/* pigz.c -- parallel implementation of gzip
 * Copyright (C) 2007-2021 Mark Adler
 * Version 2.5  23 Jan 2021  Mark Adler
 */

/*
  This software is provided 'as-is', without any express or implied
  warranty. In no event will the author be held liable for any damages
  arising from the use of this software.

  Permission is granted to anyone to use this software for any purpose,
  including commercial applications, and to alter it and redistribute it
  freely, subject to the following restrictions:

  1. 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.
  2. Altered source versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
  3. This notice may not be removed or altered from any source distribution.

  Mark Adler
  madler@alumni.caltech.edu

 */

/* Version history:
   1.0    17 Jan 2007  First version, pipe only
   1.1    28 Jan 2007  Avoid void * arithmetic (some compilers don't get that)
                       Add note about requiring zlib 1.2.3
                       Allow compression level 0 (no compression)
                       Completely rewrite parallelism -- add a write thread
                       Use deflateSetDictionary() to make use of history
                       Tune argument defaults to best performance on four cores
   1.2.1   1 Feb 2007  Add long command line options, add all gzip options
                       Add debugging options
   1.2.2  19 Feb 2007  Add list (--list) function
                       Process file names on command line, write .gz output
                       Write name and time in gzip header, set output file time
                       Implement all command line options except --recursive
                       Add --keep option to prevent deleting input files
                       Add thread tracing information with -vv used
                       Copy crc32_combine() from zlib (shared libraries issue)
   1.3    25 Feb 2007  Implement --recursive
                       Expand help to show all options
                       Show help if no arguments or output piping are provided
                       Process options in GZIP environment variable
                       Add progress indicator to write thread if --verbose
   1.4     4 Mar 2007  Add --independent to facilitate damaged file recovery
                       Reallocate jobs for new --blocksize or --processes
                       Do not delete original if writing to stdout
                       Allow --processes 1, which does no threading
                       Add NOTHREAD define to compile without threads
                       Incorporate license text from zlib in source code
   1.5    25 Mar 2007  Reinitialize jobs for new compression level
                       Copy attributes and owner from input file to output file
                       Add decompression and testing
                       Add -lt (or -ltv) to show all entries and proper lengths
                       Add decompression, testing, listing of LZW (.Z) files
                       Only generate and show trace log if DEBUG defined
                       Take "-" argument to mean read file from stdin
   1.6    30 Mar 2007  Add zlib stream compression (--zlib), and decompression
   1.7    29 Apr 2007  Decompress first entry of a zip file (if deflated)
                       Avoid empty deflate blocks at end of deflate stream
                       Show zlib check value (Adler-32) when listing
                       Don't complain when decompressing empty file
                       Warn about trailing junk for gzip and zlib streams
                       Make listings consistent, ignore gzip extra flags
                       Add zip stream compression (--zip)
   1.8    13 May 2007  Document --zip option in help output
   2.0    19 Oct 2008  Complete rewrite of thread usage and synchronization
                       Use polling threads and a pool of memory buffers
                       Remove direct pthread library use, hide in yarn.c
   2.0.1  20 Oct 2008  Check version of zlib at compile time, need >= 1.2.3
   2.1    24 Oct 2008  Decompress with read, write, inflate, and check threads
                       Remove spurious use of ctime_r(), ctime() more portable
                       Change application of job->calc lock to be a semaphore
                       Detect size of off_t at run time to select %lu vs. %llu
                       #define large file support macro even if not __linux__
                       Remove _LARGEFILE64_SOURCE, _FILE_OFFSET_BITS is enough
                       Detect file-too-large error and report, blame build
                       Replace check combination routines with those from zlib
   2.1.1  28 Oct 2008  Fix a leak for files with an integer number of blocks
                       Update for yarn 1.1 (yarn_prefix and yarn_abort)
   2.1.2  30 Oct 2008  Work around use of beta zlib in production systems
   2.1.3   8 Nov 2008  Don't use zlib combination routines, put back in pigz
   2.1.4   9 Nov 2008  Fix bug when decompressing very short files
   2.1.5  20 Jul 2009  Added 2008, 2009 to --license statement
                       Allow numeric parameter immediately after -p or -b
                       Enforce parameter after -p, -b, -s, before other options
                       Enforce numeric parameters to have only numeric digits
                       Try to determine the number of processors for -p default
                       Fix --suffix short option to be -S to match gzip [Bloch]
                       Decompress if executable named "unpigz" [Amundsen]
                       Add a little bit of testing to Makefile
   2.1.6  17 Jan 2010  Added pigz.spec to distribution for RPM systems [Brown]
                       Avoid some compiler warnings
                       Process symbolic links if piping to stdout [Hoffstätte]
                       Decompress if executable named "gunzip" [Hoffstätte]
                       Allow ".tgz" suffix [Chernookiy]
                       Fix adler32 comparison on .zz files
   2.1.7  17 Dec 2011  Avoid unused parameter warning in reenter()
                       Don't assume 2's complement ints in compress_thread()
                       Replicate gzip -cdf cat-like behavior
                       Replicate gzip -- option to suppress option decoding
                       Test output from make test instead of showing it
                       Updated pigz.spec to install unpigz, pigz.1 [Obermaier]
                       Add PIGZ environment variable [Mueller]
                       Replicate gzip suffix search when decoding or listing
                       Fix bug in load() to set in_left to zero on end of file
                       Do not check suffix when input file won't be modified
                       Decompress to stdout if name is "*cat" [Hayasaka]
                       Write data descriptor signature to be like Info-ZIP
                       Update and sort options list in help
                       Use CC variable for compiler in Makefile
                       Exit with code 2 if a warning has been issued
                       Fix thread synchronization problem when tracing
                       Change macro name MAX to MAX2 to avoid library conflicts
                       Determine number of processors on HP-UX [Lloyd]
   2.2    31 Dec 2011  Check for expansion bound busting (e.g. modified zlib)
                       Make the "threads" list head global variable volatile
                       Fix construction and printing of 32-bit check values
                       Add --rsyncable functionality
   2.2.1   1 Jan 2012  Fix bug in --rsyncable buffer management
   2.2.2   1 Jan 2012  Fix another bug in --rsyncable buffer management
   2.2.3  15 Jan 2012  Remove volatile in yarn.c
                       Reduce the number of input buffers
                       Change initial rsyncable hash to comparison value
                       Improve the efficiency of arriving at a byte boundary
                       Add thread portability #defines from yarn.c
                       Have rsyncable compression be independent of threading
                       Fix bug where constructed dictionaries not being used
   2.2.4  11 Mar 2012  Avoid some return value warnings
                       Improve the portability of printing the off_t type
                       Check for existence of compress binary before using
                       Update zlib version checking to 1.2.6 for new functions
                       Fix bug in zip (-K) output
                       Fix license in pigz.spec
                       Remove thread portability #defines in pigz.c
   2.2.5  28 Jul 2012  Avoid race condition in free_pool()
                       Change suffix to .tar when decompressing or listing .tgz
                       Print name of executable in error messages
                       Show help properly when the name is unpigz or gunzip
                       Fix permissions security problem before output is closed
   2.3     3 Mar 2013  Don't complain about missing suffix on stdout
                       Put all global variables in a structure for readability
                       Do not decompress concatenated zlib streams (just gzip)
                       Add option for compression level 11 to use zopfli
                       Fix handling of junk after compressed data
   2.3.1   9 Oct 2013  Fix builds of pigzt and pigzn to include zopfli
                       Add -lm, needed to link log function on some systems
                       Respect LDFLAGS in Makefile, use CFLAGS consistently
                       Add memory allocation tracking
                       Fix casting error in uncompressed length calculation
                       Update zopfli to Mar 10, 2013 Google state
                       Support zopfli in single thread case
                       Add -F, -I, -M, and -O options for zopfli tuning
   2.3.2  24 Jan 2015  Change whereis to which in Makefile for portability
                       Return zero exit code when only warnings are issued
                       Increase speed of unlzw (Unix compress decompression)
                       Update zopfli to current google state
                       Allow larger maximum blocksize (-b), now 512 MiB
                       Do not require that -d precede -N, -n, -T options
                       Strip any path from header name for -dN or -dNT
                       Remove use of PATH_MAX (PATH_MAX is not reliable)
                       Do not abort on inflate data error, do remaining files
                       Check gzip header CRC if present
                       Improve decompression error detection and reporting
   2.3.3  24 Jan 2015  Portability improvements
                       Update copyright years in documentation
   2.3.4   1 Oct 2016  Fix an out of bounds access due to invalid LZW input
                       Add an extra sync marker between independent blocks
                       Add zlib version for verbose version option (-vV)
                       Permit named pipes as input (e.g. made by mkfifo())
                       Fix a bug in -r directory traversal
                       Add warning for a zip file entry 4 GiB or larger
   2.4    26 Dec 2017  Portability improvements
                       Produce Zip64 format when needed for --zip (>= 4 GiB)
                       Make -no-name compatible with gzip, add --time option
                       Add -m as a short option for --no-time
                       Check run-time zlib version to handle weak linking
                       Fix a concurrent read bug in --list operation
                       Process options first, for gzip compatibility
                       Add --synchronous (-Y) option to force device write
                       Disallow an empty suffix (e.g. --suffix '')
                       Return an exit code of 1 if any issues are encountered
                       Fix sign error in compression reduction percentage
   2.5    23 Jan 2021  Add --alias/-A option to set .zip name for stdin input
                       Add --comment/-C option to add comment in .gz or .zip
                       Fix a bug that misidentified a multi-entry .zip
                       Fix a bug that did not emit double syncs for -i -p 1
                       Fix a bug in yarn that could try to access freed data
                       Do not delete multi-entry .zip files when extracting
                       Do not reject .zip entries with bit 11 set
                       Avoid a possible threads lock-order inversion
                       Ignore trailing junk after a gzip stream by default
 */

#define VERSION "pigz 2.5"

/* To-do:
    - make source portable for Windows, VMS, etc. (see gzip source code)
    - make build portable (currently good for Unixish)
 */

/*
   pigz compresses using threads to make use of multiple processors and cores.
   The input is broken up into 128 KB chunks with each compressed in parallel.
   The individual check value for each chunk is also calculated in parallel.
   The compressed data is written in order to the output, and a combined check
   value is calculated from the individual check values.

   The compressed data format generated is in the gzip, zlib, or single-entry
   zip format using the deflate compression method. The compression produces
   partial raw deflate streams which are concatenated by a single write thread
   and wrapped with the appropriate header and trailer, where the trailer
   contains the combined check value.

   Each partial raw deflate stream is terminated by an empty stored block
   (using the Z_SYNC_FLUSH option of zlib), in order to end that partial bit
   stream at a byte boundary, unless that partial stream happens to already end
   at a byte boundary (the latter requires zlib 1.2.6 or later). Ending on a
   byte boundary allows the partial streams to be concatenated simply as
   sequences of bytes. This adds a very small four to five byte overhead
   (average 3.75 bytes) to the output for each input chunk.

   The default input block size is 128K, but can be changed with the -b option.
   The number of compress threads is set by default to 8, which can be changed
   using the -p option. Specifying -p 1 avoids the use of threads entirely.
   pigz will try to determine the number of processors in the machine, in which
   case if that number is two or greater, pigz will use that as the default for
   -p instead of 8.

   The input blocks, while compressed independently, have the last 32K of the
   previous block loaded as a preset dictionary to preserve the compression
   effectiveness of deflating in a single thread. This can be turned off using
   the --independent or -i option, so that the blocks can be decompressed
   independently for partial error recovery or for random access.

   Decompression can't be parallelized over an arbitrary number of processors
   like compression can be, at least not without specially prepared deflate
   streams for that purpose. As a result, pigz uses a single thread (the main
   thread) for decompression, but will create three other threads for reading,
   writing, and check calculation, which can speed up decompression under some
   circumstances. Parallel decompression can be turned off by specifying one
   process (-dp 1 or -tp 1).

   pigz requires zlib 1.2.1 or later to allow setting the dictionary when doing
   raw deflate. Since zlib 1.2.3 corrects security vulnerabilities in zlib
   version 1.2.1 and 1.2.2, conditionals check for zlib 1.2.3 or later during
   the compilation of pigz.c. zlib 1.2.4 includes some improvements to
   Z_FULL_FLUSH and deflateSetDictionary() that permit identical output for
   pigz with and without threads, which is not possible with zlib 1.2.3. This
   may be important for uses of pigz -R where small changes in the contents
   should result in small changes in the archive for rsync. Note that due to
   the details of how the lower levels of compression result in greater speed,
   compression level 3 and below does not permit identical pigz output with and
   without threads.

   pigz uses the POSIX pthread library for thread control and communication,
   through the yarn.h interface to yarn.c. yarn.c can be replaced with
   equivalent implementations using other thread libraries. pigz can be
   compiled with NOTHREAD #defined to not use threads at all (in which case
   pigz will not be able to live up to the "parallel" in its name).
 */

/*
   Details of parallel compression implementation:

   When doing parallel compression, pigz uses the main thread to read the input
   in 'size' sized chunks (see -b), and puts those in a compression job list,
   each with a sequence number to keep track of the ordering. If it is not the
   first chunk, then that job also points to the previous input buffer, from
   which the last 32K will be used as a dictionary (unless -i is specified).
   This sets a lower limit of 32K on 'size'.

   pigz launches up to 'procs' compression threads (see -p). Each compression
   thread continues to look for jobs in the compression list and perform those
   jobs until instructed to return. When a job is pulled, the dictionary, if
   provided, will be loaded into the deflate engine and then that input buffer
   is dropped for reuse. Then the input data is compressed into an output
   buffer that grows in size if necessary to hold the compressed data. The job
   is then put into the write job list, sorted by the sequence number. The
   compress thread however continues to calculate the check value on the input
   data, either a CRC-32 or Adler-32, possibly in parallel with the write
   thread writing the output data. Once that's done, the compress thread drops
   the input buffer and also releases the lock on the check value so that the
   write thread can combine it with the previous check values. The compress
   thread has then completed that job, and goes to look for another.

   All of the compress threads are left running and waiting even after the last
   chunk is processed, so that they can support the next input to be compressed
   (more than one input file on the command line). Once pigz is done, it will
   call all the compress threads home (that'll do pig, that'll do).

   Before starting to read the input, the main thread launches the write thread
   so that it is ready pick up jobs immediately. The compress thread puts the
   write jobs in the list in sequence sorted order, so that the first job in
   the list is always has the lowest sequence number. The write thread waits
   for the next write job in sequence, and then gets that job. The job still
   holds its input buffer, from which the write thread gets the input buffer
   length for use in check value combination. Then the write thread drops that
   input buffer to allow its reuse. Holding on to the input buffer until the
   write thread starts also has the benefit that the read and compress threads
   can't get way ahead of the write thread and build up a large backlog of
   unwritten compressed data. The write thread will write the compressed data,
   drop the output buffer, and then wait for the check value to be unlocked by
   the compress thread. Then the write thread combines the check value for this
   chunk with the total check value for eventual use in the trailer. If this is
   not the last chunk, the write thread then goes back to look for the next
   output chunk in sequence. After the last chunk, the write thread returns and
   joins the main thread. Unlike the compress threads, a new write thread is
   launched for each input stream. The write thread writes the appropriate
   header and trailer around the compressed data.

   The input and output buffers are reused through their collection in pools.
   Each buffer has a use count, which when decremented to zero returns the
   buffer to the respective pool. Each input buffer has up to three parallel
   uses: as the input for compression, as the data for the check value
   calculation, and as a dictionary for compression. Each output buffer has
   only one use, which is as the output of compression followed serially as
   data to be written. The input pool is limited in the number of buffers, so
   that reading does not get way ahead of compression and eat up memory with
   more input than can be used. The limit is approximately two times the number
   of compression threads. In the case that reading is fast as compared to
   compression, that number allows a second set of buffers to be read while the
   first set of compressions are being performed. The number of output buffers
   is not directly limited, but is indirectly limited by the release of input
   buffers to about the same number.
 */

// Portability defines.
#define _FILE_OFFSET_BITS 64            // Use large file functions
#define _LARGE_FILES                    // Same thing for AIX
#define _XOPEN_SOURCE 700               // For POSIX 2008

// Included headers and what is expected from each.
#include <stdio.h>      // fflush(), fprintf(), fputs(), getchar(), putc(),
                        // puts(), printf(), vasprintf(), stderr, EOF, NULL,
                        // SEEK_END, size_t, off_t
#include <stdlib.h>     // exit(), malloc(), free(), realloc(), atol(), atoi(),
                        // getenv()
#include <stdarg.h>     // va_start(), va_arg(), va_end(), va_list
#include <string.h>     // memset(), memchr(), memcpy(), strcmp(), strcpy(),
                        // strncpy(), strlen(), strcat(), strrchr(),
                        // strerror()
#include <errno.h>      // errno, EEXIST
#include <assert.h>     // assert()
#include <time.h>       // ctime(), time(), time_t, mktime()
#include <signal.h>     // signal(), SIGINT
#include <sys/types.h>  // ssize_t
#include <sys/stat.h>   // chmod(), stat(), fstat(), lstat(), struct stat,
                        // S_IFDIR, S_IFLNK, S_IFMT, S_IFREG
#include <sys/time.h>   // utimes(), gettimeofday(), struct timeval
#include <unistd.h>     // unlink(), _exit(), read(), write(), close(),
                        // lseek(), isatty(), chown(), fsync()
#include <fcntl.h>      // open(), O_CREAT, O_EXCL, O_RDONLY, O_TRUNC,
                        // O_WRONLY, fcntl(), F_FULLFSYNC
#include <dirent.h>     // opendir(), readdir(), closedir(), DIR,
                        // struct dirent
#include <limits.h>     // UINT_MAX, INT_MAX
#if __STDC_VERSION__-0 >= 199901L || __GNUC__-0 >= 3
#  include <inttypes.h> // intmax_t, uintmax_t
   typedef uintmax_t length_t;
   typedef uint32_t crc_t;
#else
   typedef unsigned long length_t;
   typedef unsigned long crc_t;
#endif

#ifdef PIGZ_DEBUG
#  if defined(__APPLE__)
#    include <malloc/malloc.h>
#    define MALLOC_SIZE(p) malloc_size(p)
#  elif defined (__linux)
#    include <malloc.h>
#    define MALLOC_SIZE(p) malloc_usable_size(p)
#  elif defined (_WIN32) || defined(_WIN64)
#    include <malloc.h>
#    define MALLOC_SIZE(p) _msize(p)
#  else
#    define MALLOC_SIZE(p) (0)
#  endif
#endif

#ifdef __hpux
#  include <sys/param.h>
#  include <sys/pstat.h>
#endif

#ifndef S_IFLNK
#  define S_IFLNK 0
#endif

#ifdef __MINGW32__
#  define chown(p,o,g) 0
#  define utimes(p,t)  0
#  define lstat(p,s)   stat(p,s)
#  define _exit(s)     exit(s)
#endif

#include "zlib.h"       // deflateInit2(), deflateReset(), deflate(),
                        // deflateEnd(), deflateSetDictionary(), crc32(),
                        // adler32(), inflateBackInit(), inflateBack(),
                        // inflateBackEnd(), Z_DEFAULT_COMPRESSION,
                        // Z_DEFAULT_STRATEGY, Z_DEFLATED, Z_NO_FLUSH, Z_NULL,
                        // Z_OK, Z_SYNC_FLUSH, z_stream
#if !defined(ZLIB_VERNUM) || ZLIB_VERNUM < 0x1230
#  error "Need zlib version 1.2.3 or later"
#endif

#ifndef NOTHREAD
#  include "yarn.h"     // thread, launch(), join(), join_all(), lock,
                        // new_lock(), possess(), twist(), wait_for(),
                        // release(), peek_lock(), free_lock(), yarn_name
#endif

#ifndef NOZOPFLI
#  include "zopfli/src/zopfli/deflate.h"    // ZopfliDeflatePart(),
                                            // ZopfliInitOptions(),
                                            // ZopfliOptions
#endif

#include "try.h"        // try, catch, always, throw, drop, punt, ball_t

// For local functions and globals.
#define local static

// Prevent end-of-line conversions on MSDOSish operating systems.
#if defined(MSDOS) || defined(OS2) || defined(_WIN32) || defined(__CYGWIN__)
#  include <io.h>       // setmode(), O_BINARY, _commit() for _WIN32
#  define SET_BINARY_MODE(fd) setmode(fd, O_BINARY)
#else
#  define SET_BINARY_MODE(fd)
#endif

// Release an allocated pointer, if allocated, and mark as unallocated.
#define RELEASE(ptr) \
    do { \
        if ((ptr) != NULL) { \
            FREE(ptr); \
            ptr = NULL; \
        } \
    } while (0)

// Sliding dictionary size for deflate.
#define DICT 32768U

// Largest power of 2 that fits in an unsigned int. Used to limit requests to
// zlib functions that use unsigned int lengths.
#define MAXP2 (UINT_MAX - (UINT_MAX >> 1))

/* rsyncable constants -- RSYNCBITS is the number of bits in the mask for
   comparison. For random input data, there will be a hit on average every
   1<<RSYNCBITS bytes. So for an RSYNCBITS of 12, there will be an average of
   one hit every 4096 bytes, resulting in a mean block size of 4096. RSYNCMASK
   is the resulting bit mask. RSYNCHIT is what the hash value is compared to
   after applying the mask.

   The choice of 12 for RSYNCBITS is consistent with the original rsyncable
   patch for gzip which also uses a 12-bit mask. This results in a relatively
   small hit to compression, on the order of 1.5% to 3%. A mask of 13 bits can
   be used instead if a hit of less than 1% to the compression is desired, at
   the expense of more blocks transmitted for rsync updates. (Your mileage may
   vary.)

   This implementation of rsyncable uses a different hash algorithm than what
   the gzip rsyncable patch uses in order to provide better performance in
   several regards. The algorithm is simply to shift the hash value left one
   bit and exclusive-or that with the next byte. This is masked to the number
   of hash bits (RSYNCMASK) and compared to all ones except for a zero in the
   top bit (RSYNCHIT). This rolling hash has a very small window of 19 bytes
   (RSYNCBITS+7). The small window provides the benefit of much more rapid
   resynchronization after a change, than does the 4096-byte window of the gzip
   rsyncable patch.

   The comparison value is chosen to avoid matching any repeated bytes or short
   sequences. The gzip rsyncable patch on the other hand uses a sum and zero
   for comparison, which results in certain bad behaviors, such as always
   matching everywhere in a long sequence of zeros. Such sequences occur
   frequently in tar files.

   This hash efficiently discards history older than 19 bytes simply by
   shifting that data past the top of the mask -- no history needs to be
   retained to undo its impact on the hash value, as is needed for a sum.

   The choice of the comparison value (RSYNCHIT) has the virtue of avoiding
   extremely short blocks. The shortest block is five bytes (RSYNCBITS-7) from
   hit to hit, and is unlikely. Whereas with the gzip rsyncable algorithm,
   blocks of one byte are not only possible, but in fact are the most likely
   block size.

   Thanks and acknowledgement to Kevin Day for his experimentation and insights
   on rsyncable hash characteristics that led to some of the choices here.
 */
#define RSYNCBITS 12
#define RSYNCMASK ((1U << RSYNCBITS) - 1)
#define RSYNCHIT (RSYNCMASK >> 1)

// Initial pool counts and sizes -- INBUFS is the limit on the number of input
// spaces as a function of the number of processors (used to throttle the
// creation of compression jobs), OUTPOOL is the initial size of the output
// data buffer, chosen to make resizing of the buffer very unlikely and to
// allow prepending with a dictionary for use as an input buffer for zopfli.
#define INBUFS(p) (((p)<<1)+3)
#define OUTPOOL(s) ((s)+((s)>>4)+DICT)

// Input buffer size, and augmentation for re-inserting a central header.
#define BUF 32768
#define CEN 42
#define EXT (BUF + CEN)     // provide enough room to unget a header

// Globals (modified by main thread only when it's the only thread).
local struct {
    int volatile ret;       // pigz return code
    char *prog;             // name by which pigz was invoked
    int ind;                // input file descriptor
    int outd;               // output file descriptor
    char *inf;              // input file name (allocated)
    size_t inz;             // input file name allocated size
    char *outf;             // output file name (allocated)
    int verbosity;          // 0 = quiet, 1 = normal, 2 = verbose, 3 = trace
    int headis;             // 1 to store name, 2 to store date, 3 both
    int pipeout;            // write output to stdout even if file
    int keep;               // true to prevent deletion of input file
    int force;              // true to overwrite, compress links, cat
    int sync;               // true to flush output file
    int form;               // gzip = 0, zlib = 1, zip = 2 or 3
    int magic1;             // first byte of possible header when decoding
    int recurse;            // true to dive down into directory structure
    char *sufx;             // suffix to use (".gz" or user supplied)
    char *name;             // name for gzip or zip header
    char *alias;            // name for zip header when input is stdin
    char *comment;          // comment for gzip or zip header.
    time_t mtime;           // time stamp from input file for gzip header
    int list;               // true to list files instead of compress
    int first;              // true if we need to print listing header
    int decode;             // 0 to compress, 1 to decompress, 2 to test
    int level;              // compression level
#ifndef NOZOPFLI
    ZopfliOptions zopts;    // zopfli compression options
#endif
    int rsync;              // true for rsync blocking
    int procs;              // maximum number of compression threads (>= 1)
    int setdict;            // true to initialize dictionary in each thread
    size_t block;           // uncompressed input size per thread (>= 32K)
#ifndef NOTHREAD
    crc_t shift;            // pre-calculated CRC-32 shift for length block
#endif

    // saved gzip/zip header data for decompression, testing, and listing
    time_t stamp;           // time stamp from gzip header
    char *hname;            // name from header (allocated)
    char *hcomm;            // comment from header (allocated)
    unsigned long zip_crc;  // local header crc
    length_t zip_clen;      // local header compressed length
    length_t zip_ulen;      // local header uncompressed length
    int zip64;              // true if has zip64 extended information

    // globals for decompression and listing buffered reading
    unsigned char in_buf[EXT];  // input buffer
    unsigned char *in_next; // next unused byte in buffer
    size_t in_left;         // number of unused bytes in buffer
    int in_eof;             // true if reached end of file on input
    int in_short;           // true if last read didn't fill buffer
    length_t in_tot;        // total bytes read from input
    length_t out_tot;       // total bytes written to output
    unsigned long out_check;    // check value of output

#ifndef NOTHREAD
    // globals for decompression parallel reading
    unsigned char in_buf2[EXT]; // second buffer for parallel reads
    size_t in_len;          // data waiting in next buffer
    int in_which;           // -1: start, 0: in_buf2, 1: in_buf
    lock *load_state;       // value = 0 to wait, 1 to read a buffer
    thread *load_thread;    // load_read() thread for joining
#endif
} g;

// Display a complaint with the program name on stderr.
local int complain(char *fmt, ...) {
    va_list ap;

    if (g.verbosity > 0) {
        fprintf(stderr, "%s: ", g.prog);
        va_start(ap, fmt);
        vfprintf(stderr, fmt, ap);
        va_end(ap);
        putc('\n', stderr);
        fflush(stderr);
    }
    g.ret = 1;
    return 0;
}

#ifdef PIGZ_DEBUG

// Memory tracking.

#define MAXMEM 131072   // maximum number of tracked pointers

local struct mem_track_s {
    size_t num;         // current number of allocations
    size_t size;        // total size of current allocations
    size_t tot;         // maximum number of allocations
    size_t max;         // maximum size of allocations
#ifndef NOTHREAD
    lock *lock;         // lock for access across threads
#endif
    size_t have;        // number in array (possibly != num)
    void *mem[MAXMEM];  // sorted array of allocated pointers
} mem_track;

#ifndef NOTHREAD
#  define mem_track_grab(m) possess((m)->lock)
#  define mem_track_drop(m) release((m)->lock)
#else
#  define mem_track_grab(m)
#  define mem_track_drop(m)
#endif

// Return the leftmost insert location of ptr in the sorted list mem->mem[],
// which currently has mem->have elements. If ptr is already in the list, the
// returned value will point to its first occurrence. The return location will
// be one after the last element if ptr is greater than all of the elements.
local size_t search_track(struct mem_track_s *mem, void *ptr) {
    ptrdiff_t left = 0;
    ptrdiff_t right = mem->have - 1;
    while (left <= right) {
        ptrdiff_t mid = (left + right) >> 1;
        if (mem->mem[mid] < ptr)
            left = mid + 1;
        else
            right = mid - 1;
    }
    return left;
}

// Insert ptr in the sorted list mem->mem[] and update the memory allocation
// statistics.
local void insert_track(struct mem_track_s *mem, void *ptr) {
    mem_track_grab(mem);
    assert(mem->have < MAXMEM && "increase MAXMEM in source and try again");
    size_t i = search_track(mem, ptr);
    if (i < mem->have && mem->mem[i] == ptr)
        complain("mem_track: duplicate pointer %p\n", ptr);
    memmove(&mem->mem[i + 1], &mem->mem[i],
            (mem->have - i) * sizeof(void *));
    mem->mem[i] = ptr;
    mem->have++;
    mem->num++;
    mem->size += MALLOC_SIZE(ptr);
    if (mem->num > mem->tot)
        mem->tot = mem->num;
    if (mem->size > mem->max)
        mem->max = mem->size;
    mem_track_drop(mem);
}

// Find and delete ptr from the sorted list mem->mem[] and update the memory
// allocation statistics.
local void delete_track(struct mem_track_s *mem, void *ptr) {
    mem_track_grab(mem);
    size_t i = search_track(mem, ptr);
    if (i < mem->num && mem->mem[i] == ptr) {
        memmove(&mem->mem[i], &mem->mem[i + 1],
                (mem->have - (i + 1)) * sizeof(void *));
        mem->have--;
    }
    else
        complain("mem_track: missing pointer %p\n", ptr);
    mem->num--;
    mem->size -= MALLOC_SIZE(ptr);
    mem_track_drop(mem);
}

local void *malloc_track(struct mem_track_s *mem, size_t size) {
    void *ptr = malloc(size);
    if (ptr != NULL)
        insert_track(mem, ptr);
    return ptr;
}

local void *realloc_track(struct mem_track_s *mem, void *ptr, size_t size) {
    if (ptr == NULL)
        return malloc_track(mem, size);
    delete_track(mem, ptr);
    void *got = realloc(ptr, size);
    insert_track(mem, got == NULL ? ptr : got);
    return got;
}

local void free_track(struct mem_track_s *mem, void *ptr) {
    if (ptr != NULL) {
        delete_track(mem, ptr);
        free(ptr);
    }
}

#ifndef NOTHREAD
local void *yarn_malloc(size_t size) {
    return malloc_track(&mem_track, size);
}

local void yarn_free(void *ptr) {
    free_track(&mem_track, ptr);
}
#endif

local voidpf zlib_alloc(voidpf opaque, uInt items, uInt size) {
    return malloc_track(opaque, items * (size_t)size);
}

local void zlib_free(voidpf opaque, voidpf address) {
    free_track(opaque, address);
}

#define REALLOC(p, s) realloc_track(&mem_track, p, s)
#define FREE(p) free_track(&mem_track, p)
#define OPAQUE (&mem_track)
#define ZALLOC zlib_alloc
#define ZFREE zlib_free

#else // !PIGZ_DEBUG

#define REALLOC realloc
#define FREE free
#define OPAQUE Z_NULL
#define ZALLOC Z_NULL
#define ZFREE Z_NULL

#endif

// Assured memory allocation.
local void *alloc(void *ptr, size_t size) {
    ptr = REALLOC(ptr, size);
    if (ptr == NULL)
        throw(ENOMEM, "not enough memory");
    return ptr;
}

#ifdef PIGZ_DEBUG

// Logging.

// Starting time of day for tracing.
local struct timeval start;

// Trace log.
local struct log {
    struct timeval when;    // time of entry
    char *msg;              // message
    struct log *next;       // next entry
} *log_head, **log_tail = NULL;
#ifndef NOTHREAD
  local lock *log_lock = NULL;
#endif

// Maximum log entry length.
#define MAXMSG 256

// Set up log (call from main thread before other threads launched).
local void log_init(void) {
    if (log_tail == NULL) {
        mem_track.num = 0;
        mem_track.size = 0;
        mem_track.num = 0;
        mem_track.max = 0;
        mem_track.have = 0;
#ifndef NOTHREAD
        mem_track.lock = new_lock(0);
        yarn_mem(yarn_malloc, yarn_free);
        log_lock = new_lock(0);
#endif
        log_head = NULL;
        log_tail = &log_head;
    }
}

// Add entry to trace log.
local void log_add(char *fmt, ...) {
    struct timeval now;
    struct log *me;
    va_list ap;
    char msg[MAXMSG];

    gettimeofday(&now, NULL);
    me = alloc(NULL, sizeof(struct log));
    me->when = now;
    va_start(ap, fmt);
    vsnprintf(msg, MAXMSG, fmt, ap);
    va_end(ap);
    me->msg = alloc(NULL, strlen(msg) + 1);
    strcpy(me->msg, msg);
    me->next = NULL;
#ifndef NOTHREAD
    assert(log_lock != NULL);
    possess(log_lock);
#endif
    *log_tail = me;
    log_tail = &(me->next);
#ifndef NOTHREAD
    twist(log_lock, BY, +1);
#endif
}

// Pull entry from trace log and print it, return false if empty.
local int log_show(void) {
    struct log *me;
    struct timeval diff;

    if (log_tail == NULL)
        return 0;
#ifndef NOTHREAD
    possess(log_lock);
#endif
    me = log_head;
    if (me == NULL) {
#ifndef NOTHREAD
        release(log_lock);
#endif
        return 0;
    }
    log_head = me->next;
    if (me->next == NULL)
        log_tail = &log_head;
#ifndef NOTHREAD
    twist(log_lock, BY, -1);
#endif
    diff.tv_usec = me->when.tv_usec - start.tv_usec;
    diff.tv_sec = me->when.tv_sec - start.tv_sec;
    if (diff.tv_usec < 0) {
        diff.tv_usec += 1000000L;
        diff.tv_sec--;
    }
    fprintf(stderr, "trace %ld.%06ld %s\n",
            (long)diff.tv_sec, (long)diff.tv_usec, me->msg);
    fflush(stderr);
    FREE(me->msg);
    FREE(me);
    return 1;
}

// Release log resources (need to do log_init() to use again).
local void log_free(void) {
    struct log *me;

    if (log_tail != NULL) {
#ifndef NOTHREAD
        possess(log_lock);
#endif
        while ((me = log_head) != NULL) {
            log_head = me->next;
            FREE(me->msg);
            FREE(me);
        }
#ifndef NOTHREAD
        twist(log_lock, TO, 0);
        free_lock(log_lock);
        log_lock = NULL;
        yarn_mem(malloc, free);
        free_lock(mem_track.lock);
#endif
        log_tail = NULL;
    }
}

// Show entries until no more, free log.
local void log_dump(void) {
    if (log_tail == NULL)
        return;
    while (log_show())
        ;
    log_free();
    if (mem_track.num || mem_track.size)
        complain("memory leak: %zu allocs of %zu bytes total",
                 mem_track.num, mem_track.size);
    if (mem_track.max)
        fprintf(stderr, "%zu bytes of memory used in %zu allocs\n",
                mem_track.max, mem_track.tot);
}

// Debugging macro.
#define Trace(x) \
    do { \
        if (g.verbosity > 2) { \
            log_add x; \
        } \
    } while (0)

#else // !PIGZ_DEBUG

#define log_dump()
#define Trace(x)

#endif

// Abort or catch termination signal.
local void cut_short(int sig) {
    if (sig == SIGINT) {
        Trace(("termination by user"));
    }
    if (g.outd != -1 && g.outd != 1) {
        unlink(g.outf);
        RELEASE(g.outf);
        g.outd = -1;
    }
    log_dump();
    _exit(sig < 0 ? -sig : EINTR);
}

// Common code for catch block of top routine in the thread.
#define THREADABORT(ball) \
    do { \
        complain("abort: %s", (ball).why); \
        drop(ball); \
        cut_short(-(ball).code); \
    } while (0)

// Compute next size up by multiplying by about 2**(1/3) and rounding to the
// next power of 2 if close (three applications results in doubling). If small,
// go up to at least 16, if overflow, go to max size_t value.
local inline size_t grow(size_t size) {
    size_t was, top;
    int shift;

    was = size;
    size += size >> 2;
    top = size;
    for (shift = 0; top > 7; shift++)
        top >>= 1;
    if (top == 7)
        size = (size_t)1 << (shift + 3);
    if (size < 16)
        size = 16;
    if (size <= was)
        size = (size_t)0 - 1;
    return size;
}

// Copy cpy[0..len-1] to *mem + off, growing *mem if necessary, where *size is
// the allocated size of *mem. Return the number of bytes in the result.
local inline size_t vmemcpy(char **mem, size_t *size, size_t off,
                            void *cpy, size_t len) {
    size_t need;

    need = off + len;
    if (need < off)
        throw(ERANGE, "overflow");
    if (need > *size) {
        need = grow(need);
        if (off == 0) {
            RELEASE(*mem);
            *size = 0;
        }
        *mem = alloc(*mem, need);
        *size = need;
    }
    memcpy(*mem + off, cpy, len);
    return off + len;
}

// Copy the zero-terminated string cpy to *str + off, growing *str if
// necessary, where *size is the allocated size of *str. Return the length of
// the string plus one.
local inline size_t vstrcpy(char **str, size_t *size, size_t off, void *cpy) {
    return vmemcpy(str, size, off, cpy, strlen(cpy) + 1);
}

// Read up to len bytes into buf, repeating read() calls as needed.
local size_t readn(int desc, unsigned char *buf, size_t len) {
    ssize_t ret;
    size_t got;

    got = 0;
    while (len) {
        ret = read(desc, buf, len);
        if (ret < 0)
            throw(errno, "read error on %s (%s)", g.inf, strerror(errno));
        if (ret == 0)
            break;
        buf += ret;
        len -= (size_t)ret;
        got += (size_t)ret;
    }
    return got;
}

// Write len bytes, repeating write() calls as needed. Return len.
local size_t writen(int desc, void const *buf, size_t len) {
    char const *next = buf;
    size_t left = len;

    while (left) {
        size_t const max = SSIZE_MAX;
        ssize_t ret = write(desc, next, left > max ? max : left);
        if (ret < 1)
            throw(errno, "write error on %s (%s)", g.outf, strerror(errno));
        next += ret;
        left -= (size_t)ret;
    }
    return len;
}

// Convert Unix time to MS-DOS date and time, assuming the current timezone.
// (You got a better idea?)
local unsigned long time2dos(time_t t) {
    struct tm *tm;
    unsigned long dos;

    if (t == 0)
        t = time(NULL);
    tm = localtime(&t);
    if (tm->tm_year < 80 || tm->tm_year > 207)
        return 0;
    dos = (unsigned long)(tm->tm_year - 80) << 25;
    dos += (unsigned long)(tm->tm_mon + 1) << 21;
    dos += (unsigned long)tm->tm_mday << 16;
    dos += (unsigned long)tm->tm_hour << 11;
    dos += (unsigned long)tm->tm_min << 5;
    dos += (unsigned long)(tm->tm_sec + 1) >> 1;    // round to even seconds
    return dos;
}

// Value type for put() value arguments. All value arguments for put() must be
// cast to this type in order for va_arg() to pull the correct type from the
// argument list.
typedef length_t val_t;

// Write a set of header or trailer values to out, which is a file descriptor.
// The values are specified by a series of arguments in pairs, where the first
// argument in each pair is the number of bytes, and the second argument in
// each pair is the unsigned integer value to write. The first argument in each
// pair must be an int, and the second argument in each pair must be a val_t.
// The arguments are terminated by a single zero (an int). If the number of
// bytes is positive, then the value is written in little-endian order. If the
// number of bytes is negative, then the value is written in big-endian order.
// The total number of bytes written is returned. This makes the long and
// tiresome zip format headers and trailers more readable, maintainable, and
// verifiable.
local unsigned put(int out, ...) {
    // compute the total number of bytes
    unsigned count = 0;
    int n;
    va_list ap;
    va_start(ap, out);
    while ((n = va_arg(ap, int)) != 0) {
        va_arg(ap, val_t);
        count += (unsigned)abs(n);
    }
    va_end(ap);

    // allocate memory for the data
    unsigned char *wrap = alloc(NULL, count);
    unsigned char *next = wrap;

    // write the requested data to wrap[]
    va_start(ap, out);
    while ((n = va_arg(ap, int)) != 0) {
        val_t val = va_arg(ap, val_t);
        if (n < 0) {            // big endian
            n = -n << 3;
            do {
                n -= 8;
                *next++ = (unsigned char)(val >> n);
            } while (n);
        }
        else                    // little endian
            do {
                *next++ = (unsigned char)val;
                val >>= 8;
            } while (--n);
    }
    va_end(ap);

    // write wrap[] to out and return the number of bytes written
    writen(out, wrap, count);
    FREE(wrap);
    return count;
}

// Low 32-bits set to all ones.
#define LOW32 0xffffffff

// Write a gzip, zlib, or zip header using the information in the globals.
local length_t put_header(void) {
    length_t len;

    if (g.form > 1) {               // zip
        // write local header -- we don't know yet whether the lengths will fit
        // in 32 bits or not, so we have to assume that they might not and put
        // in a Zip64 extra field so that the data descriptor that appears
        // after the compressed data is interpreted with 64-bit lengths
        len = put(g.outd,
            4, (val_t)0x04034b50,   // local header signature
            2, (val_t)45,           // version needed to extract (4.5)
            2, (val_t)8,            // flags: data descriptor follows data
            2, (val_t)8,            // deflate
            4, (val_t)time2dos(g.mtime),
            4, (val_t)0,            // crc (not here)
            4, (val_t)LOW32,        // compressed length (not here)
            4, (val_t)LOW32,        // uncompressed length (not here)
            2, (val_t)(strlen(g.name == NULL ? g.alias : g.name)),  // name len
            2, (val_t)29,           // length of extra field (see below)
            0);

        // write file name (use g.alias for stdin)
        len += writen(g.outd, g.name == NULL ? g.alias : g.name,
                      strlen(g.name == NULL ? g.alias : g.name));

        // write Zip64 and extended timestamp extra field blocks (29 bytes)
        len += put(g.outd,
            2, (val_t)0x0001,       // Zip64 extended information ID
            2, (val_t)16,           // number of data bytes in this block
            8, (val_t)0,            // uncompressed length (not here)
            8, (val_t)0,            // compressed length (not here)
            2, (val_t)0x5455,       // extended timestamp ID
            2, (val_t)5,            // number of data bytes in this block
            1, (val_t)1,            // flag presence of mod time
            4, (val_t)g.mtime,      // mod time
            0);
    }
    else if (g.form) {              // zlib
        if (g.comment != NULL)
            complain("can't store comment in zlib format -- ignoring");
        unsigned head;
        head = (0x78 << 8) +        // deflate, 32K window
               (g.level >= 9 ? 3 << 6 :
                g.level == 1 ? 0 << 6:
                g.level >= 6 || g.level == Z_DEFAULT_COMPRESSION ? 1 << 6 :
                2 << 6);            // optional compression level clue
        head += 31 - (head % 31);   // make it a multiple of 31
        len = put(g.outd,
            -2, (val_t)head,        // zlib format uses big-endian order
            0);
    }
    else {                          // gzip
        len = put(g.outd,
            1, (val_t)31,
            1, (val_t)139,
            1, (val_t)8,            // deflate
            1, (val_t)((g.name != NULL ? 8 : 0) +
                       (g.comment != NULL ? 16 : 0)),
            4, (val_t)g.mtime,
            1, (val_t)(g.level >= 9 ? 2 : g.level == 1 ? 4 : 0),
            1, (val_t)3,            // unix
            0);
        if (g.name != NULL)
            len += writen(g.outd, g.name, strlen(g.name) + 1);
        if (g.comment != NULL)
            len += writen(g.outd, g.comment, strlen(g.comment) + 1);
    }
    return len;
}

// Write a gzip, zlib, or zip trailer.
local void put_trailer(length_t ulen, length_t clen,
                       unsigned long check, length_t head) {
    if (g.form > 1) {               // zip
        // write Zip64 data descriptor, as promised in the local header
        length_t desc = put(g.outd,
            4, (val_t)0x08074b50,
            4, (val_t)check,
            8, (val_t)clen,
            8, (val_t)ulen,
            0);

        // zip64 is true if either the compressed or the uncompressed length
        // does not fit in 32 bits, in which case there needs to be a Zip64
        // extra block in the central directory entry
        int zip64 = ulen >= LOW32 || clen >= LOW32;

        // write central file header
        length_t cent = put(g.outd,
            4, (val_t)0x02014b50,   // central header signature
            1, (val_t)45,           // made by 4.5 for Zip64 V1 end record
            1, (val_t)255,          // ignore external attributes
            2, (val_t)45,           // version needed to extract (4.5)
            2, (val_t)8,            // data descriptor is present
            2, (val_t)8,            // deflate
            4, (val_t)time2dos(g.mtime),
            4, (val_t)check,        // crc
            4, (val_t)(zip64 ? LOW32 : clen),   // compressed length
            4, (val_t)(zip64 ? LOW32 : ulen),   // uncompressed length
            2, (val_t)(strlen(g.name == NULL ? g.alias : g.name)),  // name len
            2, (val_t)(zip64 ? 29 : 9), // extra field size (see below)
            2, (val_t)(g.comment == NULL ? 0 : strlen(g.comment)),  // comment
            2, (val_t)0,            // disk number 0
            2, (val_t)0,            // internal file attributes
            4, (val_t)0,            // external file attributes (ignored)
            4, (val_t)0,            // offset of local header
            0);

        // write file name (use g.alias for stdin)
        cent += writen(g.outd, g.name == NULL ? g.alias : g.name,
                       strlen(g.name == NULL ? g.alias : g.name));

        // write Zip64 extra field block (20 bytes)
        if (zip64)
            cent += put(g.outd,
                2, (val_t)0x0001,   // Zip64 extended information ID
                2, (val_t)16,       // number of data bytes in this block
                8, (val_t)ulen,     // uncompressed length
                8, (val_t)clen,     // compressed length
                0);

        // write extended timestamp extra field block (9 bytes)
        cent += put(g.outd,
            2, (val_t)0x5455,       // extended timestamp signature
            2, (val_t)5,            // number of data bytes in this block
            1, (val_t)1,            // flag presence of mod time
            4, (val_t)g.mtime,      // mod time
            0);

        // write comment, if requested
        if (g.comment != NULL)
            cent += writen(g.outd, g.comment, strlen(g.comment));

        // here zip64 is true if the offset of the central directory does not
        // fit in 32 bits, in which case insert the Zip64 end records to
        // provide a 64-bit offset
        zip64 = head + clen + desc >= LOW32;
        if (zip64) {
            // write Zip64 end of central directory record and locator
            put(g.outd,
                4, (val_t)0x06064b50,   // Zip64 end of central dir sig
                8, (val_t)44,       // size of the remainder of this record
                2, (val_t)45,       // version made by
                2, (val_t)45,       // version needed to extract
                4, (val_t)0,        // number of this disk
                4, (val_t)0,        // disk with start of central directory
                8, (val_t)1,        // number of entries on this disk
                8, (val_t)1,        // total number of entries
                8, (val_t)cent,     // size of central directory
                8, (val_t)(head + clen + desc), // central dir offset
                4, (val_t)0x07064b50,   // Zip64 end locator signature
                4, (val_t)0,        // disk with Zip64 end of central dir
                8, (val_t)(head + clen + desc + cent),  // location
                4, (val_t)1,        // total number of disks
                0);
        }

        // write end of central directory record
        put(g.outd,
            4, (val_t)0x06054b50,   // end of central directory signature
            2, (val_t)0,            // number of this disk
            2, (val_t)0,            // disk with start of central directory
            2, (val_t)(zip64 ? 0xffff : 1), // entries on this disk
            2, (val_t)(zip64 ? 0xffff : 1), // total number of entries
            4, (val_t)(zip64 ? LOW32 : cent),   // size of central directory
            4, (val_t)(zip64 ? LOW32 : head + clen + desc), // offset
            2, (val_t)0,            // no zip file comment
            0);
    }
    else if (g.form)                // zlib
        put(g.outd,
            -4, (val_t)check,       // zlib format uses big-endian order
            0);
    else                            // gzip
        put(g.outd,
            4, (val_t)check,
            4, (val_t)ulen,
            0);
}

// Compute an Adler-32, allowing a size_t length.
local unsigned long adler32z(unsigned long adler,
                           unsigned char const *buf, size_t len) {
    while (len > UINT_MAX && buf != NULL) {
        adler = adler32(adler, buf, UINT_MAX);
        buf += UINT_MAX;
        len -= UINT_MAX;
    }
    return adler32(adler, buf, (unsigned)len);
}

// Compute a CRC-32, allowing a size_t length.
local unsigned long crc32z(unsigned long crc,
                           unsigned char const *buf, size_t len) {
    while (len > UINT_MAX && buf != NULL) {
        crc = crc32(crc, buf, UINT_MAX);
        buf += UINT_MAX;
        len -= UINT_MAX;
    }
    return crc32(crc, buf, (unsigned)len);
}

// Compute check value depending on format.
#define CHECK(a,b,c) (g.form == 1 ? adler32z(a,b,c) : crc32z(a,b,c))

// Return the zlib version as an integer, where each component is interpreted
// as a decimal number and converted to four hexadecimal digits. E.g.
// '1.2.11.1' -> 0x12b1, or return -1 if the string is not a valid version.
local long zlib_vernum(void) {
    char const *ver = zlibVersion();
    long num = 0;
    int left = 4;
    int comp = 0;
    do {
        if (*ver >= '0' && *ver <= '9')
            comp = 10 * comp + *ver - '0';
        else {
            num = (num << 4) + (comp > 0xf ? 0xf : comp);
            left--;
            if (*ver != '.')
                break;
            comp = 0;
        }
        ver++;
    } while (left);
    return left < 2 ? num << (left << 2) : -1;
}

#ifndef NOTHREAD
// -- threaded portions of pigz --

// -- check value combination routines for parallel calculation --

#define COMB(a,b,c) (g.form == 1 ? adler32_comb(a,b,c) : crc32_comb(a,b,c))
// Combine two crc-32's or two adler-32's (copied from zlib 1.2.3 so that pigz
// can be compatible with older versions of zlib).

// We copy the combination routines from zlib here, in order to avoid linkage
// issues with the zlib 1.2.3 builds on Sun, Ubuntu, and others.

// CRC-32 polynomial, reflected.
#define POLY 0xedb88320

// Return a(x) multiplied by b(x) modulo p(x), where p(x) is the CRC
// polynomial, reflected. For speed, this requires that a not be zero.
local crc_t multmodp(crc_t a, crc_t b) {
    crc_t m = (crc_t)1 << 31;
    crc_t p = 0;
    for (;;) {
        if (a & m) {
            p ^= b;
            if ((a & (m - 1)) == 0)
                break;
        }
        m >>= 1;
        b = b & 1 ? (b >> 1) ^ POLY : b >> 1;
    }
    return p;
}

// Table of x^2^n modulo p(x).
local const crc_t x2n_table[] = {
    0x40000000, 0x20000000, 0x08000000, 0x00800000, 0x00008000,
    0xedb88320, 0xb1e6b092, 0xa06a2517, 0xed627dae, 0x88d14467,
    0xd7bbfe6a, 0xec447f11, 0x8e7ea170, 0x6427800e, 0x4d47bae0,
    0x09fe548f, 0x83852d0f, 0x30362f1a, 0x7b5a9cc3, 0x31fec169,
    0x9fec022a, 0x6c8dedc4, 0x15d6874d, 0x5fde7a4e, 0xbad90e37,
    0x2e4e5eef, 0x4eaba214, 0xa8a472c0, 0x429a969e, 0x148d302a,
    0xc40ba6d0, 0xc4e22c3c};

// Return x^(n*2^k) modulo p(x).
local crc_t x2nmodp(size_t n, unsigned k) {
    crc_t p = (crc_t)1 << 31;       // x^0 == 1
    while (n) {
        if (n & 1)
            p = multmodp(x2n_table[k & 31], p);
        n >>= 1;
        k++;
    }
    return p;
}

// This uses the pre-computed g.shift value most of the time. Only the last
// combination requires a new x2nmodp() calculation.
local unsigned long crc32_comb(unsigned long crc1, unsigned long crc2,
                               size_t len2) {
    return multmodp(len2 == g.block ? g.shift : x2nmodp(len2, 3), crc1) ^ crc2;
}

#define BASE 65521U     // largest prime smaller than 65536
#define LOW16 0xffff    // mask lower 16 bits

local unsigned long adler32_comb(unsigned long adler1, unsigned long adler2,
                                 size_t len2) {
    unsigned long sum1;
    unsigned long sum2;
    unsigned rem;

    // the derivation of this formula is left as an exercise for the reader
    rem = (unsigned)(len2 % BASE);
    sum1 = adler1 & LOW16;
    sum2 = (rem * sum1) % BASE;
    sum1 += (adler2 & LOW16) + BASE - 1;
    sum2 += ((adler1 >> 16) & LOW16) + ((adler2 >> 16) & LOW16) + BASE - rem;
    if (sum1 >= BASE) sum1 -= BASE;
    if (sum1 >= BASE) sum1 -= BASE;
    if (sum2 >= (BASE << 1)) sum2 -= (BASE << 1);
    if (sum2 >= BASE) sum2 -= BASE;
    return sum1 | (sum2 << 16);
}

// -- pool of spaces for buffer management --

// These routines manage a pool of spaces. Each pool specifies a fixed size
// buffer to be contained in each space. Each space has a use count, which when
// decremented to zero returns the space to the pool. If a space is requested
// from the pool and the pool is empty, a space is immediately created unless a
// specified limit on the number of spaces has been reached. Only if the limit
// is reached will it wait for a space to be returned to the pool. Each space
// knows what pool it belongs to, so that it can be returned.

// A space (one buffer for each space).
struct space {
    lock *use;              // use count -- return to pool when zero
    unsigned char *buf;     // buffer of size size
    size_t size;            // current size of this buffer
    size_t len;             // for application usage (initially zero)
    struct pool *pool;      // pool to return to
    struct space *next;     // for pool linked list
};

// Pool of spaces (one pool for each type needed).
struct pool {
    lock *have;             // unused spaces available, lock for list
    struct space *head;     // linked list of available buffers
    size_t size;            // size of new buffers in this pool
    int limit;              // number of new spaces allowed, or -1
    int made;               // number of buffers made
};

// Initialize a pool (pool structure itself provided, not allocated). The limit
// is the maximum number of spaces in the pool, or -1 to indicate no limit,
// i.e., to never wait for a buffer to return to the pool.
local void new_pool(struct pool *pool, size_t size, int limit) {
    pool->have = new_lock(0);
    pool->head = NULL;
    pool->size = size;
    pool->limit = limit;
    pool->made = 0;
}

// Get a space from a pool. The use count is initially set to one, so there is
// no need to call use_space() for the first use.
local struct space *get_space(struct pool *pool) {
    struct space *space;

    // if can't create any more, wait for a space to show up
    possess(pool->have);
    if (pool->limit == 0)
        wait_for(pool->have, NOT_TO_BE, 0);

    // if a space is available, pull it from the list and return it
    if (pool->head != NULL) {
        space = pool->head;
        pool->head = space->next;
        twist(pool->have, BY, -1);      // one less in pool
        possess(space->use);
        twist(space->use, TO, 1);       // initially one user
        space->len = 0;
        return space;
    }

    // nothing available, don't want to wait, make a new space
    assert(pool->limit != 0);
    if (pool->limit > 0)
        pool->limit--;
    pool->made++;
    release(pool->have);
    space = alloc(NULL, sizeof(struct space));
    space->use = new_lock(1);           // initially one user
    space->buf = alloc(NULL, pool->size);
    space->size = pool->size;
    space->len = 0;
    space->pool = pool;                 // remember the pool this belongs to
    return space;
}

// Increase the size of the buffer in space.
local void grow_space(struct space *space) {
    size_t more;

    // compute next size up
    more = grow(space->size);
    if (more == space->size)
        throw(ERANGE, "overflow");

    // reallocate the buffer
    space->buf = alloc(space->buf, more);
    space->size = more;
}

// Increment the use count to require one more drop before returning this space
// to the pool.
local void use_space(struct space *space) {
    long use;

    possess(space->use);
    use = peek_lock(space->use);
    assert(use != 0);
    twist(space->use, BY, +1);
}

// Drop a space, returning it to the pool if the use count is zero.
local void drop_space(struct space *space) {
    long use;
    struct pool *pool;

    if (space == NULL)
        return;
    possess(space->use);
    use = peek_lock(space->use);
    assert(use != 0);
    twist(space->use, BY, -1);
    if (use == 1) {
        pool = space->pool;
        possess(pool->have);
        space->next = pool->head;
        pool->head = space;
        twist(pool->have, BY, +1);
    }
}

// Free the memory and lock resources of a pool. Return number of spaces for
// debugging and resource usage measurement.
local int free_pool(struct pool *pool) {
    int count;
    struct space *space;

    possess(pool->have);
    count = 0;
    while ((space = pool->head) != NULL) {
        pool->head = space->next;
        FREE(space->buf);
        free_lock(space->use);
        FREE(space);
        count++;
    }
    assert(count == pool->made);
    release(pool->have);
    free_lock(pool->have);
    return count;
}

// Input and output buffer pools.
local struct pool in_pool;
local struct pool out_pool;
local struct pool dict_pool;
local struct pool lens_pool;

// -- parallel compression --

// Compress or write job (passed from compress list to write list). If seq is
// equal to -1, compress_thread is instructed to return; if more is false then
// this is the last chunk, which after writing tells write_thread to return.
struct job {
    long seq;                   // sequence number
    int more;                   // true if this is not the last chunk
    struct space *in;           // input data to compress
    struct space *out;          // dictionary or resulting compressed data
    struct space *lens;         // coded list of flush block lengths
    unsigned long check;        // check value for input data
    lock *calc;                 // released when check calculation complete
    struct job *next;           // next job in the list (either list)
};

// List of compress jobs (with tail for appending to list).
local lock *compress_have = NULL;   // number of compress jobs waiting
local struct job *compress_head, **compress_tail;

// List of write jobs.
local lock *write_first;            // lowest sequence number in list
local struct job *write_head;

// Number of compression threads running.
local int cthreads = 0;

// Write thread if running.
local thread *writeth = NULL;

// Setup job lists (call from main thread).
local void setup_jobs(void) {
    // set up only if not already set up
    if (compress_have != NULL)
        return;

    // allocate locks and initialize lists
    compress_have = new_lock(0);
    compress_head = NULL;
    compress_tail = &compress_head;
    write_first = new_lock(-1);
    write_head = NULL;

    // initialize buffer pools (initial size for out_pool not critical, since
    // buffers will be grown in size if needed -- the initial size chosen to
    // make this unlikely, the same for lens_pool)
    new_pool(&in_pool, g.block, INBUFS(g.procs));
    new_pool(&out_pool, OUTPOOL(g.block), -1);
    new_pool(&dict_pool, DICT, -1);
    new_pool(&lens_pool, g.block >> (RSYNCBITS - 1), -1);
}

// Command the compress threads to all return, then join them all (call from
// main thread), free all the thread-related resources.
local void finish_jobs(void) {
    struct job job;
    int caught;

    // only do this once
    if (compress_have == NULL)
        return;

    // command all of the extant compress threads to return
    possess(compress_have);
    job.seq = -1;
    job.next = NULL;
    compress_head = &job;
    compress_tail = &(job.next);
    twist(compress_have, BY, +1);       // will wake them all up

    // join all of the compress threads, verify they all came back
    caught = join_all();
    Trace(("-- joined %d compress threads", caught));
    assert(caught == cthreads);
    cthreads = 0;

    // free the resources
    caught = free_pool(&lens_pool);
    Trace(("-- freed %d block lengths buffers", caught));
    caught = free_pool(&dict_pool);
    Trace(("-- freed %d dictionary buffers", caught));
    caught = free_pool(&out_pool);
    Trace(("-- freed %d output buffers", caught));
    caught = free_pool(&in_pool);
    Trace(("-- freed %d input buffers", caught));
    free_lock(write_first);
    free_lock(compress_have);
    compress_have = NULL;
}

// Compress all strm->avail_in bytes at strm->next_in to out->buf, updating
// out->len, grow the size of the buffer (out->size) if necessary. Respect the
// size limitations of the zlib stream data types (size_t may be larger than
// unsigned).
local void deflate_engine(z_stream *strm, struct space *out, int flush) {
    size_t room;

    do {
        room = out->size - out->len;
        if (room == 0) {
            grow_space(out);
            room = out->size - out->len;
        }
        strm->next_out = out->buf + out->len;
        strm->avail_out = room < UINT_MAX ? (unsigned)room : UINT_MAX;
        (void)deflate(strm, flush);
        out->len = (size_t)(strm->next_out - out->buf);
    } while (strm->avail_out == 0);
    assert(strm->avail_in == 0);
}

// Get the next compression job from the head of the list, compress and compute
// the check value on the input, and put a job in the write list with the
// results. Keep looking for more jobs, returning when a job is found with a
// sequence number of -1 (leave that job in the list for other incarnations to
// find).
local void compress_thread(void *dummy) {
    struct job *job;                // job pulled and working on
    struct job *here, **prior;      // pointers for inserting in write list
    unsigned long check;            // check value of input
    unsigned char *next;            // pointer for blocks, check value data
    size_t left;                    // input left to process
    size_t len;                     // remaining bytes to compress/check
#if ZLIB_VERNUM >= 0x1260
    int bits;                       // deflate pending bits
#endif
    int ret;                        // zlib return code
    ball_t err;                     // error information from throw()

    (void)dummy;

    try {
        z_stream strm;                  // deflate stream
#ifndef NOZOPFLI
        struct space *temp = NULL;
        // get temporary space for zopfli input
        if (g.level > 9)
            temp = get_space(&out_pool);
        else
#endif
        {
            // initialize the deflate stream for this thread
            strm.zfree = ZFREE;
            strm.zalloc = ZALLOC;
            strm.opaque = OPAQUE;
            ret = deflateInit2(&strm, 6, Z_DEFLATED, -15, 8, Z_DEFAULT_STRATEGY);
            if (ret == Z_MEM_ERROR)
                throw(ENOMEM, "not enough memory");
            if (ret != Z_OK)
                throw(EINVAL, "internal error");
        }

        // keep looking for work
        for (;;) {
            // get a job (like I tell my son)
            possess(compress_have);
            wait_for(compress_have, NOT_TO_BE, 0);
            job = compress_head;
            assert(job != NULL);
            if (job->seq == -1)
                break;
            compress_head = job->next;
            if (job->next == NULL)
                compress_tail = &compress_head;
            twist(compress_have, BY, -1);

            // got a job -- initialize and set the compression level (note that
            // if deflateParams() is called immediately after deflateReset(),
            // there is no need to initialize input/output for the stream)
            Trace(("-- compressing #%ld", job->seq));
#ifndef NOZOPFLI
            if (g.level <= 9) {
#endif
                (void)deflateReset(&strm);
                (void)deflateParams(&strm, g.level, Z_DEFAULT_STRATEGY);
#ifndef NOZOPFLI
            }
            else
                temp->len = 0;
#endif

            // set dictionary if provided, release that input or dictionary
            // buffer (not NULL if g.setdict is true and if this is not the
            // first work unit)
            if (job->out != NULL) {
                len = job->out->len;
                left = len < DICT ? len : DICT;
#ifndef NOZOPFLI
                if (g.level <= 9)
#endif
                    deflateSetDictionary(&strm, job->out->buf + (len - left),
                                         (unsigned)left);
#ifndef NOZOPFLI
                else {
                    memcpy(temp->buf, job->out->buf + (len - left), left);
                    temp->len = left;
                }
#endif
                drop_space(job->out);
            }

            // set up input and output
            job->out = get_space(&out_pool);
#ifndef NOZOPFLI
            if (g.level <= 9) {
#endif
                strm.next_in = job->in->buf;
                strm.next_out = job->out->buf;
#ifndef NOZOPFLI
            }
            else
                memcpy(temp->buf + temp->len, job->in->buf, job->in->len);
#endif

            // compress each block, either flushing or finishing
            next = job->lens == NULL ? NULL : job->lens->buf;
            left = job->in->len;
            job->out->len = 0;
            do {
                // decode next block length from blocks list
                len = next == NULL ? 128 : *next++;
                if (len < 128)                  // 64..32831
                    len = (len << 8) + (*next++) + 64;
                else if (len == 128)            // end of list
                    len = left;
                else if (len < 192)             // 1..63
                    len &= 0x3f;
                else if (len < 224){            // 32832..2129983
                    len = ((len & 0x1f) << 16) + ((size_t)*next++ << 8);
                    len += *next++ + 32832U;
                }
                else {                          // 2129984..539000895
                    len = ((len & 0x1f) << 24) + ((size_t)*next++ << 16);
                    len += (size_t)*next++ << 8;
                    len += (size_t)*next++ + 2129984UL;
                }
                left -= len;

#ifndef NOZOPFLI
                if (g.level <= 9) {
#endif
                    // run MAXP2-sized amounts of input through deflate -- this
                    // loop is needed for those cases where the unsigned type
                    // is smaller than the size_t type, or when len is close to
                    // the limit of the size_t type
                    while (len > MAXP2) {
                        strm.avail_in = MAXP2;
                        deflate_engine(&strm, job->out, Z_NO_FLUSH);
                        len -= MAXP2;
                    }

                    // run the last piece through deflate -- end on a byte
                    // boundary, using a sync marker if necessary, or finish
                    // the deflate stream if this is the last block
                    strm.avail_in = (unsigned)len;
                    if (left || job->more) {
#if ZLIB_VERNUM >= 0x1260
                        if (zlib_vernum() >= 0x1260) {
                            deflate_engine(&strm, job->out, Z_BLOCK);

                            // add enough empty blocks to get to a byte
                            // boundary
                            (void)deflatePending(&strm, Z_NULL, &bits);
                            if ((bits & 1) || !g.setdict)
                                deflate_engine(&strm, job->out, Z_SYNC_FLUSH);
                            else if (bits & 7) {
                                do {        // add static empty blocks
                                    bits = deflatePrime(&strm, 10, 2);
                                    assert(bits == Z_OK);
                                    (void)deflatePending(&strm, Z_NULL, &bits);
                                } while (bits & 7);
                                deflate_engine(&strm, job->out, Z_BLOCK);
                            }
                        }
                        else
#endif
                        {
                            deflate_engine(&strm, job->out, Z_SYNC_FLUSH);
                        }
                        if (!g.setdict)     // two markers when independent
                            deflate_engine(&strm, job->out, Z_FULL_FLUSH);
                    }
                    else
                        deflate_engine(&strm, job->out, Z_FINISH);
#ifndef NOZOPFLI
                }
                else {
                    // compress len bytes using zopfli, end at byte boundary
                    unsigned char bits, *out;
                    size_t outsize;

                    out = NULL;
                    outsize = 0;
                    bits = 0;
                    ZopfliDeflatePart(&g.zopts, 2, !(left || job->more),
                                      temp->buf, temp->len, temp->len + len,
                                      &bits, &out, &outsize);
                    assert(job->out->len + outsize + 5 <= job->out->size);
                    memcpy(job->out->buf + job->out->len, out, outsize);
                    free(out);
                    job->out->len += outsize;
                    if (left || job->more) {
                        bits &= 7;
                        if ((bits & 1) || !g.setdict) {
                            if (bits == 0 || bits > 5)
                                job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0xff;
                            job->out->buf[job->out->len++] = 0xff;
                        }
                        else if (bits) {
                            do {
                                job->out->buf[job->out->len - 1] += 2 << bits;
                                job->out->buf[job->out->len++] = 0;
                                bits += 2;
                            } while (bits < 8);
                        }
                        if (!g.setdict) {   // two markers when independent
                            job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0;
                            job->out->buf[job->out->len++] = 0xff;
                            job->out->buf[job->out->len++] = 0xff;
                        }
                    }
                    temp->len += len;
                }
#endif
            } while (left);
            drop_space(job->lens);
            job->lens = NULL;
            Trace(("-- compressed #%ld%s", job->seq,
                   job->more ? "" : " (last)"));

            // reserve input buffer until check value has been calculated
            use_space(job->in);

            // insert write job in list in sorted order, alert write thread
            possess(write_first);
            prior = &write_head;
            while ((here = *prior) != NULL) {
                if (here->seq > job->seq)
                    break;
                prior = &(here->next);
            }
            job->next = here;
            *prior = job;
            twist(write_first, TO, write_head->seq);

            // calculate the check value in parallel with writing, alert the
            // write thread that the calculation is complete, and drop this
            // usage of the input buffer
            len = job->in->len;
            next = job->in->buf;
            check = CHECK(0L, Z_NULL, 0);
            while (len > MAXP2) {
                check = CHECK(check, next, MAXP2);
                len -= MAXP2;
                next += MAXP2;
            }
            check = CHECK(check, next, (unsigned)len);
            drop_space(job->in);
            job->check = check;
            Trace(("-- checked #%ld%s", job->seq, job->more ? "" : " (last)"));
            possess(job->calc);
            twist(job->calc, TO, 1);

            // done with that one -- go find another job
        }

        // found job with seq == -1 -- return to join
        release(compress_have);
#ifndef NOZOPFLI
        if (g.level > 9)
            drop_space(temp);
        else
#endif
        {
            (void)deflateEnd(&strm);
        }
    }
    catch (err) {
        THREADABORT(err);
    }
}

// Collect the write jobs off of the list in sequence order and write out the
// compressed data until the last chunk is written. Also write the header and
// trailer and combine the individual check values of the input buffers.
local void write_thread(void *dummy) {
    long seq;                       // next sequence number looking for
    struct job *job;                // job pulled and working on
    size_t len;                     // input length
    int more;                       // true if more chunks to write
    length_t head;                  // header length
    length_t ulen;                  // total uncompressed size (overflow ok)
    length_t clen;                  // total compressed size (overflow ok)
    unsigned long check;            // check value of uncompressed data
    ball_t err;                     // error information from throw()

    (void)dummy;

    try {
        // build and write header
        Trace(("-- write thread running"));
        head = put_header();

        // process output of compress threads until end of input
        ulen = clen = 0;
        check = CHECK(0L, Z_NULL, 0);
        seq = 0;
        do {
            // get next write job in order
            possess(write_first);
            wait_for(write_first, TO_BE, seq);
            job = write_head;
            write_head = job->next;
            twist(write_first, TO, write_head == NULL ? -1 : write_head->seq);

            // update lengths, save uncompressed length for COMB
            more = job->more;
            len = job->in->len;
            drop_space(job->in);
            ulen += len;
            clen += job->out->len;

            // write the compressed data and drop the output buffer
            Trace(("-- writing #%ld", seq));
            writen(g.outd, job->out->buf, job->out->len);
            drop_space(job->out);
            Trace(("-- wrote #%ld%s", seq, more ? "" : " (last)"));

            // wait for check calculation to complete, then combine, once the
            // compress thread is done with the input, release it
            possess(job->calc);
            wait_for(job->calc, TO_BE, 1);
            release(job->calc);
            check = COMB(check, job->check, len);
            Trace(("-- combined #%ld%s", seq, more ? "" : " (last)"));

            // free the job
            free_lock(job->calc);
            FREE(job);

            // get the next buffer in sequence
            seq++;
        } while (more);

        // write trailer
        put_trailer(ulen, clen, check, head);

        // verify no more jobs, prepare for next use
        possess(compress_have);
        assert(compress_head == NULL && peek_lock(compress_have) == 0);
        release(compress_have);
        possess(write_first);
        assert(write_head == NULL);
        twist(write_first, TO, -1);
    }
    catch (err) {
        THREADABORT(err);
    }
}

// Encode a hash hit to the block lengths list. hit == 0 ends the list.
local void append_len(struct job *job, size_t len) {
    struct space *lens;

    assert(len < 539000896UL);
    if (job->lens == NULL)
        job->lens = get_space(&lens_pool);
    lens = job->lens;
    if (lens->size < lens->len + 3)
        grow_space(lens);
    if (len < 64)
        lens->buf[lens->len++] = (unsigned char)(len + 128);
    else if (len < 32832U) {
        len -= 64;
        lens->buf[lens->len++] = (unsigned char)(len >> 8);
        lens->buf[lens->len++] = (unsigned char)len;
    }
    else if (len < 2129984UL) {
        len -= 32832U;
        lens->buf[lens->len++] = (unsigned char)((len >> 16) + 192);
        lens->buf[lens->len++] = (unsigned char)(len >> 8);
        lens->buf[lens->len++] = (unsigned char)len;
    }
    else {
        len -= 2129984UL;
        lens->buf[lens->len++] = (unsigned char)((len >> 24) + 224);
        lens->buf[lens->len++] = (unsigned char)(len >> 16);
        lens->buf[lens->len++] = (unsigned char)(len >> 8);
        lens->buf[lens->len++] = (unsigned char)len;
    }
}

// Compress ind to outd, using multiple threads for the compression and check
// value calculations and one other thread for writing the output. Compress
// threads will be launched and left running (waiting actually) to support
// subsequent calls of parallel_compress().
local void parallel_compress(void) {
    long seq;                       // sequence number
    struct space *curr;             // input data to compress
    struct space *next;             // input data that follows curr
    struct space *hold;             // input data that follows next
    struct space *dict;             // dictionary for next compression
    struct job *job;                // job for compress, then write
    int more;                       // true if more input to read
    unsigned hash;                  // hash for rsyncable
    unsigned char *scan;            // next byte to compute hash on
    unsigned char *end;             // after end of data to compute hash on
    unsigned char *last;            // position after last hit
    size_t left;                    // last hit in curr to end of curr
    size_t len;                     // for various length computations

    // if first time or after an option change, setup the job lists
    setup_jobs();

    // start write thread
    writeth = launch(write_thread, NULL);

    // read from input and start compress threads (write thread will pick up
    // the output of the compress threads)
    seq = 0;
    next = get_space(&in_pool);
    next->len = readn(g.ind, next->buf, next->size);
    hold = NULL;
    dict = NULL;
    scan = next->buf;
    hash = RSYNCHIT;
    left = 0;
    do {
        // create a new job
        job = alloc(NULL, sizeof(struct job));
        job->calc = new_lock(0);

        // update input spaces
        curr = next;
        next = hold;
        hold = NULL;

        // get more input if we don't already have some
        if (next == NULL) {
            next = get_space(&in_pool);
            next->len = readn(g.ind, next->buf, next->size);
        }

        // if rsyncable, generate block lengths and prepare curr for job to
        // likely have less than size bytes (up to the last hash hit)
        job->lens = NULL;
        if (g.rsync && curr->len) {
            // compute the hash function starting where we last left off to
            // cover either size bytes or to EOF, whichever is less, through
            // the data in curr (and in the next loop, through next) -- save
            // the block lengths resulting from the hash hits in the job->lens
            // list
            if (left == 0) {
                // scan is in curr
                last = curr->buf;
                end = curr->buf + curr->len;
                while (scan < end) {
                    hash = ((hash << 1) ^ *scan++) & RSYNCMASK;
                    if (hash == RSYNCHIT) {
                        len = (size_t)(scan - last);
                        append_len(job, len);
                        last = scan;
                    }
                }

                // continue scan in next
                left = (size_t)(scan - last);
                scan = next->buf;
            }

            // scan in next for enough bytes to fill curr, or what is available
            // in next, whichever is less (if next isn't full, then we're at
            // the end of the file) -- the bytes in curr since the last hit,
            // stored in left, counts towards the size of the first block
            last = next->buf;
            len = curr->size - curr->len;
            if (len > next->len)
                len = next->len;
            end = next->buf + len;
            while (scan < end) {
                hash = ((hash << 1) ^ *scan++) & RSYNCMASK;
                if (hash == RSYNCHIT) {
                    len = (size_t)(scan - last) + left;
                    left = 0;
                    append_len(job, len);
                    last = scan;
                }
            }
            append_len(job, 0);

            // create input in curr for job up to last hit or entire buffer if
            // no hits at all -- save remainder in next and possibly hold
            len = (size_t)((job->lens->len == 1 ? scan : last) - next->buf);
            if (len) {
                // got hits in next, or no hits in either -- copy to curr
                memcpy(curr->buf + curr->len, next->buf, len);
                curr->len += len;
                memmove(next->buf, next->buf + len, next->len - len);
                next->len -= len;
                scan -= len;
                left = 0;
            }
            else if (job->lens->len != 1 && left && next->len) {
                // had hits in curr, but none in next, and last hit in curr
                // wasn't right at the end, so we have input there to save --
                // use curr up to the last hit, save the rest, moving next to
                // hold
                hold = next;
                next = get_space(&in_pool);
                memcpy(next->buf, curr->buf + (curr->len - left), left);
                next->len = left;
                curr->len -= left;
            }
            else {
                // else, last match happened to be right at the end of curr, or
                // we're at the end of the input compressing the rest
                left = 0;
            }
        }

        // compress curr->buf to curr->len -- compress thread will drop curr
        job->in = curr;

        // set job->more if there is more to compress after curr
        more = next->len != 0;
        job->more = more;

        // provide dictionary for this job, prepare dictionary for next job
        job->out = dict;
        if (more && g.setdict) {
            if (curr->len >= DICT || job->out == NULL) {
                dict = curr;
                use_space(dict);
            }
            else {
                dict = get_space(&dict_pool);
                len = DICT - curr->len;
                memcpy(dict->buf, job->out->buf + (job->out->len - len), len);
                memcpy(dict->buf + len, curr->buf, curr->len);
                dict->len = DICT;
            }
        }

        // preparation of job is complete
        job->seq = seq;
        Trace(("-- read #%ld%s", seq, more ? "" : " (last)"));
        if (++seq < 1)
            throw(ERANGE, "overflow");

        // start another compress thread if needed
        if (cthreads < seq && cthreads < g.procs) {
            (void)launch(compress_thread, NULL);
            cthreads++;
        }

        // put job at end of compress list, let all the compressors know
        possess(compress_have);
        job->next = NULL;
        *compress_tail = job;
        compress_tail = &(job->next);
        twist(compress_have, BY, +1);
    } while (more);
    drop_space(next);

    // wait for the write thread to complete (we leave the compress threads out
    // there and waiting in case there is another stream to compress)
    join(writeth);
    writeth = NULL;
    Trace(("-- write thread joined"));
}

#endif

// Repeated code in single_compress to compress available input and write it.
#define DEFLATE_WRITE(flush) \
    do { \
        do { \
            strm->avail_out = out_size; \
            strm->next_out = out; \
            (void)deflate(strm, flush); \
            clen += writen(g.outd, out, out_size - strm->avail_out); \
        } while (strm->avail_out == 0); \
        assert(strm->avail_in == 0); \
    } while (0)

// Do a simple compression in a single thread from ind to outd. If reset is
// true, instead free the memory that was allocated and retained for input,
// output, and deflate.
local void single_compress(int reset) {
    size_t got;                     // amount of data in in[]
    size_t more;                    // amount of data in next[] (0 if eof)
    size_t start;                   // start of data in next[]
    size_t have;                    // bytes in current block for -i
    size_t hist;                    // offset of permitted history
    int fresh;                      // if true, reset compression history
    unsigned hash;                  // hash for rsyncable
    unsigned char *scan;            // pointer for hash computation
    size_t left;                    // bytes left to compress after hash hit
    unsigned long head;             // header length
    length_t ulen;                  // total uncompressed size
    length_t clen;                  // total compressed size
    unsigned long check;            // check value of uncompressed data
    static unsigned out_size;       // size of output buffer
    static unsigned char *in, *next, *out;  // reused i/o buffers
    static z_stream *strm = NULL;   // reused deflate structure

    // if requested, just release the allocations and return
    if (reset) {
        if (strm != NULL) {
            (void)deflateEnd(strm);
            FREE(strm);
            FREE(out);
            FREE(next);
            FREE(in);
            strm = NULL;
        }
        return;
    }

    // initialize the deflate structure if this is the first time
    if (strm == NULL) {
        int ret;                    // zlib return code

        out_size = g.block > MAXP2 ? MAXP2 : (unsigned)g.block;
        in = alloc(NULL, g.block + DICT);
        next = alloc(NULL, g.block + DICT);
        out = alloc(NULL, out_size);
        strm = alloc(NULL, sizeof(z_stream));
        strm->zfree = ZFREE;
        strm->zalloc = ZALLOC;
        strm->opaque = OPAQUE;
        ret = deflateInit2(strm, 6, Z_DEFLATED, -15, 8, Z_DEFAULT_STRATEGY);
        if (ret == Z_MEM_ERROR)
            throw(ENOMEM, "not enough memory");
        if (ret != Z_OK)
            throw(EINVAL, "internal error");
    }

    // write header
    head = put_header();

    // set compression level in case it changed
#ifndef NOZOPFLI
    if (g.level <= 9) {
#endif
        (void)deflateReset(strm);
        (void)deflateParams(strm, g.level, Z_DEFAULT_STRATEGY);
#ifndef NOZOPFLI
    }
#endif

    // do raw deflate and calculate check value
    got = 0;
    more = readn(g.ind, next, g.block);
    ulen = more;
    start = 0;
    hist = 0;
    clen = 0;
    have = 0;
    check = CHECK(0L, Z_NULL, 0);
    hash = RSYNCHIT;
    do {
        // get data to compress, see if there is any more input
        if (got == 0) {
            scan = in;  in = next;  next = scan;
            strm->next_in = in + start;
            got = more;
            if (g.level > 9) {
                left = start + more - hist;
                if (left > DICT)
                    left = DICT;
                memcpy(next, in + ((start + more) - left), left);
                start = left;
                hist = 0;
            }
            else
                start = 0;
            more = readn(g.ind, next + start, g.block);
            ulen += more;
        }

        // if rsyncable, compute hash until a hit or the end of the block
        left = 0;
        if (g.rsync && got) {
            scan = strm->next_in;
            left = got;
            do {
                if (left == 0) {
                    // went to the end -- if no more or no hit in size bytes,
                    // then proceed to do a flush or finish with got bytes
                    if (more == 0 || got == g.block)
                        break;

                    // fill in[] with what's left there and as much as possible
                    // from next[] -- set up to continue hash hit search
                    if (g.level > 9) {
                        left = (size_t)(strm->next_in - in) - hist;
                        if (left > DICT)
                            left = DICT;
                    }
                    memmove(in, strm->next_in - left, left + got);
                    hist = 0;
                    strm->next_in = in + left;
                    scan = in + left + got;
                    left = more > g.block - got ? g.block - got : more;
                    memcpy(scan, next + start, left);
                    got += left;
                    more -= left;
                    start += left;

                    // if that emptied the next buffer, try to refill it
                    if (more == 0) {
                        more = readn(g.ind, next, g.block);
                        ulen += more;
                        start = 0;
                    }
                }
                left--;
                hash = ((hash << 1) ^ *scan++) & RSYNCMASK;
            } while (hash != RSYNCHIT);
            got -= left;
        }

        // clear history for --independent option
        fresh = 0;
        if (!g.setdict) {
            have += got;
            if (have > g.block) {
                fresh = 1;
                have = got;
            }
        }

#ifndef NOZOPFLI
        if (g.level <= 9) {
#endif
            // clear history if requested
            if (fresh)
                (void)deflateReset(strm);

            // compress MAXP2-size chunks in case unsigned type is small
            while (got > MAXP2) {
                strm->avail_in = MAXP2;
                check = CHECK(check, strm->next_in, strm->avail_in);
                DEFLATE_WRITE(Z_NO_FLUSH);
                got -= MAXP2;
            }

            // compress the remainder, emit a block, finish if end of input
            strm->avail_in = (unsigned)got;
            got = left;
            check = CHECK(check, strm->next_in, strm->avail_in);
            if (more || got) {
#if ZLIB_VERNUM >= 0x1260
                if (zlib_vernum() >= 0x1260) {
                    int bits;

                    DEFLATE_WRITE(Z_BLOCK);
                    (void)deflatePending(strm, Z_NULL, &bits);
                    if ((bits & 1) || !g.setdict)
                        DEFLATE_WRITE(Z_SYNC_FLUSH);
                    else if (bits & 7) {
                        do {
                            bits = deflatePrime(strm, 10, 2);
                            assert(bits == Z_OK);
                            (void)deflatePending(strm, Z_NULL, &bits);
                        } while (bits & 7);
                        DEFLATE_WRITE(Z_NO_FLUSH);
                    }
                }
                else
                    DEFLATE_WRITE(Z_SYNC_FLUSH);
#else
                DEFLATE_WRITE(Z_SYNC_FLUSH);
#endif
                if (!g.setdict)             // two markers when independent
                    DEFLATE_WRITE(Z_FULL_FLUSH);
            }
            else
                DEFLATE_WRITE(Z_FINISH);
#ifndef NOZOPFLI
        }
        else {
            // compress got bytes using zopfli, bring to byte boundary
            unsigned char bits, *def;
            size_t size, off;

            // discard history if requested
            off = (size_t)(strm->next_in - in);
            if (fresh)
                hist = off;

            def = NULL;
            size = 0;
            bits = 0;
            ZopfliDeflatePart(&g.zopts, 2, !(more || left),
                              in + hist, off - hist, (off - hist) + got,
                              &bits, &def, &size);
            bits &= 7;
            if (more || left) {
                if ((bits & 1) || !g.setdict) {
                    writen(g.outd, def, size);
                    if (bits == 0 || bits > 5)
                        writen(g.outd, (unsigned char *)"\0", 1);
                    writen(g.outd, (unsigned char *)"\0\0\xff\xff", 4);
                }
                else {
                    assert(size > 0);
                    writen(g.outd, def, size - 1);
                    if (bits)
                        do {
                            def[size - 1] += 2 << bits;
                            writen(g.outd, def + size - 1, 1);
                            def[size - 1] = 0;
                            bits += 2;
                        } while (bits < 8);
                    writen(g.outd, def + size - 1, 1);
                }
                if (!g.setdict)             // two markers when independent
                    writen(g.outd, (unsigned char *)"\0\0\0\xff\xff", 5);
            }
            else
                writen(g.outd, def, size);
            free(def);
            while (got > MAXP2) {
                check = CHECK(check, strm->next_in, MAXP2);
                strm->next_in += MAXP2;
                got -= MAXP2;
            }
            check = CHECK(check, strm->next_in, (unsigned)got);
            strm->next_in += got;
            got = left;
        }
#endif

        // do until no more input
    } while (more || got);

    // write trailer
    put_trailer(ulen, clen, check, head);
}

// --- decompression ---

#ifndef NOTHREAD
// Parallel read thread. If the state is 1, then read a buffer and set the
// state to 0 when done, if the state is > 1, then end this thread.
local void load_read(void *dummy) {
    size_t len;
    ball_t err;                     // error information from throw()

    (void)dummy;

    Trace(("-- launched decompress read thread"));
    try {
        do {
            possess(g.load_state);
            wait_for(g.load_state, NOT_TO_BE, 0);
            if (peek_lock(g.load_state) > 1) {
                release(g.load_state);
                break;
            }
            g.in_len = len = readn(g.ind, g.in_which ? g.in_buf : g.in_buf2,
                                   BUF);
            Trace(("-- decompress read thread read %lu bytes", len));
            twist(g.load_state, TO, 0);
        } while (len == BUF);
    }
    catch (err) {
        THREADABORT(err);
    }
    Trace(("-- exited decompress read thread"));
}

// Wait for load_read() to complete the current read operation. If the
// load_read() thread is not active, then return immediately.
local void load_wait(void) {
    if (g.in_which == -1)
        return;
    possess(g.load_state);
    wait_for(g.load_state, TO_BE, 0);
    release(g.load_state);
}
#endif

// load() is called when the input has been consumed in order to provide more
// input data: load the input buffer with BUF or fewer bytes (fewer if at end
// of file) from the file g.ind, set g.in_next to point to the g.in_left bytes
// read, update g.in_tot, and return g.in_left. g.in_eof is set to true when
// g.in_left has gone to zero and there is no more data left to read.
local size_t load(void) {
    // if already detected end of file, do nothing
    if (g.in_short) {
        g.in_eof = 1;
        g.in_left = 0;
        return 0;
    }

#ifndef NOTHREAD
    // if first time in or procs == 1, read a buffer to have something to
    // return, otherwise wait for the previous read job to complete
    if (g.procs > 1) {
        // if first time, fire up the read thread, ask for a read
        if (g.in_which == -1) {
            g.in_which = 1;
            g.load_state = new_lock(1);
            g.load_thread = launch(load_read, NULL);
        }

        // wait for the previously requested read to complete
        load_wait();

        // set up input buffer with the data just read
        g.in_next = g.in_which ? g.in_buf : g.in_buf2;
        g.in_left = g.in_len;

        // if not at end of file, alert read thread to load next buffer,
        // alternate between g.in_buf and g.in_buf2
        if (g.in_len == BUF) {
            g.in_which = 1 - g.in_which;
            possess(g.load_state);
            twist(g.load_state, TO, 1);
        }

        // at end of file -- join read thread (already exited), clean up
        else {
            join(g.load_thread);
            free_lock(g.load_state);
            g.in_which = -1;
        }
    }
    else
#endif
    {
        // don't use threads -- simply read a buffer into g.in_buf
        g.in_left = readn(g.ind, g.in_next = g.in_buf, BUF);
    }

    // note end of file
    if (g.in_left < BUF) {
        g.in_short = 1;

        // if we got bupkis, now is the time to mark eof
        if (g.in_left == 0)
            g.in_eof = 1;
    }

    // update the total and return the available bytes
    g.in_tot += g.in_left;
    return g.in_left;
}

// Terminate the load() operation. Empty buffer, mark end, close file (if not
// stdin), and free the name and comment obtained from the header, if present.
local void load_end(void) {
#ifndef NOTHREAD
    // if the read thread is running, then end it
    if (g.in_which != -1) {
        // wait for the previously requested read to complete and send the
        // thread a message to exit
        possess(g.load_state);
        wait_for(g.load_state, TO_BE, 0);
        twist(g.load_state, TO, 2);

        // join the thread (which has exited or will very shortly) and clean up
        join(g.load_thread);
        free_lock(g.load_state);
        g.in_which = -1;
    }
#endif
    g.in_left = 0;
    g.in_short = 1;
    g.in_eof = 1;
    if (g.ind != 0)
        close(g.ind);
    RELEASE(g.hname);
    RELEASE(g.hcomm);
}

// Initialize for reading new input.
local void in_init(void) {
    g.in_left = 0;
    g.in_eof = 0;
    g.in_short = 0;
    g.in_tot = 0;
#ifndef NOTHREAD
    g.in_which = -1;
#endif
}

// Buffered reading macros for decompression and listing.
#define GET() (g.in_left == 0 && (g.in_eof || load() == 0) ? 0 : \
               (g.in_left--, *g.in_next++))
#define GET2() (tmp2 = GET(), tmp2 + ((unsigned)(GET()) << 8))
#define GET4() (tmp4 = GET2(), tmp4 + ((unsigned long)(GET2()) << 16))
#define SKIP(dist) \
    do { \
        size_t togo = (dist); \
        while (togo > g.in_left) { \
            togo -= g.in_left; \
            if (load() == 0) \
                return -3; \
        } \
        g.in_left -= togo; \
        g.in_next += togo; \
    } while (0)

// GET(), GET2(), GET4() and SKIP() equivalents, with crc update.
#define GETC() (g.in_left == 0 && (g.in_eof || load() == 0) ? 0 : \
                (g.in_left--, crc = crc32z(crc, g.in_next, 1), *g.in_next++))
#define GET2C() (tmp2 = GETC(), tmp2 + ((unsigned)(GETC()) << 8))
#define GET4C() (tmp4 = GET2C(), tmp4 + ((unsigned long)(GET2C()) << 16))
#define SKIPC(dist) \
    do { \
        size_t togo = (dist); \
        while (togo > g.in_left) { \
            crc = crc32z(crc, g.in_next, g.in_left); \
            togo -= g.in_left; \
            if (load() == 0) \
                return -3; \
        } \
        crc = crc32z(crc, g.in_next, togo); \
        g.in_left -= togo; \
        g.in_next += togo; \
    } while (0)

// Get a zero-terminated string into allocated memory, with crc update.
#define GETZC(str) \
    do { \
        unsigned char *end; \
        size_t copy, have, size = 0; \
        have = 0; \
        do { \
            if (g.in_left == 0 && load() == 0) \
                return -3; \
            end = memchr(g.in_next, 0, g.in_left); \
            copy = end == NULL ? g.in_left : (size_t)(end - g.in_next) + 1; \
            have = vmemcpy(&str, &size, have, g.in_next, copy); \
            g.in_left -= copy; \
            g.in_next += copy; \
        } while (end == NULL); \
        crc = crc32z(crc, (unsigned char *)str, have); \
    } while (0)

// Pull LSB order or MSB order integers from an unsigned char buffer.
#define PULL2L(p) ((p)[0] + ((unsigned)((p)[1]) << 8))
#define PULL4L(p) (PULL2L(p) + ((unsigned long)(PULL2L((p) + 2)) << 16))
#define PULL2M(p) (((unsigned)((p)[0]) << 8) + (p)[1])
#define PULL4M(p) (((unsigned long)(PULL2M(p)) << 16) + PULL2M((p) + 2))

// Convert MS-DOS date and time to a Unix time, assuming current timezone.
// (You got a better idea?)
local time_t dos2time(unsigned long dos) {
    struct tm tm;

    if (dos == 0)
        return time(NULL);
    tm.tm_year = ((int)(dos >> 25) & 0x7f) + 80;
    tm.tm_mon  = ((int)(dos >> 21) & 0xf) - 1;
    tm.tm_mday = (int)(dos >> 16) & 0x1f;
    tm.tm_hour = (int)(dos >> 11) & 0x1f;
    tm.tm_min  = (int)(dos >> 5) & 0x3f;
    tm.tm_sec  = (int)(dos << 1) & 0x3e;
    tm.tm_isdst = -1;           // figure out if DST or not
    return mktime(&tm);
}

// Convert an unsigned 32-bit integer to signed, even if long > 32 bits.
local long tolong(unsigned long val) {
    return (long)(val & 0x7fffffffUL) - (long)(val & 0x80000000UL);
}

// Process zip extra field to extract zip64 lengths and Unix mod time.
local int read_extra(unsigned len, int save) {
    unsigned id, size, tmp2;
    unsigned long tmp4;

    // process extra blocks
    while (len >= 4) {
        id = GET2();
        size = GET2();
        if (g.in_eof)
            return -1;
        len -= 4;
        if (size > len)
            break;
        len -= size;
        if (id == 0x0001) {
            // Zip64 Extended Information Extra Field
            g.zip64 = 1;
            if (g.zip_ulen == LOW32 && size >= 8) {
                g.zip_ulen = GET4();
                SKIP(4);
                size -= 8;
            }
            if (g.zip_clen == LOW32 && size >= 8) {
                g.zip_clen = GET4();
                SKIP(4);
                size -= 8;
            }
        }
        if (save) {
            if ((id == 0x000d || id == 0x5855) && size >= 8) {
                // PKWare Unix or Info-ZIP Type 1 Unix block
                SKIP(4);
                g.stamp = tolong(GET4());
                size -= 8;
            }
            if (id == 0x5455 && size >= 5) {
                // Extended Timestamp block
                size--;
                if (GET() & 1) {
                    g.stamp = tolong(GET4());
                    size -= 4;
                }
            }
        }
        SKIP(size);
    }
    SKIP(len);
    return 0;
}

// Read a gzip, zip, zlib, or Unix compress header from ind and return the
// compression method in the range 0..257. 8 is deflate, 256 is a zip method
// greater than 255, and 257 is LZW (compress). The only methods decompressed
// by pigz are 8 and 257. On error, return negative: -1 is immediate EOF, -2 is
// not a recognized compressed format (considering only the first two bytes of
// input), -3 is premature EOF within the header, -4 is unexpected header flag
// values, -5 is the zip central directory, and -6 is a failed gzip header crc
// check. If -2 is returned, the input pointer has been reset to the beginning.
// If the return value is not negative, then get_header() sets g.form to
// indicate gzip (0), zlib (1), or zip (2, or 3 if the entry is followed by a
// data descriptor), and the input points to the first byte of compressed data.
local int get_header(int save) {
    unsigned magic;             // magic header
    unsigned method;            // compression method
    unsigned flags;             // header flags
    unsigned fname, extra;      // name and extra field lengths
    unsigned tmp2;              // for macro
    unsigned long tmp4;         // for macro
    unsigned long crc;          // gzip header crc

    // clear return information
    if (save) {
        g.stamp = 0;
        RELEASE(g.hname);
        RELEASE(g.hcomm);
    }

    // see if it's a gzip, zlib, or lzw file
    g.magic1 = GET();
    if (g.in_eof) {
        g.magic1 = -1;
        return -1;
    }
    magic = (unsigned)g.magic1 << 8;
    magic += GET();
    if (g.in_eof)
        return -2;
    if (magic % 31 == 0 && (magic & 0x8f20) == 0x0800) {
        // it's zlib
        g.form = 1;
        return 8;
    }
    if (magic == 0x1f9d) {          // it's lzw
        g.form = -1;
        return 257;
    }
    if (magic == 0x504b) {          // it's zip
        magic = GET2();             // the rest of the signature
        if (g.in_eof)
            return -3;
        if (magic == 0x0201 || magic == 0x0806)
            return -5;              // central header or archive extra
        if (magic != 0x0403)
            return -4;              // not a local header
        g.zip64 = 0;
        SKIP(2);
        flags = GET2();
        if (flags & 0xf7f0)
            return -4;
        method = GET();             // return low byte of method or 256
        if (GET() != 0 || flags & 1)
            method = 256;           // unknown or encrypted
        if (save)
            g.stamp = dos2time(GET4());
        else
            SKIP(4);
        g.zip_crc = GET4();
        g.zip_clen = GET4();
        g.zip_ulen = GET4();
        fname = GET2();
        extra = GET2();
        if (save) {
            char *next;

            if (g.in_eof)
                return -3;
            next = g.hname = alloc(NULL, fname + 1);
            while (fname > g.in_left) {
                memcpy(next, g.in_next, g.in_left);
                fname -= g.in_left;
                next += g.in_left;
                if (load() == 0)
                    return -3;
            }
            memcpy(next, g.in_next, fname);
            g.in_left -= fname;
            g.in_next += fname;
            next += fname;
            *next = 0;
        }
        else
            SKIP(fname);
        read_extra(extra, save);
        g.form = 2 + ((flags & 8) >> 3);
        return g.in_eof ? -3 : (int)method;
    }
    if (magic != 0x1f8b) {          // not gzip
        g.in_left++;                // return the second byte
        g.in_next--;
        return -2;
    }

    // it's gzip -- get method and flags
    crc = 0xf6e946c9;       // crc of 0x1f 0x8b
    method = GETC();
    flags = GETC();
    if (flags & 0xe0)
        return -4;

    // get time stamp
    if (save)
        g.stamp = tolong(GET4C());
    else
        SKIPC(4);

    // skip extra field and OS
    SKIPC(2);

    // skip extra field, if present
    if (flags & 4)
        SKIPC(GET2C());

    // read file name, if present, into allocated memory
    if (flags & 8) {
        if (save)
            GETZC(g.hname);
        else
            while (GETC() != 0)
                ;
    }

    // read comment, if present, into allocated memory
    if (flags & 16) {
        if (save)
            GETZC(g.hcomm);
        else
            while (GETC() != 0)
                ;
    }

    // check header crc
    if ((flags & 2) && GET2() != (crc & 0xffff))
        return -6;

    // return gzip compression method
    g.form = 0;
    return g.in_eof ? -3 : (int)method;
}

// Process the remainder of a zip file after the first entry. Return true if
// the next signature is another local file header. If listing verbosely, then
// search the remainder of the zip file for the central file header
// corresponding to the first zip entry, and save the file comment, if any.
local int more_zip_entries(void) {
    unsigned long sig;
    int ret, n;
    unsigned char *first;
    unsigned tmp2;              // for macro
    unsigned long tmp4;         // for macro
    unsigned char const central[] = {0x50, 0x4b, 1, 2};

    sig = GET4();
    ret = !g.in_eof && sig == 0x04034b50;   // true if another entry follows
    if (!g.list || g.verbosity < 2)
        return ret;

    // if it was a central file header signature, then already four bytes
    // into a central directory header -- otherwise search for the next one
    n = sig == 0x02014b50 ? 4 : 0;  // number of bytes into central header
    for (;;) {
        // assure that more input is available
        if (g.in_left == 0 && load() == 0)      // never found it!
            return ret;
        if (n == 0) {
            // look for first byte in central signature
            first = memchr(g.in_next, central[0], g.in_left);
            if (first == NULL) {
                // not found -- go get the next buffer and keep looking
                g.in_left = 0;
            }
            else {
                // found -- continue search at next byte
                n++;
                g.in_left -= first - g.in_next + 1;
                g.in_next = first + 1;
            }
        }
        else if (n < 4) {
            // look for the remaining bytes in the central signature
            if (g.in_next[0] == central[n]) {
                n++;
                g.in_next++;
                g.in_left--;
            }
            else
                n = 0;      // mismatch -- restart search with this byte
        }
        else {
            // Now in a suspected central file header, just past the signature.
            // Read the rest of the fixed-length portion of the header.
            unsigned char head[CEN];
            size_t need = CEN, part = 0, len, i;

            if (need > g.in_left) {     // will only need to do this once
                part = g.in_left;
                memcpy(head + CEN - need, g.in_next, part);
                need -= part;
                g.in_left = 0;
                if (load() == 0)                // never found it!
                    return ret;
            }
            memcpy(head + CEN - need, g.in_next, need);

            // Determine to sufficient probability that this is the droid we're
            // looking for, by checking the CRC and the local header offset.
            if (PULL4L(head + 12) == g.out_check && PULL4L(head + 38) == 0) {
                // Update the number of bytes consumed from the current buffer.
                g.in_next += need;
                g.in_left -= need;

                // Get the comment length.
                len = PULL2L(head + 28);
                if (len == 0)                   // no comment
                    return ret;

                // Skip the file name and extra field.
                SKIP(PULL2L(head + 24) + (unsigned long)PULL2L(head + 26));

                // Save the comment field.
                need = len;
                g.hcomm = alloc(NULL, len + 1);
                while (need > g.in_left) {
                    memcpy(g.hcomm + len - need, g.in_next, g.in_left);
                    need -= g.in_left;
                    g.in_left = 0;
                    if (load() == 0) {          // premature EOF
                        RELEASE(g.hcomm);
                        return ret;
                    }
                }
                memcpy(g.hcomm + len - need, g.in_next, need);
                g.in_next += need;
                g.in_left -= need;
                for (i = 0; i < len; i++)
                    if (g.hcomm[i] == 0)
                        g.hcomm[i] = ' ';
                g.hcomm[len] = 0;
                return ret;
            }
            else {
                // Nope, false alarm. Restart the search at the first byte
                // after what we thought was the central file header signature.
                if (part) {
                    // Move buffer data up and insert the part of the header
                    // data read from the previous buffer.
                    memmove(g.in_next + part, g.in_next, g.in_left);
                    memcpy(g.in_next, head, part);
                    g.in_left += part;
                }
                n = 0;
            }
        }
    }
}

// --- list contents of compressed input (gzip, zlib, or lzw) ---

// Find standard compressed file suffix, return length of suffix.
local size_t compressed_suffix(char *nm) {
    size_t len;

    len = strlen(nm);
    if (len > 4) {
        nm += len - 4;
        len = 4;
        if (strcmp(nm, ".zip") == 0 || strcmp(nm, ".ZIP") == 0 ||
            strcmp(nm, ".tgz") == 0)
            return 4;
    }
    if (len > 3) {
        nm += len - 3;
        len = 3;
        if (strcmp(nm, ".gz") == 0 || strcmp(nm, "-gz") == 0 ||
            strcmp(nm, ".zz") == 0 || strcmp(nm, "-zz") == 0)
            return 3;
    }
    if (len > 2) {
        nm += len - 2;
        if (strcmp(nm, ".z") == 0 || strcmp(nm, "-z") == 0 ||
            strcmp(nm, "_z") == 0 || strcmp(nm, ".Z") == 0)
            return 2;
    }
    return 0;
}

// Listing file name lengths for -l and -lv.
#define NAMEMAX1 48     // name display limit at verbosity 1
#define NAMEMAX2 16     // name display limit at verbosity 2

// Print gzip, lzw, zlib, or zip file information.
local void show_info(int method, unsigned long check, length_t len, int cont) {
    size_t max;             // maximum name length for current verbosity
    size_t n;               // name length without suffix
    time_t now;             // for getting current year
    char mod[26];           // modification time in text
    char tag[NAMEMAX1+1];   // header or file name, possibly truncated

    // create abbreviated name from header file name or actual file name
    max = g.verbosity > 1 ? NAMEMAX2 : NAMEMAX1;
    memset(tag, 0, max + 1);
    if (cont)
        strncpy(tag, "<...>", max + 1);
    else if (g.hname == NULL) {
        n = strlen(g.inf) - compressed_suffix(g.inf);
        memcpy(tag, g.inf, n > max + 1 ? max + 1 : n);
        if (strcmp(g.inf + n, ".tgz") == 0 && n < max + 1)
            strncpy(tag + n, ".tar", max + 1 - n);
    }
    else
        strncpy(tag, g.hname, max + 1);
    if (tag[max])
        strcpy(tag + max - 3, "...");

    // convert time stamp to text
    if (g.stamp) {
        strcpy(mod, ctime(&g.stamp));
        now = time(NULL);
        if (strcmp(mod + 20, ctime(&now) + 20) != 0)
            strcpy(mod + 11, mod + 19);
    }
    else
        strcpy(mod + 4, "------ -----");
    mod[16] = 0;

    // if first time, print header
    if (g.first) {
        if (g.verbosity > 1)
            fputs("method    check    timestamp    ", stdout);
        if (g.verbosity > 0)
            puts("compressed   original reduced  name");
        g.first = 0;
    }

    // print information
    if (g.verbosity > 1) {
        if (g.form == 3 && !g.decode)
            printf("zip%3d  --------  %s  ", method, mod + 4);
        else if (g.form > 1)
            printf("zip%3d  %08lx  %s  ", method, check, mod + 4);
        else if (g.form == 1)
            printf("zlib%2d  %08lx  %s  ", method, check, mod + 4);
        else if (method == 257)
            printf("lzw     --------  %s  ", mod + 4);
        else
            printf("gzip%2d  %08lx  %s  ", method, check, mod + 4);
    }
    if (g.verbosity > 0) {
        // compute reduction percent -- allow divide-by-zero, displays as -inf%
        double red = 100. * (len - (double)g.in_tot) / len;
        if ((g.form == 3 && !g.decode) ||
            (method == 8 && g.in_tot > (len + (len >> 10) + 12)) ||
            (method == 257 && g.in_tot > len + (len >> 1) + 3))
#if __STDC_VERSION__-0 >= 199901L || __GNUC__-0 >= 3
            printf("%10jd %10jd?  unk    %s\n",
                   (intmax_t)g.in_tot, (intmax_t)len, tag);
        else
            printf("%10jd %10jd %6.1f%%  %s\n",
                   (intmax_t)g.in_tot, (intmax_t)len, red, tag);
#else
            printf(sizeof(off_t) == sizeof(long) ?
                   "%10ld %10ld?  unk    %s\n" : "%10lld %10lld?  unk    %s\n",
                   g.in_tot, len, tag);
        else
            printf(sizeof(off_t) == sizeof(long) ?
                   "%10ld %10ld %6.1f%%  %s\n" : "%10lld %10lld %6.1f%%  %s\n",
                   g.in_tot, len, red, tag);
#endif
    }
    if (g.verbosity > 1 && g.hcomm != NULL)
        puts(g.hcomm);
}

// List content information about the gzip file at ind (only works if the gzip
// file contains a single gzip stream with no junk at the end, and only works
// well if the uncompressed length is less than 4 GB).
local void list_info(void) {
    int method;             // get_header() return value
    size_t n;               // available trailer bytes
    off_t at;               // used to calculate compressed length
    unsigned char tail[8];  // trailer containing check and length
    unsigned long check;    // check value
    length_t len;           // length from trailer

    // initialize input buffer
    in_init();

    // read header information and position input after header
    method = get_header(1);
    if (method < 0) {
        complain(method == -6 ? "skipping: %s corrupt: header crc error" :
                 method == -1 ? "skipping: %s empty" :
                 "skipping: %s unrecognized format", g.inf);
        return;
    }

#ifndef NOTHREAD
    // wait for read thread to complete current read() operation, to permit
    // seeking and reading on g.ind here in the main thread
    load_wait();
#endif

    // list zip file
    if (g.form > 1) {
        more_zip_entries();         // get first entry comment, if any
        g.in_tot = g.zip_clen;
        show_info(method, g.zip_crc, g.zip_ulen, 0);
        return;
    }

    // list zlib file
    if (g.form == 1) {
        at = lseek(g.ind, 0, SEEK_END);
        if (at == -1) {
            check = 0;
            do {
                len = g.in_left < 4 ? g.in_left : 4;
                g.in_next += g.in_left - len;
                while (len--)
                    check = (check << 8) + *g.in_next++;
            } while (load() != 0);
            check &= LOW32;
        }
        else {
            g.in_tot = (length_t)at;
            lseek(g.ind, -4, SEEK_END);
            readn(g.ind, tail, 4);
            check = PULL4M(tail);
        }
        g.in_tot -= 6;
        show_info(method, check, 0, 0);
        return;
    }

    // list lzw file
    if (method == 257) {
        at = lseek(g.ind, 0, SEEK_END);
        if (at == -1)
            while (load() != 0)
                ;
        else
            g.in_tot = (length_t)at;
        g.in_tot -= 3;
        show_info(method, 0, 0, 0);
        return;
    }

    // skip to end to get trailer (8 bytes), compute compressed length
    if (g.in_short) {                   // whole thing already read
        if (g.in_left < 8) {
            complain("skipping: %s not a valid gzip file", g.inf);
            return;
        }
        g.in_tot = g.in_left - 8;       // compressed size
        memcpy(tail, g.in_next + (g.in_left - 8), 8);
    }
    else if ((at = lseek(g.ind, -8, SEEK_END)) != -1) {
        g.in_tot = (length_t)at - g.in_tot + g.in_left; // compressed size
        readn(g.ind, tail, 8);          // get trailer
    }
    else {                              // can't seek
        len = g.in_tot - g.in_left;     // save header size
        do {
            n = g.in_left < 8 ? g.in_left : 8;
            memcpy(tail, g.in_next + (g.in_left - n), n);
            load();
        } while (g.in_left == BUF);     // read until end
        if (g.in_left < 8) {
            if (n + g.in_left < 8) {
                complain("skipping: %s not a valid gzip file", g.inf);
                return;
            }
            if (g.in_left) {
                if (n + g.in_left > 8)
                    memcpy(tail, tail + n - (8 - g.in_left), 8 - g.in_left);
                memcpy(tail + 8 - g.in_left, g.in_next, g.in_left);
            }
        }
        else
            memcpy(tail, g.in_next + (g.in_left - 8), 8);
        g.in_tot -= len + 8;
    }
    if (g.in_tot < 2) {
        complain("skipping: %s not a valid gzip file", g.inf);
        return;
    }

    // convert trailer to check and uncompressed length (modulo 2^32)
    check = PULL4L(tail);
    len = PULL4L(tail + 4);

    // list information about contents
    show_info(method, check, len, 0);
}

// --- copy input to output (when acting like cat) ---

local void cat(void) {
    // copy the first header byte read, if any
    if (g.magic1 != -1) {
        unsigned char buf[1] = {g.magic1};
        g.out_tot += writen(g.outd, buf, 1);
    }

    // copy the remainder of the input to the output
    while (g.in_left) {
        g.out_tot += writen(g.outd, g.in_next, g.in_left);
        g.in_left = 0;
        load();
    }
}

// --- decompress deflate input ---

// Call-back input function for inflateBack().
local unsigned inb(void *desc, unsigned char **buf) {
    (void)desc;
    if (g.in_left == 0)
        load();
    *buf = g.in_next;
    unsigned len = g.in_left > UINT_MAX ? UINT_MAX : (unsigned)g.in_left;
    g.in_next += len;
    g.in_left -= len;
    return len;
}

// Output buffers and window for infchk() and unlzw().
#define OUTSIZE 32768U      // must be at least 32K for inflateBack() window
local unsigned char out_buf[OUTSIZE];

#ifndef NOTHREAD
// Output data for parallel write and check.
local unsigned char out_copy[OUTSIZE];
local size_t out_len;

// outb threads states.
local lock *outb_write_more = NULL;
local lock *outb_check_more;

// Output write thread.
local void outb_write(void *dummy) {
    size_t len;
    ball_t err;                     // error information from throw()

    (void)dummy;

    Trace(("-- launched decompress write thread"));
    try {
        do {
            possess(outb_write_more);
            wait_for(outb_write_more, TO_BE, 1);
            len = out_len;
            if (len && g.decode == 1)
                writen(g.outd, out_copy, len);
            Trace(("-- decompress wrote %lu bytes", len));
            twist(outb_write_more, TO, 0);
        } while (len);
    }
    catch (err) {
        THREADABORT(err);
    }
    Trace(("-- exited decompress write thread"));
}

// Output check thread.
local void outb_check(void *dummy) {
    size_t len;
    ball_t err;                     // error information from throw()

    (void)dummy;

    Trace(("-- launched decompress check thread"));
    try {
        do {
            possess(outb_check_more);
            wait_for(outb_check_more, TO_BE, 1);
            len = out_len;
            g.out_check = CHECK(g.out_check, out_copy, len);
            Trace(("-- decompress checked %lu bytes", len));
            twist(outb_check_more, TO, 0);
        } while (len);
    }
    catch (err) {
        THREADABORT(err);
    }
    Trace(("-- exited decompress check thread"));
}
#endif

// Call-back output function for inflateBack(). Wait for the last write and
// check calculation to complete, copy the write buffer, and then alert the
// write and check threads and return for more decompression while that's going
// on (or just write and check if no threads or if proc == 1).
local int outb(void *desc, unsigned char *buf, unsigned len) {
    (void)desc;

#ifndef NOTHREAD
    static thread *wr, *ch;

    if (g.procs > 1) {
        // if first time, initialize state and launch threads
        if (outb_write_more == NULL) {
            outb_write_more = new_lock(0);
            outb_check_more = new_lock(0);
            wr = launch(outb_write, NULL);
            ch = launch(outb_check, NULL);
        }

        // wait for previous write and check threads to complete
        possess(outb_check_more);
        wait_for(outb_check_more, TO_BE, 0);
        possess(outb_write_more);
        wait_for(outb_write_more, TO_BE, 0);

        // copy the output and alert the worker bees
        out_len = len;
        g.out_tot += len;
        memcpy(out_copy, buf, len);
        twist(outb_write_more, TO, 1);
        twist(outb_check_more, TO, 1);

        // if requested with len == 0, clean up -- terminate and join write and
        // check threads, free lock
        if (len == 0 && outb_write_more != NULL) {
            join(ch);
            join(wr);
            free_lock(outb_check_more);
            free_lock(outb_write_more);
            outb_write_more = NULL;
        }

        // return for more decompression while last buffer is being written and
        // having its check value calculated -- we wait for those to finish the
        // next time this function is called
        return 0;
    }
#endif

    // if just one process or no threads, then do it without threads
    if (len) {
        if (g.decode == 1)
            writen(g.outd, buf, len);
        g.out_check = CHECK(g.out_check, buf, len);
        g.out_tot += len;
    }
    return 0;
}

// Zip file data descriptor signature. This signature may or may not precede
// the CRC and lengths, with either resulting in a valid zip file! There is
// some odd code below that tries to detect and accommodate both cases.
#define SIG 0x08074b50

// Inflate for decompression or testing. Decompress from ind to outd unless
// decode != 1, in which case just test ind, and then also list if list != 0;
// look for and decode multiple, concatenated gzip and/or zlib streams; read
// and check the gzip, zlib, or zip trailer.
local void infchk(void) {
    int ret, cont, more;
    unsigned long check, len;
    z_stream strm;
    unsigned tmp2;
    unsigned long tmp4;
    length_t clen;

    cont = more = 0;
    do {
        // header already read -- set up for decompression
        g.in_tot = g.in_left;       // track compressed data length
        g.out_tot = 0;
        g.out_check = CHECK(0L, Z_NULL, 0);
        strm.zalloc = ZALLOC;
        strm.zfree = ZFREE;
        strm.opaque = OPAQUE;
        ret = inflateBackInit(&strm, 15, out_buf);
        if (ret == Z_MEM_ERROR)
            throw(ENOMEM, "not enough memory");
        if (ret != Z_OK)
            throw(EINVAL, "internal error");

        // decompress, compute lengths and check value
        strm.avail_in = 0;
        strm.next_in = Z_NULL;
        ret = inflateBack(&strm, inb, NULL, outb, NULL);
        inflateBackEnd(&strm);
        if (ret == Z_DATA_ERROR)
            throw(EDOM, "%s: corrupted -- invalid deflate data (%s)",
                  g.inf, strm.msg);
        if (ret == Z_BUF_ERROR)
            throw(EDOM, "%s: corrupted -- incomplete deflate data", g.inf);
        if (ret != Z_STREAM_END)
            throw(EINVAL, "internal error");
        g.in_left += strm.avail_in;
        g.in_next = strm.next_in;
        outb(NULL, NULL, 0);        // finish off final write and check

        // compute compressed data length
        clen = g.in_tot - g.in_left;

        // read and check trailer
        if (g.form > 1) {           // zip local trailer (if any)
            if (g.form == 3) {      // data descriptor follows
                // get data descriptor values, assuming no signature
                g.zip_crc = GET4();
                g.zip_clen = GET4();
                g.zip_ulen = GET4();        // ZIP64 -> high clen, not ulen

                // deduce whether or not a signature precedes the values
                if (g.zip_crc == SIG &&         // might be the signature
                    // if the expected CRC is not SIG, then it's a signature
                    (g.out_check != SIG ||      // assume signature
                     // now we're in a very rare case where CRC == SIG -- the
                     // first four bytes could be the signature or the CRC
                     (g.zip_clen == SIG &&      // if not, then no signature
                      // now we have the first two words are SIG and the
                      // expected CRC is SIG, so it could be a signature and
                      // the CRC, or it could be the CRC and a compressed
                      // length that is *also* SIG (!) -- so check the low 32
                      // bits of the expected compressed length for SIG
                      ((clen & LOW32) != SIG || // assume signature and CRC
                       // now the expected CRC *and* the expected low 32 bits
                       // of the compressed length are SIG -- this is so
                       // incredibly unlikely, clearly someone is messing with
                       // us, but we continue ... if the next four bytes are
                       // not SIG, then there is not a signature -- check those
                       // bytes, currently in g.zip_ulen:
                       (g.zip_ulen == SIG &&    // if not, then no signature
                        // we have three SIGs in a row in the descriptor, and
                        // both the expected CRC and the expected clen are SIG
                        // -- the first one is a signature if we don't expect
                        // the third word to be SIG, which is either the low 32
                        // bits of ulen, or if ZIP64, the high 32 bits of clen:
                        (g.zip64 ? clen >> 32 : g.out_tot) != SIG
                        // if that last compare was equal, then the expected
                        // values for the CRC, the low 32 bits of clen, *and*
                        // the low 32 bits of ulen are all SIG (!!), or in the
                        // case of ZIP64, even crazier, the CRC and *both*
                        // 32-bit halves of clen are all SIG (clen > 500
                        // petabytes!!!) ... we can no longer discriminate the
                        // hypotheses, so we will assume no signature
                        ))))) {
                    // first four bytes were actually the descriptor -- shift
                    // the values down and get another four bytes
                    g.zip_crc = g.zip_clen;
                    g.zip_clen = g.zip_ulen;
                    g.zip_ulen = GET4();
                }

                // if ZIP64, then ulen is really the high word of clen -- get
                // the actual ulen and skip its high word as well (we only
                // compare the low 32 bits of the lengths to verify)
                if (g.zip64) {
                    g.zip_ulen = GET4();
                    (void)GET4();
                }
                if (g.in_eof)
                    throw(EDOM, "%s: corrupted entry -- missing trailer",
                          g.inf);
            }
            check = g.zip_crc;
            if (check != g.out_check)
                throw(EDOM, "%s: corrupted entry -- crc32 mismatch", g.inf);
            if (g.zip_clen != (clen & LOW32) ||
                g.zip_ulen != (g.out_tot & LOW32))
                throw(EDOM, "%s: corrupted entry -- length mismatch",
                      g.inf);
            more = more_zip_entries();  // see if more entries, get comment
        }
        else if (g.form == 1) {     // zlib (big-endian) trailer
            check = (unsigned long)(GET()) << 24;
            check += (unsigned long)(GET()) << 16;
            check += (unsigned)(GET()) << 8;
            check += GET();
            if (g.in_eof)
                throw(EDOM, "%s: corrupted -- missing trailer", g.inf);
            if (check != g.out_check)
                throw(EDOM, "%s: corrupted -- adler32 mismatch", g.inf);
        }
        else {                      // gzip trailer
            check = GET4();
            len = GET4();
            if (g.in_eof)
                throw(EDOM, "%s: corrupted -- missing trailer", g.inf);
            if (check != g.out_check)
                throw(EDOM, "%s: corrupted -- crc32 mismatch", g.inf);
            if (len != (g.out_tot & LOW32))
                throw(EDOM, "%s: corrupted -- length mismatch", g.inf);
        }

        // show file information if requested
        if (g.list) {
            g.in_tot = clen;
            show_info(8, check, g.out_tot, cont);
            cont = 1;
        }

        // if a gzip entry follows a gzip entry, decompress it (don't replace
        // saved header information from first entry)
    } while (g.form == 0 && (ret = get_header(0)) == 8);

    // gzip -cdf copies junk after gzip stream directly to output
    if (g.form == 0 && ret == -2 && g.force && g.pipeout && g.decode != 2 &&
        !g.list)
        cat();

    // check for more entries in zip file
    else if (more) {
        complain("warning: %s: entries after the first were ignored", g.inf);
        g.keep = 1;         // don't delete the .zip file
    }

    // check for non-gzip after gzip stream, or anything after zlib stream
    else if ((g.verbosity > 1 && g.form == 0 && ret != -1) ||
             (g.form == 1 && (GET(), !g.in_eof)))
        complain("warning: %s: trailing junk was ignored", g.inf);
}

// --- decompress Unix compress (LZW) input ---

// Type for accumulating bits. 23 bits will be used to accumulate up to 16-bit
// symbols.
typedef unsigned long bits_t;

#define NOMORE() (g.in_left == 0 && (g.in_eof || load() == 0))
#define NEXT() (g.in_left--, (unsigned)*g.in_next++)

// Decompress a compress (LZW) file from ind to outd. The compress magic header
// (two bytes) has already been read and verified.
local void unlzw(void) {
    unsigned bits;              // current bits per code (9..16)
    unsigned mask;              // mask for current bits codes = (1<<bits)-1
    bits_t buf;                 // bit buffer (need 23 bits)
    unsigned left;              // bits left in buf (0..7 after code pulled)
    length_t mark;              // offset where last change in bits began
    unsigned code;              // code, table traversal index
    unsigned max;               // maximum bits per code for this stream
    unsigned flags;             // compress flags, then block compress flag
    unsigned end;               // last valid entry in prefix/suffix tables
    unsigned prev;              // previous code
    unsigned final;             // last character written for previous code
    unsigned stack;             // next position for reversed string
    unsigned outcnt;            // bytes in output buffer
    // memory for unlzw() -- the first 256 entries of prefix[] and suffix[] are
    // never used, so could have offset the index but it's faster to waste a
    // little memory
    uint_least16_t prefix[65536];       // index to LZW prefix string
    unsigned char suffix[65536];        // one-character LZW suffix
    unsigned char match[65280 + 2];     // buffer for reversed match

    // process remainder of compress header -- a flags byte
    g.out_tot = 0;
    if (NOMORE())
        throw(EDOM, "%s: lzw premature end", g.inf);
    flags = NEXT();
    if (flags & 0x60)
        throw(EDOM, "%s: unknown lzw flags set", g.inf);
    max = flags & 0x1f;
    if (max < 9 || max > 16)
        throw(EDOM, "%s: lzw bits out of range", g.inf);
    if (max == 9)                           // 9 doesn't really mean 9
        max = 10;
    flags &= 0x80;                          // true if block compress

    // mark the start of the compressed data for computing the first flush
    mark = g.in_tot - g.in_left;

    // clear table, start at nine bits per symbol
    bits = 9;
    mask = 0x1ff;
    end = flags ? 256 : 255;

    // set up: get first 9-bit code, which is the first decompressed byte, but
    // don't create a table entry until the next code
    if (NOMORE())                           // no compressed data is ok
        return;
    buf = NEXT();
    if (NOMORE())
        throw(EDOM, "%s: lzw premature end", g.inf);  // need nine bits
    buf += NEXT() << 8;
    final = prev = buf & mask;              // code
    buf >>= bits;
    left = 16 - bits;
    if (prev > 255)
        throw(EDOM, "%s: invalid lzw code", g.inf);
    out_buf[0] = (unsigned char)final;      // write first decompressed byte
    outcnt = 1;

    // decode codes
    stack = 0;
    for (;;) {
        // if the table will be full after this, increment the code size
        if (end >= mask && bits < max) {
            // flush unused input bits and bytes to next 8*bits bit boundary
            // (this is a vestigial aspect of the compressed data format
            // derived from an implementation that made use of a special VAX
            // machine instruction!)
            {
                unsigned rem = ((g.in_tot - g.in_left) - mark) % bits;
                if (rem) {
                    rem = bits - rem;
                    if (NOMORE())
                        break;              // end of compressed data
                    while (rem > g.in_left) {
                        rem -= g.in_left;
                        if (load() == 0)
                            throw(EDOM, "%s: lzw premature end", g.inf);
                    }
                    g.in_left -= rem;
                    g.in_next += rem;
                }
            }
            buf = 0;
            left = 0;

            // mark this new location for computing the next flush
            mark = g.in_tot - g.in_left;

            // go to the next number of bits per symbol
            bits++;
            mask <<= 1;
            mask++;
        }

        // get a code of bits bits
        if (NOMORE())
            break;                          // end of compressed data
        buf += (bits_t)(NEXT()) << left;
        left += 8;
        if (left < bits) {
            if (NOMORE())
                throw(EDOM, "%s: lzw premature end", g.inf);
            buf += (bits_t)(NEXT()) << left;
            left += 8;
        }
        code = buf & mask;
        buf >>= bits;
        left -= bits;

        // process clear code (256)
        if (code == 256 && flags) {
            // flush unused input bits and bytes to next 8*bits bit boundary
            {
                unsigned rem = ((g.in_tot - g.in_left) - mark) % bits;
                if (rem) {
                    rem = bits - rem;
                    while (rem > g.in_left) {
                        rem -= g.in_left;
                        if (load() == 0)
                            throw(EDOM, "%s: lzw premature end", g.inf);
                    }
                    g.in_left -= rem;
                    g.in_next += rem;
                }
            }
            buf = 0;
            left = 0;

            // mark this new location for computing the next flush
            mark = g.in_tot - g.in_left;

            // go back to nine bits per symbol
            bits = 9;                       // initialize bits and mask
            mask = 0x1ff;
            end = 255;                      // empty table
            continue;                       // get next code
        }

        // special code to reuse last match
        {
            unsigned temp = code;           // save the current code
            if (code > end) {
                // be picky on the allowed code here, and make sure that the
                // code we drop through (prev) will be a valid index so that
                // random input does not cause an exception
                if (code != end + 1 || prev > end)
                    throw(EDOM, "%s: invalid lzw code", g.inf);
                match[stack++] = (unsigned char)final;
                code = prev;
            }

            // walk through linked list to generate output in reverse order
            while (code >= 256) {
                match[stack++] = suffix[code];
                code = prefix[code];
            }
            match[stack++] = (unsigned char)code;
            final = code;

            // link new table entry
            if (end < mask) {
                end++;
                prefix[end] = (uint_least16_t)prev;
                suffix[end] = (unsigned char)final;
            }

            // set previous code for next iteration
            prev = temp;
        }

        // write output in forward order
        while (stack > OUTSIZE - outcnt) {
            while (outcnt < OUTSIZE)
                out_buf[outcnt++] = match[--stack];
            g.out_tot += outcnt;
            if (g.decode == 1)
                writen(g.outd, out_buf, outcnt);
            outcnt = 0;
        }
        do {
            out_buf[outcnt++] = match[--stack];
        } while (stack);
    }

    // write any remaining buffered output
    g.out_tot += outcnt;
    if (outcnt && g.decode == 1)
        writen(g.outd, out_buf, outcnt);
}

// --- file processing ---

// Extract file name from path.
local char *justname(char *path) {
    char *p;

    p = strrchr(path, '/');
    return p == NULL ? path : p + 1;
}

// Copy file attributes, from -> to, as best we can. This is best effort, so no
// errors are reported. The mode bits, including suid, sgid, and the sticky bit
// are copied (if allowed), the owner's user id and group id are copied (again
// if allowed), and the access and modify times are copied.
local int copymeta(char *from, char *to) {
    struct stat st;
    struct timeval times[2];

    // get all of from's Unix meta data, return if not a regular file
    if (stat(from, &st) != 0 || (st.st_mode & S_IFMT) != S_IFREG)
        return -4;

    // set to's mode bits, ignore errors
    int ret = chmod(to, st.st_mode & 07777);

    // copy owner's user and group, ignore errors
    ret += chown(to, st.st_uid, st.st_gid);

    // copy access and modify times, ignore errors
    times[0].tv_sec = st.st_atime;
    times[0].tv_usec = 0;
    times[1].tv_sec = st.st_mtime;
    times[1].tv_usec = 0;
    ret += utimes(to, times);
    return ret;
}

// Set the access and modify times of fd to t.
local void touch(char *path, time_t t) {
    struct timeval times[2];

    times[0].tv_sec = t;
    times[0].tv_usec = 0;
    times[1].tv_sec = t;
    times[1].tv_usec = 0;
    (void)utimes(path, times);
}

// Request that all data buffered by the operating system for g.outd be written
// to the permanent storage device. If fsync(fd) is used (POSIX), then all of
// the data is sent to the device, but will likely be buffered in volatile
// memory on the device itself, leaving open a window of vulnerability.
// fcntl(fd, F_FULLSYNC) on the other hand, available in macOS only, will
// request and wait for the device to write out its buffered data to permanent
// storage. On Windows, _commit() is used.
local void out_push(void) {
    if (g.outd == -1)
        return;
#if defined(F_FULLSYNC)
    int ret = fcntl(g.outd, F_FULLSYNC);
#elif defined(_WIN32)
    int ret = _commit(g.outd);
#else
    int ret = fsync(g.outd);
#endif
    if (ret == -1)
        throw(errno, "sync error on %s (%s)", g.outf, strerror(errno));
}

// Process provided input file, or stdin if path is NULL. process() can call
// itself for recursive directory processing.
local void process(char *path) {
    volatile int method = -1;       // get_header() return value
    size_t len;                     // length of base name (minus suffix)
    struct stat st;                 // to get file type and mod time
    ball_t err;                     // error information from throw()
    // all compressed suffixes for decoding search, in length order
    static char *sufs[] = {".z", "-z", "_z", ".Z", ".gz", "-gz", ".zz", "-zz",
                           ".zip", ".ZIP", ".tgz", NULL};

    // open input file with name in, descriptor ind -- set name and mtime
    if (path == NULL) {
        vstrcpy(&g.inf, &g.inz, 0, "<stdin>");
        g.ind = 0;
        g.name = NULL;
        g.mtime = g.headis & 2 ?
                  (fstat(g.ind, &st) ? time(NULL) : st.st_mtime) : 0;
        len = 0;
    }
    else {
        // set input file name (already set if recursed here)
        if (path != g.inf)
            vstrcpy(&g.inf, &g.inz, 0, path);
        len = strlen(g.inf);

        // try to stat input file -- if not there and decoding, look for that
        // name with compressed suffixes
        if (lstat(g.inf, &st)) {
            if (errno == ENOENT && (g.list || g.decode)) {
                char **sufx = sufs;
                do {
                    if (*sufx == NULL)
                        break;
                    vstrcpy(&g.inf, &g.inz, len, *sufx++);
                    errno = 0;
                } while (lstat(g.inf, &st) && errno == ENOENT);
            }
#if defined(EOVERFLOW) && defined(EFBIG)
            if (errno == EOVERFLOW || errno == EFBIG)
                throw(EDOM, "%s too large -- "
                      "not compiled with large file support", g.inf);
#endif
            if (errno) {
                g.inf[len] = 0;
                complain("skipping: %s does not exist", g.inf);
                return;
            }
            len = strlen(g.inf);
        }

        // only process regular files or named pipes, but allow symbolic links
        // if -f, recurse into directory if -r
        if ((st.st_mode & S_IFMT) != S_IFREG &&
            (st.st_mode & S_IFMT) != S_IFIFO &&
            (st.st_mode & S_IFMT) != S_IFLNK &&
            (st.st_mode & S_IFMT) != S_IFDIR) {
            complain("skipping: %s is a special file or device", g.inf);
            return;
        }
        if ((st.st_mode & S_IFMT) == S_IFLNK && !g.force && !g.pipeout) {
            complain("skipping: %s is a symbolic link", g.inf);
            return;
        }
        if ((st.st_mode & S_IFMT) == S_IFDIR && !g.recurse) {
            complain("skipping: %s is a directory", g.inf);
            return;
        }

        // recurse into directory (assumes Unix)
        if ((st.st_mode & S_IFMT) == S_IFDIR) {
            char *roll = NULL;
            size_t size = 0, off = 0, base;
            DIR *here;
            struct dirent *next;

            // accumulate list of entries (need to do this, since readdir()
            // behavior not defined if directory modified between calls)
            here = opendir(g.inf);
            if (here == NULL)
                return;
            while ((next = readdir(here)) != NULL) {
                if (next->d_name[0] == 0 ||
                    (next->d_name[0] == '.' && (next->d_name[1] == 0 ||
                     (next->d_name[1] == '.' && next->d_name[2] == 0))))
                    continue;
                off = vstrcpy(&roll, &size, off, next->d_name);
            }
            closedir(here);
            vstrcpy(&roll, &size, off, "");

            // run process() for each entry in the directory
            base = len && g.inf[len - 1] != (unsigned char)'/' ?
                   vstrcpy(&g.inf, &g.inz, len, "/") - 1 : len;
            for (off = 0; roll[off]; off += strlen(roll + off) + 1) {
                vstrcpy(&g.inf, &g.inz, base, roll + off);
                process(g.inf);
            }
            g.inf[len] = 0;

            // release list of entries
            FREE(roll);
            return;
        }

        // don't compress .gz (or provided suffix) files, unless -f
        if (!(g.force || g.list || g.decode) && len >= strlen(g.sufx) &&
                strcmp(g.inf + len - strlen(g.sufx), g.sufx) == 0) {
            complain("skipping: %s ends with %s", g.inf, g.sufx);
            return;
        }

        // create output file only if input file has compressed suffix
        if (g.decode == 1 && !g.pipeout && !g.list) {
            size_t suf = compressed_suffix(g.inf);
            if (suf == 0) {
                complain("skipping: %s does not have compressed suffix",
                         g.inf);
                return;
            }
            len -= suf;
        }

        // open input file
        g.ind = open(g.inf, O_RDONLY, 0);
        if (g.ind < 0)
            throw(errno, "read error on %s (%s)", g.inf, strerror(errno));

        // prepare gzip header information for compression
        g.name = g.headis & 1 ? justname(g.inf) : NULL;
        g.mtime = g.headis & 2 ? st.st_mtime : 0;
    }
    SET_BINARY_MODE(g.ind);

    // if decoding or testing, try to read gzip header
    if (g.decode) {
        in_init();
        method = get_header(1);
        if (method != 8 && method != 257 &&
                // gzip -cdf acts like cat on uncompressed input
                !((method == -1 || method == -2) && g.force && g.pipeout &&
                  g.decode != 2 && !g.list)) {
            load_end();
            complain(method == -6 ? "skipping: %s corrupt: header crc error" :
                     method == -1 ? "skipping: %s empty" :
                     method < 0 ? "skipping: %s unrecognized format" :
                     "skipping: %s unknown compression method", g.inf);
            return;
        }

        // if requested, test input file (possibly a test list)
        if (g.decode == 2) {
            try {
                if (method == 8)
                    infchk();
                else {
                    unlzw();
                    if (g.list) {
                        g.in_tot -= 3;
                        show_info(method, 0, g.out_tot, 0);
                    }
                }
            }
            catch (err) {
                if (err.code != EDOM)
                    punt(err);
                complain("skipping: %s", err.why);
                drop(err);
                outb(NULL, NULL, 0);
            }
            load_end();
            return;
        }
    }

    // if requested, just list information about input file
    if (g.list) {
        list_info();
        load_end();
        return;
    }

    // create output file out, descriptor outd
    if (path == NULL || g.pipeout) {
        // write to stdout
        g.outf = alloc(NULL, strlen("<stdout>") + 1);
        strcpy(g.outf, "<stdout>");
        g.outd = 1;
        if (!g.decode && !g.force && isatty(g.outd))
            throw(EINVAL, "trying to write compressed data to a terminal"
                          " (use -f to force)");
    }
    else {
        char *to = g.inf, *sufx = "";
        size_t pre = 0;

        // select parts of the output file name
        if (g.decode) {
            // for -dN or -dNT, use the path from the input file and the name
            // from the header, stripping any path in the header name
            if ((g.headis & 1) != 0 && g.hname != NULL) {
                pre = (size_t)(justname(g.inf) - g.inf);
                to = justname(g.hname);
                len = strlen(to);
            }
            // for -d or -dNn, replace abbreviated suffixes
            else if (strcmp(to + len, ".tgz") == 0)
                sufx = ".tar";
        }
        else
            // add appropriate suffix when compressing
            sufx = g.sufx;

        // create output file and open to write, overwriting any existing file
        // of the same name only if requested with --force or -f
        g.outf = alloc(NULL, pre + len + strlen(sufx) + 1);
        memcpy(g.outf, g.inf, pre);
        memcpy(g.outf + pre, to, len);
        strcpy(g.outf + pre + len, sufx);
        g.outd = open(g.outf, O_CREAT | O_TRUNC | O_WRONLY |
                              (g.force ? 0 : O_EXCL), 0600);

        // if it exists and wasn't forced, give the user a chance to overwrite
        if (g.outd < 0 && errno == EEXIST) {
            int overwrite = 0;
            if (isatty(0) && g.verbosity) {
                // get a response from the user -- the first non-blank
                // character has to be a "y" or a "Y" to permit an overwrite
                fprintf(stderr, "%s exists -- overwrite (y/n)? ", g.outf);
                fflush(stderr);
                int ch, first = 1;
                do {
                    ch = getchar();
                    if (first == 1) {
                        if (ch == ' ' || ch == '\t')
                            continue;
                        if (ch == 'y' || ch == 'Y')
                            overwrite = 1;
                        first = 0;
                    }
                } while (ch != EOF && ch != '\n' && ch != '\r');
            }
            if (!overwrite) {
                complain("skipping: %s exists", g.outf);
                RELEASE(g.outf);
                load_end();
                return;
            }
            g.outd = open(g.outf, O_CREAT | O_TRUNC | O_WRONLY, 0600);
        }

        // if some other error, give up
        if (g.outd < 0)
            throw(errno, "write error on %s (%s)", g.outf, strerror(errno));
    }
    SET_BINARY_MODE(g.outd);

    // process ind to outd
    if (g.verbosity > 1)
        fprintf(stderr, "%s to %s ", g.inf, g.outf);
    if (g.decode) {
        try {
            if (method == 8)
                infchk();
            else if (method == 257)
                unlzw();
            else
                cat();
        }
        catch (err) {
            if (err.code != EDOM)
                punt(err);
            complain("skipping: %s", err.why);
            drop(err);
            outb(NULL, NULL, 0);
            if (g.outd != -1 && g.outd != 1) {
                close(g.outd);
                g.outd = -1;
                unlink(g.outf);
                RELEASE(g.outf);
            }
        }
    }
#ifndef NOTHREAD
    else if (g.procs > 1)
        parallel_compress();
#endif
    else
        single_compress(0);
    if (g.verbosity > 1) {
        putc('\n', stderr);
        fflush(stderr);
    }

    // finish up, copy attributes, set times, delete original
    load_end();
    if (g.outd != -1 && g.outd != 1) {
        if (g.sync)
            out_push();         // push to permanent storage
        if (close(g.outd))
            throw(errno, "write error on %s (%s)", g.outf, strerror(errno));
        g.outd = -1;            // now prevent deletion on interrupt
        if (g.ind != 0) {
            copymeta(g.inf, g.outf);
            if (!g.keep)
                unlink(g.inf);
        }
        if (g.decode && (g.headis & 2) != 0 && g.stamp)
            touch(g.outf, g.stamp);
    }
    RELEASE(g.outf);
}

local char *helptext[] = {
"Usage: pigz [options] [files ...]",
"  will compress files in place, adding the suffix '.gz'. If no files are",
#ifdef NOTHREAD
"  specified, stdin will be compressed to stdout. pigz does what gzip does.",
#else
"  specified, stdin will be compressed to stdout. pigz does what gzip does,",
"  but spreads the work over multiple processors and cores when compressing.",
#endif
"",
"Options:",
#ifdef NOZOPFLI
"  -0 to -9             Compression level",
#else
"  -0 to -9, -11        Compression level (level 11, zopfli, is much slower)",
#endif
"  --fast, --best       Compression levels 1 and 9 respectively",
"  -A, --alias xxx      Use xxx as the name for any --zip entry from stdin",
"  -b, --blocksize mmm  Set compression block size to mmmK (default 128K)",
"  -c, --stdout         Write all processed output to stdout (won't delete)",
"  -C, --comment ccc    Put comment ccc in the gzip or zip header",
"  -d, --decompress     Decompress the compressed input",
"  -f, --force          Force overwrite, compress .gz, links, and to terminal",
#ifndef NOZOPFLI
"  -F  --first          Do iterations first, before block split for -11",
#endif
"  -h, --help           Display a help screen and quit",
"  -i, --independent    Compress blocks independently for damage recovery",
#ifndef NOZOPFLI
"  -I, --iterations n   Number of iterations for -11 optimization",
"  -J, --maxsplits n    Maximum number of split blocks for -11",
#endif
"  -k, --keep           Do not delete original file after processing",
"  -K, --zip            Compress to PKWare zip (.zip) single entry format",
"  -l, --list           List the contents of the compressed input",
"  -L, --license        Display the pigz license and quit",
"  -m, --no-time        Do not store or restore mod time",
"  -M, --time           Store or restore mod time",
"  -n, --no-name        Do not store or restore file name or mod time",
"  -N, --name           Store or restore file name and mod time",
#ifndef NOZOPFLI
"  -O  --oneblock       Do not split into smaller blocks for -11",
#endif
#ifndef NOTHREAD
"  -p, --processes n    Allow up to n compression threads (default is the",
"                       number of online processors, or 8 if unknown)",
#endif
"  -q, --quiet          Print no messages, even on error",
"  -r, --recursive      Process the contents of all subdirectories",
"  -R, --rsyncable      Input-determined block locations for rsync",
"  -S, --suffix .sss    Use suffix .sss instead of .gz (for compression)",
"  -t, --test           Test the integrity of the compressed input",
#ifdef PIGZ_DEBUG
"  -v, --verbose        Provide more verbose output (-vv to debug)",
#else
"  -v, --verbose        Provide more verbose output",
#endif
"  -V  --version        Show the version of pigz",
"  -Y  --synchronous    Force output file write to permanent storage",
"  -z, --zlib           Compress to zlib (.zz) instead of gzip format",
"  --                   All arguments after \"--\" are treated as files"
};

// Display the help text above.
local void help(void) {
    int n;

    if (g.verbosity == 0)
        return;
    for (n = 0; n < (int)(sizeof(helptext) / sizeof(char *)); n++)
        fprintf(stderr, "%s\n", helptext[n]);
    fflush(stderr);
    exit(0);
}

#ifndef NOTHREAD

// Try to determine the number of processors.
local int nprocs(int n) {
#  ifdef _SC_NPROCESSORS_ONLN
    n = (int)sysconf(_SC_NPROCESSORS_ONLN);
#  else
#    ifdef _SC_NPROC_ONLN
    n = (int)sysconf(_SC_NPROC_ONLN);
#    else
#      ifdef __hpux
    struct pst_dynamic psd;

    if (pstat_getdynamic(&psd, sizeof(psd), (size_t)1, 0) != -1)
        n = psd.psd_proc_cnt;
#      endif
#    endif
#  endif
    return n;
}

#endif

// Set option defaults.
local void defaults(void) {
    g.level = Z_DEFAULT_COMPRESSION;
#ifndef NOZOPFLI
    // default zopfli options as set by ZopfliInitOptions():
    //  verbose = 0
    //  numiterations = 15
    //  blocksplitting = 1
    //  blocksplittinglast = 0
    //  blocksplittingmax = 15
    ZopfliInitOptions(&g.zopts);
#endif
    g.block = 131072UL;             // 128K
#ifdef NOTHREAD
    g.procs = 1;
#else
    g.procs = nprocs(8);
    g.shift = x2nmodp(g.block, 3);
#endif
    g.rsync = 0;                    // don't do rsync blocking
    g.setdict = 1;                  // initialize dictionary each thread
    g.verbosity = 1;                // normal message level
    g.headis = 3;                   // store name and time (low bits == 11),
                                    // restore neither (next bits == 00),
                                    // where 01 is name and 10 is time
    g.pipeout = 0;                  // don't force output to stdout
    g.sufx = ".gz";                 // compressed file suffix
    g.comment = NULL;               // no comment
    g.decode = 0;                   // compress
    g.list = 0;                     // compress
    g.keep = 0;                     // delete input file once compressed
    g.force = 0;                    // don't overwrite, don't compress links
    g.sync = 0;                     // don't force a flush on output
    g.recurse = 0;                  // don't go into directories
    g.form = 0;                     // use gzip format
}

// Long options conversion to short options.
local char *longopts[][2] = {
    {"LZW", "Z"}, {"lzw", "Z"}, {"alias", "A"}, {"ascii", "a"}, {"best", "9"},
    {"bits", "Z"}, {"blocksize", "b"}, {"decompress", "d"}, {"fast", "1"},
    {"force", "f"}, {"comment", "C"},
#ifndef NOZOPFLI
    {"first", "F"}, {"iterations", "I"}, {"maxsplits", "J"}, {"oneblock", "O"},
#endif
    {"help", "h"}, {"independent", "i"}, {"keep", "k"}, {"license", "L"},
    {"list", "l"}, {"name", "N"}, {"no-name", "n"}, {"no-time", "m"},
    {"processes", "p"}, {"quiet", "q"}, {"recursive", "r"}, {"rsyncable", "R"},
    {"silent", "q"}, {"stdout", "c"}, {"suffix", "S"}, {"synchronous", "Y"},
    {"test", "t"}, {"time", "M"}, {"to-stdout", "c"}, {"uncompress", "d"},
    {"verbose", "v"}, {"version", "V"}, {"zip", "K"}, {"zlib", "z"}};
#define NLOPTS (sizeof(longopts) / (sizeof(char *) << 1))

// Either new buffer size, new compression level, or new number of processes.
// Get rid of old buffers and threads to force the creation of new ones with
// the new settings.
local void new_opts(void) {
    single_compress(1);
#ifndef NOTHREAD
    finish_jobs();
#endif
}

// Verify that arg is only digits, and if so, return the decimal value.
local size_t num(char *arg) {
    char *str = arg;
    size_t val = 0;

    if (*str == 0)
        throw(EINVAL, "internal error: empty parameter");
    do {
        if (*str < '0' || *str > '9' ||
            (val && ((~(size_t)0) - (size_t)(*str - '0')) / val < 10))
            throw(EINVAL, "invalid numeric parameter: %s", arg);
        val = val * 10 + (size_t)(*str - '0');
    } while (*++str);
    return val;
}

// Process an argument, return true if it is an option (not a filename)
local int option(char *arg) {
    static int get = 0;     // if not zero, look for option parameter
    char bad[3] = "-X";     // for error messages (X is replaced)

    // if no argument or dash option, check status of get
    if (get && (arg == NULL || *arg == '-')) {
        bad[1] = "bpSIJAC"[get - 1];
        throw(EINVAL, "missing parameter after %s", bad);
    }
    if (arg == NULL)
        return 1;

    // process long option or short options
    if (*arg == '-') {
        // a single dash will be interpreted as stdin
        if (*++arg == 0)
            return 0;

        // process long option (fall through with equivalent short option)
        if (*arg == '-') {
            int j;

            arg++;
            for (j = NLOPTS - 1; j >= 0; j--)
                if (strcmp(arg, longopts[j][0]) == 0) {
                    arg = longopts[j][1];
                    break;
                }
            if (j < 0)
                throw(EINVAL, "invalid option: %s", arg - 2);
        }

        // process short options (more than one allowed after dash)
        do {
            // if looking for a parameter, don't process more single character
            // options until we have the parameter
            if (get) {
                if (get == 3)
                    throw(EINVAL,
                          "invalid usage: -S must be followed by space");
                if (get == 7)
                    throw(EINVAL,
                          "invalid usage: -C must be followed by space");
                break;      // allow -*nnn to fall to parameter code
            }

            // process next single character option or compression level
            bad[1] = *arg;
            switch (*arg) {
            case '0': case '1': case '2': case '3': case '4':
            case '5': case '6': case '7': case '8': case '9':
                g.level = *arg - '0';
                while (arg[1] >= '0' && arg[1] <= '9') {
                    if (g.level && (INT_MAX - (arg[1] - '0')) / g.level < 10)
                        throw(EINVAL, "only levels 0..9 and 11 are allowed");
                    g.level = g.level * 10 + *++arg - '0';
                }
                if (g.level == 10 || g.level > 11)
                    throw(EINVAL, "only levels 0..9 and 11 are allowed");
                break;
            case 'A':  get = 6;  break;
            case 'C':  get = 7;  break;
#ifndef NOZOPFLI
            case 'F':  g.zopts.blocksplittinglast = 1;  break;
            case 'I':  get = 4;  break;
            case 'J':  get = 5;  break;
#endif
            case 'K':  g.form = 2;  g.sufx = ".zip";  break;
            case 'L':
                puts(VERSION);
                puts("Copyright (C) 2007-2021 Mark Adler");
                puts("Subject to the terms of the zlib license.");
                puts("No warranty is provided or implied.");
                exit(0);
                break;          // avoid warning
            case 'M':  g.headis |= 0xa;  break;
            case 'N':  g.headis = 0xf;  break;
#ifndef NOZOPFLI
            case 'O':  g.zopts.blocksplitting = 0;  break;
#endif
            case 'R':  g.rsync = 1;  break;
            case 'S':  get = 3;  break;
                // -T defined below as an alternative for -m
            case 'V':
                puts(VERSION);
                if (g.verbosity > 1)
                    printf("zlib %s\n", zlibVersion());
                exit(0);
                break;          // avoid warning
            case 'Y':  g.sync = 1;  break;
            case 'Z':
                throw(EINVAL, "invalid option: LZW output not supported: %s",
                      bad);
                break;          // avoid warning
            case 'a':
                throw(EINVAL, "invalid option: no ascii conversion: %s",
                      bad);
                break;          // avoid warning
            case 'b':  get = 1;  break;
            case 'c':  g.pipeout = 1;  break;
            case 'd':  if (!g.decode) g.headis >>= 2;  g.decode = 1;  break;
            case 'f':  g.force = 1;  break;
            case 'h':  help();  break;
            case 'i':  g.setdict = 0;  break;
            case 'k':  g.keep = 1;  break;
            case 'l':  g.list = 1;  break;
            case 'n':  g.headis = 0;  break;
            case 'T':
            case 'm':  g.headis &= ~0xa;  break;
            case 'p':  get = 2;  break;
            case 'q':  g.verbosity = 0;  break;
            case 'r':  g.recurse = 1;  break;
            case 't':  g.decode = 2;  break;
            case 'v':  g.verbosity++;  break;
            case 'z':  g.form = 1;  g.sufx = ".zz";  break;
            default:
                throw(EINVAL, "invalid option: %s", bad);
            }
        } while (*++arg);
        if (*arg == 0)
            return 1;
    }

    // process option parameter for -b, -p, -A, -S, -I, or -J
    if (get) {
        size_t n;

        if (get == 1) {
            n = num(arg);
            g.block = n << 10;                  // chunk size
#ifndef NOTHREAD
            g.shift = x2nmodp(g.block, 3);
#endif
            if (g.block < DICT)
                throw(EINVAL, "block size too small (must be >= 32K)");
            if (n != g.block >> 10 ||
                OUTPOOL(g.block) < g.block ||
                (ssize_t)OUTPOOL(g.block) < 0 ||
                g.block > (1UL << 29))          // limited by append_len()
                throw(EINVAL, "block size too large: %s", arg);
        }
        else if (get == 2) {
            n = num(arg);
            g.procs = (int)n;                   // # processes
            if (g.procs < 1)
                throw(EINVAL, "invalid number of processes: %s", arg);
            if ((size_t)g.procs != n || INBUFS(g.procs) < 1)
                throw(EINVAL, "too many processes: %s", arg);
#ifdef NOTHREAD
            if (g.procs > 1)
                throw(EINVAL, "compiled without threads");
#endif
        }
        else if (get == 3) {
            if (*arg == 0)
                throw(EINVAL, "suffix cannot be empty");
            g.sufx = arg;                       // gz suffix
        }
#ifndef NOZOPFLI
        else if (get == 4)
            g.zopts.numiterations = (int)num(arg);  // optimize iterations
        else if (get == 5)
            g.zopts.blocksplittingmax = (int)num(arg);  // max block splits
        else if (get == 6)
            g.alias = arg;                      // zip name for stdin
#endif
        else if (get == 7)
            g.comment = arg;                    // header comment
        get = 0;
        return 1;
    }

    // neither an option nor parameter
    return 0;
}

#ifndef NOTHREAD
// handle error received from yarn function
local void cut_yarn(int err) {
    throw(err, "internal threads error");
}
#endif

// Process command line arguments.
int main(int argc, char **argv) {
    int n;                          // general index
    int nop;                        // index before which "-" means stdin
    int done;                       // number of named files processed
    size_t k;                       // program name length
    char *opts, *p;                 // environment default options, marker
    ball_t err;                     // error information from throw()

    g.ret = 0;
    try {
        // initialize globals
        g.inf = NULL;
        g.inz = 0;
#ifndef NOTHREAD
        g.in_which = -1;
#endif
        g.alias = "-";
        g.outf = NULL;
        g.first = 1;
        g.hname = NULL;
        g.hcomm = NULL;

        // save pointer to program name for error messages
        p = strrchr(argv[0], '/');
        p = p == NULL ? argv[0] : p + 1;
        g.prog = *p ? p : "pigz";

        // prepare for interrupts and logging
        signal(SIGINT, cut_short);
#ifndef NOTHREAD
        yarn_prefix = g.prog;           // prefix for yarn error messages
        yarn_abort = cut_yarn;          // call on thread error
#endif
#ifdef PIGZ_DEBUG
        gettimeofday(&start, NULL);     // starting time for log entries
        log_init();                     // initialize logging
#endif

        // set all options to defaults
        defaults();

        // check zlib version
        if (zlib_vernum() < 0x1230)
           throw(EINVAL, "zlib version less than 1.2.3");

        // create CRC table, in case zlib compiled with dynamic tables
        get_crc_table();

        // process user environment variable defaults in GZIP
        opts = getenv("GZIP");
        if (opts != NULL) {
            while (*opts) {
                while (*opts == ' ' || *opts == '\t')
                    opts++;
                p = opts;
                while (*p && *p != ' ' && *p != '\t')
                    p++;
                n = *p;
                *p = 0;
                if (!option(opts))
                    throw(EINVAL, "cannot provide files in "
                                  "GZIP environment variable");
                opts = p + (n ? 1 : 0);
            }
            option(NULL);           // check for missing parameter
        }

        // process user environment variable defaults in PIGZ as well
        opts = getenv("PIGZ");
        if (opts != NULL) {
            while (*opts) {
                while (*opts == ' ' || *opts == '\t')
                    opts++;
                p = opts;
                while (*p && *p != ' ' && *p != '\t')
                    p++;
                n = *p;
                *p = 0;
                if (!option(opts))
                    throw(EINVAL, "cannot provide files in "
                                  "PIGZ environment variable");
                opts = p + (n ? 1 : 0);
            }
            option(NULL);           // check for missing parameter
        }

        // decompress if named "unpigz" or "gunzip", to stdout if "*cat"
        if (strcmp(g.prog, "unpigz") == 0 || strcmp(g.prog, "gunzip") == 0) {
            if (!g.decode)
                g.headis >>= 2;
            g.decode = 1;
        }
        if ((k = strlen(g.prog)) > 2 && strcmp(g.prog + k - 3, "cat") == 0) {
            if (!g.decode)
                g.headis >>= 2;
            g.decode = 1;
            g.pipeout = 1;
        }

        // if no arguments and compressed data to/from terminal, show help
        if (argc < 2 && isatty(g.decode ? 0 : 1))
            help();

        // process all command-line options first
        nop = argc;
        for (n = 1; n < argc; n++)
            if (strcmp(argv[n], "--") == 0) {
                nop = n;                // after this, "-" is the name "-"
                argv[n] = NULL;         // remove option
                break;                  // ignore options after "--"
            }
            else if (option(argv[n]))   // process argument
                argv[n] = NULL;         // remove if option
        option(NULL);                   // check for missing parameter

        // process command-line filenames
        done = 0;
        for (n = 1; n < argc; n++)
            if (argv[n] != NULL) {
                if (done == 1 && g.pipeout && !g.decode && !g.list &&
                    g.form > 1)
                    complain("warning: output will be concatenated zip files"
                             " -- %s will not be able to extract", g.prog);
                process(n < nop && strcmp(argv[n], "-") == 0 ? NULL : argv[n]);
                done++;
            }

        // list stdin or compress stdin to stdout if no file names provided
        if (done == 0)
            process(NULL);
    }
    always {
        // release resources
        RELEASE(g.inf);
        g.inz = 0;
        new_opts();
    }
    catch (err) {
        THREADABORT(err);
    }

    // show log (if any)
    log_dump();
    return g.ret;
}