sort: -R now uses less memory on long lines with internal NULs

[coreutils.git] / src / sort.c

/* sort - sort lines of text (with all kinds of options).
   Copyright (C) 1988, 1991-2010 Free Software Foundation, Inc.

   This program is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation, either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <http://www.gnu.org/licenses/>.

   Written December 1988 by Mike Haertel.
   The author may be reached (Email) at the address mike@gnu.ai.mit.edu,
   or (US mail) as Mike Haertel c/o Free Software Foundation.

   Ørn E. Hansen added NLS support in 1997.  */

#include <config.h>

#include <getopt.h>
#include <pthread.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <signal.h>
#include "system.h"
#include "argmatch.h"
#include "error.h"
#include "fadvise.h"
#include "filevercmp.h"
#include "hard-locale.h"
#include "hash.h"
#include "heap.h"
#include "md5.h"
#include "mbswidth.h"
#include "nproc.h"
#include "physmem.h"
#include "posixver.h"
#include "quote.h"
#include "quotearg.h"
#include "randread.h"
#include "readtokens0.h"
#include "stdio--.h"
#include "stdlib--.h"
#include "strnumcmp.h"
#include "xmemcoll.h"
#include "xnanosleep.h"
#include "xstrtol.h"

#if HAVE_SYS_RESOURCE_H
# include <sys/resource.h>
#endif
#ifndef RLIMIT_DATA
struct rlimit { size_t rlim_cur; };
# define getrlimit(Resource, Rlp) (-1)
#endif

/* The official name of this program (e.g., no `g' prefix).  */
#define PROGRAM_NAME "sort"

#define AUTHORS \
  proper_name ("Mike Haertel"), \
  proper_name ("Paul Eggert")

#if HAVE_LANGINFO_CODESET
# include <langinfo.h>
#endif

/* Use SA_NOCLDSTOP as a proxy for whether the sigaction machinery is
   present.  */
#ifndef SA_NOCLDSTOP
# define SA_NOCLDSTOP 0
/* No sigprocmask.  Always 'return' zero. */
# define sigprocmask(How, Set, Oset) (0)
# define sigset_t int
# if ! HAVE_SIGINTERRUPT
#  define siginterrupt(sig, flag) /* empty */
# endif
#endif

#if !defined OPEN_MAX && defined NR_OPEN
# define OPEN_MAX NR_OPEN
#endif
#if !defined OPEN_MAX
# define OPEN_MAX 20
#endif

#define UCHAR_LIM (UCHAR_MAX + 1)

#ifndef DEFAULT_TMPDIR
# define DEFAULT_TMPDIR "/tmp"
#endif

/* Maximum number of lines to merge every time a NODE is taken from
   the MERGE_QUEUE.  Node is at LEVEL in the binary merge tree,
   and is responsible for merging TOTAL lines. */
#define MAX_MERGE(total, level) ((total) / ((2 << level) * (2 << level)) + 1)

/* Heuristic value for the number of lines for which it is worth
   creating a subthread, during an internal merge sort, on a machine
   that has processors galore.  Currently this number is just a guess.
   This value must be at least 4.  We don't know of any machine where
   this number has any practical effect.  */
enum { SUBTHREAD_LINES_HEURISTIC = 4 };

/* Exit statuses.  */
enum
  {
    /* POSIX says to exit with status 1 if invoked with -c and the
       input is not properly sorted.  */
    SORT_OUT_OF_ORDER = 1,

    /* POSIX says any other irregular exit must exit with a status
       code greater than 1.  */
    SORT_FAILURE = 2
  };

enum
  {
    /* The number of times we should try to fork a compression process
       (we retry if the fork call fails).  We don't _need_ to compress
       temp files, this is just to reduce disk access, so this number
       can be small.  Each retry doubles in duration.  */
    MAX_FORK_TRIES_COMPRESS = 4,

    /* The number of times we should try to fork a decompression process.
       If we can't fork a decompression process, we can't sort, so this
       number should be big.  Each retry doubles in duration.  */
    MAX_FORK_TRIES_DECOMPRESS = 9
  };

enum
  {
    /* Level of the end-of-merge node, one level above the root. */
    MERGE_END = 0,

    /* Level of the root node in merge tree. */
    MERGE_ROOT = 1
  };

/* The representation of the decimal point in the current locale.  */
static int decimal_point;

/* Thousands separator; if -1, then there isn't one.  */
static int thousands_sep;

/* Nonzero if the corresponding locales are hard.  */
static bool hard_LC_COLLATE;
#if HAVE_NL_LANGINFO
static bool hard_LC_TIME;
#endif

#define NONZERO(x) ((x) != 0)

/* The kind of blanks for '-b' to skip in various options. */
enum blanktype { bl_start, bl_end, bl_both };

/* The character marking end of line. Default to \n. */
static char eolchar = '\n';

/* Lines are held in core as counted strings. */
struct line
{
  char *text;                   /* Text of the line. */
  size_t length;                /* Length including final newline. */
  char *keybeg;                 /* Start of first key. */
  char *keylim;                 /* Limit of first key. */
};

/* Input buffers. */
struct buffer
{
  char *buf;                    /* Dynamically allocated buffer,
                                   partitioned into 3 regions:
                                   - input data;
                                   - unused area;
                                   - an array of lines, in reverse order.  */
  size_t used;                  /* Number of bytes used for input data.  */
  size_t nlines;                /* Number of lines in the line array.  */
  size_t alloc;                 /* Number of bytes allocated. */
  size_t left;                  /* Number of bytes left from previous reads. */
  size_t line_bytes;            /* Number of bytes to reserve for each line. */
  bool eof;                     /* An EOF has been read.  */
};

struct keyfield
{
  size_t sword;                 /* Zero-origin 'word' to start at. */
  size_t schar;                 /* Additional characters to skip. */
  size_t eword;                 /* Zero-origin last 'word' of key. */
  size_t echar;                 /* Additional characters in field. */
  bool const *ignore;           /* Boolean array of characters to ignore. */
  char const *translate;        /* Translation applied to characters. */
  bool skipsblanks;             /* Skip leading blanks when finding start.  */
  bool skipeblanks;             /* Skip leading blanks when finding end.  */
  bool numeric;                 /* Flag for numeric comparison.  Handle
                                   strings of digits with optional decimal
                                   point, but no exponential notation. */
  bool random;                  /* Sort by random hash of key.  */
  bool general_numeric;         /* Flag for general, numeric comparison.
                                   Handle numbers in exponential notation. */
  bool human_numeric;           /* Flag for sorting by human readable
                                   units with either SI xor IEC prefixes. */
  bool month;                   /* Flag for comparison by month name. */
  bool reverse;                 /* Reverse the sense of comparison. */
  bool version;                 /* sort by version number */
  bool obsolete_used;           /* obsolescent key option format is used. */
  struct keyfield *next;        /* Next keyfield to try. */
};

struct month
{
  char const *name;
  int val;
};

/* Binary merge tree node. */
struct merge_node
{
  struct line *lo;              /* Lines to merge from LO child node. */
  struct line *hi;              /* Lines to merge from HI child ndoe. */
  struct line *end_lo;          /* End of available lines from LO. */
  struct line *end_hi;          /* End of available lines from HI. */
  struct line **dest;           /* Pointer to destination of merge. */
  size_t nlo;                   /* Total Lines remaining from LO. */
  size_t nhi;                   /* Total lines remaining from HI. */
  size_t level;                 /* Level in merge tree. */
  struct merge_node *parent;    /* Parent node. */
  bool queued;                  /* Node is already in heap. */
  pthread_spinlock_t *lock;     /* Lock for node operations. */
};

/* Priority queue of merge nodes. */
struct merge_node_queue
{
  struct heap *priority_queue;  /* Priority queue of merge tree nodes. */
  pthread_mutex_t mutex;        /* Lock for queue operations. */
  pthread_cond_t cond;          /* Conditional wait for empty queue to populate
                                   when popping. */
};

/* FIXME: None of these tables work with multibyte character sets.
   Also, there are many other bugs when handling multibyte characters.
   One way to fix this is to rewrite `sort' to use wide characters
   internally, but doing this with good performance is a bit
   tricky.  */

/* Table of blanks.  */
static bool blanks[UCHAR_LIM];

/* Table of non-printing characters. */
static bool nonprinting[UCHAR_LIM];

/* Table of non-dictionary characters (not letters, digits, or blanks). */
static bool nondictionary[UCHAR_LIM];

/* Translation table folding lower case to upper.  */
static unsigned char fold_toupper[UCHAR_LIM];

#define MONTHS_PER_YEAR 12

/* Table mapping month names to integers.
   Alphabetic order allows binary search. */
static struct month monthtab[] =
{
  {"APR", 4},
  {"AUG", 8},
  {"DEC", 12},
  {"FEB", 2},
  {"JAN", 1},
  {"JUL", 7},
  {"JUN", 6},
  {"MAR", 3},
  {"MAY", 5},
  {"NOV", 11},
  {"OCT", 10},
  {"SEP", 9}
};

/* During the merge phase, the number of files to merge at once. */
#define NMERGE_DEFAULT 16

/* Minimum size for a merge or check buffer.  */
#define MIN_MERGE_BUFFER_SIZE (2 + sizeof (struct line))

/* Minimum sort size; the code might not work with smaller sizes.  */
#define MIN_SORT_SIZE (nmerge * MIN_MERGE_BUFFER_SIZE)

/* The number of bytes needed for a merge or check buffer, which can
   function relatively efficiently even if it holds only one line.  If
   a longer line is seen, this value is increased.  */
static size_t merge_buffer_size = MAX (MIN_MERGE_BUFFER_SIZE, 256 * 1024);

/* The approximate maximum number of bytes of main memory to use, as
   specified by the user.  Zero if the user has not specified a size.  */
static size_t sort_size;

/* The guessed size for non-regular files.  */
#define INPUT_FILE_SIZE_GUESS (1024 * 1024)

/* Array of directory names in which any temporary files are to be created. */
static char const **temp_dirs;

/* Number of temporary directory names used.  */
static size_t temp_dir_count;

/* Number of allocated slots in temp_dirs.  */
static size_t temp_dir_alloc;

/* Flag to reverse the order of all comparisons. */
static bool reverse;

/* Flag for stable sort.  This turns off the last ditch bytewise
   comparison of lines, and instead leaves lines in the same order
   they were read if all keys compare equal.  */
static bool stable;

/* If TAB has this value, blanks separate fields.  */
enum { TAB_DEFAULT = CHAR_MAX + 1 };

/* Tab character separating fields.  If TAB_DEFAULT, then fields are
   separated by the empty string between a non-blank character and a blank
   character. */
static int tab = TAB_DEFAULT;

/* Flag to remove consecutive duplicate lines from the output.
   Only the last of a sequence of equal lines will be output. */
static bool unique;

/* Nonzero if any of the input files are the standard input. */
static bool have_read_stdin;

/* List of key field comparisons to be tried.  */
static struct keyfield *keylist;

/* Program used to (de)compress temp files.  Must accept -d.  */
static char const *compress_program;

/* Annotate the output with extra info to aid the user.  */
static bool debug;

/* Maximum number of files to merge in one go.  If more than this
   number are present, temp files will be used. */
static unsigned int nmerge = NMERGE_DEFAULT;

/* Report MESSAGE for FILE, then clean up and exit.
   If FILE is null, it represents standard output.  */

static void die (char const *, char const *) ATTRIBUTE_NORETURN;
static void
die (char const *message, char const *file)
{
  error (0, errno, "%s: %s", message, file ? file : _("standard output"));
  exit (SORT_FAILURE);
}

void
usage (int status)
{
  if (status != EXIT_SUCCESS)
    fprintf (stderr, _("Try `%s --help' for more information.\n"),
             program_name);
  else
    {
      printf (_("\
Usage: %s [OPTION]... [FILE]...\n\
  or:  %s [OPTION]... --files0-from=F\n\
"),
              program_name, program_name);
      fputs (_("\
Write sorted concatenation of all FILE(s) to standard output.\n\
\n\
"), stdout);
      fputs (_("\
Mandatory arguments to long options are mandatory for short options too.\n\
"), stdout);
      fputs (_("\
Ordering options:\n\
\n\
"), stdout);
      fputs (_("\
  -b, --ignore-leading-blanks  ignore leading blanks\n\
  -d, --dictionary-order      consider only blanks and alphanumeric characters\n\
  -f, --ignore-case           fold lower case to upper case characters\n\
"), stdout);
      fputs (_("\
  -g, --general-numeric-sort  compare according to general numerical value\n\
  -i, --ignore-nonprinting    consider only printable characters\n\
  -M, --month-sort            compare (unknown) < `JAN' < ... < `DEC'\n\
"), stdout);
      fputs (_("\
  -h, --human-numeric-sort    compare human readable numbers (e.g., 2K 1G)\n\
"), stdout);
      fputs (_("\
  -n, --numeric-sort          compare according to string numerical value\n\
  -R, --random-sort           sort by random hash of keys\n\
      --random-source=FILE    get random bytes from FILE\n\
  -r, --reverse               reverse the result of comparisons\n\
"), stdout);
      fputs (_("\
      --sort=WORD             sort according to WORD:\n\
                                general-numeric -g, human-numeric -h, month -M,\n\
                                numeric -n, random -R, version -V\n\
  -V, --version-sort          natural sort of (version) numbers within text\n\
\n\
"), stdout);
      fputs (_("\
Other options:\n\
\n\
"), stdout);
      fputs (_("\
      --batch-size=NMERGE   merge at most NMERGE inputs at once;\n\
                            for more use temp files\n\
"), stdout);
      fputs (_("\
  -c, --check, --check=diagnose-first  check for sorted input; do not sort\n\
  -C, --check=quiet, --check=silent  like -c, but do not report first bad line\n\
      --compress-program=PROG  compress temporaries with PROG;\n\
                              decompress them with PROG -d\n\
"), stdout);
      fputs (_("\
      --debug               annotate the part of the line used to sort,\n\
                              and warn about questionable usage to stderr\n\
      --files0-from=F       read input from the files specified by\n\
                            NUL-terminated names in file F;\n\
                            If F is - then read names from standard input\n\
"), stdout);
      fputs (_("\
  -k, --key=POS1[,POS2]     start a key at POS1 (origin 1), end it at POS2\n\
                            (default end of line).  See POS syntax below\n\
  -m, --merge               merge already sorted files; do not sort\n\
"), stdout);
      fputs (_("\
  -o, --output=FILE         write result to FILE instead of standard output\n\
  -s, --stable              stabilize sort by disabling last-resort comparison\n\
  -S, --buffer-size=SIZE    use SIZE for main memory buffer\n\
"), stdout);
      printf (_("\
  -t, --field-separator=SEP  use SEP instead of non-blank to blank transition\n\
  -T, --temporary-directory=DIR  use DIR for temporaries, not $TMPDIR or %s;\n\
                              multiple options specify multiple directories\n\
      --parallel=N          limit the number of sorts run concurrently to N\n\
  -u, --unique              with -c, check for strict ordering;\n\
                              without -c, output only the first of an equal run\n\
"), DEFAULT_TMPDIR);
      fputs (_("\
  -z, --zero-terminated     end lines with 0 byte, not newline\n\
"), stdout);
      fputs (HELP_OPTION_DESCRIPTION, stdout);
      fputs (VERSION_OPTION_DESCRIPTION, stdout);
      fputs (_("\
\n\
POS is F[.C][OPTS], where F is the field number and C the character position\n\
in the field; both are origin 1.  If neither -t nor -b is in effect, characters\n\
in a field are counted from the beginning of the preceding whitespace.  OPTS is\n\
one or more single-letter ordering options, which override global ordering\n\
options for that key.  If no key is given, use the entire line as the key.\n\
\n\
SIZE may be followed by the following multiplicative suffixes:\n\
"), stdout);
      fputs (_("\
% 1% of memory, b 1, K 1024 (default), and so on for M, G, T, P, E, Z, Y.\n\
\n\
With no FILE, or when FILE is -, read standard input.\n\
\n\
*** WARNING ***\n\
The locale specified by the environment affects sort order.\n\
Set LC_ALL=C to get the traditional sort order that uses\n\
native byte values.\n\
"), stdout );
      emit_ancillary_info ();
    }

  exit (status);
}

/* For long options that have no equivalent short option, use a
   non-character as a pseudo short option, starting with CHAR_MAX + 1.  */
enum
{
  CHECK_OPTION = CHAR_MAX + 1,
  COMPRESS_PROGRAM_OPTION,
  DEBUG_PROGRAM_OPTION,
  FILES0_FROM_OPTION,
  NMERGE_OPTION,
  RANDOM_SOURCE_OPTION,
  SORT_OPTION,
  PARALLEL_OPTION
};

static char const short_options[] = "-bcCdfghik:mMno:rRsS:t:T:uVy:z";

static struct option const long_options[] =
{
  {"ignore-leading-blanks", no_argument, NULL, 'b'},
  {"check", optional_argument, NULL, CHECK_OPTION},
  {"compress-program", required_argument, NULL, COMPRESS_PROGRAM_OPTION},
  {"debug", no_argument, NULL, DEBUG_PROGRAM_OPTION},
  {"dictionary-order", no_argument, NULL, 'd'},
  {"ignore-case", no_argument, NULL, 'f'},
  {"files0-from", required_argument, NULL, FILES0_FROM_OPTION},
  {"general-numeric-sort", no_argument, NULL, 'g'},
  {"ignore-nonprinting", no_argument, NULL, 'i'},
  {"key", required_argument, NULL, 'k'},
  {"merge", no_argument, NULL, 'm'},
  {"month-sort", no_argument, NULL, 'M'},
  {"numeric-sort", no_argument, NULL, 'n'},
  {"human-numeric-sort", no_argument, NULL, 'h'},
  {"version-sort", no_argument, NULL, 'V'},
  {"random-sort", no_argument, NULL, 'R'},
  {"random-source", required_argument, NULL, RANDOM_SOURCE_OPTION},
  {"sort", required_argument, NULL, SORT_OPTION},
  {"output", required_argument, NULL, 'o'},
  {"reverse", no_argument, NULL, 'r'},
  {"stable", no_argument, NULL, 's'},
  {"batch-size", required_argument, NULL, NMERGE_OPTION},
  {"buffer-size", required_argument, NULL, 'S'},
  {"field-separator", required_argument, NULL, 't'},
  {"temporary-directory", required_argument, NULL, 'T'},
  {"unique", no_argument, NULL, 'u'},
  {"zero-terminated", no_argument, NULL, 'z'},
  {"parallel", required_argument, NULL, PARALLEL_OPTION},
  {GETOPT_HELP_OPTION_DECL},
  {GETOPT_VERSION_OPTION_DECL},
  {NULL, 0, NULL, 0},
};

#define CHECK_TABLE \
  _ct_("quiet",          'C') \
  _ct_("silent",         'C') \
  _ct_("diagnose-first", 'c')

static char const *const check_args[] =
{
#define _ct_(_s, _c) _s,
  CHECK_TABLE NULL
#undef  _ct_
};
static char const check_types[] =
{
#define _ct_(_s, _c) _c,
  CHECK_TABLE
#undef  _ct_
};

#define SORT_TABLE \
  _st_("general-numeric", 'g') \
  _st_("human-numeric",   'h') \
  _st_("month",           'M') \
  _st_("numeric",         'n') \
  _st_("random",          'R') \
  _st_("version",         'V')

static char const *const sort_args[] =
{
#define _st_(_s, _c) _s,
  SORT_TABLE NULL
#undef  _st_
};
static char const sort_types[] =
{
#define _st_(_s, _c) _c,
  SORT_TABLE
#undef  _st_
};

/* The set of signals that are caught.  */
static sigset_t caught_signals;

/* Critical section status.  */
struct cs_status
{
  bool valid;
  sigset_t sigs;
};

/* Enter a critical section.  */
static struct cs_status
cs_enter (void)
{
  struct cs_status status;
  status.valid = (sigprocmask (SIG_BLOCK, &caught_signals, &status.sigs) == 0);
  return status;
}

/* Leave a critical section.  */
static void
cs_leave (struct cs_status status)
{
  if (status.valid)
    {
      /* Ignore failure when restoring the signal mask. */
      sigprocmask (SIG_SETMASK, &status.sigs, NULL);
    }
}

/* The list of temporary files. */
struct tempnode
{
  struct tempnode *volatile next;
  pid_t pid;     /* If compressed, the pid of compressor, else zero */
  char name[1];  /* Actual size is 1 + file name length.  */
};
static struct tempnode *volatile temphead;
static struct tempnode *volatile *temptail = &temphead;

struct sortfile
{
  char const *name;
  pid_t pid;     /* If compressed, the pid of compressor, else zero */
};

/* A table where we store compression process states.  We clean up all
   processes in a timely manner so as not to exhaust system resources,
   so we store the info on whether the process is still running, or has
   been reaped here.  */
static Hash_table *proctab;

enum { INIT_PROCTAB_SIZE = 47 };

enum procstate { ALIVE, ZOMBIE };

/* A proctab entry.  The COUNT field is there in case we fork a new
   compression process that has the same PID as an old zombie process
   that is still in the table (because the process to decompress the
   temp file it was associated with hasn't started yet).  */
struct procnode
{
  pid_t pid;
  enum procstate state;
  size_t count;
};

static size_t
proctab_hasher (void const *entry, size_t tabsize)
{
  struct procnode const *node = entry;
  return node->pid % tabsize;
}

static bool
proctab_comparator (void const *e1, void const *e2)
{
  struct procnode const *n1 = e1, *n2 = e2;
  return n1->pid == n2->pid;
}

/* The total number of forked processes (compressors and decompressors)
   that have not been reaped yet. */
static size_t nprocs;

/* The number of child processes we'll allow before we try to reap some. */
enum { MAX_PROCS_BEFORE_REAP = 2 };

/* If 0 < PID, wait for the child process with that PID to exit.
   If PID is -1, clean up a random child process which has finished and
   return the process ID of that child.  If PID is -1 and no processes
   have quit yet, return 0 without waiting.  */

static pid_t
reap (pid_t pid)
{
  int status;
  pid_t cpid = waitpid (pid, &status, pid < 0 ? WNOHANG : 0);

  if (cpid < 0)
    error (SORT_FAILURE, errno, _("waiting for %s [-d]"),
           compress_program);
  else if (0 < cpid)
    {
      if (! WIFEXITED (status) || WEXITSTATUS (status))
        error (SORT_FAILURE, 0, _("%s [-d] terminated abnormally"),
               compress_program);
      --nprocs;
    }

  return cpid;
}

/* Add the PID of a running compression process to proctab, or update
   the entry COUNT and STATE fields if it's already there.  This also
   creates the table for us the first time it's called.  */

static void
register_proc (pid_t pid)
{
  struct procnode test, *node;

  if (! proctab)
    {
      proctab = hash_initialize (INIT_PROCTAB_SIZE, NULL,
                                 proctab_hasher,
                                 proctab_comparator,
                                 free);
      if (! proctab)
        xalloc_die ();
    }

  test.pid = pid;
  node = hash_lookup (proctab, &test);
  if (node)
    {
      node->state = ALIVE;
      ++node->count;
    }
  else
    {
      node = xmalloc (sizeof *node);
      node->pid = pid;
      node->state = ALIVE;
      node->count = 1;
      if (hash_insert (proctab, node) == NULL)
        xalloc_die ();
    }
}

/* This is called when we reap a random process.  We don't know
   whether we have reaped a compression process or a decompression
   process until we look in the table.  If there's an ALIVE entry for
   it, then we have reaped a compression process, so change the state
   to ZOMBIE.  Otherwise, it's a decompression processes, so ignore it.  */

static void
update_proc (pid_t pid)
{
  struct procnode test, *node;

  test.pid = pid;
  node = hash_lookup (proctab, &test);
  if (node)
    node->state = ZOMBIE;
}

/* This is for when we need to wait for a compression process to exit.
   If it has a ZOMBIE entry in the table then it's already dead and has
   been reaped.  Note that if there's an ALIVE entry for it, it still may
   already have died and been reaped if a second process was created with
   the same PID.  This is probably exceedingly rare, but to be on the safe
   side we will have to wait for any compression process with this PID.  */

static void
wait_proc (pid_t pid)
{
  struct procnode test, *node;

  test.pid = pid;
  node = hash_lookup (proctab, &test);
  if (node->state == ALIVE)
    reap (pid);

  node->state = ZOMBIE;
  if (! --node->count)
    {
      hash_delete (proctab, node);
      free (node);
    }
}

/* Keep reaping finished children as long as there are more to reap.
   This doesn't block waiting for any of them, it only reaps those
   that are already dead.  */

static void
reap_some (void)
{
  pid_t pid;

  while (0 < nprocs && (pid = reap (-1)))
    update_proc (pid);
}

/* Clean up any remaining temporary files.  */

static void
cleanup (void)
{
  struct tempnode const *node;

  for (node = temphead; node; node = node->next)
    unlink (node->name);
  temphead = NULL;
}

/* Cleanup actions to take when exiting.  */

static void
exit_cleanup (void)
{
  if (temphead)
    {
      /* Clean up any remaining temporary files in a critical section so
         that a signal handler does not try to clean them too.  */
      struct cs_status cs = cs_enter ();
      cleanup ();
      cs_leave (cs);
    }

  close_stdout ();
}

/* Create a new temporary file, returning its newly allocated tempnode.
   Store into *PFD the file descriptor open for writing.
   If the creation fails, return NULL and store -1 into *PFD if the
   failure is due to file descriptor exhaustion and
   SURVIVE_FD_EXHAUSTION; otherwise, die.  */

static struct tempnode *
create_temp_file (int *pfd, bool survive_fd_exhaustion)
{
  static char const slashbase[] = "/sortXXXXXX";
  static size_t temp_dir_index;
  int fd;
  int saved_errno;
  char const *temp_dir = temp_dirs[temp_dir_index];
  size_t len = strlen (temp_dir);
  struct tempnode *node =
    xmalloc (offsetof (struct tempnode, name) + len + sizeof slashbase);
  char *file = node->name;
  struct cs_status cs;

  memcpy (file, temp_dir, len);
  memcpy (file + len, slashbase, sizeof slashbase);
  node->next = NULL;
  node->pid = 0;
  if (++temp_dir_index == temp_dir_count)
    temp_dir_index = 0;

  /* Create the temporary file in a critical section, to avoid races.  */
  cs = cs_enter ();
  fd = mkstemp (file);
  if (0 <= fd)
    {
      *temptail = node;
      temptail = &node->next;
    }
  saved_errno = errno;
  cs_leave (cs);
  errno = saved_errno;

  if (fd < 0)
    {
      if (! (survive_fd_exhaustion && errno == EMFILE))
        error (SORT_FAILURE, errno, _("cannot create temporary file in %s"),
               quote (temp_dir));
      free (node);
      node = NULL;
    }

  *pfd = fd;
  return node;
}

/* Return a stream for FILE, opened with mode HOW.  A null FILE means
   standard output; HOW should be "w".  When opening for input, "-"
   means standard input.  To avoid confusion, do not return file
   descriptors STDIN_FILENO, STDOUT_FILENO, or STDERR_FILENO when
   opening an ordinary FILE.  Return NULL if unsuccessful.

   fadvise() is used to specify an access pattern for input files.
   There are a few hints we could possibly provide,
   and after careful testing it was decided that
   specifying POSIX_FADV_SEQUENTIAL was not detrimental
   to any cases.  On Linux 2.6.31, this option doubles
   the size of read ahead performed and thus was seen to
   benefit these cases:
     Merging
     Sorting with a smaller internal buffer
     Reading from faster flash devices

   In _addition_ one could also specify other hints...

   POSIX_FADV_WILLNEED was tested, but Linux 2.6.31
   at least uses that to _synchronously_ prepopulate the cache
   with the specified range.  While sort does need to
   read all of its input before outputting, a synchronous
   read of the whole file up front precludes any processing
   that sort could do in parallel with the system doing
   read ahead of the data. This was seen to have negative effects
   in a couple of cases:
     Merging
     Sorting with a smaller internal buffer
   Note this option was seen to shorten the runtime for sort
   on a multicore system with lots of RAM and other processes
   competing for CPU.  It could be argued that more explicit
   scheduling hints with `nice` et. al. are more appropriate
   for this situation.

   POSIX_FADV_NOREUSE is a possibility as it could lower
   the priority of input data in the cache as sort will
   only need to process it once.  However its functionality
   has changed over Linux kernel versions and as of 2.6.31
   it does nothing and thus we can't depend on what it might
   do in future.

   POSIX_FADV_DONTNEED is not appropriate for user specified
   input files, but for temp files we do want to drop the
   cache immediately after processing.  This is done implicitly
   however when the files are unlinked.  */

static FILE *
stream_open (char const *file, char const *how)
{
  if (!file)
    return stdout;
  if (*how == 'r')
    {
      FILE *fp;
      if (STREQ (file, "-"))
        {
          have_read_stdin = true;
          fp = stdin;
        }
      else
        fp = fopen (file, how);
      fadvise (fp, FADVISE_SEQUENTIAL);
      return fp;
    }
  return fopen (file, how);
}

/* Same as stream_open, except always return a non-null value; die on
   failure.  */

static FILE *
xfopen (char const *file, char const *how)
 {
  FILE *fp = stream_open (file, how);
  if (!fp)
    die (_("open failed"), file);
  return fp;
}

/* Close FP, whose name is FILE, and report any errors.  */

static void
xfclose (FILE *fp, char const *file)
{
  switch (fileno (fp))
    {
    case STDIN_FILENO:
      /* Allow reading stdin from tty more than once.  */
      if (feof (fp))
        clearerr (fp);
      break;

    case STDOUT_FILENO:
      /* Don't close stdout just yet.  close_stdout does that.  */
      if (fflush (fp) != 0)
        die (_("fflush failed"), file);
      break;

    default:
      if (fclose (fp) != 0)
        die (_("close failed"), file);
      break;
    }
}

static void
dup2_or_die (int oldfd, int newfd)
{
  if (dup2 (oldfd, newfd) < 0)
    error (SORT_FAILURE, errno, _("dup2 failed"));
}

/* Fork a child process for piping to and do common cleanup.  The
   TRIES parameter tells us how many times to try to fork before
   giving up.  Return the PID of the child, or -1 (setting errno)
   on failure. */

static pid_t
pipe_fork (int pipefds[2], size_t tries)
{
#if HAVE_WORKING_FORK
  struct tempnode *saved_temphead;
  int saved_errno;
  double wait_retry = 0.25;
  pid_t pid IF_LINT ( = -1);
  struct cs_status cs;

  if (pipe (pipefds) < 0)
    return -1;

  while (tries--)
    {
      /* This is so the child process won't delete our temp files
         if it receives a signal before exec-ing.  */
      cs = cs_enter ();
      saved_temphead = temphead;
      temphead = NULL;

      pid = fork ();
      saved_errno = errno;
      if (pid)
        temphead = saved_temphead;

      cs_leave (cs);
      errno = saved_errno;

      if (0 <= pid || errno != EAGAIN)
        break;
      else
        {
          xnanosleep (wait_retry);
          wait_retry *= 2;
          reap_some ();
        }
    }

  if (pid < 0)
    {
      saved_errno = errno;
      close (pipefds[0]);
      close (pipefds[1]);
      errno = saved_errno;
    }
  else if (pid == 0)
    {
      close (STDIN_FILENO);
      close (STDOUT_FILENO);
    }
  else
    ++nprocs;

  return pid;

#else  /* ! HAVE_WORKING_FORK */
  return -1;
#endif
}

/* Create a temporary file and start a compression program to filter output
   to that file.  Set *PFP to the file handle and if PPID is non-NULL,
   set *PPID to the PID of the newly-created process.  If the creation
   fails, return NULL if the failure is due to file descriptor
   exhaustion and SURVIVE_FD_EXHAUSTION; otherwise, die.  */

static char *
maybe_create_temp (FILE **pfp, pid_t *ppid, bool survive_fd_exhaustion)
{
  int tempfd;
  struct tempnode *node = create_temp_file (&tempfd, survive_fd_exhaustion);
  char *name;

  if (! node)
    return NULL;

  name = node->name;

  if (compress_program)
    {
      int pipefds[2];

      node->pid = pipe_fork (pipefds, MAX_FORK_TRIES_COMPRESS);
      if (0 < node->pid)
        {
          close (tempfd);
          close (pipefds[0]);
          tempfd = pipefds[1];

          register_proc (node->pid);
        }
      else if (node->pid == 0)
        {
          close (pipefds[1]);
          dup2_or_die (tempfd, STDOUT_FILENO);
          close (tempfd);
          dup2_or_die (pipefds[0], STDIN_FILENO);
          close (pipefds[0]);

          if (execlp (compress_program, compress_program, (char *) NULL) < 0)
            error (SORT_FAILURE, errno, _("couldn't execute %s"),
                   compress_program);
        }
      else
        node->pid = 0;
    }

  *pfp = fdopen (tempfd, "w");
  if (! *pfp)
    die (_("couldn't create temporary file"), name);

  if (ppid)
    *ppid = node->pid;

  return name;
}

/* Create a temporary file and start a compression program to filter output
   to that file.  Set *PFP to the file handle and if *PPID is non-NULL,
   set it to the PID of the newly-created process.  Die on failure.  */

static char *
create_temp (FILE **pfp, pid_t *ppid)
{
  return maybe_create_temp (pfp, ppid, false);
}

/* Open a compressed temp file and start a decompression process through
   which to filter the input.  PID must be the valid processes ID of the
   process used to compress the file.  Return NULL (setting errno to
   EMFILE) if we ran out of file descriptors, and die on any other
   kind of failure.  */

static FILE *
open_temp (char const *name, pid_t pid)
{
  int tempfd, pipefds[2];
  FILE *fp = NULL;

  wait_proc (pid);

  tempfd = open (name, O_RDONLY);
  if (tempfd < 0)
    return NULL;

  switch (pipe_fork (pipefds, MAX_FORK_TRIES_DECOMPRESS))
    {
    case -1:
      if (errno != EMFILE)
        error (SORT_FAILURE, errno, _("couldn't create process for %s -d"),
               compress_program);
      close (tempfd);
      errno = EMFILE;
      break;

    case 0:
      close (pipefds[0]);
      dup2_or_die (tempfd, STDIN_FILENO);
      close (tempfd);
      dup2_or_die (pipefds[1], STDOUT_FILENO);
      close (pipefds[1]);

      execlp (compress_program, compress_program, "-d", (char *) NULL);
      error (SORT_FAILURE, errno, _("couldn't execute %s -d"),
             compress_program);

    default:
      close (tempfd);
      close (pipefds[1]);

      fp = fdopen (pipefds[0], "r");
      if (! fp)
        {
          int saved_errno = errno;
          close (pipefds[0]);
          errno = saved_errno;
        }
      break;
    }

  return fp;
}

/* Append DIR to the array of temporary directory names.  */
static void
add_temp_dir (char const *dir)
{
  if (temp_dir_count == temp_dir_alloc)
    temp_dirs = X2NREALLOC (temp_dirs, &temp_dir_alloc);

  temp_dirs[temp_dir_count++] = dir;
}

/* Remove NAME from the list of temporary files.  */

static void
zaptemp (char const *name)
{
  struct tempnode *volatile *pnode;
  struct tempnode *node;
  struct tempnode *next;
  int unlink_status;
  int unlink_errno = 0;
  struct cs_status cs;

  for (pnode = &temphead; (node = *pnode)->name != name; pnode = &node->next)
    continue;

  /* Unlink the temporary file in a critical section to avoid races.  */
  next = node->next;
  cs = cs_enter ();
  unlink_status = unlink (name);
  unlink_errno = errno;
  *pnode = next;
  cs_leave (cs);

  if (unlink_status != 0)
    error (0, unlink_errno, _("warning: cannot remove: %s"), name);
  if (! next)
    temptail = pnode;
  free (node);
}

#if HAVE_NL_LANGINFO

static int
struct_month_cmp (void const *m1, void const *m2)
{
  struct month const *month1 = m1;
  struct month const *month2 = m2;
  return strcmp (month1->name, month2->name);
}

#endif

/* Initialize the character class tables. */

static void
inittables (void)
{
  size_t i;

  for (i = 0; i < UCHAR_LIM; ++i)
    {
      blanks[i] = !! isblank (i);
      nonprinting[i] = ! isprint (i);
      nondictionary[i] = ! isalnum (i) && ! isblank (i);
      fold_toupper[i] = toupper (i);
    }

#if HAVE_NL_LANGINFO
  /* If we're not in the "C" locale, read different names for months.  */
  if (hard_LC_TIME)
    {
      for (i = 0; i < MONTHS_PER_YEAR; i++)
        {
          char const *s;
          size_t s_len;
          size_t j, k;
          char *name;

          s = nl_langinfo (ABMON_1 + i);
          s_len = strlen (s);
          monthtab[i].name = name = xmalloc (s_len + 1);
          monthtab[i].val = i + 1;

          for (j = k = 0; j < s_len; j++)
            if (! isblank (to_uchar (s[j])))
              name[k++] = fold_toupper[to_uchar (s[j])];
          name[k] = '\0';
        }
      qsort (monthtab, MONTHS_PER_YEAR, sizeof *monthtab, struct_month_cmp);
    }
#endif
}

/* Specify how many inputs may be merged at once.
   This may be set on the command-line with the
   --batch-size option. */
static void
specify_nmerge (int oi, char c, char const *s)
{
  uintmax_t n;
  struct rlimit rlimit;
  enum strtol_error e = xstrtoumax (s, NULL, 10, &n, NULL);

  /* Try to find out how many file descriptors we'll be able
     to open.  We need at least nmerge + 3 (STDIN_FILENO,
     STDOUT_FILENO and STDERR_FILENO). */
  unsigned int max_nmerge = ((getrlimit (RLIMIT_NOFILE, &rlimit) == 0
                              ? rlimit.rlim_cur
                              : OPEN_MAX)
                             - 3);

  if (e == LONGINT_OK)
    {
      nmerge = n;
      if (nmerge != n)
        e = LONGINT_OVERFLOW;
      else
        {
          if (nmerge < 2)
            {
              error (0, 0, _("invalid --%s argument %s"),
                     long_options[oi].name, quote (s));
              error (SORT_FAILURE, 0,
                     _("minimum --%s argument is %s"),
                     long_options[oi].name, quote ("2"));
            }
          else if (max_nmerge < nmerge)
            {
              e = LONGINT_OVERFLOW;
            }
          else
            return;
        }
    }

  if (e == LONGINT_OVERFLOW)
    {
      char max_nmerge_buf[INT_BUFSIZE_BOUND (max_nmerge)];
      error (0, 0, _("--%s argument %s too large"),
             long_options[oi].name, quote (s));
      error (SORT_FAILURE, 0,
             _("maximum --%s argument with current rlimit is %s"),
             long_options[oi].name,
             uinttostr (max_nmerge, max_nmerge_buf));
    }
  else
    xstrtol_fatal (e, oi, c, long_options, s);
}

/* Specify the amount of main memory to use when sorting.  */
static void
specify_sort_size (int oi, char c, char const *s)
{
  uintmax_t n;
  char *suffix;
  enum strtol_error e = xstrtoumax (s, &suffix, 10, &n, "EgGkKmMPtTYZ");

  /* The default unit is KiB.  */
  if (e == LONGINT_OK && ISDIGIT (suffix[-1]))
    {
      if (n <= UINTMAX_MAX / 1024)
        n *= 1024;
      else
        e = LONGINT_OVERFLOW;
    }

  /* A 'b' suffix means bytes; a '%' suffix means percent of memory.  */
  if (e == LONGINT_INVALID_SUFFIX_CHAR && ISDIGIT (suffix[-1]) && ! suffix[1])
    switch (suffix[0])
      {
      case 'b':
        e = LONGINT_OK;
        break;

      case '%':
        {
          double mem = physmem_total () * n / 100;

          /* Use "<", not "<=", to avoid problems with rounding.  */
          if (mem < UINTMAX_MAX)
            {
              n = mem;
              e = LONGINT_OK;
            }
          else
            e = LONGINT_OVERFLOW;
        }
        break;
      }

  if (e == LONGINT_OK)
    {
      /* If multiple sort sizes are specified, take the maximum, so
         that option order does not matter.  */
      if (n < sort_size)
        return;

      sort_size = n;
      if (sort_size == n)
        {
          sort_size = MAX (sort_size, MIN_SORT_SIZE);
          return;
        }

      e = LONGINT_OVERFLOW;
    }

  xstrtol_fatal (e, oi, c, long_options, s);
}

/* Specify the number of threads to spawn during internal sort.  */
static unsigned long int
specify_nthreads (int oi, char c, char const *s)
{
  unsigned long int nthreads;
  enum strtol_error e = xstrtoul (s, NULL, 10, &nthreads, "");
  if (e == LONGINT_OVERFLOW)
    return ULONG_MAX;
  if (e != LONGINT_OK)
    xstrtol_fatal (e, oi, c, long_options, s);
  if (nthreads == 0)
    error (SORT_FAILURE, 0, _("number in parallel must be nonzero"));
  return nthreads;
}


/* Return the default sort size.  */
static size_t
default_sort_size (void)
{
  /* Let MEM be available memory or 1/8 of total memory, whichever
     is greater.  */
  double avail = physmem_available ();
  double total = physmem_total ();
  double mem = MAX (avail, total / 8);
  struct rlimit rlimit;

  /* Let SIZE be MEM, but no more than the maximum object size or
     system resource limits.  Avoid the MIN macro here, as it is not
     quite right when only one argument is floating point.  Don't
     bother to check for values like RLIM_INFINITY since in practice
     they are not much less than SIZE_MAX.  */
  size_t size = SIZE_MAX;
  if (mem < size)
    size = mem;
  if (getrlimit (RLIMIT_DATA, &rlimit) == 0 && rlimit.rlim_cur < size)
    size = rlimit.rlim_cur;
#ifdef RLIMIT_AS
  if (getrlimit (RLIMIT_AS, &rlimit) == 0 && rlimit.rlim_cur < size)
    size = rlimit.rlim_cur;
#endif

  /* Leave a large safety margin for the above limits, as failure can
     occur when they are exceeded.  */
  size /= 2;

#ifdef RLIMIT_RSS
  /* Leave a 1/16 margin for RSS to leave room for code, stack, etc.
     Exceeding RSS is not fatal, but can be quite slow.  */
  if (getrlimit (RLIMIT_RSS, &rlimit) == 0 && rlimit.rlim_cur / 16 * 15 < size)
    size = rlimit.rlim_cur / 16 * 15;
#endif

  /* Use no less than the minimum.  */
  return MAX (size, MIN_SORT_SIZE);
}

/* Return the sort buffer size to use with the input files identified
   by FPS and FILES, which are alternate names of the same files.
   NFILES gives the number of input files; NFPS may be less.  Assume
   that each input line requires LINE_BYTES extra bytes' worth of line
   information.  Do not exceed the size bound specified by the user
   (or a default size bound, if the user does not specify one).  */

static size_t
sort_buffer_size (FILE *const *fps, size_t nfps,
                  char *const *files, size_t nfiles,
                  size_t line_bytes)
{
  /* A bound on the input size.  If zero, the bound hasn't been
     determined yet.  */
  static size_t size_bound;

  /* In the worst case, each input byte is a newline.  */
  size_t worst_case_per_input_byte = line_bytes + 1;

  /* Keep enough room for one extra input line and an extra byte.
     This extra room might be needed when preparing to read EOF.  */
  size_t size = worst_case_per_input_byte + 1;

  size_t i;

  for (i = 0; i < nfiles; i++)
    {
      struct stat st;
      off_t file_size;
      size_t worst_case;

      if ((i < nfps ? fstat (fileno (fps[i]), &st)
           : STREQ (files[i], "-") ? fstat (STDIN_FILENO, &st)
           : stat (files[i], &st))
          != 0)
        die (_("stat failed"), files[i]);

      if (S_ISREG (st.st_mode))
        file_size = st.st_size;
      else
        {
          /* The file has unknown size.  If the user specified a sort
             buffer size, use that; otherwise, guess the size.  */
          if (sort_size)
            return sort_size;
          file_size = INPUT_FILE_SIZE_GUESS;
        }

      if (! size_bound)
        {
          size_bound = sort_size;
          if (! size_bound)
            size_bound = default_sort_size ();
        }

      /* Add the amount of memory needed to represent the worst case
         where the input consists entirely of newlines followed by a
         single non-newline.  Check for overflow.  */
      worst_case = file_size * worst_case_per_input_byte + 1;
      if (file_size != worst_case / worst_case_per_input_byte
          || size_bound - size <= worst_case)
        return size_bound;
      size += worst_case;
    }

  return size;
}

/* Initialize BUF.  Reserve LINE_BYTES bytes for each line; LINE_BYTES
   must be at least sizeof (struct line).  Allocate ALLOC bytes
   initially.  */

static void
initbuf (struct buffer *buf, size_t line_bytes, size_t alloc)
{
  /* Ensure that the line array is properly aligned.  If the desired
     size cannot be allocated, repeatedly halve it until allocation
     succeeds.  The smaller allocation may hurt overall performance,
     but that's better than failing.  */
  while (true)
    {
      alloc += sizeof (struct line) - alloc % sizeof (struct line);
      buf->buf = malloc (alloc);
      if (buf->buf)
        break;
      alloc /= 2;
      if (alloc <= line_bytes + 1)
        xalloc_die ();
    }

  buf->line_bytes = line_bytes;
  buf->alloc = alloc;
  buf->used = buf->left = buf->nlines = 0;
  buf->eof = false;
}

/* Return one past the limit of the line array.  */

static inline struct line *
buffer_linelim (struct buffer const *buf)
{
  return (struct line *) (buf->buf + buf->alloc);
}

/* Return a pointer to the first character of the field specified
   by KEY in LINE. */

static char *
begfield (struct line const *line, struct keyfield const *key)
{
  char *ptr = line->text, *lim = ptr + line->length - 1;
  size_t sword = key->sword;
  size_t schar = key->schar;

  /* The leading field separator itself is included in a field when -t
     is absent.  */

  if (tab != TAB_DEFAULT)
    while (ptr < lim && sword--)
      {
        while (ptr < lim && *ptr != tab)
          ++ptr;
        if (ptr < lim)
          ++ptr;
      }
  else
    while (ptr < lim && sword--)
      {
        while (ptr < lim && blanks[to_uchar (*ptr)])
          ++ptr;
        while (ptr < lim && !blanks[to_uchar (*ptr)])
          ++ptr;
      }

  /* If we're ignoring leading blanks when computing the Start
     of the field, skip past them here.  */
  if (key->skipsblanks)
    while (ptr < lim && blanks[to_uchar (*ptr)])
      ++ptr;

  /* Advance PTR by SCHAR (if possible), but no further than LIM.  */
  ptr = MIN (lim, ptr + schar);

  return ptr;
}

/* Return the limit of (a pointer to the first character after) the field
   in LINE specified by KEY. */

static char *
limfield (struct line const *line, struct keyfield const *key)
{
  char *ptr = line->text, *lim = ptr + line->length - 1;
  size_t eword = key->eword, echar = key->echar;

  if (echar == 0)
    eword++; /* Skip all of end field.  */

  /* Move PTR past EWORD fields or to one past the last byte on LINE,
     whichever comes first.  If there are more than EWORD fields, leave
     PTR pointing at the beginning of the field having zero-based index,
     EWORD.  If a delimiter character was specified (via -t), then that
     `beginning' is the first character following the delimiting TAB.
     Otherwise, leave PTR pointing at the first `blank' character after
     the preceding field.  */
  if (tab != TAB_DEFAULT)
    while (ptr < lim && eword--)
      {
        while (ptr < lim && *ptr != tab)
          ++ptr;
        if (ptr < lim && (eword || echar))
          ++ptr;
      }
  else
    while (ptr < lim && eword--)
      {
        while (ptr < lim && blanks[to_uchar (*ptr)])
          ++ptr;
        while (ptr < lim && !blanks[to_uchar (*ptr)])
          ++ptr;
      }

#ifdef POSIX_UNSPECIFIED
  /* The following block of code makes GNU sort incompatible with
     standard Unix sort, so it's ifdef'd out for now.
     The POSIX spec isn't clear on how to interpret this.
     FIXME: request clarification.

     From: kwzh@gnu.ai.mit.edu (Karl Heuer)
     Date: Thu, 30 May 96 12:20:41 -0400
     [Translated to POSIX 1003.1-2001 terminology by Paul Eggert.]

     [...]I believe I've found another bug in `sort'.

     $ cat /tmp/sort.in
     a b c 2 d
     pq rs 1 t
     $ textutils-1.15/src/sort -k1.7,1.7 </tmp/sort.in
     a b c 2 d
     pq rs 1 t
     $ /bin/sort -k1.7,1.7 </tmp/sort.in
     pq rs 1 t
     a b c 2 d

     Unix sort produced the answer I expected: sort on the single character
     in column 7.  GNU sort produced different results, because it disagrees
     on the interpretation of the key-end spec "M.N".  Unix sort reads this
     as "skip M-1 fields, then N-1 characters"; but GNU sort wants it to mean
     "skip M-1 fields, then either N-1 characters or the rest of the current
     field, whichever comes first".  This extra clause applies only to
     key-ends, not key-starts.
     */

  /* Make LIM point to the end of (one byte past) the current field.  */
  if (tab != TAB_DEFAULT)
    {
      char *newlim;
      newlim = memchr (ptr, tab, lim - ptr);
      if (newlim)
        lim = newlim;
    }
  else
    {
      char *newlim;
      newlim = ptr;
      while (newlim < lim && blanks[to_uchar (*newlim)])
        ++newlim;
      while (newlim < lim && !blanks[to_uchar (*newlim)])
        ++newlim;
      lim = newlim;
    }
#endif

  if (echar != 0) /* We need to skip over a portion of the end field.  */
    {
      /* If we're ignoring leading blanks when computing the End
         of the field, skip past them here.  */
      if (key->skipeblanks)
        while (ptr < lim && blanks[to_uchar (*ptr)])
          ++ptr;

      /* Advance PTR by ECHAR (if possible), but no further than LIM.  */
      ptr = MIN (lim, ptr + echar);
    }

  return ptr;
}

/* Fill BUF reading from FP, moving buf->left bytes from the end
   of buf->buf to the beginning first.  If EOF is reached and the
   file wasn't terminated by a newline, supply one.  Set up BUF's line
   table too.  FILE is the name of the file corresponding to FP.
   Return true if some input was read.  */

static bool
fillbuf (struct buffer *buf, FILE *fp, char const *file)
{
  struct keyfield const *key = keylist;
  char eol = eolchar;
  size_t line_bytes = buf->line_bytes;
  size_t mergesize = merge_buffer_size - MIN_MERGE_BUFFER_SIZE;

  if (buf->eof)
    return false;

  if (buf->used != buf->left)
    {
      memmove (buf->buf, buf->buf + buf->used - buf->left, buf->left);
      buf->used = buf->left;
      buf->nlines = 0;
    }

  while (true)
    {
      char *ptr = buf->buf + buf->used;
      struct line *linelim = buffer_linelim (buf);
      struct line *line = linelim - buf->nlines;
      size_t avail = (char *) linelim - buf->nlines * line_bytes - ptr;
      char *line_start = buf->nlines ? line->text + line->length : buf->buf;

      while (line_bytes + 1 < avail)
        {
          /* Read as many bytes as possible, but do not read so many
             bytes that there might not be enough room for the
             corresponding line array.  The worst case is when the
             rest of the input file consists entirely of newlines,
             except that the last byte is not a newline.  */
          size_t readsize = (avail - 1) / (line_bytes + 1);
          size_t bytes_read = fread (ptr, 1, readsize, fp);
          char *ptrlim = ptr + bytes_read;
          char *p;
          avail -= bytes_read;

          if (bytes_read != readsize)
            {
              if (ferror (fp))
                die (_("read failed"), file);
              if (feof (fp))
                {
                  buf->eof = true;
                  if (buf->buf == ptrlim)
                    return false;
                  if (line_start != ptrlim && ptrlim[-1] != eol)
                    *ptrlim++ = eol;
                }
            }

          /* Find and record each line in the just-read input.  */
          while ((p = memchr (ptr, eol, ptrlim - ptr)))
            {
              /* Delimit the line with NUL. This eliminates the need to
                 temporarily replace the last byte with NUL when calling
                 xmemcoll(), which increases performance.  */
              *p = '\0';
              ptr = p + 1;
              line--;
              line->text = line_start;
              line->length = ptr - line_start;
              mergesize = MAX (mergesize, line->length);
              avail -= line_bytes;

              if (key)
                {
                  /* Precompute the position of the first key for
                     efficiency.  */
                  line->keylim = (key->eword == SIZE_MAX
                                  ? p
                                  : limfield (line, key));

                  if (key->sword != SIZE_MAX)
                    line->keybeg = begfield (line, key);
                  else
                    {
                      if (key->skipsblanks)
                        while (blanks[to_uchar (*line_start)])
                          line_start++;
                      line->keybeg = line_start;
                    }
                }

              line_start = ptr;
            }

          ptr = ptrlim;
          if (buf->eof)
            break;
        }

      buf->used = ptr - buf->buf;
      buf->nlines = buffer_linelim (buf) - line;
      if (buf->nlines != 0)
        {
          buf->left = ptr - line_start;
          merge_buffer_size = mergesize + MIN_MERGE_BUFFER_SIZE;
          return true;
        }

      {
        /* The current input line is too long to fit in the buffer.
           Double the buffer size and try again, keeping it properly
           aligned.  */
        size_t line_alloc = buf->alloc / sizeof (struct line);
        buf->buf = x2nrealloc (buf->buf, &line_alloc, sizeof (struct line));
        buf->alloc = line_alloc * sizeof (struct line);
      }
    }
}

/* Return an integer that represents the order of magnitude of the
   unit following the number.  The number may contain thousands
   separators and a decimal point, but it may not contain leading blanks.
   Negative numbers get negative orders; zero numbers have a zero order.
   Store the address of the end of the number into *ENDPTR.  */

static int
find_unit_order (char const *number, char **endptr)
{
  static char const orders[UCHAR_LIM] =
    {
#if ! ('K' == 75 && 'M' == 77 && 'G' == 71 && 'T' == 84 && 'P' == 80 \
       && 'E' == 69 && 'Z' == 90 && 'Y' == 89 && 'k' == 107)
      /* This initializer syntax works on all C99 hosts.  For now, use
         it only on non-ASCII hosts, to ease the pain of porting to
         pre-C99 ASCII hosts.  */
      ['K']=1, ['M']=2, ['G']=3, ['T']=4, ['P']=5, ['E']=6, ['Z']=7, ['Y']=8,
      ['k']=1,
#else
      /* Generate the following table with this command:
         perl -e 'my %a=(k=>1, K=>1, M=>2, G=>3, T=>4, P=>5, E=>6, Z=>7, Y=>8);
         foreach my $i (0..255) {my $c=chr($i); $a{$c} ||= 0;print "$a{$c}, "}'\
         |fmt  */
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, 0, 3,
      0, 0, 0, 1, 0, 2, 0, 0, 5, 0, 0, 0, 4, 0, 0, 0, 0, 8, 7, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
#endif
    };

  bool minus_sign = (*number == '-');
  char const *p = number + minus_sign;
  int nonzero = 0;
  unsigned char ch;

  /* Scan to end of number.
     Decimals or separators not followed by digits stop the scan.
     Numbers ending in decimals or separators are thus considered
     to be lacking in units.
     FIXME: add support for multibyte thousands_sep and decimal_point.  */

  do
    {
      while (ISDIGIT (ch = *p++))
        nonzero |= ch - '0';
    }
  while (ch == thousands_sep);

  if (ch == decimal_point)
    while (ISDIGIT (ch = *p++))
      nonzero |= ch - '0';

  int order = (nonzero ? orders[ch] : 0);
  *endptr = (char *) p - !order;
  return (minus_sign ? -order : order);
}

/* Compare numbers A and B ending in units with SI or IEC prefixes
       <none/unknown> < K/k < M < G < T < P < E < Z < Y
   This may temporarily modify the strings.  Store into *EA the end
   of the string A.  */

static int
human_numcompare (char *a, char *b, char **ea)
{
  char *endb;

  while (blanks[to_uchar (*a)])
    a++;
  while (blanks[to_uchar (*b)])
    b++;

  int diff = find_unit_order (a, ea) - find_unit_order (b, &endb);
  return (diff ? diff : strnumcmp (a, b, decimal_point, thousands_sep));
}

/* Compare strings A and B as numbers without explicitly converting them to
   machine numbers.  Comparatively slow for short strings, but asymptotically
   hideously fast. */

static int
numcompare (char const *a, char const *b, char **ea)
{
  while (blanks[to_uchar (*a)])
    a++;
  while (blanks[to_uchar (*b)])
    b++;

  if (debug)
    {
      /* Approximate strnumcmp extents with find_unit_order.  */
      int order = find_unit_order (a, ea);
      *ea -= !!order;
    }

  return strnumcmp (a, b, decimal_point, thousands_sep);
}

static int
general_numcompare (char const *sa, char const *sb, char **ea)
{
  /* FIXME: maybe add option to try expensive FP conversion
     only if A and B can't be compared more cheaply/accurately.  */

#if HAVE_C99_STRTOLD
# define long_double long double
#else
# define long_double double
# undef strtold
# define strtold strtod
#endif

  char *eb;
  long_double a = strtold (sa, ea);
  long_double b = strtold (sb, &eb);

  /* Put conversion errors at the start of the collating sequence.  */
  if (sa == *ea)
    return sb == eb ? 0 : -1;
  if (sb == eb)
    return 1;

  /* Sort numbers in the usual way, where -0 == +0.  Put NaNs after
     conversion errors but before numbers; sort them by internal
     bit-pattern, for lack of a more portable alternative.  */
  return (a < b ? -1
          : a > b ? 1
          : a == b ? 0
          : b == b ? -1
          : a == a ? 1
          : memcmp (&a, &b, sizeof a));
}

/* Return an integer in 1..12 of the month name MONTH with length LEN.
   Return 0 if the name in S is not recognized.  */

static int
getmonth (char const *month, size_t len, char const **ea)
{
  size_t lo = 0;
  size_t hi = MONTHS_PER_YEAR;
  char const *monthlim = month + len;

  while (true)
    {
      if (month == monthlim)
        return 0;
      if (!blanks[to_uchar (*month)])
        break;
      ++month;
    }

  do
    {
      size_t ix = (lo + hi) / 2;
      char const *m = month;
      char const *n = monthtab[ix].name;

      for (;; m++, n++)
        {
          if (!*n)
            {
              if (ea)
                *ea = m;
              return monthtab[ix].val;
            }
          if (m == monthlim || fold_toupper[to_uchar (*m)] < to_uchar (*n))
            {
              hi = ix;
              break;
            }
          else if (fold_toupper[to_uchar (*m)] > to_uchar (*n))
            {
              lo = ix + 1;
              break;
            }
        }
    }
  while (lo < hi);

  return 0;
}

/* A randomly chosen MD5 state, used for random comparison.  */
static struct md5_ctx random_md5_state;

/* Initialize the randomly chosen MD5 state.  */

static void
random_md5_state_init (char const *random_source)
{
  unsigned char buf[MD5_DIGEST_SIZE];
  struct randread_source *r = randread_new (random_source, sizeof buf);
  if (! r)
    die (_("open failed"), random_source);
  randread (r, buf, sizeof buf);
  if (randread_free (r) != 0)
    die (_("close failed"), random_source);
  md5_init_ctx (&random_md5_state);
  md5_process_bytes (buf, sizeof buf, &random_md5_state);
}

/* This is like strxfrm, except it reports any error and exits.  */

static size_t
xstrxfrm (char *restrict dest, char const *restrict src, size_t destsize)
{
  errno = 0;
  size_t translated_size = strxfrm (dest, src, destsize);

  if (errno)
    {
      error (0, errno, _("string transformation failed"));
      error (0, 0, _("set LC_ALL='C' to work around the problem"));
      error (SORT_FAILURE, 0,
             _("the untransformed string was %s"),
             quotearg_n_style (0, locale_quoting_style, src));
    }

  return translated_size;
}

/* Compare the keys TEXTA (of length LENA) and TEXTB (of length LENB)
   using one or more random hash functions.  */

static int
compare_random (char *restrict texta, size_t lena,
                char *restrict textb, size_t lenb)
{
  /* XFRM_DIFF records the equivalent of memcmp on the transformed
     data.  This is used to break ties if there is an checksum
     collision, and this is good enough given the astronomically low
     probability of a collision.  */
  int xfrm_diff = 0;

  char stackbuf[4000];
  char *buf = stackbuf;
  size_t bufsize = sizeof stackbuf;
  uint32_t dig[2][MD5_DIGEST_SIZE / sizeof (uint32_t)];
  struct md5_ctx s[2];
  s[0] = s[1] = random_md5_state;

  if (hard_LC_COLLATE)
    {
      /* Null-terminate the keys so that strxfrm knows where to stop.  */
      char *lima = texta + lena; char enda = *lima; *lima = '\0';
      char *limb = textb + lenb; char endb = *limb; *limb = '\0';

      while (true)
        {
          /* Transform the text into the basis of comparison, so that byte
             strings that would otherwise considered to be equal are
             considered equal here even if their bytes differ.

             Each time through this loop, transform one
             null-terminated string's worth from TEXTA or from TEXTB
             or both.  That way, there's no need to store the
             transformation of the whole line, if it contains many
             null-terminated strings.  */

          /* Store the transformed data into a big-enough buffer.  */

          size_t sizea =
            (texta < lima ? xstrxfrm (buf, texta, bufsize) + 1 : 0);
          bool a_fits = sizea <= bufsize;
          size_t sizeb =
            (textb < limb
             ? (xstrxfrm ((a_fits ? buf + sizea : NULL), textb,
                          (a_fits ? bufsize - sizea : 0))
                + 1)
             : 0);

          if (! (a_fits && sizea + sizeb <= bufsize))
            {
              bufsize = sizea + sizeb;
              if (bufsize < SIZE_MAX / 3)
                bufsize = bufsize * 3 / 2;
              buf = (buf == stackbuf
                     ? xmalloc (bufsize)
                     : xrealloc (buf, bufsize));
              if (texta < lima)
                strxfrm (buf, texta, sizea);
              if (textb < limb)
                strxfrm (buf + sizea, textb, sizeb);
            }

          /* Advance past NULs to the next part of each input string,
             exiting the loop if both strings are exhausted.  When
             exiting the loop, prepare to finish off the tiebreaker
             comparison properly.  */
          if (texta < lima)
            texta += strlen (texta) + 1;
          if (textb < limb)
            textb += strlen (textb) + 1;
          if (! (texta < lima || textb < limb))
            {
              lena = sizea; texta = buf;
              lenb = sizeb; textb = buf + sizea;
              break;
            }

          /* Accumulate the transformed data in the corresponding
             checksums.  */
          md5_process_bytes (buf, sizea, &s[0]);
          md5_process_bytes (buf + sizea, sizeb, &s[1]);

          /* Update the tiebreaker comparison of the transformed data.  */
          if (! xfrm_diff)
            {
              xfrm_diff = memcmp (buf, buf + sizea, MIN (sizea, sizeb));
              if (! xfrm_diff)
                xfrm_diff = (sizea > sizeb) - (sizea < sizeb);
            }
        }

      *lima = enda;
      *limb = endb;
    }

  /* Compute and compare the checksums.  */
  md5_process_bytes (texta, lena, &s[0]); md5_finish_ctx (&s[0], dig[0]);
  md5_process_bytes (textb, lenb, &s[1]); md5_finish_ctx (&s[1], dig[1]);
  int diff = memcmp (dig[0], dig[1], sizeof dig[0]);

  /* Fall back on the tiebreaker if the checksums collide.  */
  if (! diff)
    {
      if (! xfrm_diff)
        {
          xfrm_diff = memcmp (texta, textb, MIN (lena, lenb));
          if (! xfrm_diff)
            xfrm_diff = (lena > lenb) - (lena < lenb);
        }

      diff = xfrm_diff;
    }

  if (buf != stackbuf)
    free (buf);

  return diff;
}

/* Compare the keys TEXTA (of length LENA) and TEXTB (of length LENB)
   using filevercmp. See lib/filevercmp.h for function description. */

static int
compare_version (char *restrict texta, size_t lena,
                 char *restrict textb, size_t lenb)
{
  int diff;

  /* It is necessary to save the character after the end of the field.
     "filevercmp" works with NUL terminated strings.  Our blocks of
     text are not necessarily terminated with a NUL byte. */
  char sv_a = texta[lena];
  char sv_b = textb[lenb];

  texta[lena] = '\0';
  textb[lenb] = '\0';

  diff = filevercmp (texta, textb);

  texta[lena] = sv_a;
  textb[lenb] = sv_b;

  return diff;
}

/* Return the printable width of the block of memory starting at
   TEXT and ending just before LIM, counting each tab as one byte.
   FIXME: Should we generally be counting non printable chars?  */

static size_t
debug_width (char const *text, char const *lim)
{
  size_t width = mbsnwidth (text, lim - text, 0);
  while (text < lim)
    width += (*text++ == '\t');
  return width;
}

/* For debug mode, "underline" a key at the
   specified offset and screen width.  */

static void
mark_key (size_t offset, size_t width)
{
  while (offset--)
    putchar (' ');

  if (!width)
    printf (_("^ no match for key\n"));
  else
    {
      do
        putchar ('_');
      while (--width);

      putchar ('\n');
    }
}

/* For debug mode, determine the screen offset and width
   to highlight for a key, and then output the highlight.  */

static void
debug_key (char const *sline, char const *sfield, char const *efield,
           size_t flen, bool skipb)
{
  char const *sa = sfield;

  if (skipb) /* This key type implicitly skips leading blanks.  */
    {
      while (sa < efield && blanks[to_uchar (*sa)])
        {
          sa++;
          if (flen)
            flen--; /* This assumes TABs same width as SPACEs.  */
        }
    }

  size_t offset = debug_width (sline, sfield) + (sa - sfield);
  size_t width = debug_width (sa, sa + flen);
  mark_key (offset, width);
}

/* Testing if a key is numeric is done in various places.  */

static inline bool
key_numeric (struct keyfield const *key)
{
  return key->numeric || key->general_numeric || key->human_numeric;
}

/* Return whether sorting options specified for key.  */

static bool
default_key_compare (struct keyfield const *key)
{
  return ! (key->ignore
            || key->translate
            || key->skipsblanks
            || key->skipeblanks
            || key_numeric (key)
            || key->month
            || key->version
            || key->random
            /* || key->reverse */
           );
}

/* Convert a key to the short options used to specify it.  */

static void
key_to_opts (struct keyfield const *key, char *opts)
{
  if (key->skipsblanks || key->skipeblanks)
    *opts++ = 'b';/* either disables global -b  */
  if (key->ignore == nondictionary)
    *opts++ = 'd';
  if (key->translate)
    *opts++ = 'f';
  if (key->general_numeric)
    *opts++ = 'g';
  if (key->human_numeric)
    *opts++ = 'h';
  if (key->ignore == nonprinting)
    *opts++ = 'i';
  if (key->month)
    *opts++ = 'M';
  if (key->numeric)
    *opts++ = 'n';
  if (key->random)
    *opts++ = 'R';
  if (key->reverse)
    *opts++ = 'r';
  if (key->version)
    *opts++ = 'V';
  *opts = '\0';
}

/* Output data independent key warnings to stderr.  */

static void
key_warnings (struct keyfield const *gkey, bool gkey_only)
{
  struct keyfield const *key;
  struct keyfield ugkey = *gkey;
  unsigned long keynum = 1;

  for (key = keylist; key; key = key->next, keynum++)
    {
      if (key->obsolete_used)
        {
          size_t sword = key->sword;
          size_t eword = key->eword;
          char tmp[INT_BUFSIZE_BOUND (sword)];
          /* obsolescent syntax +A.x -B.y is equivalent to:
               -k A+1.x+1,B.y   (when y = 0)
               -k A+1.x+1,B+1.y (when y > 0)  */
          char obuf[INT_BUFSIZE_BOUND (sword) * 2 + 4]; /* +# -#  */
          char nbuf[INT_BUFSIZE_BOUND (sword) * 2 + 5]; /* -k #,#  */
          char *po = obuf;
          char *pn = nbuf;

          if (sword == SIZE_MAX)
            sword++;

          po = stpcpy (stpcpy (po, "+"), umaxtostr (sword, tmp));
          pn = stpcpy (stpcpy (pn, "-k "), umaxtostr (sword + 1, tmp));
          if (key->eword != SIZE_MAX)
            {
              po = stpcpy (stpcpy (po, " -"), umaxtostr (eword + 1, tmp));
              pn = stpcpy (stpcpy (pn, ","),
                           umaxtostr (eword + 1
                                      + (key->echar == SIZE_MAX), tmp));
            }
          error (0, 0, _("obsolescent key `%s' used; consider `%s' instead"),
                 obuf, nbuf);
        }

      /* Warn about field specs that will never match.  */
      if (key->sword != SIZE_MAX && key->eword < key->sword)
        error (0, 0, _("key %lu has zero width and will be ignored"), keynum);

      /* Warn about significant leading blanks.  */
      bool implicit_skip = key_numeric (key) || key->month;
      bool maybe_space_aligned = !hard_LC_COLLATE && default_key_compare (key)
                                 && !(key->schar || key->echar);
      bool line_offset = key->eword == 0 && key->echar != 0; /* -k1.x,1.y  */
      if (!gkey_only && tab == TAB_DEFAULT && !line_offset
          && ((!key->skipsblanks && !(implicit_skip || maybe_space_aligned))
              || (!key->skipsblanks && key->schar)
              || (!key->skipeblanks && key->echar)))
        error (0, 0, _("leading blanks are significant in key %lu; "
                       "consider also specifying `b'"), keynum);

      /* Warn about numeric comparisons spanning fields,
         as field delimiters could be interpreted as part
         of the number (maybe only in other locales).  */
      if (!gkey_only && key_numeric (key))
        {
          size_t sword = key->sword + 1;
          size_t eword = key->eword + 1;
          if (!sword)
            sword++;
          if (sword != eword)
            error (0, 0, _("key %lu is numeric and spans multiple fields"),
                   keynum);
        }

      /* Flag global options not copied or specified in any key.  */
      if (ugkey.ignore && (ugkey.ignore == key->ignore))
        ugkey.ignore = NULL;
      if (ugkey.translate && (ugkey.translate == key->translate))
        ugkey.translate = NULL;
      ugkey.skipsblanks &= !key->skipsblanks;
      ugkey.skipeblanks &= !key->skipeblanks;
      ugkey.month &= !key->month;
      ugkey.numeric &= !key->numeric;
      ugkey.general_numeric &= !key->general_numeric;
      ugkey.human_numeric &= !key->human_numeric;
      ugkey.random &= !key->random;
      ugkey.version &= !key->version;
      ugkey.reverse &= !key->reverse;
    }

  /* Warn about ignored global options flagged above.
     Note if gkey is the only one in the list, all flags are cleared.  */
  if (!default_key_compare (&ugkey)
      || (ugkey.reverse && (stable || unique) && keylist))
    {
      bool ugkey_reverse = ugkey.reverse;
      if (!(stable || unique))
        ugkey.reverse = false;
      /* The following is too big, but guaranteed to be "big enough".  */
      char opts[sizeof short_options];
      key_to_opts (&ugkey, opts);
      error (0, 0,
             ngettext ("option `-%s' is ignored",
                       "options `-%s' are ignored",
                       select_plural (strlen (opts))), opts);
      ugkey.reverse = ugkey_reverse;
    }
  if (ugkey.reverse && !(stable || unique) && keylist)
    error (0, 0, _("option `-r' only applies to last-resort comparison"));
}

/* Compare two lines A and B trying every key in sequence until there
   are no more keys or a difference is found. */

static int
keycompare (struct line const *a, struct line const *b, bool show_debug)
{
  struct keyfield *key = keylist;

  /* For the first iteration only, the key positions have been
     precomputed for us. */
  char *texta = a->keybeg;
  char *textb = b->keybeg;
  char *lima = a->keylim;
  char *limb = b->keylim;

  int diff;

  while (true)
    {
      char const *translate = key->translate;
      bool const *ignore = key->ignore;
      bool skipb = false; /* Whether key type auto skips leading blanks.  */

      /* Treat field ends before field starts as empty fields.  */
      lima = MAX (texta, lima);
      limb = MAX (textb, limb);

      /* Find the lengths. */
      size_t lena = lima - texta;
      size_t lenb = limb - textb;

      /* Actually compare the fields. */

      if (key->random)
        diff = compare_random (texta, lena, textb, lenb);
      else if (key_numeric (key))
        {
          char savea = *lima, saveb = *limb;
          char *ea = lima;

          *lima = *limb = '\0';
          diff = (key->numeric ? numcompare (texta, textb, &ea)
                  : key->general_numeric ? general_numcompare (texta, textb,
                                                               &ea)
                  : human_numcompare (texta, textb, &ea));
          if (show_debug)
            {
              lena = ea - texta;
              skipb = true;
            }
          *lima = savea, *limb = saveb;
        }
      else if (key->version)
        diff = compare_version (texta, lena, textb, lenb);
      else if (key->month)
        {
          char const *ea = lima;

          int amon = getmonth (texta, lena, &ea);
          diff = amon - getmonth (textb, lenb, NULL);

          if (show_debug)
            {
              lena = amon ? ea - texta : 0;
              skipb = true;
            }
        }
      /* Sorting like this may become slow, so in a simple locale the user
         can select a faster sort that is similar to ascii sort.  */
      else if (hard_LC_COLLATE)
        {
          /* FIXME: for debug, account for skipped chars, while handling mb chars.
             Generally perhaps xmemfrm could be used to determine chars that are
             excluded from the collating order?  */
          if (ignore || translate)
            {
              char buf[4000];
              size_t size = lena + 1 + lenb + 1;
              char *copy_a = (size <= sizeof buf ? buf : xmalloc (size));
              char *copy_b = copy_a + lena + 1;
              size_t new_len_a, new_len_b, i;

              /* Ignore and/or translate chars before comparing.  */
              for (new_len_a = new_len_b = i = 0; i < MAX (lena, lenb); i++)
                {
                  if (i < lena)
                    {
                      copy_a[new_len_a] = (translate
                                           ? translate[to_uchar (texta[i])]
                                           : texta[i]);
                      if (!ignore || !ignore[to_uchar (texta[i])])
                        ++new_len_a;
                    }
                  if (i < lenb)
                    {
                      copy_b[new_len_b] = (translate
                                           ? translate[to_uchar (textb[i])]
                                           : textb [i]);
                      if (!ignore || !ignore[to_uchar (textb[i])])
                        ++new_len_b;
                    }
                }

              copy_a[new_len_a++] = '\0';
              copy_b[new_len_b++] = '\0';
              diff = xmemcoll0 (copy_a, new_len_a, copy_b, new_len_b);

              if (sizeof buf < size)
                free (copy_a);
            }
          else if (lena == 0)
            diff = - NONZERO (lenb);
          else if (lenb == 0)
            goto greater;
          else
            diff = xmemcoll (texta, lena, textb, lenb);
        }
      else if (ignore)
        {
          char *savea = texta;

#define CMP_WITH_IGNORE(A, B)                                           \
  do                                                                    \
    {                                                                   \
          while (true)                                                  \
            {                                                           \
              while (texta < lima && ignore[to_uchar (*texta)])         \
                ++texta;                                                \
              while (textb < limb && ignore[to_uchar (*textb)])         \
                ++textb;                                                \
              if (! (texta < lima && textb < limb))                     \
                break;                                                  \
              diff = to_uchar (A) - to_uchar (B);                       \
              if (diff)                                                 \
                goto not_equal;                                         \
              ++texta;                                                  \
              ++textb;                                                  \
            }                                                           \
                                                                        \
          diff = (texta < lima) - (textb < limb);                       \
    }                                                                   \
  while (0)

          if (translate)
            CMP_WITH_IGNORE (translate[to_uchar (*texta)],
                             translate[to_uchar (*textb)]);
          else
            CMP_WITH_IGNORE (*texta, *textb);

          /* We only need to restore this for debug_key
             in which case the keys being compared are equal.  */
          texta = savea;
        }
      else if (lena == 0)
        diff = - NONZERO (lenb);
      else if (lenb == 0)
        goto greater;
      else
        {
          if (translate)
            {
              char *savea = texta;

              while (texta < lima && textb < limb)
                {
                  diff = (to_uchar (translate[to_uchar (*texta++)])
                          - to_uchar (translate[to_uchar (*textb++)]));
                  if (diff)
                    goto not_equal;
                }

              /* We only need to restore this for debug_key
                 in which case the keys being compared are equal.  */
              texta = savea;
            }
          else
            {
              diff = memcmp (texta, textb, MIN (lena, lenb));
              if (diff)
                goto not_equal;
            }
          diff = lena < lenb ? -1 : lena != lenb;
        }

      if (diff)
        goto not_equal;

      if (show_debug)
        debug_key (a->text, texta, lima, lena, skipb);

      key = key->next;
      if (! key)
        break;

      /* Find the beginning and limit of the next field.  */
      if (key->eword != SIZE_MAX)
        lima = limfield (a, key), limb = limfield (b, key);
      else
        lima = a->text + a->length - 1, limb = b->text + b->length - 1;

      if (key->sword != SIZE_MAX)
        texta = begfield (a, key), textb = begfield (b, key);
      else
        {
          texta = a->text, textb = b->text;
          if (key->skipsblanks)
            {
              while (texta < lima && blanks[to_uchar (*texta)])
                ++texta;
              while (textb < limb && blanks[to_uchar (*textb)])
                ++textb;
            }
        }
    }

  return 0;

 greater:
  diff = 1;
 not_equal:
  return key->reverse ? -diff : diff;
}

/* Compare two lines A and B, returning negative, zero, or positive
   depending on whether A compares less than, equal to, or greater than B. */

static int
compare (struct line const *a, struct line const *b, bool show_debug)
{
  int diff;
  size_t alen, blen;

  /* First try to compare on the specified keys (if any).
     The only two cases with no key at all are unadorned sort,
     and unadorned sort -r. */
  if (keylist)
    {
      diff = keycompare (a, b, show_debug);
      if (diff || unique || stable)
        return diff;
    }

  /* If the keys all compare equal (or no keys were specified)
     fall through to the default comparison.  */
  alen = a->length - 1, blen = b->length - 1;

  if (show_debug)
    debug_key (a->text, a->text, a->text + alen, alen, false);

  if (alen == 0)
    diff = - NONZERO (blen);
  else if (blen == 0)
    diff = 1;
  else if (hard_LC_COLLATE)
    {
      /* Note xmemcoll0 is a performance enhancement as
         it will not unconditionally write '\0' after the
         passed in buffers, which was seen to give around
         a 3% increase in performance for short lines.  */
      diff = xmemcoll0 (a->text, alen + 1, b->text, blen + 1);
    }
  else if (! (diff = memcmp (a->text, b->text, MIN (alen, blen))))
    diff = alen < blen ? -1 : alen != blen;

  return reverse ? -diff : diff;
}

static void
write_bytes (struct line const *line, FILE *fp, char const *output_file)
{
  char *buf = line->text;
  size_t n_bytes = line->length;
  char *ebuf = buf + n_bytes;

  /* Convert TABs to '>' and \0 to \n when -z specified.  */
  if (debug && fp == stdout)
    {
      char const *c = buf;

      while (c < ebuf)
        {
          char wc = *c++;
          if (wc == '\t')
            wc = '>';
          else if (c == ebuf)
            wc = '\n';
          if (fputc (wc, fp) == EOF)
            die (_("write failed"), output_file);
        }

      compare (line, line, true);
    }
  else
    {
      ebuf[-1] = eolchar;
      if (fwrite (buf, 1, n_bytes, fp) != n_bytes)
        die (_("write failed"), output_file);
      ebuf[-1] = '\0';
    }
}

/* Check that the lines read from FILE_NAME come in order.  Return
   true if they are in order.  If CHECKONLY == 'c', also print a
   diagnostic (FILE_NAME, line number, contents of line) to stderr if
   they are not in order.  */

static bool
check (char const *file_name, char checkonly)
{
  FILE *fp = xfopen (file_name, "r");
  struct buffer buf;            /* Input buffer. */
  struct line temp;             /* Copy of previous line. */
  size_t alloc = 0;
  uintmax_t line_number = 0;
  struct keyfield const *key = keylist;
  bool nonunique = ! unique;
  bool ordered = true;

  initbuf (&buf, sizeof (struct line),
           MAX (merge_buffer_size, sort_size));
  temp.text = NULL;

  while (fillbuf (&buf, fp, file_name))
    {
      struct line const *line = buffer_linelim (&buf);
      struct line const *linebase = line - buf.nlines;

      /* Make sure the line saved from the old buffer contents is
         less than or equal to the first line of the new buffer. */
      if (alloc && nonunique <= compare (&temp, line - 1, false))
        {
        found_disorder:
          {
            if (checkonly == 'c')
              {
                struct line const *disorder_line = line - 1;
                uintmax_t disorder_line_number =
                  buffer_linelim (&buf) - disorder_line + line_number;
                char hr_buf[INT_BUFSIZE_BOUND (disorder_line_number)];
                fprintf (stderr, _("%s: %s:%s: disorder: "),
                         program_name, file_name,
                         umaxtostr (disorder_line_number, hr_buf));
                if (debug)
                  fputc ('\n', stderr);
                write_bytes (disorder_line, debug ? stdout : stderr,
                             debug ? _("standard out") : _("standard error"));
              }

            ordered = false;
            break;
          }
        }

      /* Compare each line in the buffer with its successor.  */
      while (linebase < --line)
        if (nonunique <= compare (line, line - 1, false))
          goto found_disorder;

      line_number += buf.nlines;

      /* Save the last line of the buffer.  */
      if (alloc < line->length)
        {
          do
            {
              alloc *= 2;
              if (! alloc)
                {
                  alloc = line->length;
                  break;
                }
            }
          while (alloc < line->length);

          temp.text = xrealloc (temp.text, alloc);
        }
      memcpy (temp.text, line->text, line->length);
      temp.length = line->length;
      if (key)
        {
          temp.keybeg = temp.text + (line->keybeg - line->text);
          temp.keylim = temp.text + (line->keylim - line->text);
        }
    }

  xfclose (fp, file_name);
  free (buf.buf);
  free (temp.text);
  return ordered;
}

/* Open FILES (there are NFILES of them) and store the resulting array
   of stream pointers into (*PFPS).  Allocate the array.  Return the
   number of successfully opened files, setting errno if this value is
   less than NFILES.  */

static size_t
open_input_files (struct sortfile *files, size_t nfiles, FILE ***pfps)
{
  FILE **fps = *pfps = xnmalloc (nfiles, sizeof *fps);
  int i;

  /* Open as many input files as we can.  */
  for (i = 0; i < nfiles; i++)
    {
      fps[i] = (files[i].pid
                ? open_temp (files[i].name, files[i].pid)
                : stream_open (files[i].name, "r"));
      if (!fps[i])
        break;
    }

  return i;
}

/* Merge lines from FILES onto OFP.  NTEMPS is the number of temporary
   files (all of which are at the start of the FILES array), and
   NFILES is the number of files; 0 <= NTEMPS <= NFILES <= NMERGE.
   FPS is the vector of open stream corresponding to the files.
   Close input and output streams before returning.
   OUTPUT_FILE gives the name of the output file.  If it is NULL,
   the output file is standard output.  */

static void
mergefps (struct sortfile *files, size_t ntemps, size_t nfiles,
          FILE *ofp, char const *output_file, FILE **fps)
{
  struct buffer *buffer = xnmalloc (nfiles, sizeof *buffer);
                                /* Input buffers for each file. */
  struct line saved;            /* Saved line storage for unique check. */
  struct line const *savedline = NULL;
                                /* &saved if there is a saved line. */
  size_t savealloc = 0;         /* Size allocated for the saved line. */
  struct line const **cur = xnmalloc (nfiles, sizeof *cur);
                                /* Current line in each line table. */
  struct line const **base = xnmalloc (nfiles, sizeof *base);
                                /* Base of each line table.  */
  size_t *ord = xnmalloc (nfiles, sizeof *ord);
                                /* Table representing a permutation of fps,
                                   such that cur[ord[0]] is the smallest line
                                   and will be next output. */
  size_t i;
  size_t j;
  size_t t;
  struct keyfield const *key = keylist;
  saved.text = NULL;

  /* Read initial lines from each input file. */
  for (i = 0; i < nfiles; )
    {
      initbuf (&buffer[i], sizeof (struct line),
               MAX (merge_buffer_size, sort_size / nfiles));
      if (fillbuf (&buffer[i], fps[i], files[i].name))
        {
          struct line const *linelim = buffer_linelim (&buffer[i]);
          cur[i] = linelim - 1;
          base[i] = linelim - buffer[i].nlines;
          i++;
        }
      else
        {
          /* fps[i] is empty; eliminate it from future consideration.  */
          xfclose (fps[i], files[i].name);
          if (i < ntemps)
            {
              ntemps--;
              zaptemp (files[i].name);
            }
          free (buffer[i].buf);
          --nfiles;
          for (j = i; j < nfiles; ++j)
            {
              files[j] = files[j + 1];
              fps[j] = fps[j + 1];
            }
        }
    }

  /* Set up the ord table according to comparisons among input lines.
     Since this only reorders two items if one is strictly greater than
     the other, it is stable. */
  for (i = 0; i < nfiles; ++i)
    ord[i] = i;
  for (i = 1; i < nfiles; ++i)
    if (0 < compare (cur[ord[i - 1]], cur[ord[i]], false))
      t = ord[i - 1], ord[i - 1] = ord[i], ord[i] = t, i = 0;

  /* Repeatedly output the smallest line until no input remains. */
  while (nfiles)
    {
      struct line const *smallest = cur[ord[0]];

      /* If uniquified output is turned on, output only the first of
         an identical series of lines. */
      if (unique)
        {
          if (savedline && compare (savedline, smallest, false))
            {
              savedline = NULL;
              write_bytes (&saved, ofp, output_file);
            }
          if (!savedline)
            {
              savedline = &saved;
              if (savealloc < smallest->length)
                {
                  do
                    if (! savealloc)
                      {
                        savealloc = smallest->length;
                        break;
                      }
                  while ((savealloc *= 2) < smallest->length);

                  saved.text = xrealloc (saved.text, savealloc);
                }
              saved.length = smallest->length;
              memcpy (saved.text, smallest->text, saved.length);
              if (key)
                {
                  saved.keybeg =
                    saved.text + (smallest->keybeg - smallest->text);
                  saved.keylim =
                    saved.text + (smallest->keylim - smallest->text);
                }
            }
        }
      else
        write_bytes (smallest, ofp, output_file);

      /* Check if we need to read more lines into core. */
      if (base[ord[0]] < smallest)
        cur[ord[0]] = smallest - 1;
      else
        {
          if (fillbuf (&buffer[ord[0]], fps[ord[0]], files[ord[0]].name))
            {
              struct line const *linelim = buffer_linelim (&buffer[ord[0]]);
              cur[ord[0]] = linelim - 1;
              base[ord[0]] = linelim - buffer[ord[0]].nlines;
            }
          else
            {
              /* We reached EOF on fps[ord[0]].  */
              for (i = 1; i < nfiles; ++i)
                if (ord[i] > ord[0])
                  --ord[i];
              --nfiles;
              xfclose (fps[ord[0]], files[ord[0]].name);
              if (ord[0] < ntemps)
                {
                  ntemps--;
                  zaptemp (files[ord[0]].name);
                }
              free (buffer[ord[0]].buf);
              for (i = ord[0]; i < nfiles; ++i)
                {
                  fps[i] = fps[i + 1];
                  files[i] = files[i + 1];
                  buffer[i] = buffer[i + 1];
                  cur[i] = cur[i + 1];
                  base[i] = base[i + 1];
                }
              for (i = 0; i < nfiles; ++i)
                ord[i] = ord[i + 1];
              continue;
            }
        }

      /* The new line just read in may be larger than other lines
         already in main memory; push it back in the queue until we
         encounter a line larger than it.  Optimize for the common
         case where the new line is smallest.  */
      {
        size_t lo = 1;
        size_t hi = nfiles;
        size_t probe = lo;
        size_t ord0 = ord[0];
        size_t count_of_smaller_lines;

        while (lo < hi)
          {
            int cmp = compare (cur[ord0], cur[ord[probe]], false);
            if (cmp < 0 || (cmp == 0 && ord0 < ord[probe]))
              hi = probe;
            else
              lo = probe + 1;
            probe = (lo + hi) / 2;
          }

        count_of_smaller_lines = lo - 1;
        for (j = 0; j < count_of_smaller_lines; j++)
          ord[j] = ord[j + 1];
        ord[count_of_smaller_lines] = ord0;
      }

      /* Free up some resources every once in a while.  */
      if (MAX_PROCS_BEFORE_REAP < nprocs)
        reap_some ();
    }

  if (unique && savedline)
    {
      write_bytes (&saved, ofp, output_file);
      free (saved.text);
    }

  xfclose (ofp, output_file);
  free (fps);
  free (buffer);
  free (ord);
  free (base);
  free (cur);
}

/* Merge lines from FILES onto OFP.  NTEMPS is the number of temporary
   files (all of which are at the start of the FILES array), and
   NFILES is the number of files; 0 <= NTEMPS <= NFILES <= NMERGE.
   Close input and output files before returning.
   OUTPUT_FILE gives the name of the output file.

   Return the number of files successfully merged.  This number can be
   less than NFILES if we ran low on file descriptors, but in this
   case it is never less than 2.  */

static size_t
mergefiles (struct sortfile *files, size_t ntemps, size_t nfiles,
            FILE *ofp, char const *output_file)
{
  FILE **fps;
  size_t nopened = open_input_files (files, nfiles, &fps);
  if (nopened < nfiles && nopened < 2)
    die (_("open failed"), files[nopened].name);
  mergefps (files, ntemps, nopened, ofp, output_file, fps);
  return nopened;
}

/* Merge into T (of size NLINES) the two sorted arrays of lines
   LO (with NLINES / 2 members), and
   T - (NLINES / 2) (with NLINES - NLINES / 2 members).
   T and LO point just past their respective arrays, and the arrays
   are in reverse order.  NLINES must be at least 2.  */

static void
mergelines (struct line *restrict t, size_t nlines,
            struct line const *restrict lo)
{
  size_t nlo = nlines / 2;
  size_t nhi = nlines - nlo;
  struct line *hi = t - nlo;

  while (true)
    if (compare (lo - 1, hi - 1, false) <= 0)
      {
        *--t = *--lo;
        if (! --nlo)
          {
            /* HI must equal T now, and there is no need to copy from
               HI to T. */
            return;
          }
      }
    else
      {
        *--t = *--hi;
        if (! --nhi)
          {
            do
              *--t = *--lo;
            while (--nlo);

            return;
          }
      }
}

/* Sort the array LINES with NLINES members, using TEMP for temporary space.
   NLINES must be at least 2.
   The input and output arrays are in reverse order, and LINES and
   TEMP point just past the end of their respective arrays.

   Use a recursive divide-and-conquer algorithm, in the style
   suggested by Knuth volume 3 (2nd edition), exercise 5.2.4-23.  Use
   the optimization suggested by exercise 5.2.4-10; this requires room
   for only 1.5*N lines, rather than the usual 2*N lines.  Knuth
   writes that this memory optimization was originally published by
   D. A. Bell, Comp J. 1 (1958), 75.  */

static void
sequential_sort (struct line *restrict lines, size_t nlines,
                 struct line *restrict temp, bool to_temp)
{
  if (nlines == 2)
    {
      /* Declare `swap' as int, not bool, to work around a bug
         <http://lists.gnu.org/archive/html/bug-coreutils/2005-10/msg00086.html>
         in the IBM xlc 6.0.0.0 compiler in 64-bit mode.  */
      int swap = (0 < compare (&lines[-1], &lines[-2], false));
      if (to_temp)
        {
          temp[-1] = lines[-1 - swap];
          temp[-2] = lines[-2 + swap];
        }
      else if (swap)
        {
          temp[-1] = lines[-1];
          lines[-1] = lines[-2];
          lines[-2] = temp[-1];
        }
    }
  else
    {
      size_t nlo = nlines / 2;
      size_t nhi = nlines - nlo;
      struct line *lo = lines;
      struct line *hi = lines - nlo;

      sequential_sort (hi, nhi, temp - (to_temp ? nlo : 0), to_temp);
      if (1 < nlo)
        sequential_sort (lo, nlo, temp, !to_temp);
      else if (!to_temp)
        temp[-1] = lo[-1];

      struct line *dest;
      struct line const *sorted_lo;
      if (to_temp)
        {
          dest = temp;
          sorted_lo = lines;
        }
      else
        {
          dest = lines;
          sorted_lo = temp;
        }
      mergelines (dest, nlines, sorted_lo);
    }
}

/* Compare two NODEs for priority. The NODE with the higher (numerically
   lower) level has priority. If tie, the NODE with the most remaining
   lines has priority. */

static int
compare_nodes (void const *a, void const *b)
{
  struct merge_node const *nodea = a;
  struct merge_node const *nodeb = b;
  if (nodea->level == nodeb->level)
      return (nodea->nlo + nodea->nhi) < (nodeb->nlo + nodeb->nhi);
  return nodea->level < nodeb->level;
}

/* Lock a merge tree NODE.
   Note spin locks were seen to perform better than mutexes
   as long as the number of threads is limited to the
   number of processors.  */

static inline void
lock_node (struct merge_node *node)
{
  pthread_spin_lock (node->lock);
}

/* Unlock a merge tree NODE. */

static inline void
unlock_node (struct merge_node *node)
{
  pthread_spin_unlock (node->lock);
}

/* Destroy merge QUEUE. */

static void
queue_destroy (struct merge_node_queue *queue)
{
  heap_free (queue->priority_queue);
  pthread_cond_destroy (&queue->cond);
  pthread_mutex_destroy (&queue->mutex);
}

/* Initialize merge QUEUE, allocating space for a maximum of RESERVE nodes.
   Though it's highly unlikely all nodes are in the heap at the same time,
   RESERVE should accommodate all of them. Counting a NULL dummy head for the
   heap, RESERVE should be 2 * NTHREADS. */

static void
queue_init (struct merge_node_queue *queue, size_t reserve)
{
  queue->priority_queue = heap_alloc (compare_nodes, reserve);
  pthread_mutex_init (&queue->mutex, NULL);
  pthread_cond_init (&queue->cond, NULL);
}

/* Insert NODE into priority QUEUE. Assume caller either holds lock on NODE
   or does not need to lock NODE. */

static void
queue_insert (struct merge_node_queue *queue, struct merge_node *node)
{
  pthread_mutex_lock (&queue->mutex);
  heap_insert (queue->priority_queue, node);
  node->queued = true;
  pthread_mutex_unlock (&queue->mutex);
  pthread_cond_signal (&queue->cond);
}

/* Pop NODE off priority QUEUE. Guarantee a non-null, spinlocked NODE. */

static struct merge_node *
queue_pop (struct merge_node_queue *queue)
{
  struct merge_node *node;
  pthread_mutex_lock (&queue->mutex);
  while (! (node = heap_remove_top (queue->priority_queue)))
    pthread_cond_wait (&queue->cond, &queue->mutex);
  pthread_mutex_unlock (&queue->mutex);
  lock_node (node);
  node->queued = false;
  return node;
}

/* If UNQIUE is set, checks to make sure line isn't a duplicate before
   outputting. If UNIQUE is not set, output the passed in line. Note that
   this function does not actually save the line, nor any key information,
   thus is only appropriate for internal sort. */

static void
write_unique (struct line const *line, FILE *tfp, char const *temp_output)
{
  static struct line const *saved = NULL;

  if (!unique)
    write_bytes (line, tfp, temp_output);
  else if (!saved || compare (line, saved, false))
    {
      saved = line;
      write_bytes (line, tfp, temp_output);
    }
}

/* Merge the lines currently available to a NODE in the binary
   merge tree, up to a maximum specified by MAX_MERGE. */

static size_t
mergelines_node (struct merge_node *restrict node, size_t total_lines,
                 FILE *tfp, char const *temp_output)
{
  struct line *lo_orig = node->lo;
  struct line *hi_orig = node->hi;
  size_t to_merge = MAX_MERGE (total_lines, node->level);
  size_t merged_lo;
  size_t merged_hi;

  if (node->level > MERGE_ROOT)
    {
      /* Merge to destination buffer. */
      struct line *dest = *node->dest;
      while (node->lo != node->end_lo && node->hi != node->end_hi && to_merge--)
        if (compare (node->lo - 1, node->hi - 1, false) <= 0)
          *--dest = *--node->lo;
        else
          *--dest = *--node->hi;

      merged_lo = lo_orig - node->lo;
      merged_hi = hi_orig - node->hi;

      if (node->nhi == merged_hi)
        while (node->lo != node->end_lo && to_merge--)
          *--dest = *--node->lo;
      else if (node->nlo == merged_lo)
        while (node->hi != node->end_hi && to_merge--)
          *--dest = *--node->hi;
    }
  else
    {
      /* Merge directly to output. */
      while (node->lo != node->end_lo && node->hi != node->end_hi && to_merge--)
        {
          if (compare (node->lo - 1, node->hi - 1, false) <= 0)
            write_unique (--node->lo, tfp, temp_output);
          else
            write_unique (--node->hi, tfp, temp_output);
        }

      merged_lo = lo_orig - node->lo;
      merged_hi = hi_orig - node->hi;

      if (node->nhi == merged_hi)
        {
          while (node->lo != node->end_lo && to_merge--)
            write_unique (--node->lo, tfp, temp_output);
        }
      else if (node->nlo == merged_lo)
        {
          while (node->hi != node->end_hi && to_merge--)
            write_unique (--node->hi, tfp, temp_output);
        }
      node->dest -= lo_orig - node->lo + hi_orig - node->hi;
    }

  /* Update NODE. */
  merged_lo = lo_orig - node->lo;
  merged_hi = hi_orig - node->hi;
  node->nlo -= merged_lo;
  node->nhi -= merged_hi;
  return merged_lo + merged_hi;
}

/* Insert NODE into QUEUE if it passes insertion checks. */

static void
check_insert (struct merge_node *node, struct merge_node_queue *queue)
{
  size_t lo_avail = node->lo - node->end_lo;
  size_t hi_avail = node->hi - node->end_hi;

  /* Conditions for insertion:
     1. NODE is not already in heap.
     2. NODE has available lines from both it's children, OR one child has
          available lines, but the other has exhausted all its lines. */
  if ((!node->queued)
      && ((lo_avail && (hi_avail || !(node->nhi)))
          || (hi_avail && !(node->nlo))))
    {
      queue_insert (queue, node);
    }
}

/* Update parent merge tree NODE. */

static void
update_parent (struct merge_node *node, size_t merged,
               struct merge_node_queue *queue)
{
  if (node->level > MERGE_ROOT)
    {
      lock_node (node->parent);
      *node->dest -= merged;
      check_insert (node->parent, queue);
      unlock_node (node->parent);
    }
  else if (node->nlo + node->nhi == 0)
    {
      /* If the MERGE_ROOT NODE has finished merging, insert the
         MERGE_END node.  */
      queue_insert (queue, node->parent);
    }
}

/* Repeatedly pop QUEUE for a NODE with lines to merge, and merge at least
   some of those lines, until the MERGE_END node is popped. */

static void
merge_loop (struct merge_node_queue *queue,
            size_t total_lines, FILE *tfp, char const *temp_output)
{
  while (1)
    {
      struct merge_node *node = queue_pop (queue);

      if (node->level == MERGE_END)
        {
          unlock_node (node);
          /* Reinsert so other threads can pop it. */
          queue_insert (queue, node);
          break;
        }
      size_t merged_lines = mergelines_node (node, total_lines, tfp,
                                             temp_output);
      check_insert (node, queue);
      update_parent (node, merged_lines, queue);

      unlock_node (node);
    }
}


static void sortlines (struct line *restrict, struct line *restrict,
                       unsigned long int, size_t,
                       struct merge_node *, bool,
                       struct merge_node_queue *,
                       FILE *, char const *);

/* Thread arguments for sortlines_thread. */

struct thread_args
{
  struct line *lines;
  struct line *dest;
  unsigned long int nthreads;
  size_t const total_lines;
  struct merge_node *const parent;
  bool lo_child;
  struct merge_node_queue *const merge_queue;
  FILE *tfp;
  char const *output_temp;
};

/* Like sortlines, except with a signature acceptable to pthread_create.  */

static void *
sortlines_thread (void *data)
{
  struct thread_args const *args = data;
  sortlines (args->lines, args->dest, args->nthreads, args->total_lines,
             args->parent, args->lo_child, args->merge_queue,
             args->tfp, args->output_temp);
  return NULL;
}

/* There are three phases to the algorithm: node creation, sequential sort,
   and binary merge.

   During node creation, sortlines recursively visits each node in the
   binary merge tree and creates a NODE structure corresponding to all the
   future line merging NODE is responsible for. For each call to
   sortlines, half the available threads are assigned to each recursive
   call, until a leaf node having only 1 available thread is reached.

   Each leaf node then performs two sequential sorts, one on each half of
   the lines it is responsible for. It records in its NODE structure that
   there are two sorted sublists available to merge from, and inserts its
   NODE into the priority queue.

   The binary merge phase then begins. Each thread drops into a loop
   where the thread retrieves a NODE from the priority queue, merges lines
   available to that NODE, and potentially insert NODE or its parent back
   into the queue if there are sufficient available lines for them to
   merge. This continues until all lines at all nodes of the merge tree
   have been merged. */

static void
sortlines (struct line *restrict lines, struct line *restrict dest,
           unsigned long int nthreads, size_t total_lines,
           struct merge_node *parent, bool lo_child,
           struct merge_node_queue *merge_queue,
           FILE *tfp, char const *temp_output)
{
  /* Create merge tree NODE. */
  size_t nlines = (lo_child)? parent->nlo : parent->nhi;
  size_t nlo = nlines / 2;
  size_t nhi = nlines - nlo;
  struct line *lo = dest - total_lines;
  struct line *hi = lo - nlo;
  struct line **parent_end = (lo_child)? &parent->end_lo : &parent->end_hi;
  pthread_spinlock_t lock;
  pthread_spin_init (&lock, PTHREAD_PROCESS_PRIVATE);
  struct merge_node node = {lo, hi, lo, hi, parent_end, nlo, nhi,
                            parent->level + 1, parent, false, &lock};

  /* Calculate thread arguments. */
  unsigned long int lo_threads = nthreads / 2;
  unsigned long int hi_threads = nthreads - lo_threads;
  pthread_t thread;
  struct thread_args args = {lines, lo, lo_threads, total_lines, &node,
                             true, merge_queue, tfp, temp_output};

  if (nthreads > 1 && SUBTHREAD_LINES_HEURISTIC <= nlines
      && pthread_create (&thread, NULL, sortlines_thread, &args) == 0)
    {
      sortlines (lines - nlo, hi, hi_threads, total_lines, &node, false,
                 merge_queue, tfp, temp_output);
      pthread_join (thread, NULL);
    }
  else
    {
      /* Nthreads = 1, this is a leaf NODE, or pthread_create failed.
         Sort with 1 thread. */
      struct line *temp = lines - total_lines;
      if (1 < nhi)
        sequential_sort (lines - nlo, nhi, temp - nlo / 2, false);
      if (1 < nlo)
        sequential_sort (lines, nlo, temp, false);

      /* Update merge NODE. No need to lock yet. */
      node.lo = lines;
      node.hi = lines - nlo;
      node.end_lo = lines - nlo;
      node.end_hi = lines - nlo - nhi;

      queue_insert (merge_queue, &node);
      merge_loop (merge_queue, total_lines, tfp, temp_output);
    }
}

/* Scan through FILES[NTEMPS .. NFILES-1] looking for a file that is
   the same as OUTFILE.  If found, merge the found instances (and perhaps
   some other files) into a temporary file so that it can in turn be
   merged into OUTFILE without destroying OUTFILE before it is completely
   read.  Return the new value of NFILES, which differs from the old if
   some merging occurred.

   This test ensures that an otherwise-erroneous use like
   "sort -m -o FILE ... FILE ..." copies FILE before writing to it.
   It's not clear that POSIX requires this nicety.
   Detect common error cases, but don't try to catch obscure cases like
   "cat ... FILE ... | sort -m -o FILE"
   where traditional "sort" doesn't copy the input and where
   people should know that they're getting into trouble anyway.
   Catching these obscure cases would slow down performance in
   common cases.  */

static size_t
avoid_trashing_input (struct sortfile *files, size_t ntemps,
                      size_t nfiles, char const *outfile)
{
  size_t i;
  bool got_outstat = false;
  struct stat outstat;

  for (i = ntemps; i < nfiles; i++)
    {
      bool is_stdin = STREQ (files[i].name, "-");
      bool same;
      struct stat instat;

      if (outfile && STREQ (outfile, files[i].name) && !is_stdin)
        same = true;
      else
        {
          if (! got_outstat)
            {
              if ((outfile
                   ? stat (outfile, &outstat)
                   : fstat (STDOUT_FILENO, &outstat))
                  != 0)
                break;
              got_outstat = true;
            }

          same = (((is_stdin
                    ? fstat (STDIN_FILENO, &instat)
                    : stat (files[i].name, &instat))
                   == 0)
                  && SAME_INODE (instat, outstat));
        }

      if (same)
        {
          FILE *tftp;
          pid_t pid;
          char *temp = create_temp (&tftp, &pid);
          size_t num_merged = 0;
          do
            {
              num_merged += mergefiles (&files[i], 0, nfiles - i, tftp, temp);
              files[i].name = temp;
              files[i].pid = pid;

              if (i + num_merged < nfiles)
                memmove (&files[i + 1], &files[i + num_merged],
                         num_merged * sizeof *files);
              ntemps += 1;
              nfiles -= num_merged - 1;;
              i += num_merged;
            }
          while (i < nfiles);
        }
    }

  return nfiles;
}

/* Merge the input FILES.  NTEMPS is the number of files at the
   start of FILES that are temporary; it is zero at the top level.
   NFILES is the total number of files.  Put the output in
   OUTPUT_FILE; a null OUTPUT_FILE stands for standard output.  */

static void
merge (struct sortfile *files, size_t ntemps, size_t nfiles,
       char const *output_file)
{
  while (nmerge < nfiles)
    {
      /* Number of input files processed so far.  */
      size_t in;

      /* Number of output files generated so far.  */
      size_t out;

      /* nfiles % NMERGE; this counts input files that are left over
         after all full-sized merges have been done.  */
      size_t remainder;

      /* Number of easily-available slots at the next loop iteration.  */
      size_t cheap_slots;

      /* Do as many NMERGE-size merges as possible. In the case that
         nmerge is bogus, increment by the maximum number of file
         descriptors allowed.  */
      for (out = in = 0; nmerge <= nfiles - in; out++)
        {
          FILE *tfp;
          pid_t pid;
          char *temp = create_temp (&tfp, &pid);
          size_t num_merged = mergefiles (&files[in], MIN (ntemps, nmerge),
                                          nmerge, tfp, temp);
          ntemps -= MIN (ntemps, num_merged);
          files[out].name = temp;
          files[out].pid = pid;
          in += num_merged;
        }

      remainder = nfiles - in;
      cheap_slots = nmerge - out % nmerge;

      if (cheap_slots < remainder)
        {
          /* So many files remain that they can't all be put into the last
             NMERGE-sized output window.  Do one more merge.  Merge as few
             files as possible, to avoid needless I/O.  */
          size_t nshortmerge = remainder - cheap_slots + 1;
          FILE *tfp;
          pid_t pid;
          char *temp = create_temp (&tfp, &pid);
          size_t num_merged = mergefiles (&files[in], MIN (ntemps, nshortmerge),
                                          nshortmerge, tfp, temp);
          ntemps -= MIN (ntemps, num_merged);
          files[out].name = temp;
          files[out++].pid = pid;
          in += num_merged;
        }

      /* Put the remaining input files into the last NMERGE-sized output
         window, so they will be merged in the next pass.  */
      memmove (&files[out], &files[in], (nfiles - in) * sizeof *files);
      ntemps += out;
      nfiles -= in - out;
    }

  nfiles = avoid_trashing_input (files, ntemps, nfiles, output_file);

  /* We aren't guaranteed that this final mergefiles will work, therefore we
     try to merge into the output, and then merge as much as we can into a
     temp file if we can't. Repeat.  */

  while (true)
    {
      /* Merge directly into the output file if possible.  */
      FILE **fps;
      size_t nopened = open_input_files (files, nfiles, &fps);

      if (nopened == nfiles)
        {
          FILE *ofp = stream_open (output_file, "w");
          if (ofp)
            {
              mergefps (files, ntemps, nfiles, ofp, output_file, fps);
              break;
            }
          if (errno != EMFILE || nopened <= 2)
            die (_("open failed"), output_file);
        }
      else if (nopened <= 2)
        die (_("open failed"), files[nopened].name);

      /* We ran out of file descriptors.  Close one of the input
         files, to gain a file descriptor.  Then create a temporary
         file with our spare file descriptor.  Retry if that failed
         (e.g., some other process could open a file between the time
         we closed and tried to create).  */
      FILE *tfp;
      pid_t pid;
      char *temp;
      do
        {
          nopened--;
          xfclose (fps[nopened], files[nopened].name);
          temp = maybe_create_temp (&tfp, &pid, ! (nopened <= 2));
        }
      while (!temp);

      /* Merge into the newly allocated temporary.  */
      mergefps (&files[0], MIN (ntemps, nopened), nopened, tfp, temp, fps);
      ntemps -= MIN (ntemps, nopened);
      files[0].name = temp;
      files[0].pid = pid;

      memmove (&files[1], &files[nopened], (nfiles - nopened) * sizeof *files);
      ntemps++;
      nfiles -= nopened - 1;
    }
}

/* Sort NFILES FILES onto OUTPUT_FILE. */

static void
sort (char * const *files, size_t nfiles, char const *output_file,
      unsigned long int nthreads)
{
  struct buffer buf;
  size_t ntemps = 0;
  bool output_file_created = false;

  buf.alloc = 0;

  while (nfiles)
    {
      char const *temp_output;
      char const *file = *files;
      FILE *fp = xfopen (file, "r");
      FILE *tfp;

      size_t bytes_per_line;
      if (nthreads > 1)
        {
          /* Get log P. */
          unsigned long int tmp = 1;
          size_t mult = 1;
          while (tmp < nthreads)
            {
              tmp *= 2;
              mult++;
            }
          bytes_per_line = (mult * sizeof (struct line));
        }
      else
        bytes_per_line = sizeof (struct line) * 3 / 2;

      if (! buf.alloc)
        initbuf (&buf, bytes_per_line,
                 sort_buffer_size (&fp, 1, files, nfiles, bytes_per_line));
      buf.eof = false;
      files++;
      nfiles--;

      while (fillbuf (&buf, fp, file))
        {
          struct line *line;

          if (buf.eof && nfiles
              && (bytes_per_line + 1
                  < (buf.alloc - buf.used - bytes_per_line * buf.nlines)))
            {
              /* End of file, but there is more input and buffer room.
                 Concatenate the next input file; this is faster in
                 the usual case.  */
              buf.left = buf.used;
              break;
            }

          line = buffer_linelim (&buf);
          if (buf.eof && !nfiles && !ntemps && !buf.left)
            {
              xfclose (fp, file);
              tfp = xfopen (output_file, "w");
              temp_output = output_file;
              output_file_created = true;
            }
          else
            {
              ++ntemps;
              temp_output = create_temp (&tfp, NULL);
            }
          if (1 < buf.nlines)
            {
              struct merge_node_queue merge_queue;
              queue_init (&merge_queue, 2 * nthreads);

              pthread_spinlock_t lock;
              pthread_spin_init (&lock, PTHREAD_PROCESS_PRIVATE);
              struct merge_node node =
                {NULL, NULL, NULL, NULL, NULL, buf.nlines,
                 buf.nlines, MERGE_END, NULL, false, &lock};

              sortlines (line, line, nthreads, buf.nlines, &node, true,
                         &merge_queue, tfp, temp_output);
              queue_destroy (&merge_queue);
            }
          else
            write_unique (line - 1, tfp, temp_output);

          xfclose (tfp, temp_output);

          /* Free up some resources every once in a while.  */
          if (MAX_PROCS_BEFORE_REAP < nprocs)
            reap_some ();

          if (output_file_created)
            goto finish;
        }
      xfclose (fp, file);
    }

 finish:
  free (buf.buf);

  if (! output_file_created)
    {
      size_t i;
      struct tempnode *node = temphead;
      struct sortfile *tempfiles = xnmalloc (ntemps, sizeof *tempfiles);
      for (i = 0; node; i++)
        {
          tempfiles[i].name = node->name;
          tempfiles[i].pid = node->pid;
          node = node->next;
        }
      merge (tempfiles, ntemps, ntemps, output_file);
      free (tempfiles);
    }
}

/* Insert a malloc'd copy of key KEY_ARG at the end of the key list.  */

static void
insertkey (struct keyfield *key_arg)
{
  struct keyfield **p;
  struct keyfield *key = xmemdup (key_arg, sizeof *key);

  for (p = &keylist; *p; p = &(*p)->next)
    continue;
  *p = key;
  key->next = NULL;
}

/* Report a bad field specification SPEC, with extra info MSGID.  */

static void badfieldspec (char const *, char const *)
     ATTRIBUTE_NORETURN;
static void
badfieldspec (char const *spec, char const *msgid)
{
  error (SORT_FAILURE, 0, _("%s: invalid field specification %s"),
         _(msgid), quote (spec));
  abort ();
}

/* Report incompatible options.  */

static void incompatible_options (char const *) ATTRIBUTE_NORETURN;
static void
incompatible_options (char const *opts)
{
  error (SORT_FAILURE, 0, _("options `-%s' are incompatible"), opts);
  abort ();
}

/* Check compatibility of ordering options.  */

static void
check_ordering_compatibility (void)
{
  struct keyfield *key;

  for (key = keylist; key; key = key->next)
    if ((1 < (key->random + key->numeric + key->general_numeric + key->month
              + key->version + !!key->ignore + key->human_numeric))
        || (key->random && key->translate))
      {
        /* The following is too big, but guaranteed to be "big enough".  */
        char opts[sizeof short_options];
        /* Clear flags we're not interested in.  */
        key->skipsblanks = key->skipeblanks = key->reverse = false;
        key_to_opts (key, opts);
        incompatible_options (opts);
      }
}

/* Parse the leading integer in STRING and store the resulting value
   (which must fit into size_t) into *VAL.  Return the address of the
   suffix after the integer.  If the value is too large, silently
   substitute SIZE_MAX.  If MSGID is NULL, return NULL after
   failure; otherwise, report MSGID and exit on failure.  */

static char const *
parse_field_count (char const *string, size_t *val, char const *msgid)
{
  char *suffix;
  uintmax_t n;

  switch (xstrtoumax (string, &suffix, 10, &n, ""))
    {
    case LONGINT_OK:
    case LONGINT_INVALID_SUFFIX_CHAR:
      *val = n;
      if (*val == n)
        break;
      /* Fall through.  */
    case LONGINT_OVERFLOW:
    case LONGINT_OVERFLOW | LONGINT_INVALID_SUFFIX_CHAR:
      *val = SIZE_MAX;
      break;

    case LONGINT_INVALID:
      if (msgid)
        error (SORT_FAILURE, 0, _("%s: invalid count at start of %s"),
               _(msgid), quote (string));
      return NULL;
    }

  return suffix;
}

/* Handle interrupts and hangups. */

static void
sighandler (int sig)
{
  if (! SA_NOCLDSTOP)
    signal (sig, SIG_IGN);

  cleanup ();

  signal (sig, SIG_DFL);
  raise (sig);
}

/* Set the ordering options for KEY specified in S.
   Return the address of the first character in S that
   is not a valid ordering option.
   BLANKTYPE is the kind of blanks that 'b' should skip. */

static char *
set_ordering (char const *s, struct keyfield *key, enum blanktype blanktype)
{
  while (*s)
    {
      switch (*s)
        {
        case 'b':
          if (blanktype == bl_start || blanktype == bl_both)
            key->skipsblanks = true;
          if (blanktype == bl_end || blanktype == bl_both)
            key->skipeblanks = true;
          break;
        case 'd':
          key->ignore = nondictionary;
          break;
        case 'f':
          key->translate = fold_toupper;
          break;
        case 'g':
          key->general_numeric = true;
          break;
        case 'h':
          key->human_numeric = true;
          break;
        case 'i':
          /* Option order should not matter, so don't let -i override
             -d.  -d implies -i, but -i does not imply -d.  */
          if (! key->ignore)
            key->ignore = nonprinting;
          break;
        case 'M':
          key->month = true;
          break;
        case 'n':
          key->numeric = true;
          break;
        case 'R':
          key->random = true;
          break;
        case 'r':
          key->reverse = true;
          break;
        case 'V':
          key->version = true;
          break;
        default:
          return (char *) s;
        }
      ++s;
    }
  return (char *) s;
}

static struct keyfield *
key_init (struct keyfield *key)
{
  memset (key, 0, sizeof *key);
  key->eword = SIZE_MAX;
  return key;
}

int
main (int argc, char **argv)
{
  struct keyfield *key;
  struct keyfield key_buf;
  struct keyfield gkey;
  bool gkey_only = false;
  char const *s;
  int c = 0;
  char checkonly = 0;
  bool mergeonly = false;
  char *random_source = NULL;
  bool need_random = false;
  unsigned long int nthreads = 0;
  size_t nfiles = 0;
  bool posixly_correct = (getenv ("POSIXLY_CORRECT") != NULL);
  bool obsolete_usage = (posix2_version () < 200112);
  char **files;
  char *files_from = NULL;
  struct Tokens tok;
  char const *outfile = NULL;

  initialize_main (&argc, &argv);
  set_program_name (argv[0]);
  setlocale (LC_ALL, "");
  bindtextdomain (PACKAGE, LOCALEDIR);
  textdomain (PACKAGE);

  initialize_exit_failure (SORT_FAILURE);

  hard_LC_COLLATE = hard_locale (LC_COLLATE);
#if HAVE_NL_LANGINFO
  hard_LC_TIME = hard_locale (LC_TIME);
#endif

  /* Get locale's representation of the decimal point.  */
  {
    struct lconv const *locale = localeconv ();

    /* If the locale doesn't define a decimal point, or if the decimal
       point is multibyte, use the C locale's decimal point.  FIXME:
       add support for multibyte decimal points.  */
    decimal_point = to_uchar (locale->decimal_point[0]);
    if (! decimal_point || locale->decimal_point[1])
      decimal_point = '.';

    /* FIXME: add support for multibyte thousands separators.  */
    thousands_sep = to_uchar (*locale->thousands_sep);
    if (! thousands_sep || locale->thousands_sep[1])
      thousands_sep = -1;
  }

  have_read_stdin = false;
  inittables ();

  {
    size_t i;
    static int const sig[] =
      {
        /* The usual suspects.  */
        SIGALRM, SIGHUP, SIGINT, SIGPIPE, SIGQUIT, SIGTERM,
#ifdef SIGPOLL
        SIGPOLL,
#endif
#ifdef SIGPROF
        SIGPROF,
#endif
#ifdef SIGVTALRM
        SIGVTALRM,
#endif
#ifdef SIGXCPU
        SIGXCPU,
#endif
#ifdef SIGXFSZ
        SIGXFSZ,
#endif
      };
    enum { nsigs = ARRAY_CARDINALITY (sig) };

#if SA_NOCLDSTOP
    struct sigaction act;

    sigemptyset (&caught_signals);
    for (i = 0; i < nsigs; i++)
      {
        sigaction (sig[i], NULL, &act);
        if (act.sa_handler != SIG_IGN)
          sigaddset (&caught_signals, sig[i]);
      }

    act.sa_handler = sighandler;
    act.sa_mask = caught_signals;
    act.sa_flags = 0;

    for (i = 0; i < nsigs; i++)
      if (sigismember (&caught_signals, sig[i]))
        sigaction (sig[i], &act, NULL);
#else
    for (i = 0; i < nsigs; i++)
      if (signal (sig[i], SIG_IGN) != SIG_IGN)
        {
          signal (sig[i], sighandler);
          siginterrupt (sig[i], 1);
        }
#endif
  }
  signal (SIGCHLD, SIG_DFL); /* Don't inherit CHLD handling from parent.  */

  /* The signal mask is known, so it is safe to invoke exit_cleanup.  */
  atexit (exit_cleanup);

  key_init (&gkey);
  gkey.sword = SIZE_MAX;

  files = xnmalloc (argc, sizeof *files);

  while (true)
    {
      /* Parse an operand as a file after "--" was seen; or if
         pedantic and a file was seen, unless the POSIX version
         predates 1003.1-2001 and -c was not seen and the operand is
         "-o FILE" or "-oFILE".  */
      int oi = -1;

      if (c == -1
          || (posixly_correct && nfiles != 0
              && ! (obsolete_usage
                    && ! checkonly
                    && optind != argc
                    && argv[optind][0] == '-' && argv[optind][1] == 'o'
                    && (argv[optind][2] || optind + 1 != argc)))
          || ((c = getopt_long (argc, argv, short_options,
                                long_options, &oi))
              == -1))
        {
          if (argc <= optind)
            break;
          files[nfiles++] = argv[optind++];
        }
      else switch (c)
        {
        case 1:
          key = NULL;
          if (optarg[0] == '+')
            {
              bool minus_pos_usage = (optind != argc && argv[optind][0] == '-'
                                      && ISDIGIT (argv[optind][1]));
              obsolete_usage |= minus_pos_usage && !posixly_correct;
              if (obsolete_usage)
                {
                  /* Treat +POS1 [-POS2] as a key if possible; but silently
                     treat an operand as a file if it is not a valid +POS1.  */
                  key = key_init (&key_buf);
                  s = parse_field_count (optarg + 1, &key->sword, NULL);
                  if (s && *s == '.')
                    s = parse_field_count (s + 1, &key->schar, NULL);
                  if (! (key->sword || key->schar))
                    key->sword = SIZE_MAX;
                  if (! s || *set_ordering (s, key, bl_start))
                    key = NULL;
                  else
                    {
                      if (minus_pos_usage)
                        {
                          char const *optarg1 = argv[optind++];
                          s = parse_field_count (optarg1 + 1, &key->eword,
                                             N_("invalid number after `-'"));
                          if (*s == '.')
                            s = parse_field_count (s + 1, &key->echar,
                                               N_("invalid number after `.'"));
                          if (!key->echar && key->eword)
                            {
                              /* obsolescent syntax +A.x -B.y is equivalent to:
                                   -k A+1.x+1,B.y   (when y = 0)
                                   -k A+1.x+1,B+1.y (when y > 0)
                                 So eword is decremented as in the -k case
                                 only when the end field (B) is specified and
                                 echar (y) is 0.  */
                              key->eword--;
                            }
                          if (*set_ordering (s, key, bl_end))
                            badfieldspec (optarg1,
                                      N_("stray character in field spec"));
                        }
                      key->obsolete_used = true;
                      insertkey (key);
                    }
                }
            }
          if (! key)
            files[nfiles++] = optarg;
          break;

        case SORT_OPTION:
          c = XARGMATCH ("--sort", optarg, sort_args, sort_types);
          /* Fall through. */
        case 'b':
        case 'd':
        case 'f':
        case 'g':
        case 'h':
        case 'i':
        case 'M':
        case 'n':
        case 'r':
        case 'R':
        case 'V':
          {
            char str[2];
            str[0] = c;
            str[1] = '\0';
            set_ordering (str, &gkey, bl_both);
          }
          break;

        case CHECK_OPTION:
          c = (optarg
               ? XARGMATCH ("--check", optarg, check_args, check_types)
               : 'c');
          /* Fall through.  */
        case 'c':
        case 'C':
          if (checkonly && checkonly != c)
            incompatible_options ("cC");
          checkonly = c;
          break;

        case COMPRESS_PROGRAM_OPTION:
          if (compress_program && !STREQ (compress_program, optarg))
            error (SORT_FAILURE, 0, _("multiple compress programs specified"));
          compress_program = optarg;
          break;

        case DEBUG_PROGRAM_OPTION:
          debug = true;
          break;

        case FILES0_FROM_OPTION:
          files_from = optarg;
          break;

        case 'k':
          key = key_init (&key_buf);

          /* Get POS1. */
          s = parse_field_count (optarg, &key->sword,
                                 N_("invalid number at field start"));
          if (! key->sword--)
            {
              /* Provoke with `sort -k0' */
              badfieldspec (optarg, N_("field number is zero"));
            }
          if (*s == '.')
            {
              s = parse_field_count (s + 1, &key->schar,
                                     N_("invalid number after `.'"));
              if (! key->schar--)
                {
                  /* Provoke with `sort -k1.0' */
                  badfieldspec (optarg, N_("character offset is zero"));
                }
            }
          if (! (key->sword || key->schar))
            key->sword = SIZE_MAX;
          s = set_ordering (s, key, bl_start);
          if (*s != ',')
            {
              key->eword = SIZE_MAX;
              key->echar = 0;
            }
          else
            {
              /* Get POS2. */
              s = parse_field_count (s + 1, &key->eword,
                                     N_("invalid number after `,'"));
              if (! key->eword--)
                {
                  /* Provoke with `sort -k1,0' */
                  badfieldspec (optarg, N_("field number is zero"));
                }
              if (*s == '.')
                {
                  s = parse_field_count (s + 1, &key->echar,
                                         N_("invalid number after `.'"));
                }
              s = set_ordering (s, key, bl_end);
            }
          if (*s)
            badfieldspec (optarg, N_("stray character in field spec"));
          insertkey (key);
          break;

        case 'm':
          mergeonly = true;
          break;

        case NMERGE_OPTION:
          specify_nmerge (oi, c, optarg);
          break;

        case 'o':
          if (outfile && !STREQ (outfile, optarg))
            error (SORT_FAILURE, 0, _("multiple output files specified"));
          outfile = optarg;
          break;

        case RANDOM_SOURCE_OPTION:
          if (random_source && !STREQ (random_source, optarg))
            error (SORT_FAILURE, 0, _("multiple random sources specified"));
          random_source = optarg;
          break;

        case 's':
          stable = true;
          break;

        case 'S':
          specify_sort_size (oi, c, optarg);
          break;

        case 't':
          {
            char newtab = optarg[0];
            if (! newtab)
              error (SORT_FAILURE, 0, _("empty tab"));
            if (optarg[1])
              {
                if (STREQ (optarg, "\\0"))
                  newtab = '\0';
                else
                  {
                    /* Provoke with `sort -txx'.  Complain about
                       "multi-character tab" instead of "multibyte tab", so
                       that the diagnostic's wording does not need to be
                       changed once multibyte characters are supported.  */
                    error (SORT_FAILURE, 0, _("multi-character tab %s"),
                           quote (optarg));
                  }
              }
            if (tab != TAB_DEFAULT && tab != newtab)
              error (SORT_FAILURE, 0, _("incompatible tabs"));
            tab = newtab;
          }
          break;

        case 'T':
          add_temp_dir (optarg);
          break;

        case PARALLEL_OPTION:
          nthreads = specify_nthreads (oi, c, optarg);
          break;

        case 'u':
          unique = true;
          break;

        case 'y':
          /* Accept and ignore e.g. -y0 for compatibility with Solaris 2.x
             through Solaris 7.  It is also accepted by many non-Solaris
             "sort" implementations, e.g., AIX 5.2, HP-UX 11i v2, IRIX 6.5.
             -y is marked as obsolete starting with Solaris 8 (1999), but is
             still accepted as of Solaris 10 prerelease (2004).

             Solaris 2.5.1 "sort -y 100" reads the input file "100", but
             emulate Solaris 8 and 9 "sort -y 100" which ignores the "100",
             and which in general ignores the argument after "-y" if it
             consists entirely of digits (it can even be empty).  */
          if (optarg == argv[optind - 1])
            {
              char const *p;
              for (p = optarg; ISDIGIT (*p); p++)
                continue;
              optind -= (*p != '\0');
            }
          break;

        case 'z':
          eolchar = 0;
          break;

        case_GETOPT_HELP_CHAR;

        case_GETOPT_VERSION_CHAR (PROGRAM_NAME, AUTHORS);

        default:
          usage (SORT_FAILURE);
        }
    }

  if (files_from)
    {
      FILE *stream;

      /* When using --files0-from=F, you may not specify any files
         on the command-line.  */
      if (nfiles)
        {
          error (0, 0, _("extra operand %s"), quote (files[0]));
          fprintf (stderr, "%s\n",
                   _("file operands cannot be combined with --files0-from"));
          usage (SORT_FAILURE);
        }

      if (STREQ (files_from, "-"))
        stream = stdin;
      else
        {
          stream = fopen (files_from, "r");
          if (stream == NULL)
            error (SORT_FAILURE, errno, _("cannot open %s for reading"),
                   quote (files_from));
        }

      readtokens0_init (&tok);

      if (! readtokens0 (stream, &tok) || fclose (stream) != 0)
        error (SORT_FAILURE, 0, _("cannot read file names from %s"),
               quote (files_from));

      if (tok.n_tok)
        {
          size_t i;
          free (files);
          files = tok.tok;
          nfiles = tok.n_tok;
          for (i = 0; i < nfiles; i++)
          {
              if (STREQ (files[i], "-"))
                error (SORT_FAILURE, 0, _("when reading file names from stdin, "
                                          "no file name of %s allowed"),
                       quote (files[i]));
              else if (files[i][0] == '\0')
                {
                  /* Using the standard `filename:line-number:' prefix here is
                     not totally appropriate, since NUL is the separator, not NL,
                     but it might be better than nothing.  */
                  unsigned long int file_number = i + 1;
                  error (SORT_FAILURE, 0,
                         _("%s:%lu: invalid zero-length file name"),
                         quotearg_colon (files_from), file_number);
                }
          }
        }
      else
        error (SORT_FAILURE, 0, _("no input from %s"),
               quote (files_from));
    }

  /* Inheritance of global options to individual keys. */
  for (key = keylist; key; key = key->next)
    {
      if (default_key_compare (key) && !key->reverse)
        {
          key->ignore = gkey.ignore;
          key->translate = gkey.translate;
          key->skipsblanks = gkey.skipsblanks;
          key->skipeblanks = gkey.skipeblanks;
          key->month = gkey.month;
          key->numeric = gkey.numeric;
          key->general_numeric = gkey.general_numeric;
          key->human_numeric = gkey.human_numeric;
          key->version = gkey.version;
          key->random = gkey.random;
          key->reverse = gkey.reverse;
        }

      need_random |= key->random;
    }

  if (!keylist && !default_key_compare (&gkey))
    {
      gkey_only = true;
      insertkey (&gkey);
      need_random |= gkey.random;
    }

  check_ordering_compatibility ();

  /* Disable this combination so that users are less likely
     to inadvertantly update a file with debugging enabled.
     Also it simplifies the code for handling temp files.  */
  if (debug && outfile)
    error (SORT_FAILURE, 0, _("options -o and --debug are incompatible"));

  if (debug)
    {
      /* Always output the locale in debug mode, since this
         is such a common source of confusion.  */
      if (hard_LC_COLLATE)
        error (0, 0, _("using %s sorting rules"),
               quote (setlocale (LC_COLLATE, NULL)));
      else
        error (0, 0, _("using simple byte comparison"));
      key_warnings (&gkey, gkey_only);
    }

  reverse = gkey.reverse;

  if (need_random)
    random_md5_state_init (random_source);

  if (temp_dir_count == 0)
    {
      char const *tmp_dir = getenv ("TMPDIR");
      add_temp_dir (tmp_dir ? tmp_dir : DEFAULT_TMPDIR);
    }

  if (nfiles == 0)
    {
      static char *minus = (char *) "-";
      nfiles = 1;
      free (files);
      files = &minus;
    }

  /* Need to re-check that we meet the minimum requirement for memory
     usage with the final value for NMERGE. */
  if (0 < sort_size)
    sort_size = MAX (sort_size, MIN_SORT_SIZE);

  if (checkonly)
    {
      if (nfiles > 1)
        error (SORT_FAILURE, 0, _("extra operand %s not allowed with -%c"),
               quote (files[1]), checkonly);

      if (outfile)
        {
          static char opts[] = {0, 'o', 0};
          opts[0] = checkonly;
          incompatible_options (opts);
        }

      /* POSIX requires that sort return 1 IFF invoked with -c or -C and the
         input is not properly sorted.  */
      exit (check (files[0], checkonly) ? EXIT_SUCCESS : SORT_OUT_OF_ORDER);
    }

  if (mergeonly)
    {
      struct sortfile *sortfiles = xcalloc (nfiles, sizeof *sortfiles);
      size_t i;

      for (i = 0; i < nfiles; ++i)
        sortfiles[i].name = files[i];

      merge (sortfiles, 0, nfiles, outfile);
      IF_LINT (free (sortfiles));
    }
  else
    {
      /* If NTHREADS > number of cores on the machine, spinlocking
         could be wasteful.  */
      unsigned long int np2 = num_processors (NPROC_CURRENT_OVERRIDABLE);
      if (!nthreads || nthreads > np2)
        nthreads = np2;

      sort (files, nfiles, outfile, nthreads);
    }

  if (have_read_stdin && fclose (stdin) == EOF)
    die (_("close failed"), "-");

  exit (EXIT_SUCCESS);
}