enhancements from my DCC2000 paper.
+
+* Advance the version number to 1.0, so as to counteract the
+  (false-in-this-case) impression some people have that programs 
+  with version numbers less than 1.0 are in some way, experimental,
+  pre-release versions.
+
+* Create an initial Makefile-libbz2_so to build a shared library.
+  Yes, I know I should really use libtool et al ...
+
+* Make the program exit with 2 instead of 0 when decompression
+  fails due to a bad magic number (ie, an invalid bzip2 header).
+  Also exit with 1 (as the manual claims :-) whenever a diagnostic
+  message would have been printed AND the corresponding operation 
+  is aborted, for example
+     bzip2: Output file xx already exists.
+  When a diagnostic message is printed but the operation is not 
+  aborted, for example
+     bzip2: Can't guess original name for wurble -- using wurble.out
+  then the exit value 0 is returned, unless some other problem is
+  also detected.
+
+  I think it corresponds more closely to what the manual claims now.
+
+
+1.0.1
+~~~~~
+* Modified dlltest.c so it uses the new BZ2_ naming scheme.
+* Modified makefile-msc to fix minor build probs on Win2k.
+* Updated README.COMPILATION.PROBLEMS.
+
+There are no functionality changes or bug fixes relative to version
+1.0.0.  This is just a documentation update + a fix for minor Win32
+build problems.  For almost everyone, upgrading from 1.0.0 to 1.0.1 is
+utterly pointless.  Don't bother.
+
+
+1.0.2
+~~~~~
+A bug fix release, addressing various minor issues which have appeared
+in the 18 or so months since 1.0.1 was released.  Most of the fixes
+are to do with file-handling or documentation bugs.  To the best of my
+knowledge, there have been no data-loss-causing bugs reported in the
+compression/decompression engine of 1.0.0 or 1.0.1.
+
+Note that this release does not improve the rather crude build system
+for Unix platforms.  The general plan here is to autoconfiscate/
+libtoolise 1.0.2 soon after release, and release the result as 1.1.0
+or perhaps 1.2.0.  That, however, is still just a plan at this point.
+
+Here are the changes in 1.0.2.  Bug-reporters and/or patch-senders in
+parentheses.
+
+* Fix an infinite segfault loop in 1.0.1 when a directory is
+  encountered in -f (force) mode.
+     (Trond Eivind Glomsrod, Nicholas Nethercote, Volker Schmidt)
+
+* Avoid double fclose() of output file on certain I/O error paths.
+     (Solar Designer)
+
+* Don't fail with internal error 1007 when fed a long stream (> 48MB)
+  of byte 251.  Also print useful message suggesting that 1007s may be
+  caused by bad memory.
+     (noticed by Juan Pedro Vallejo, fixed by me)
+
+* Fix uninitialised variable silly bug in demo prog dlltest.c.
+     (Jorj Bauer)
+
+* Remove 512-MB limitation on recovered file size for bzip2recover
+  on selected platforms which support 64-bit ints.  At the moment
+  all GCC supported platforms, and Win32.
+     (me, Alson van der Meulen)
+
+* Hard-code header byte values, to give correct operation on platforms
+  using EBCDIC as their native character set (IBM's OS/390).
+     (Leland Lucius)
+
+* Copy file access times correctly.
+     (Marty Leisner)
+
+* Add distclean and check targets to Makefile.
+     (Michael Carmack)
+
+* Parameterise use of ar and ranlib in Makefile.  Also add $(LDFLAGS).
+     (Rich Ireland, Bo Thorsen)
+
+* Pass -p (create parent dirs as needed) to mkdir during make install.
+     (Jeremy Fusco)
+
+* Dereference symlinks when copying file permissions in -f mode.
+     (Volker Schmidt)
+
+* Majorly simplify implementation of uInt64_qrm10.
+     (Bo Lindbergh)
+
+* Check the input file still exists before deleting the output one,
+  when aborting in cleanUpAndFail().
+     (Joerg Prante, Robert Linden, Matthias Krings)
+
+Also a bunch of patches courtesy of Philippe Troin, the Debian maintainer
+of bzip2:
+
+* Wrapper scripts (with manpages): bzdiff, bzgrep, bzmore.
+
+* Spelling changes and minor enhancements in bzip2.1.
+
+* Avoid race condition between creating the output file and setting its
+  interim permissions safely, by using fopen_output_safely().
+  No changes to bzip2recover since there is no issue with file
+  permissions there.
+
+* do not print senseless report with -v when compressing an empty
+  file.
+
+* bzcat -f works on non-bzip2 files.
+
+* do not try to escape shell meta-characters on unix (the shell takes
+  care of these).
+
+* added --fast and --best aliases for -1 -9 for gzip compatibility.
+
+
+1.0.3 (15 Feb 05)
+~~~~~~~~~~~~~~~~~
+Fixes some minor bugs since the last version, 1.0.2.
+
+* Further robustification against corrupted compressed data.
+  There are currently no known bitstreams which can cause the
+  decompressor to crash, loop or access memory which does not
+  belong to it.  If you are using bzip2 or the library to 
+  decompress bitstreams from untrusted sources, an upgrade
+  to 1.0.3 is recommended.  This fixes CAN-2005-1260.
+
+* The documentation has been converted to XML, from which html
+  and pdf can be derived.
+
+* Various minor bugs in the documentation have been fixed.
+
+* Fixes for various compilation warnings with newer versions of
+  gcc, and on 64-bit platforms.
+
+* The BZ_NO_STDIO cpp symbol was not properly observed in 1.0.2.
+  This has been fixed.
+
+
+1.0.4 (20 Dec 06)
+~~~~~~~~~~~~~~~~~
+Fixes some minor bugs since the last version, 1.0.3.
+
+* Fix file permissions race problem (CAN-2005-0953).
+
+* Avoid possible segfault in BZ2_bzclose.  From Coverity's NetBSD
+  scan.
+
+* 'const'/prototype cleanups in the C code.
+
+* Change default install location to /usr/local, and handle multiple
+  'make install's without error.
+
+* Sanitise file names more carefully in bzgrep.  Fixes CAN-2005-0758
+  to the extent that applies to bzgrep.
+
+* Use 'mktemp' rather than 'tempfile' in bzdiff.
+
+* Tighten up a couple of assertions in blocksort.c following automated
+  analysis.
+
+* Fix minor doc/comment bugs.
+
+
+1.0.5 (10 Dec 07)
+~~~~~~~~~~~~~~~~~
+Security fix only.  Fixes CERT-FI 20469 as it applies to bzip2.
+
+
+1.0.6 (6 Sept 10)
+~~~~~~~~~~~~~~~~~
+
+* Security fix for CVE-2010-0405.  This was reported by Mikolaj
+  Izdebski.
+
+* Make the documentation build on Ubuntu 10.04
+
+--------------------------------------------------------------------------
+
+This program, "bzip2", the associated library "libbzip2", and all
+documentation, are copyright (C) 1996-2010 Julian R Seward.  All
+rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions
+are met:
+
+1. Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+2. The origin of this software must not be misrepresented; you must 
+   not claim that you wrote the original software.  If you use this 
+   software in a product, an acknowledgment in the product 
+   documentation would be appreciated but is not required.
+
+3. Altered source versions must be plainly marked as such, and must
+   not be misrepresented as being the original software.
+
+4. The name of the author may not be used to endorse or promote 
+   products derived from this software without specific prior written 
+   permission.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
+OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+Julian Seward, jseward@bzip.org
+bzip2/libbzip2 version 1.0.6 of 6 September 2010
+
+--------------------------------------------------------------------------

bzip2/blocksort.c

+
+/*-------------------------------------------------------------*/
+/*--- Block sorting machinery                               ---*/
+/*---                                           blocksort.c ---*/
+/*-------------------------------------------------------------*/
+
+/* ------------------------------------------------------------------
+   This file is part of bzip2/libbzip2, a program and library for
+   lossless, block-sorting data compression.
+
+   bzip2/libbzip2 version 1.0.6 of 6 September 2010
+   Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
+
+   Please read the WARNING, DISCLAIMER and PATENTS sections in the 
+   README file.
+
+   This program is released under the terms of the license contained
+   in the file LICENSE.
+   ------------------------------------------------------------------ */
+
+
+#include "bzlib_private.h"
+
+/*---------------------------------------------*/
+/*--- Fallback O(N log(N)^2) sorting        ---*/
+/*--- algorithm, for repetitive blocks      ---*/
+/*---------------------------------------------*/
+
+/*---------------------------------------------*/
+static 
+__inline__
+void fallbackSimpleSort ( UInt32* fmap, 
+                          UInt32* eclass, 
+                          Int32   lo, 
+                          Int32   hi )
+{
+   Int32 i, j, tmp;
+   UInt32 ec_tmp;
+
+   if (lo == hi) return;
+
+   if (hi - lo > 3) {
+      for ( i = hi-4; i >= lo; i-- ) {
+         tmp = fmap[i];
+         ec_tmp = eclass[tmp];
+         for ( j = i+4; j <= hi && ec_tmp > eclass[fmap[j]]; j += 4 )
+            fmap[j-4] = fmap[j];
+         fmap[j-4] = tmp;
+      }
+   }
+
+   for ( i = hi-1; i >= lo; i-- ) {
+      tmp = fmap[i];
+      ec_tmp = eclass[tmp];
+      for ( j = i+1; j <= hi && ec_tmp > eclass[fmap[j]]; j++ )
+         fmap[j-1] = fmap[j];
+      fmap[j-1] = tmp;
+   }
+}
+
+
+/*---------------------------------------------*/
+#define fswap(zz1, zz2) \
+   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
+
+#define fvswap(zzp1, zzp2, zzn)       \
+{                                     \
+   Int32 yyp1 = (zzp1);               \
+   Int32 yyp2 = (zzp2);               \
+   Int32 yyn  = (zzn);                \
+   while (yyn > 0) {                  \
+      fswap(fmap[yyp1], fmap[yyp2]);  \
+      yyp1++; yyp2++; yyn--;          \
+   }                                  \
+}
+
+
+#define fmin(a,b) ((a) < (b)) ? (a) : (b)
+
+#define fpush(lz,hz) { stackLo[sp] = lz; \
+                       stackHi[sp] = hz; \
+                       sp++; }
+
+#define fpop(lz,hz) { sp--;              \
+                      lz = stackLo[sp];  \
+                      hz = stackHi[sp]; }
+
+#define FALLBACK_QSORT_SMALL_THRESH 10
+#define FALLBACK_QSORT_STACK_SIZE   100
+
+
+static
+void fallbackQSort3 ( UInt32* fmap, 
+                      UInt32* eclass,
+                      Int32   loSt, 
+                      Int32   hiSt )
+{
+   Int32 unLo, unHi, ltLo, gtHi, n, m;
+   Int32 sp, lo, hi;
+   UInt32 med, r, r3;
+   Int32 stackLo[FALLBACK_QSORT_STACK_SIZE];
+   Int32 stackHi[FALLBACK_QSORT_STACK_SIZE];
+
+   r = 0;
+
+   sp = 0;
+   fpush ( loSt, hiSt );
+
+   while (sp > 0) {
+
+      AssertH ( sp < FALLBACK_QSORT_STACK_SIZE - 1, 1004 );
+
+      fpop ( lo, hi );
+      if (hi - lo < FALLBACK_QSORT_SMALL_THRESH) {
+         fallbackSimpleSort ( fmap, eclass, lo, hi );
+         continue;
+      }
+
+      /* Random partitioning.  Median of 3 sometimes fails to
+         avoid bad cases.  Median of 9 seems to help but 
+         looks rather expensive.  This too seems to work but
+         is cheaper.  Guidance for the magic constants 
+         7621 and 32768 is taken from Sedgewick's algorithms
+         book, chapter 35.
+      */
+      r = ((r * 7621) + 1) % 32768;
+      r3 = r % 3;
+      if (r3 == 0) med = eclass[fmap[lo]]; else
+      if (r3 == 1) med = eclass[fmap[(lo+hi)>>1]]; else
+                   med = eclass[fmap[hi]];
+
+      unLo = ltLo = lo;
+      unHi = gtHi = hi;
+
+      while (1) {
+         while (1) {
+            if (unLo > unHi) break;
+            n = (Int32)eclass[fmap[unLo]] - (Int32)med;
+            if (n == 0) { 
+               fswap(fmap[unLo], fmap[ltLo]); 
+               ltLo++; unLo++; 
+               continue; 
+            };
+            if (n > 0) break;
+            unLo++;
+         }
+         while (1) {
+            if (unLo > unHi) break;
+            n = (Int32)eclass[fmap[unHi]] - (Int32)med;
+            if (n == 0) { 
+               fswap(fmap[unHi], fmap[gtHi]); 
+               gtHi--; unHi--; 
+               continue; 
+            };
+            if (n < 0) break;
+            unHi--;
+         }
+         if (unLo > unHi) break;
+         fswap(fmap[unLo], fmap[unHi]); unLo++; unHi--;
+      }
+
+      AssertD ( unHi == unLo-1, "fallbackQSort3(2)" );
+
+      if (gtHi < ltLo) continue;
+
+      n = fmin(ltLo-lo, unLo-ltLo); fvswap(lo, unLo-n, n);
+      m = fmin(hi-gtHi, gtHi-unHi); fvswap(unLo, hi-m+1, m);
+
+      n = lo + unLo - ltLo - 1;
+      m = hi - (gtHi - unHi) + 1;
+
+      if (n - lo > hi - m) {
+         fpush ( lo, n );
+         fpush ( m, hi );
+      } else {
+         fpush ( m, hi );
+         fpush ( lo, n );
+      }
+   }
+}
+
+#undef fmin
+#undef fpush
+#undef fpop
+#undef fswap
+#undef fvswap
+#undef FALLBACK_QSORT_SMALL_THRESH
+#undef FALLBACK_QSORT_STACK_SIZE
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > 0
+      eclass exists for [0 .. nblock-1]
+      ((UChar*)eclass) [0 .. nblock-1] holds block
+      ptr exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)eclass) [0 .. nblock-1] holds block
+      All other areas of eclass destroyed
+      fmap [0 .. nblock-1] holds sorted order
+      bhtab [ 0 .. 2+(nblock/32) ] destroyed
+*/
+
+#define       SET_BH(zz)  bhtab[(zz) >> 5] |= (1 << ((zz) & 31))
+#define     CLEAR_BH(zz)  bhtab[(zz) >> 5] &= ~(1 << ((zz) & 31))
+#define     ISSET_BH(zz)  (bhtab[(zz) >> 5] & (1 << ((zz) & 31)))
+#define      WORD_BH(zz)  bhtab[(zz) >> 5]
+#define UNALIGNED_BH(zz)  ((zz) & 0x01f)
+
+static
+void fallbackSort ( UInt32* fmap, 
+                    UInt32* eclass, 
+                    UInt32* bhtab,
+                    Int32   nblock,
+                    Int32   verb )
+{
+   Int32 ftab[257];
+   Int32 ftabCopy[256];
+   Int32 H, i, j, k, l, r, cc, cc1;
+   Int32 nNotDone;
+   Int32 nBhtab;
+   UChar* eclass8 = (UChar*)eclass;
+
+   /*--
+      Initial 1-char radix sort to generate
+      initial fmap and initial BH bits.
+   --*/
+   if (verb >= 4)
+      VPrintf0 ( "        bucket sorting ...\n" );
+   for (i = 0; i < 257;    i++) ftab[i] = 0;
+   for (i = 0; i < nblock; i++) ftab[eclass8[i]]++;
+   for (i = 0; i < 256;    i++) ftabCopy[i] = ftab[i];
+   for (i = 1; i < 257;    i++) ftab[i] += ftab[i-1];
+
+   for (i = 0; i < nblock; i++) {
+      j = eclass8[i];
+      k = ftab[j] - 1;
+      ftab[j] = k;
+      fmap[k] = i;
+   }
+
+   nBhtab = 2 + (nblock / 32);
+   for (i = 0; i < nBhtab; i++) bhtab[i] = 0;
+   for (i = 0; i < 256; i++) SET_BH(ftab[i]);
+
+   /*--
+      Inductively refine the buckets.  Kind-of an
+      "exponential radix sort" (!), inspired by the
+      Manber-Myers suffix array construction algorithm.
+   --*/
+
+   /*-- set sentinel bits for block-end detection --*/
+   for (i = 0; i < 32; i++) { 
+      SET_BH(nblock + 2*i);
+      CLEAR_BH(nblock + 2*i + 1);
+   }
+
+   /*-- the log(N) loop --*/
+   H = 1;
+   while (1) {
+
+      if (verb >= 4) 
+         VPrintf1 ( "        depth %6d has ", H );
+
+      j = 0;
+      for (i = 0; i < nblock; i++) {
+         if (ISSET_BH(i)) j = i;
+         k = fmap[i] - H; if (k < 0) k += nblock;
+         eclass[k] = j;
+      }
+
+      nNotDone = 0;
+      r = -1;
+      while (1) {
+
+	 /*-- find the next non-singleton bucket --*/
+         k = r + 1;
+         while (ISSET_BH(k) && UNALIGNED_BH(k)) k++;
+         if (ISSET_BH(k)) {
+            while (WORD_BH(k) == 0xffffffff) k += 32;
+            while (ISSET_BH(k)) k++;
+         }
+         l = k - 1;
+         if (l >= nblock) break;
+         while (!ISSET_BH(k) && UNALIGNED_BH(k)) k++;
+         if (!ISSET_BH(k)) {
+            while (WORD_BH(k) == 0x00000000) k += 32;
+            while (!ISSET_BH(k)) k++;
+         }
+         r = k - 1;
+         if (r >= nblock) break;
+
+         /*-- now [l, r] bracket current bucket --*/
+         if (r > l) {
+            nNotDone += (r - l + 1);
+            fallbackQSort3 ( fmap, eclass, l, r );
+
+            /*-- scan bucket and generate header bits-- */
+            cc = -1;
+            for (i = l; i <= r; i++) {
+               cc1 = eclass[fmap[i]];
+               if (cc != cc1) { SET_BH(i); cc = cc1; };
+            }
+         }
+      }
+
+      if (verb >= 4) 
+         VPrintf1 ( "%6d unresolved strings\n", nNotDone );
+
+      H *= 2;
+      if (H > nblock || nNotDone == 0) break;
+   }
+
+   /*-- 
+      Reconstruct the original block in
+      eclass8 [0 .. nblock-1], since the
+      previous phase destroyed it.
+   --*/
+   if (verb >= 4)
+      VPrintf0 ( "        reconstructing block ...\n" );
+   j = 0;
+   for (i = 0; i < nblock; i++) {
+      while (ftabCopy[j] == 0) j++;
+      ftabCopy[j]--;
+      eclass8[fmap[i]] = (UChar)j;
+   }
+   AssertH ( j < 256, 1005 );
+}
+
+#undef       SET_BH
+#undef     CLEAR_BH
+#undef     ISSET_BH
+#undef      WORD_BH
+#undef UNALIGNED_BH
+
+
+/*---------------------------------------------*/
+/*--- The main, O(N^2 log(N)) sorting       ---*/
+/*--- algorithm.  Faster for "normal"       ---*/
+/*--- non-repetitive blocks.                ---*/
+/*---------------------------------------------*/
+
+/*---------------------------------------------*/
+static
+__inline__
+Bool mainGtU ( UInt32  i1, 
+               UInt32  i2,
+               UChar*  block, 
+               UInt16* quadrant,
+               UInt32  nblock,
+               Int32*  budget )
+{
+   Int32  k;
+   UChar  c1, c2;
+   UInt16 s1, s2;
+
+   AssertD ( i1 != i2, "mainGtU" );
+   /* 1 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 2 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 3 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 4 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 5 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 6 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 7 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 8 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 9 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 10 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 11 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+   /* 12 */
+   c1 = block[i1]; c2 = block[i2];
+   if (c1 != c2) return (c1 > c2);
+   i1++; i2++;
+
+   k = nblock + 8;
+
+   do {
+      /* 1 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 2 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 3 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 4 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 5 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 6 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 7 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+      /* 8 */
+      c1 = block[i1]; c2 = block[i2];
+      if (c1 != c2) return (c1 > c2);
+      s1 = quadrant[i1]; s2 = quadrant[i2];
+      if (s1 != s2) return (s1 > s2);
+      i1++; i2++;
+
+      if (i1 >= nblock) i1 -= nblock;
+      if (i2 >= nblock) i2 -= nblock;
+
+      k -= 8;
+      (*budget)--;
+   }
+      while (k >= 0);
+
+   return False;
+}
+
+
+/*---------------------------------------------*/
+/*--
+   Knuth's increments seem to work better
+   than Incerpi-Sedgewick here.  Possibly
+   because the number of elems to sort is
+   usually small, typically <= 20.
+--*/
+static
+Int32 incs[14] = { 1, 4, 13, 40, 121, 364, 1093, 3280,
+                   9841, 29524, 88573, 265720,
+                   797161, 2391484 };
+
+static
+void mainSimpleSort ( UInt32* ptr,
+                      UChar*  block,
+                      UInt16* quadrant,
+                      Int32   nblock,
+                      Int32   lo, 
+                      Int32   hi, 
+                      Int32   d,
+                      Int32*  budget )
+{
+   Int32 i, j, h, bigN, hp;
+   UInt32 v;
+
+   bigN = hi - lo + 1;
+   if (bigN < 2) return;
+
+   hp = 0;
+   while (incs[hp] < bigN) hp++;
+   hp--;
+
+   for (; hp >= 0; hp--) {
+      h = incs[hp];
+
+      i = lo + h;
+      while (True) {
+
+         /*-- copy 1 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         /*-- copy 2 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         /*-- copy 3 --*/
+         if (i > hi) break;
+         v = ptr[i];
+         j = i;
+         while ( mainGtU ( 
+                    ptr[j-h]+d, v+d, block, quadrant, nblock, budget 
+                 ) ) {
+            ptr[j] = ptr[j-h];
+            j = j - h;
+            if (j <= (lo + h - 1)) break;
+         }
+         ptr[j] = v;
+         i++;
+
+         if (*budget < 0) return;
+      }
+   }
+}
+
+
+/*---------------------------------------------*/
+/*--
+   The following is an implementation of
+   an elegant 3-way quicksort for strings,
+   described in a paper "Fast Algorithms for
+   Sorting and Searching Strings", by Robert
+   Sedgewick and Jon L. Bentley.
+--*/
+
+#define mswap(zz1, zz2) \
+   { Int32 zztmp = zz1; zz1 = zz2; zz2 = zztmp; }
+
+#define mvswap(zzp1, zzp2, zzn)       \
+{                                     \
+   Int32 yyp1 = (zzp1);               \
+   Int32 yyp2 = (zzp2);               \
+   Int32 yyn  = (zzn);                \
+   while (yyn > 0) {                  \
+      mswap(ptr[yyp1], ptr[yyp2]);    \
+      yyp1++; yyp2++; yyn--;          \
+   }                                  \
+}
+
+static 
+__inline__
+UChar mmed3 ( UChar a, UChar b, UChar c )
+{
+   UChar t;
+   if (a > b) { t = a; a = b; b = t; };
+   if (b > c) { 
+      b = c;
+      if (a > b) b = a;
+   }
+   return b;
+}
+
+#define mmin(a,b) ((a) < (b)) ? (a) : (b)
+
+#define mpush(lz,hz,dz) { stackLo[sp] = lz; \
+                          stackHi[sp] = hz; \
+                          stackD [sp] = dz; \
+                          sp++; }
+
+#define mpop(lz,hz,dz) { sp--;             \
+                         lz = stackLo[sp]; \
+                         hz = stackHi[sp]; \
+                         dz = stackD [sp]; }
+
+
+#define mnextsize(az) (nextHi[az]-nextLo[az])
+
+#define mnextswap(az,bz)                                        \
+   { Int32 tz;                                                  \
+     tz = nextLo[az]; nextLo[az] = nextLo[bz]; nextLo[bz] = tz; \
+     tz = nextHi[az]; nextHi[az] = nextHi[bz]; nextHi[bz] = tz; \
+     tz = nextD [az]; nextD [az] = nextD [bz]; nextD [bz] = tz; }
+
+
+#define MAIN_QSORT_SMALL_THRESH 20
+#define MAIN_QSORT_DEPTH_THRESH (BZ_N_RADIX + BZ_N_QSORT)
+#define MAIN_QSORT_STACK_SIZE 100
+
+static
+void mainQSort3 ( UInt32* ptr,
+                  UChar*  block,
+                  UInt16* quadrant,
+                  Int32   nblock,
+                  Int32   loSt, 
+                  Int32   hiSt, 
+                  Int32   dSt,
+                  Int32*  budget )
+{
+   Int32 unLo, unHi, ltLo, gtHi, n, m, med;
+   Int32 sp, lo, hi, d;
+
+   Int32 stackLo[MAIN_QSORT_STACK_SIZE];
+   Int32 stackHi[MAIN_QSORT_STACK_SIZE];
+   Int32 stackD [MAIN_QSORT_STACK_SIZE];
+
+   Int32 nextLo[3];
+   Int32 nextHi[3];
+   Int32 nextD [3];
+
+   sp = 0;
+   mpush ( loSt, hiSt, dSt );
+
+   while (sp > 0) {
+
+      AssertH ( sp < MAIN_QSORT_STACK_SIZE - 2, 1001 );
+
+      mpop ( lo, hi, d );
+      if (hi - lo < MAIN_QSORT_SMALL_THRESH || 
+          d > MAIN_QSORT_DEPTH_THRESH) {
+         mainSimpleSort ( ptr, block, quadrant, nblock, lo, hi, d, budget );
+         if (*budget < 0) return;
+         continue;
+      }
+
+      med = (Int32) 
+            mmed3 ( block[ptr[ lo         ]+d],
+                    block[ptr[ hi         ]+d],
+                    block[ptr[ (lo+hi)>>1 ]+d] );
+
+      unLo = ltLo = lo;
+      unHi = gtHi = hi;
+
+      while (True) {
+         while (True) {
+            if (unLo > unHi) break;
+            n = ((Int32)block[ptr[unLo]+d]) - med;
+            if (n == 0) { 
+               mswap(ptr[unLo], ptr[ltLo]); 
+               ltLo++; unLo++; continue; 
+            };
+            if (n >  0) break;
+            unLo++;
+         }
+         while (True) {
+            if (unLo > unHi) break;
+            n = ((Int32)block[ptr[unHi]+d]) - med;
+            if (n == 0) { 
+               mswap(ptr[unHi], ptr[gtHi]); 
+               gtHi--; unHi--; continue; 
+            };
+            if (n <  0) break;
+            unHi--;
+         }
+         if (unLo > unHi) break;
+         mswap(ptr[unLo], ptr[unHi]); unLo++; unHi--;
+      }
+
+      AssertD ( unHi == unLo-1, "mainQSort3(2)" );
+
+      if (gtHi < ltLo) {
+         mpush(lo, hi, d+1 );
+         continue;
+      }
+
+      n = mmin(ltLo-lo, unLo-ltLo); mvswap(lo, unLo-n, n);
+      m = mmin(hi-gtHi, gtHi-unHi); mvswap(unLo, hi-m+1, m);
+
+      n = lo + unLo - ltLo - 1;
+      m = hi - (gtHi - unHi) + 1;
+
+      nextLo[0] = lo;  nextHi[0] = n;   nextD[0] = d;
+      nextLo[1] = m;   nextHi[1] = hi;  nextD[1] = d;
+      nextLo[2] = n+1; nextHi[2] = m-1; nextD[2] = d+1;
+
+      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
+      if (mnextsize(1) < mnextsize(2)) mnextswap(1,2);
+      if (mnextsize(0) < mnextsize(1)) mnextswap(0,1);
+
+      AssertD (mnextsize(0) >= mnextsize(1), "mainQSort3(8)" );
+      AssertD (mnextsize(1) >= mnextsize(2), "mainQSort3(9)" );
+
+      mpush (nextLo[0], nextHi[0], nextD[0]);
+      mpush (nextLo[1], nextHi[1], nextD[1]);
+      mpush (nextLo[2], nextHi[2], nextD[2]);
+   }
+}
+
+#undef mswap
+#undef mvswap
+#undef mpush
+#undef mpop
+#undef mmin
+#undef mnextsize
+#undef mnextswap
+#undef MAIN_QSORT_SMALL_THRESH
+#undef MAIN_QSORT_DEPTH_THRESH
+#undef MAIN_QSORT_STACK_SIZE
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > N_OVERSHOOT
+      block32 exists for [0 .. nblock-1 +N_OVERSHOOT]
+      ((UChar*)block32) [0 .. nblock-1] holds block
+      ptr exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)block32) [0 .. nblock-1] holds block
+      All other areas of block32 destroyed
+      ftab [0 .. 65536 ] destroyed
+      ptr [0 .. nblock-1] holds sorted order
+      if (*budget < 0), sorting was abandoned
+*/
+
+#define BIGFREQ(b) (ftab[((b)+1) << 8] - ftab[(b) << 8])
+#define SETMASK (1 << 21)
+#define CLEARMASK (~(SETMASK))
+
+static
+void mainSort ( UInt32* ptr, 
+                UChar*  block,
+                UInt16* quadrant, 
+                UInt32* ftab,
+                Int32   nblock,
+                Int32   verb,
+                Int32*  budget )
+{
+   Int32  i, j, k, ss, sb;
+   Int32  runningOrder[256];
+   Bool   bigDone[256];
+   Int32  copyStart[256];
+   Int32  copyEnd  [256];
+   UChar  c1;
+   Int32  numQSorted;
+   UInt16 s;
+   if (verb >= 4) VPrintf0 ( "        main sort initialise ...\n" );
+
+   /*-- set up the 2-byte frequency table --*/
+   for (i = 65536; i >= 0; i--) ftab[i] = 0;
+
+   j = block[0] << 8;
+   i = nblock-1;
+   for (; i >= 3; i -= 4) {
+      quadrant[i] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
+      ftab[j]++;
+      quadrant[i-1] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-1]) << 8);
+      ftab[j]++;
+      quadrant[i-2] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-2]) << 8);
+      ftab[j]++;
+      quadrant[i-3] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i-3]) << 8);
+      ftab[j]++;
+   }
+   for (; i >= 0; i--) {
+      quadrant[i] = 0;
+      j = (j >> 8) | ( ((UInt16)block[i]) << 8);
+      ftab[j]++;
+   }
+
+   /*-- (emphasises close relationship of block & quadrant) --*/
+   for (i = 0; i < BZ_N_OVERSHOOT; i++) {
+      block   [nblock+i] = block[i];
+      quadrant[nblock+i] = 0;
+   }
+
+   if (verb >= 4) VPrintf0 ( "        bucket sorting ...\n" );
+
+   /*-- Complete the initial radix sort --*/
+   for (i = 1; i <= 65536; i++) ftab[i] += ftab[i-1];
+
+   s = block[0] << 8;
+   i = nblock-1;
+   for (; i >= 3; i -= 4) {
+      s = (s >> 8) | (block[i] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i;
+      s = (s >> 8) | (block[i-1] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-1;
+      s = (s >> 8) | (block[i-2] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-2;
+      s = (s >> 8) | (block[i-3] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i-3;
+   }
+   for (; i >= 0; i--) {
+      s = (s >> 8) | (block[i] << 8);
+      j = ftab[s] -1;
+      ftab[s] = j;
+      ptr[j] = i;
+   }
+
+   /*--
+      Now ftab contains the first loc of every small bucket.
+      Calculate the running order, from smallest to largest
+      big bucket.
+   --*/
+   for (i = 0; i <= 255; i++) {
+      bigDone     [i] = False;
+      runningOrder[i] = i;
+   }
+
+   {
+      Int32 vv;
+      Int32 h = 1;
+      do h = 3 * h + 1; while (h <= 256);
+      do {
+         h = h / 3;
+         for (i = h; i <= 255; i++) {
+            vv = runningOrder[i];
+            j = i;
+            while ( BIGFREQ(runningOrder[j-h]) > BIGFREQ(vv) ) {
+               runningOrder[j] = runningOrder[j-h];
+               j = j - h;
+               if (j <= (h - 1)) goto zero;
+            }
+            zero:
+            runningOrder[j] = vv;
+         }
+      } while (h != 1);
+   }
+
+   /*--
+      The main sorting loop.
+   --*/
+
+   numQSorted = 0;
+
+   for (i = 0; i <= 255; i++) {
+
+      /*--
+         Process big buckets, starting with the least full.
+         Basically this is a 3-step process in which we call
+         mainQSort3 to sort the small buckets [ss, j], but
+         also make a big effort to avoid the calls if we can.
+      --*/
+      ss = runningOrder[i];
+
+      /*--
+         Step 1:
+         Complete the big bucket [ss] by quicksorting
+         any unsorted small buckets [ss, j], for j != ss.  
+         Hopefully previous pointer-scanning phases have already
+         completed many of the small buckets [ss, j], so
+         we don't have to sort them at all.
+      --*/
+      for (j = 0; j <= 255; j++) {
+         if (j != ss) {
+            sb = (ss << 8) + j;
+            if ( ! (ftab[sb] & SETMASK) ) {
+               Int32 lo = ftab[sb]   & CLEARMASK;
+               Int32 hi = (ftab[sb+1] & CLEARMASK) - 1;
+               if (hi > lo) {
+                  if (verb >= 4)
+                     VPrintf4 ( "        qsort [0x%x, 0x%x]   "
+                                "done %d   this %d\n",
+                                ss, j, numQSorted, hi - lo + 1 );
+                  mainQSort3 ( 
+                     ptr, block, quadrant, nblock, 
+                     lo, hi, BZ_N_RADIX, budget 
+                  );   
+                  numQSorted += (hi - lo + 1);
+                  if (*budget < 0) return;
+               }
+            }
+            ftab[sb] |= SETMASK;
+         }
+      }
+
+      AssertH ( !bigDone[ss], 1006 );
+
+      /*--
+         Step 2:
+         Now scan this big bucket [ss] so as to synthesise the
+         sorted order for small buckets [t, ss] for all t,
+         including, magically, the bucket [ss,ss] too.
+         This will avoid doing Real Work in subsequent Step 1's.
+      --*/
+      {
+         for (j = 0; j <= 255; j++) {
+            copyStart[j] =  ftab[(j << 8) + ss]     & CLEARMASK;
+            copyEnd  [j] = (ftab[(j << 8) + ss + 1] & CLEARMASK) - 1;
+         }
+         for (j = ftab[ss << 8] & CLEARMASK; j < copyStart[ss]; j++) {
+            k = ptr[j]-1; if (k < 0) k += nblock;
+            c1 = block[k];
+            if (!bigDone[c1])
+               ptr[ copyStart[c1]++ ] = k;
+         }
+         for (j = (ftab[(ss+1) << 8] & CLEARMASK) - 1; j > copyEnd[ss]; j--) {
+            k = ptr[j]-1; if (k < 0) k += nblock;
+            c1 = block[k];
+            if (!bigDone[c1]) 
+               ptr[ copyEnd[c1]-- ] = k;
+         }
+      }
+
+      AssertH ( (copyStart[ss]-1 == copyEnd[ss])
+                || 
+                /* Extremely rare case missing in bzip2-1.0.0 and 1.0.1.
+                   Necessity for this case is demonstrated by compressing 
+                   a sequence of approximately 48.5 million of character 
+                   251; 1.0.0/1.0.1 will then die here. */
+                (copyStart[ss] == 0 && copyEnd[ss] == nblock-1),
+                1007 )
+
+      for (j = 0; j <= 255; j++) ftab[(j << 8) + ss] |= SETMASK;
+
+      /*--
+         Step 3:
+         The [ss] big bucket is now done.  Record this fact,
+         and update the quadrant descriptors.  Remember to
+         update quadrants in the overshoot area too, if
+         necessary.  The "if (i < 255)" test merely skips
+         this updating for the last bucket processed, since
+         updating for the last bucket is pointless.
+
+         The quadrant array provides a way to incrementally
+         cache sort orderings, as they appear, so as to 
+         make subsequent comparisons in fullGtU() complete
+         faster.  For repetitive blocks this makes a big
+         difference (but not big enough to be able to avoid
+         the fallback sorting mechanism, exponential radix sort).
+
+         The precise meaning is: at all times:
+
+            for 0 <= i < nblock and 0 <= j <= nblock
+
+            if block[i] != block[j], 
+
+               then the relative values of quadrant[i] and 
+                    quadrant[j] are meaningless.
+
+               else {
+                  if quadrant[i] < quadrant[j]
+                     then the string starting at i lexicographically
+                     precedes the string starting at j
+
+                  else if quadrant[i] > quadrant[j]
+                     then the string starting at j lexicographically
+                     precedes the string starting at i
+
+                  else
+                     the relative ordering of the strings starting
+                     at i and j has not yet been determined.
+               }
+      --*/
+      bigDone[ss] = True;
+
+      if (i < 255) {
+         Int32 bbStart  = ftab[ss << 8] & CLEARMASK;
+         Int32 bbSize   = (ftab[(ss+1) << 8] & CLEARMASK) - bbStart;
+         Int32 shifts   = 0;
+
+         while ((bbSize >> shifts) > 65534) shifts++;
+
+         for (j = bbSize-1; j >= 0; j--) {
+            Int32 a2update     = ptr[bbStart + j];
+            UInt16 qVal        = (UInt16)(j >> shifts);
+            quadrant[a2update] = qVal;
+            if (a2update < BZ_N_OVERSHOOT)
+               quadrant[a2update + nblock] = qVal;
+         }
+         AssertH ( ((bbSize-1) >> shifts) <= 65535, 1002 );
+      }
+
+   }
+
+   if (verb >= 4)
+      VPrintf3 ( "        %d pointers, %d sorted, %d scanned\n",
+                 nblock, numQSorted, nblock - numQSorted );
+}
+
+#undef BIGFREQ
+#undef SETMASK
+#undef CLEARMASK
+
+
+/*---------------------------------------------*/
+/* Pre:
+      nblock > 0
+      arr2 exists for [0 .. nblock-1 +N_OVERSHOOT]
+      ((UChar*)arr2)  [0 .. nblock-1] holds block
+      arr1 exists for [0 .. nblock-1]
+
+   Post:
+      ((UChar*)arr2) [0 .. nblock-1] holds block
+      All other areas of block destroyed
+      ftab [ 0 .. 65536 ] destroyed
+      arr1 [0 .. nblock-1] holds sorted order
+*/
+void BZ2_blockSort ( EState* s )
+{
+   UInt32* ptr    = s->ptr; 
+   UChar*  block  = s->block;
+   UInt32* ftab   = s->ftab;
+   Int32   nblock = s->nblock;
+   Int32   verb   = s->verbosity;
+   Int32   wfact  = s->workFactor;
+   UInt16* quadrant;
+   Int32   budget;
+   Int32   budgetInit;
+   Int32   i;
+
+   if (nblock < 10000) {
+      fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
+   } else {
+      /* Calculate the location for quadrant, remembering to get
+         the alignment right.  Assumes that &(block[0]) is at least
+         2-byte aligned -- this should be ok since block is really
+         the first section of arr2.
+      */
+      i = nblock+BZ_N_OVERSHOOT;
+      if (i & 1) i++;
+      quadrant = (UInt16*)(&(block[i]));
+
+      /* (wfact-1) / 3 puts the default-factor-30
+         transition point at very roughly the same place as 
+         with v0.1 and v0.9.0.  
+         Not that it particularly matters any more, since the
+         resulting compressed stream is now the same regardless
+         of whether or not we use the main sort or fallback sort.
+      */
+      if (wfact < 1  ) wfact = 1;
+      if (wfact > 100) wfact = 100;
+      budgetInit = nblock * ((wfact-1) / 3);
+      budget = budgetInit;
+
+      mainSort ( ptr, block, quadrant, ftab, nblock, verb, &budget );
+      if (verb >= 3) 
+         VPrintf3 ( "      %d work, %d block, ratio %5.2f\n",
+                    budgetInit - budget,
+                    nblock, 
+                    (float)(budgetInit - budget) /
+                    (float)(nblock==0 ? 1 : nblock) ); 
+      if (budget < 0) {
+         if (verb >= 2) 
+            VPrintf0 ( "    too repetitive; using fallback"
+                       " sorting algorithm\n" );
+         fallbackSort ( s->arr1, s->arr2, ftab, nblock, verb );
+      }
+   }
+
+   s->origPtr = -1;
+   for (i = 0; i < s->nblock; i++)
+      if (ptr[i] == 0)
+         { s->origPtr = i; break; };
+
+   AssertH( s->origPtr != -1, 1003 );
+}
+
+
+/*-------------------------------------------------------------*/
+/*--- end                                       blocksort.c ---*/
+/*-------------------------------------------------------------*/
+
+/*-------------------------------------------------------------*/
+/*--- Library top-level functions.                          ---*/
+/*---                                               bzlib.c ---*/
+/*-------------------------------------------------------------*/
+
+/* ------------------------------------------------------------------
+   This file is part of bzip2/libbzip2, a program and library for
+   lossless, block-sorting data compression.
+
+   bzip2/libbzip2 version 1.0.6 of 6 September 2010
+   Copyright (C) 1996-2010 Julian Seward <jseward@bzip.org>
+
+   Please read the WARNING, DISCLAIMER and PATENTS sections in the 
+   README file.
+
+   This program is released under the terms of the license contained
+   in the file LICENSE.
+   ------------------------------------------------------------------ */
+
+/* CHANGES
+   0.9.0    -- original version.
+   0.9.0a/b -- no changes in this file.
+   0.9.0c   -- made zero-length BZ_FLUSH work correctly in bzCompress().
+     fixed bzWrite/bzRead to ignore zero-length requests.
+     fixed bzread to correctly handle read requests after EOF.
+     wrong parameter order in call to bzDecompressInit in
+     bzBuffToBuffDecompress.  Fixed.
+*/
+
+#include "bzlib_private.h"
+
+
+/*---------------------------------------------------*/
+/*--- Compression stuff                           ---*/
+/*---------------------------------------------------*/
+
+
+/*---------------------------------------------------*/
+#ifndef BZ_NO_STDIO
+void BZ2_bz__AssertH__fail ( int errcode )
+{
+   fprintf(stderr, 
+      "\n\nbzip2/libbzip2: internal error number %d.\n"
+      "This is a bug in bzip2/libbzip2, %s.\n"
+      "Please report it to me at: jseward@bzip.org.  If this happened\n"
+      "when you were using some program which uses libbzip2 as a\n"
+      "component, you should also report this bug to the author(s)\n"
+      "of that program.  Please make an effort to report this bug;\n"
+      "timely and accurate bug reports eventually lead to higher\n"
+      "quality software.  Thanks.  Julian Seward, 10 December 2007.\n\n",
+      errcode,
+      BZ2_bzlibVersion()
+   );
+
+   if (errcode == 1007) {
+   fprintf(stderr,
+      "\n*** A special note about internal error number 1007 ***\n"
+      "\n"
+      "Experience suggests that a common cause of i.e. 1007\n"
+      "is unreliable memory or other hardware.  The 1007 assertion\n"
+      "just happens to cross-check the results of huge numbers of\n"
+      "memory reads/writes, and so acts (unintendedly) as a stress\n"
+      "test of your memory system.\n"
+      "\n"
+      "I suggest the following: try compressing the file again,\n"
+      "possibly monitoring progress in detail with the -vv flag.\n"
+      "\n"
+      "* If the error cannot be reproduced, and/or happens at different\n"
+      "  points in compression, you may have a flaky memory system.\n"
+      "  Try a memory-test program.  I have used Memtest86\n"
+      "  (www.memtest86.com).  At the time of writing it is free (GPLd).\n"
+      "  Memtest86 tests memory much more thorougly than your BIOSs\n"
+      "  power-on test, and may find failures that the BIOS doesn't.\n"
+      "\n"
+      "* If the error can be repeatably reproduced, this is a bug in\n"
+      "  bzip2, and I would very much like to hear about it.  Please\n"
+      "  let me know, and, ideally, save a copy of the file causing the\n"
+      "  problem -- without which I will be unable to investigate it.\n"
+      "\n"
+   );
+   }
+
+   exit(3);
+}
+#endif
+
+
+/*---------------------------------------------------*/
+static
+int bz_config_ok ( void )
+{
+   if (sizeof(int)   != 4) return 0;
+   if (sizeof(short) != 2) return 0;
+   if (sizeof(char)  != 1) return 0;
+   return 1;
+}
+
+
+/*---------------------------------------------------*/
+static
+void* default_bzalloc ( void* opaque, Int32 items, Int32 size )
+{
+   void* v = malloc ( items * size );
+   return v;
+}
+
+static
+void default_bzfree ( void* opaque, void* addr )
+{
+   if (addr != NULL) free ( addr );
+}
+
+
+/*---------------------------------------------------*/
+static
+void prepare_new_block ( EState* s )
+{
+   Int32 i;
+   s->nblock = 0;
+   s->numZ = 0;
+   s->state_out_pos = 0;
+   BZ_INITIALISE_CRC ( s->blockCRC );
+   for (i = 0; i < 256; i++) s->inUse[i] = False;
+   s->blockNo++;
+}
+
+
+/*---------------------------------------------------*/
+static
+void init_RL ( EState* s )
+{
+   s->state_in_ch  = 256;
+   s->state_in_len = 0;
+}
+
+
+static
+Bool isempty_RL ( EState* s )
+{
+   if (s->state_in_ch < 256 && s->state_in_len > 0)
+      return False; else
+      return True;
+}
+
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzCompressInit) 
+                    ( bz_stream* strm, 
+                     int        blockSize100k,
+                     int        verbosity,
+                     int        workFactor )
+{
+   Int32   n;
+   EState* s;
+
+   if (!bz_config_ok()) return BZ_CONFIG_ERROR;
+
+   if (strm == NULL || 
+       blockSize100k < 1 || blockSize100k > 9 ||
+       workFactor < 0 || workFactor > 250)
+     return BZ_PARAM_ERROR;
+
+   if (workFactor == 0) workFactor = 30;
+   if (strm->bzalloc == NULL) strm->bzalloc = default_bzalloc;
+   if (strm->bzfree == NULL) strm->bzfree = default_bzfree;
+
+   s = BZALLOC( sizeof(EState) );
+   if (s == NULL) return BZ_MEM_ERROR;
+   s->strm = strm;
+
+   s->arr1 = NULL;
+   s->arr2 = NULL;
+   s->ftab = NULL;
+
+   n       = 100000 * blockSize100k;
+   s->arr1 = BZALLOC( n                  * sizeof(UInt32) );
+   s->arr2 = BZALLOC( (n+BZ_N_OVERSHOOT) * sizeof(UInt32) );
+   s->ftab = BZALLOC( 65537              * sizeof(UInt32) );
+
+   if (s->arr1 == NULL || s->arr2 == NULL || s->ftab == NULL) {
+      if (s->arr1 != NULL) BZFREE(s->arr1);
+      if (s->arr2 != NULL) BZFREE(s->arr2);
+      if (s->ftab != NULL) BZFREE(s->ftab);
+      if (s       != NULL) BZFREE(s);
+      return BZ_MEM_ERROR;
+   }
+
+   s->blockNo           = 0;
+   s->state             = BZ_S_INPUT;
+   s->mode              = BZ_M_RUNNING;
+   s->combinedCRC       = 0;
+   s->blockSize100k     = blockSize100k;
+   s->nblockMAX         = 100000 * blockSize100k - 19;
+   s->verbosity         = verbosity;
+   s->workFactor        = workFactor;
+
+   s->block             = (UChar*)s->arr2;
+   s->mtfv              = (UInt16*)s->arr1;
+   s->zbits             = NULL;
+   s->ptr               = (UInt32*)s->arr1;
+
+   strm->state          = s;
+   strm->total_in_lo32  = 0;
+   strm->total_in_hi32  = 0;
+   strm->total_out_lo32 = 0;
+   strm->total_out_hi32 = 0;
+   init_RL ( s );
+   prepare_new_block ( s );
+   return BZ_OK;
+}
+
+
+/*---------------------------------------------------*/
+static
+void add_pair_to_block ( EState* s )
+{
+   Int32 i;
+   UChar ch = (UChar)(s->state_in_ch);
+   for (i = 0; i < s->state_in_len; i++) {
+      BZ_UPDATE_CRC( s->blockCRC, ch );
+   }
+   s->inUse[s->state_in_ch] = True;
+   switch (s->state_in_len) {
+      case 1:
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         break;
+      case 2:
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         break;
+      case 3:
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         break;
+      default:
+         s->inUse[s->state_in_len-4] = True;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = (UChar)ch; s->nblock++;
+         s->block[s->nblock] = ((UChar)(s->state_in_len-4));
+         s->nblock++;
+         break;
+   }
+}
+
+
+/*---------------------------------------------------*/
+static
+void flush_RL ( EState* s )
+{
+   if (s->state_in_ch < 256) add_pair_to_block ( s );
+   init_RL ( s );
+}
+
+
+/*---------------------------------------------------*/
+#define ADD_CHAR_TO_BLOCK(zs,zchh0)               \
+{                                                 \
+   UInt32 zchh = (UInt32)(zchh0);                 \
+   /*-- fast track the common case --*/           \
+   if (zchh != zs->state_in_ch &&                 \
+       zs->state_in_len == 1) {                   \
+      UChar ch = (UChar)(zs->state_in_ch);        \
+      BZ_UPDATE_CRC( zs->blockCRC, ch );          \
+      zs->inUse[zs->state_in_ch] = True;          \
+      zs->block[zs->nblock] = (UChar)ch;          \
+      zs->nblock++;                               \
+      zs->state_in_ch = zchh;                     \
+   }                                              \
+   else                                           \
+   /*-- general, uncommon cases --*/              \
+   if (zchh != zs->state_in_ch ||                 \
+      zs->state_in_len == 255) {                  \
+      if (zs->state_in_ch < 256)                  \
+         add_pair_to_block ( zs );                \
+      zs->state_in_ch = zchh;                     \
+      zs->state_in_len = 1;                       \
+   } else {                                       \
+      zs->state_in_len++;                         \
+   }                                              \
+}
+
+
+/*---------------------------------------------------*/
+static
+Bool copy_input_until_stop ( EState* s )
+{
+   Bool progress_in = False;
+
+   if (s->mode == BZ_M_RUNNING) {
+
+      /*-- fast track the common case --*/
+      while (True) {
+         /*-- block full? --*/
+         if (s->nblock >= s->nblockMAX) break;
+         /*-- no input? --*/
+         if (s->strm->avail_in == 0) break;
+         progress_in = True;
+         ADD_CHAR_TO_BLOCK ( s, (UInt32)(*((UChar*)(s->strm->next_in))) ); 
+         s->strm->next_in++;
+         s->strm->avail_in--;
+         s->strm->total_in_lo32++;
+         if (s->strm->total_in_lo32 == 0) s->strm->total_in_hi32++;
+      }
+
+   } else {
+
+      /*-- general, uncommon case --*/
+      while (True) {
+         /*-- block full? --*/
+         if (s->nblock >= s->nblockMAX) break;
+         /*-- no input? --*/
+         if (s->strm->avail_in == 0) break;
+         /*-- flush/finish end? --*/
+         if (s->avail_in_expect == 0) break;
+         progress_in = True;
+         ADD_CHAR_TO_BLOCK ( s, (UInt32)(*((UChar*)(s->strm->next_in))) ); 
+         s->strm->next_in++;
+         s->strm->avail_in--;
+         s->strm->total_in_lo32++;
+         if (s->strm->total_in_lo32 == 0) s->strm->total_in_hi32++;
+         s->avail_in_expect--;
+      }
+   }
+   return progress_in;
+}
+
+
+/*---------------------------------------------------*/
+static
+Bool copy_output_until_stop ( EState* s )
+{
+   Bool progress_out = False;
+
+   while (True) {
+
+      /*-- no output space? --*/
+      if (s->strm->avail_out == 0) break;
+
+      /*-- block done? --*/
+      if (s->state_out_pos >= s->numZ) break;
+
+      progress_out = True;
+      *(s->strm->next_out) = s->zbits[s->state_out_pos];
+      s->state_out_pos++;
+      s->strm->avail_out--;
+      s->strm->next_out++;
+      s->strm->total_out_lo32++;
+      if (s->strm->total_out_lo32 == 0) s->strm->total_out_hi32++;
+   }
+
+   return progress_out;
+}
+
+
+/*---------------------------------------------------*/
+static
+Bool handle_compress ( bz_stream* strm )
+{
+   Bool progress_in  = False;
+   Bool progress_out = False;
+   EState* s = strm->state;
+   
+   while (True) {
+
+      if (s->state == BZ_S_OUTPUT) {
+         progress_out |= copy_output_until_stop ( s );
+         if (s->state_out_pos < s->numZ) break;
+         if (s->mode == BZ_M_FINISHING && 
+             s->avail_in_expect == 0 &&
+             isempty_RL(s)) break;
+         prepare_new_block ( s );
+         s->state = BZ_S_INPUT;
+         if (s->mode == BZ_M_FLUSHING && 
+             s->avail_in_expect == 0 &&
+             isempty_RL(s)) break;
+      }
+
+      if (s->state == BZ_S_INPUT) {
+         progress_in |= copy_input_until_stop ( s );
+         if (s->mode != BZ_M_RUNNING && s->avail_in_expect == 0) {
+            flush_RL ( s );
+            BZ2_compressBlock ( s, (Bool)(s->mode == BZ_M_FINISHING) );
+            s->state = BZ_S_OUTPUT;
+         }
+         else
+         if (s->nblock >= s->nblockMAX) {
+            BZ2_compressBlock ( s, False );
+            s->state = BZ_S_OUTPUT;
+         }
+         else
+         if (s->strm->avail_in == 0) {
+            break;
+         }
+      }
+
+   }
+
+   return progress_in || progress_out;
+}
+
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzCompress) ( bz_stream *strm, int action )
+{
+   Bool progress;
+   EState* s;
+   if (strm == NULL) return BZ_PARAM_ERROR;
+   s = strm->state;
+   if (s == NULL) return BZ_PARAM_ERROR;
+   if (s->strm != strm) return BZ_PARAM_ERROR;
+
+   preswitch:
+   switch (s->mode) {
+
+      case BZ_M_IDLE:
+         return BZ_SEQUENCE_ERROR;
+
+      case BZ_M_RUNNING:
+         if (action == BZ_RUN) {
+            progress = handle_compress ( strm );
+            return progress ? BZ_RUN_OK : BZ_PARAM_ERROR;
+         } 
+         else
+	 if (action == BZ_FLUSH) {
+            s->avail_in_expect = strm->avail_in;
+            s->mode = BZ_M_FLUSHING;
+            goto preswitch;
+         }
+         else
+         if (action == BZ_FINISH) {
+            s->avail_in_expect = strm->avail_in;
+            s->mode = BZ_M_FINISHING;
+            goto preswitch;
+         }
+         else 
+            return BZ_PARAM_ERROR;
+
+      case BZ_M_FLUSHING:
+         if (action != BZ_FLUSH) return BZ_SEQUENCE_ERROR;
+         if (s->avail_in_expect != s->strm->avail_in) 
+            return BZ_SEQUENCE_ERROR;
+         progress = handle_compress ( strm );
+         if (s->avail_in_expect > 0 || !isempty_RL(s) ||
+             s->state_out_pos < s->numZ) return BZ_FLUSH_OK;
+         s->mode = BZ_M_RUNNING;
+         return BZ_RUN_OK;
+
+      case BZ_M_FINISHING:
+         if (action != BZ_FINISH) return BZ_SEQUENCE_ERROR;
+         if (s->avail_in_expect != s->strm->avail_in) 
+            return BZ_SEQUENCE_ERROR;
+         progress = handle_compress ( strm );
+         if (!progress) return BZ_SEQUENCE_ERROR;
+         if (s->avail_in_expect > 0 || !isempty_RL(s) ||
+             s->state_out_pos < s->numZ) return BZ_FINISH_OK;
+         s->mode = BZ_M_IDLE;
+         return BZ_STREAM_END;
+   }
+   return BZ_OK; /*--not reached--*/
+}
+
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzCompressEnd)  ( bz_stream *strm )
+{
+   EState* s;
+   if (strm == NULL) return BZ_PARAM_ERROR;
+   s = strm->state;
+   if (s == NULL) return BZ_PARAM_ERROR;
+   if (s->strm != strm) return BZ_PARAM_ERROR;
+
+   if (s->arr1 != NULL) BZFREE(s->arr1);
+   if (s->arr2 != NULL) BZFREE(s->arr2);
+   if (s->ftab != NULL) BZFREE(s->ftab);
+   BZFREE(strm->state);
+
+   strm->state = NULL;   
+
+   return BZ_OK;
+}
+
+
+/*---------------------------------------------------*/
+/*--- Decompression stuff                         ---*/
+/*---------------------------------------------------*/
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzDecompressInit) 
+                     ( bz_stream* strm, 
+                       int        verbosity,
+                       int        small )
+{
+   DState* s;
+
+   if (!bz_config_ok()) return BZ_CONFIG_ERROR;
+
+   if (strm == NULL) return BZ_PARAM_ERROR;
+   if (small != 0 && small != 1) return BZ_PARAM_ERROR;
+   if (verbosity < 0 || verbosity > 4) return BZ_PARAM_ERROR;
+
+   if (strm->bzalloc == NULL) strm->bzalloc = default_bzalloc;
+   if (strm->bzfree == NULL) strm->bzfree = default_bzfree;
+
+   s = BZALLOC( sizeof(DState) );
+   if (s == NULL) return BZ_MEM_ERROR;
+   s->strm                  = strm;
+   strm->state              = s;
+   s->state                 = BZ_X_MAGIC_1;
+   s->bsLive                = 0;
+   s->bsBuff                = 0;
+   s->calculatedCombinedCRC = 0;
+   strm->total_in_lo32      = 0;
+   strm->total_in_hi32      = 0;
+   strm->total_out_lo32     = 0;
+   strm->total_out_hi32     = 0;
+   s->smallDecompress       = (Bool)small;
+   s->ll4                   = NULL;
+   s->ll16                  = NULL;
+   s->tt                    = NULL;
+   s->currBlockNo           = 0;
+   s->verbosity             = verbosity;
+
+   return BZ_OK;
+}
+
+
+/*---------------------------------------------------*/
+/* Return  True iff data corruption is discovered.
+   Returns False if there is no problem.
+*/
+static
+Bool unRLE_obuf_to_output_FAST ( DState* s )
+{
+   UChar k1;
+
+   if (s->blockRandomised) {
+
+      while (True) {
+         /* try to finish existing run */
+         while (True) {
+            if (s->strm->avail_out == 0) return False;
+            if (s->state_out_len == 0) break;
+            *( (UChar*)(s->strm->next_out) ) = s->state_out_ch;
+            BZ_UPDATE_CRC ( s->calculatedBlockCRC, s->state_out_ch );
+            s->state_out_len--;
+            s->strm->next_out++;
+            s->strm->avail_out--;
+            s->strm->total_out_lo32++;
+            if (s->strm->total_out_lo32 == 0) s->strm->total_out_hi32++;
+         }
+
+         /* can a new run be started? */
+         if (s->nblock_used == s->save_nblock+1) return False;
+               
+         /* Only caused by corrupt data stream? */
+         if (s->nblock_used > s->save_nblock+1)
+            return True;
+   
+         s->state_out_len = 1;
+         s->state_out_ch = s->k0;
+         BZ_GET_FAST(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 2;
+         BZ_GET_FAST(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 3;
+         BZ_GET_FAST(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         BZ_GET_FAST(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         s->state_out_len = ((Int32)k1) + 4;
+         BZ_GET_FAST(s->k0); BZ_RAND_UPD_MASK; 
+         s->k0 ^= BZ_RAND_MASK; s->nblock_used++;
+      }
+
+   } else {
+
+      /* restore */
+      UInt32        c_calculatedBlockCRC = s->calculatedBlockCRC;
+      UChar         c_state_out_ch       = s->state_out_ch;
+      Int32         c_state_out_len      = s->state_out_len;
+      Int32         c_nblock_used        = s->nblock_used;
+      Int32         c_k0                 = s->k0;
+      UInt32*       c_tt                 = s->tt;
+      UInt32        c_tPos               = s->tPos;
+      char*         cs_next_out          = s->strm->next_out;
+      unsigned int  cs_avail_out         = s->strm->avail_out;
+      Int32         ro_blockSize100k     = s->blockSize100k;
+      /* end restore */
+
+      UInt32       avail_out_INIT = cs_avail_out;
+      Int32        s_save_nblockPP = s->save_nblock+1;
+      unsigned int total_out_lo32_old;
+
+      while (True) {
+
+         /* try to finish existing run */
+         if (c_state_out_len > 0) {
+            while (True) {
+               if (cs_avail_out == 0) goto return_notr;
+               if (c_state_out_len == 1) break;
+               *( (UChar*)(cs_next_out) ) = c_state_out_ch;
+               BZ_UPDATE_CRC ( c_calculatedBlockCRC, c_state_out_ch );
+               c_state_out_len--;
+               cs_next_out++;
+               cs_avail_out--;
+            }
+            s_state_out_len_eq_one:
+            {
+               if (cs_avail_out == 0) { 
+                  c_state_out_len = 1; goto return_notr;
+               };
+               *( (UChar*)(cs_next_out) ) = c_state_out_ch;
+               BZ_UPDATE_CRC ( c_calculatedBlockCRC, c_state_out_ch );
+               cs_next_out++;
+               cs_avail_out--;
+            }
+         }   
+         /* Only caused by corrupt data stream? */
+         if (c_nblock_used > s_save_nblockPP)
+            return True;
+
+         /* can a new run be started? */
+         if (c_nblock_used == s_save_nblockPP) {
+            c_state_out_len = 0; goto return_notr;
+         };   
+         c_state_out_ch = c_k0;
+         BZ_GET_FAST_C(k1); c_nblock_used++;
+         if (k1 != c_k0) { 
+            c_k0 = k1; goto s_state_out_len_eq_one; 
+         };
+         if (c_nblock_used == s_save_nblockPP) 
+            goto s_state_out_len_eq_one;
+   
+         c_state_out_len = 2;
+         BZ_GET_FAST_C(k1); c_nblock_used++;
+         if (c_nblock_used == s_save_nblockPP) continue;
+         if (k1 != c_k0) { c_k0 = k1; continue; };
+   
+         c_state_out_len = 3;
+         BZ_GET_FAST_C(k1); c_nblock_used++;
+         if (c_nblock_used == s_save_nblockPP) continue;
+         if (k1 != c_k0) { c_k0 = k1; continue; };
+   
+         BZ_GET_FAST_C(k1); c_nblock_used++;
+         c_state_out_len = ((Int32)k1) + 4;
+         BZ_GET_FAST_C(c_k0); c_nblock_used++;
+      }
+
+      return_notr:
+      total_out_lo32_old = s->strm->total_out_lo32;
+      s->strm->total_out_lo32 += (avail_out_INIT - cs_avail_out);
+      if (s->strm->total_out_lo32 < total_out_lo32_old)
+         s->strm->total_out_hi32++;
+
+      /* save */
+      s->calculatedBlockCRC = c_calculatedBlockCRC;
+      s->state_out_ch       = c_state_out_ch;
+      s->state_out_len      = c_state_out_len;
+      s->nblock_used        = c_nblock_used;
+      s->k0                 = c_k0;
+      s->tt                 = c_tt;
+      s->tPos               = c_tPos;
+      s->strm->next_out     = cs_next_out;
+      s->strm->avail_out    = cs_avail_out;
+      /* end save */
+   }
+   return False;
+}
+
+
+
+/*---------------------------------------------------*/
+__inline__ Int32 BZ2_indexIntoF ( Int32 indx, Int32 *cftab )
+{
+   Int32 nb, na, mid;
+   nb = 0;
+   na = 256;
+   do {
+      mid = (nb + na) >> 1;
+      if (indx >= cftab[mid]) nb = mid; else na = mid;
+   }
+   while (na - nb != 1);
+   return nb;
+}
+
+
+/*---------------------------------------------------*/
+/* Return  True iff data corruption is discovered.
+   Returns False if there is no problem.
+*/
+static
+Bool unRLE_obuf_to_output_SMALL ( DState* s )
+{
+   UChar k1;
+
+   if (s->blockRandomised) {
+
+      while (True) {
+         /* try to finish existing run */
+         while (True) {
+            if (s->strm->avail_out == 0) return False;
+            if (s->state_out_len == 0) break;
+            *( (UChar*)(s->strm->next_out) ) = s->state_out_ch;
+            BZ_UPDATE_CRC ( s->calculatedBlockCRC, s->state_out_ch );
+            s->state_out_len--;
+            s->strm->next_out++;
+            s->strm->avail_out--;
+            s->strm->total_out_lo32++;
+            if (s->strm->total_out_lo32 == 0) s->strm->total_out_hi32++;
+         }
+   
+         /* can a new run be started? */
+         if (s->nblock_used == s->save_nblock+1) return False;
+
+         /* Only caused by corrupt data stream? */
+         if (s->nblock_used > s->save_nblock+1)
+            return True;
+   
+         s->state_out_len = 1;
+         s->state_out_ch = s->k0;
+         BZ_GET_SMALL(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 2;
+         BZ_GET_SMALL(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 3;
+         BZ_GET_SMALL(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         BZ_GET_SMALL(k1); BZ_RAND_UPD_MASK; 
+         k1 ^= BZ_RAND_MASK; s->nblock_used++;
+         s->state_out_len = ((Int32)k1) + 4;
+         BZ_GET_SMALL(s->k0); BZ_RAND_UPD_MASK; 
+         s->k0 ^= BZ_RAND_MASK; s->nblock_used++;
+      }
+
+   } else {
+
+      while (True) {
+         /* try to finish existing run */
+         while (True) {
+            if (s->strm->avail_out == 0) return False;
+            if (s->state_out_len == 0) break;
+            *( (UChar*)(s->strm->next_out) ) = s->state_out_ch;
+            BZ_UPDATE_CRC ( s->calculatedBlockCRC, s->state_out_ch );
+            s->state_out_len--;
+            s->strm->next_out++;
+            s->strm->avail_out--;
+            s->strm->total_out_lo32++;
+            if (s->strm->total_out_lo32 == 0) s->strm->total_out_hi32++;
+         }
+   
+         /* can a new run be started? */
+         if (s->nblock_used == s->save_nblock+1) return False;
+
+         /* Only caused by corrupt data stream? */
+         if (s->nblock_used > s->save_nblock+1)
+            return True;
+   
+         s->state_out_len = 1;
+         s->state_out_ch = s->k0;
+         BZ_GET_SMALL(k1); s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 2;
+         BZ_GET_SMALL(k1); s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         s->state_out_len = 3;
+         BZ_GET_SMALL(k1); s->nblock_used++;
+         if (s->nblock_used == s->save_nblock+1) continue;
+         if (k1 != s->k0) { s->k0 = k1; continue; };
+   
+         BZ_GET_SMALL(k1); s->nblock_used++;
+         s->state_out_len = ((Int32)k1) + 4;
+         BZ_GET_SMALL(s->k0); s->nblock_used++;
+      }
+
+   }
+}
+
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzDecompress) ( bz_stream *strm )
+{
+   Bool    corrupt;
+   DState* s;
+   if (strm == NULL) return BZ_PARAM_ERROR;
+   s = strm->state;
+   if (s == NULL) return BZ_PARAM_ERROR;
+   if (s->strm != strm) return BZ_PARAM_ERROR;
+
+   while (True) {
+      if (s->state == BZ_X_IDLE) return BZ_SEQUENCE_ERROR;
+      if (s->state == BZ_X_OUTPUT) {
+         if (s->smallDecompress)
+            corrupt = unRLE_obuf_to_output_SMALL ( s ); else
+            corrupt = unRLE_obuf_to_output_FAST  ( s );
+         if (corrupt) return BZ_DATA_ERROR;
+         if (s->nblock_used == s->save_nblock+1 && s->state_out_len == 0) {
+            BZ_FINALISE_CRC ( s->calculatedBlockCRC );
+            if (s->verbosity >= 3) 
+               VPrintf2 ( " {0x%08x, 0x%08x}", s->storedBlockCRC, 
+                          s->calculatedBlockCRC );
+            if (s->verbosity >= 2) VPrintf0 ( "]" );
+            if (s->calculatedBlockCRC != s->storedBlockCRC)
+               return BZ_DATA_ERROR;
+            s->calculatedCombinedCRC 
+               = (s->calculatedCombinedCRC << 1) | 
+                    (s->calculatedCombinedCRC >> 31);
+            s->calculatedCombinedCRC ^= s->calculatedBlockCRC;
+            s->state = BZ_X_BLKHDR_1;
+         } else {
+            return BZ_OK;
+         }
+      }
+      if (s->state >= BZ_X_MAGIC_1) {
+         Int32 r = BZ2_decompress ( s );
+         if (r == BZ_STREAM_END) {
+            if (s->verbosity >= 3)
+               VPrintf2 ( "\n    combined CRCs: stored = 0x%08x, computed = 0x%08x", 
+                          s->storedCombinedCRC, s->calculatedCombinedCRC );
+            if (s->calculatedCombinedCRC != s->storedCombinedCRC)
+               return BZ_DATA_ERROR;
+            return r;
+         }
+         if (s->state != BZ_X_OUTPUT) return r;
+      }
+   }
+
+   AssertH ( 0, 6001 );
+
+   return 0;  /*NOTREACHED*/
+}
+
+
+/*---------------------------------------------------*/
+int BZ_API(BZ2_bzDecompressEnd)  ( bz_stream *strm )
+{
+   DState* s;
+   if (strm == NULL) return BZ_PARAM_ERROR;
+   s = strm->state;
+   if (s == NULL) return BZ_PARAM_ERROR;
+   if (s->strm != strm) return BZ_PARAM_ERROR;
+
+   if (s->tt   != NULL) BZFREE(s->tt);
+   if (s->ll16 != NULL) BZFREE(s->ll16);
+   if (s->ll4  != NULL) BZFREE(s->ll4);
+
+   BZFREE(strm->state);
+   strm->state = NULL;
+
+   return BZ_OK;
+}
+
+
+#ifndef BZ_NO_STDIO
+/*---------------------------------------------------*/
+/*--- File I/O stuff                              ---*/
+/*---------------------------------------------------*/
+
+#define BZ_SETERR(eee)                    \
+{                                         \
+   if (bzerror != NULL) *bzerror = eee;   \
+   if (bzf != NULL) bzf->lastErr = eee;   \
+}
+
+typedef 
+   struct {
+      FILE*     handle;
+      Char      buf[BZ_MAX_UNUSED];
+      Int32     bufN;
+      Bool      writing;
+      bz_stream strm;
+      Int32     lastErr;
+      Bool      initialisedOk;