test

git-svn-id: http://svn.net-core.org/repos/t-engine4@4373 51575b47-30f0-44d4-a5cc-537603b46e54

@@ -22,7 +22,7 @@ project "TEngine"
 	language "C"
 	targetname("t-engine")
 	files { "../src/*.c", }
-	links { "physfs", "lua".._OPTIONS.lua, "fov", "luasocket", "luaprofiler", "lualanes", "lpeg", "tcodimport", "lxp", "expatstatic", "luamd5", "luazlib", "luabitop" }
+	links { "physfs", "lua".._OPTIONS.lua, "fov", "luasocket", "luaprofiler", "lualanes", "lpeg", "tcodimport", "lxp", "expatstatic", "luamd5", "luazlib", "luabitop", "te4-bzip" }
 	defines { "_DEFAULT_VIDEOMODE_FLAGS_='SDL_HWSURFACE|SDL_DOUBLEBUF'" }
 	defines { [[TENGINE_HOME_PATH='".t-engine"']], "TE4CORE_VERSION="..TE4CORE_VERSION }
 
@@ -272,3 +272,10 @@ project "utf8proc"
 	targetname "utf8proc"
 
 	files { "../src/utf8proc/utf8proc.c", }
+
+project "te4-bzip"
+	kind "StaticLib"
+	language "C"
+	targetname "te4-bzip"
+
+	files { "../src/bzip2/*.c", }

@@ -45,13 +45,19 @@ Still, this is a golden age. Civilisations are healing the wounds of thousands o
 You are an adventurer, set out to discover wonders, explore old places, and venture into the unknown for wealth and glory.
 ]]
 starter = "mod.load"
+
+-- List of additional team files required
+teams = {
+	{ "#name#-#version#-music.team", "optional", {"/data/music/"} },
+	{ "#name#-#version#-gfx-shockbolt.team", "optional", {"/data/gfx/shockbolt/"} },
+}
+
 loading_wait_ticks = 260
 profile_stats_fields = {"artifacts", "characters", "deaths", "uniques", "scores", "lore", "escorts"}
 allow_userchat = true -- We can talk to the online community
 no_get_name = true -- Name setting for new characters is done by the module itself
 
 -- Define the fields that are sync'ed online, and how they are sync'ed
--- '
 profile_defs = {
 	allow_build = { {name="index:string:30"}, receive=function(data, save) save[data.name] = true end, export=function(env) for k, _ in pairs(env) do add{name=k} end end },
 	lore = { {name="index:string:30"}, receive=function(data, save) save.lore = save.lore or {} save.lore[data.name] = true end, export=function(env) for k, v in pairs(env.lore or {}) do add{name=k} end end },
@@ -82,6 +88,7 @@ profile_defs = {
 	},
 }
 
+-- Formatter for scores
 score_formatters = {
 	["Maj'Eyal"] = {
 		alive="#LIGHT_GREEN#{score} : #BLUE#{name}#LAST# the #LIGHT_RED#level {level} {subrace} {subclass}#LAST# is still alive and well on #GREEN#{where}#LAST##WHITE#",

@@ -17,6 +17,7 @@ solution "TEngine"
 		"src/utf8proc",
 		"src/physfs",
 		"src/physfs/zlib123",
+		"src/bzip2",
 		"/opt/SDL-1.3/include/SDL/",
 		"/usr/include/GL",
 	}

+/*-
+ * Copyright 2003-2005 Colin Percival
+ * All rights reserved
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted providing 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. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * 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.
+ */
+
+#include "bzlib.h"
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+static off_t offtin(u_char *buf)
+{
+	off_t y;
+
+	y=buf[7]&0x7F;
+	y=y*256;y+=buf[6];
+	y=y*256;y+=buf[5];
+	y=y*256;y+=buf[4];
+	y=y*256;y+=buf[3];
+	y=y*256;y+=buf[2];
+	y=y*256;y+=buf[1];
+	y=y*256;y+=buf[0];
+
+	if(buf[7]&0x80) y=-y;
+
+	return y;
+}
+
+#define returnerr(ret, str, f1, f2) { printf(str, f1, f2); perror("\nError in bspatch\n"); return (ret); }
+
+int bspatch(char *infile, char *outfile, char *patchfile)
+{
+	FILE * f, * cpf, * dpf, * epf;
+	BZFILE * cpfbz2, * dpfbz2, * epfbz2;
+	int cbz2err, dbz2err, ebz2err;
+	int fd;
+	ssize_t oldsize,newsize;
+	ssize_t bzctrllen,bzdatalen;
+	u_char header[32],buf[8];
+	u_char *old, *new;
+	off_t oldpos,newpos;
+	off_t ctrl[3];
+	off_t lenread;
+	off_t i;
+
+	/* Open patch file */
+	if ((f = fopen(patchfile, "r")) == NULL)
+		returnerr(1, "fopen(%s)", patchfile, 0);
+
+	/*
+	File format:
+		0	8	"BSDIFF40"
+		8	8	X
+		16	8	Y
+		24	8	sizeof(newfile)
+		32	X	bzip2(control block)
+		32+X	Y	bzip2(diff block)
+		32+X+Y	???	bzip2(extra block)
+	with control block a set of triples (x,y,z) meaning "add x bytes
+	from oldfile to x bytes from the diff block; copy y bytes from the
+	extra block; seek forwards in oldfile by z bytes".
+	*/
+
+	/* Read header */
+	if (fread(header, 1, 32, f) < 32) {
+		if (feof(f))
+			errx(1, "Corrupt patch\n");
+		returnerr(1, "fread(%s)", patchfile, 0);
+	}
+
+	/* Check for appropriate magic */
+	if (memcmp(header, "BSDIFF40", 8) != 0)
+		errx(1, "Corrupt patch\n");
+
+	/* Read lengths from header */
+	bzctrllen=offtin(header+8);
+	bzdatalen=offtin(header+16);
+	newsize=offtin(header+24);
+	if((bzctrllen<0) || (bzdatalen<0) || (newsize<0))
+		errx(1,"Corrupt patch\n");
+
+	/* Close patch file and re-open it via libbzip2 at the right places */
+	if (fclose(f))
+		returnerr(1, "fclose(%s)", patchfile, 0);
+	if ((cpf = fopen(patchfile, "r")) == NULL)
+		returnerr(1, "fopen(%s)", patchfile, 0);
+	if (fseeko(cpf, 32, SEEK_SET))
+		returnerr(1, "fseeko(%s, %lld)", patchfile,
+		    (long long)32);
+	if ((cpfbz2 = BZ2_bzReadOpen(&cbz2err, cpf, 0, 0, NULL, 0)) == NULL)
+		errx(1, "BZ2_bzReadOpen, bz2err = %d", cbz2err);
+	if ((dpf = fopen(patchfile, "r")) == NULL)
+		returnerr(1, "fopen(%s)", patchfile, 0);
+	if (fseeko(dpf, 32 + bzctrllen, SEEK_SET))
+		returnerr(1, "fseeko(%s, %lld)", patchfile,
+		    (long long)(32 + bzctrllen));
+	if ((dpfbz2 = BZ2_bzReadOpen(&dbz2err, dpf, 0, 0, NULL, 0)) == NULL)
+		errx(1, "BZ2_bzReadOpen, bz2err = %d", dbz2err);
+	if ((epf = fopen(patchfile, "r")) == NULL)
+		returnerr(1, "fopen(%s)", patchfile, 0);
+	if (fseeko(epf, 32 + bzctrllen + bzdatalen, SEEK_SET))
+		returnerr(1, "fseeko(%s, %lld)", patchfile,
+		    (long long)(32 + bzctrllen + bzdatalen));
+	if ((epfbz2 = BZ2_bzReadOpen(&ebz2err, epf, 0, 0, NULL, 0)) == NULL)
+		errx(1, "BZ2_bzReadOpen, bz2err = %d", ebz2err);
+
+	if(((fd=open(infile,O_RDONLY,0))<0) ||
+		((oldsize=lseek(fd,0,SEEK_END))==-1) ||
+		((old=malloc(oldsize+1))==NULL) ||
+		(lseek(fd,0,SEEK_SET)!=0) ||
+		(read(fd,old,oldsize)!=oldsize) ||
+		(close(fd)==-1)) returnerr(1,"%s",infile, 0);
+	if((new=malloc(newsize+1))==NULL) returnerr(1,"not enough memory", 0, 0);
+
+	oldpos=0;newpos=0;
+	while(newpos<newsize) {
+		/* Read control data */
+		for(i=0;i<=2;i++) {
+			lenread = BZ2_bzRead(&cbz2err, cpfbz2, buf, 8);
+			if ((lenread < 8) || ((cbz2err != BZ_OK) &&
+			    (cbz2err != BZ_STREAM_END)))
+				errx(1, "Corrupt patch\n");
+			ctrl[i]=offtin(buf);
+		};
+
+		/* Sanity-check */
+		if(newpos+ctrl[0]>newsize)
+			errx(1,"Corrupt patch\n");
+
+		/* Read diff string */
+		lenread = BZ2_bzRead(&dbz2err, dpfbz2, new + newpos, ctrl[0]);
+		if ((lenread < ctrl[0]) ||
+		    ((dbz2err != BZ_OK) && (dbz2err != BZ_STREAM_END)))
+			errx(1, "Corrupt patch\n");
+
+		/* Add old data to diff string */
+		for(i=0;i<ctrl[0];i++)
+			if((oldpos+i>=0) && (oldpos+i<oldsize))
+				new[newpos+i]+=old[oldpos+i];
+
+		/* Adjust pointers */
+		newpos+=ctrl[0];
+		oldpos+=ctrl[0];
+
+		/* Sanity-check */
+		if(newpos+ctrl[1]>newsize)
+			errx(1,"Corrupt patch\n");
+
+		/* Read extra string */
+		lenread = BZ2_bzRead(&ebz2err, epfbz2, new + newpos, ctrl[1]);
+		if ((lenread < ctrl[1]) ||
+		    ((ebz2err != BZ_OK) && (ebz2err != BZ_STREAM_END)))
+			errx(1, "Corrupt patch\n");
+
+		/* Adjust pointers */
+		newpos+=ctrl[1];
+		oldpos+=ctrl[2];
+	};
+
+	/* Clean up the bzip2 reads */
+	BZ2_bzReadClose(&cbz2err, cpfbz2);
+	BZ2_bzReadClose(&dbz2err, dpfbz2);
+	BZ2_bzReadClose(&ebz2err, epfbz2);
+	if (fclose(cpf) || fclose(dpf) || fclose(epf))
+		returnerr(1, "fclose(%s)", patchfile, 0);
+
+	/* Write the new file */
+	if(((fd=open(outfile,O_CREAT|O_TRUNC|O_WRONLY,0666))<0) ||
+		(write(fd,new,newsize)!=newsize) || (close(fd)==-1))
+		returnerr(1,"%s",outfile, 0);
+
+	free(new);
+	free(old);
+
+	return 0;
+}

+extern int bspatch(char *infile, char *outfile, char *patchfile);

+ ------------------------------------------------------------------
+ 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
+
+--------------------------------------------------------------------------

+
+/*-------------------------------------------------------------*/
+/*--- 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