1 /* sort - sort lines of text (with all kinds of options).
2 Copyright (C) 1988, 1991-2008 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>.
17 Written December 1988 by Mike Haertel.
18 The author may be reached (Email) at the address mike@gnu.ai.mit.edu,
19 or (US mail) as Mike Haertel c/o Free Software Foundation.
21 Ørn E. Hansen added NLS support in 1997. */
26 #include <sys/types.h>
32 #include "hard-locale.h"
40 #include "readtokens0.h"
43 #include "strnumcmp.h"
48 #if HAVE_SYS_RESOURCE_H
49 # include <sys/resource.h>
52 struct rlimit { size_t rlim_cur; };
53 # define getrlimit(Resource, Rlp) (-1)
56 /* The official name of this program (e.g., no `g' prefix). */
57 #define PROGRAM_NAME "sort"
60 proper_name ("Mike Haertel"), \
61 proper_name ("Paul Eggert")
63 #if HAVE_LANGINFO_CODESET
64 # include <langinfo.h>
67 /* Use SA_NOCLDSTOP as a proxy for whether the sigaction machinery is
70 # define SA_NOCLDSTOP 0
71 /* No sigprocmask. Always 'return' zero. */
72 # define sigprocmask(How, Set, Oset) (0)
74 # if ! HAVE_SIGINTERRUPT
75 # define siginterrupt(sig, flag) /* empty */
79 #if !defined OPEN_MAX && defined NR_OPEN
80 # define OPEN_MAX NR_OPEN
86 #define UCHAR_LIM (UCHAR_MAX + 1)
88 #ifndef DEFAULT_TMPDIR
89 # define DEFAULT_TMPDIR "/tmp"
95 /* POSIX says to exit with status 1 if invoked with -c and the
96 input is not properly sorted. */
97 SORT_OUT_OF_ORDER = 1,
99 /* POSIX says any other irregular exit must exit with a status
100 code greater than 1. */
106 /* The number of times we should try to fork a compression process
107 (we retry if the fork call fails). We don't _need_ to compress
108 temp files, this is just to reduce disk access, so this number
110 MAX_FORK_TRIES_COMPRESS = 2,
112 /* The number of times we should try to fork a decompression process.
113 If we can't fork a decompression process, we can't sort, so this
114 number should be big. */
115 MAX_FORK_TRIES_DECOMPRESS = 8
118 /* The representation of the decimal point in the current locale. */
119 static int decimal_point;
121 /* Thousands separator; if -1, then there isn't one. */
122 static int thousands_sep;
124 /* Nonzero if the corresponding locales are hard. */
125 static bool hard_LC_COLLATE;
127 static bool hard_LC_TIME;
130 #define NONZERO(x) ((x) != 0)
132 /* The kind of blanks for '-b' to skip in various options. */
133 enum blanktype { bl_start, bl_end, bl_both };
135 /* The character marking end of line. Default to \n. */
136 static char eolchar = '\n';
138 /* Lines are held in core as counted strings. */
141 char *text; /* Text of the line. */
142 size_t length; /* Length including final newline. */
143 char *keybeg; /* Start of first key. */
144 char *keylim; /* Limit of first key. */
150 char *buf; /* Dynamically allocated buffer,
151 partitioned into 3 regions:
154 - an array of lines, in reverse order. */
155 size_t used; /* Number of bytes used for input data. */
156 size_t nlines; /* Number of lines in the line array. */
157 size_t alloc; /* Number of bytes allocated. */
158 size_t left; /* Number of bytes left from previous reads. */
159 size_t line_bytes; /* Number of bytes to reserve for each line. */
160 bool eof; /* An EOF has been read. */
165 size_t sword; /* Zero-origin 'word' to start at. */
166 size_t schar; /* Additional characters to skip. */
167 size_t eword; /* Zero-origin first word after field. */
168 size_t echar; /* Additional characters in field. */
169 bool const *ignore; /* Boolean array of characters to ignore. */
170 char const *translate; /* Translation applied to characters. */
171 bool skipsblanks; /* Skip leading blanks when finding start. */
172 bool skipeblanks; /* Skip leading blanks when finding end. */
173 bool numeric; /* Flag for numeric comparison. Handle
174 strings of digits with optional decimal
175 point, but no exponential notation. */
176 bool random; /* Sort by random hash of key. */
177 bool general_numeric; /* Flag for general, numeric comparison.
178 Handle numbers in exponential notation. */
179 bool month; /* Flag for comparison by month name. */
180 bool reverse; /* Reverse the sense of comparison. */
181 bool version; /* sort by version number */
182 struct keyfield *next; /* Next keyfield to try. */
191 /* FIXME: None of these tables work with multibyte character sets.
192 Also, there are many other bugs when handling multibyte characters.
193 One way to fix this is to rewrite `sort' to use wide characters
194 internally, but doing this with good performance is a bit
197 /* Table of blanks. */
198 static bool blanks[UCHAR_LIM];
200 /* Table of non-printing characters. */
201 static bool nonprinting[UCHAR_LIM];
203 /* Table of non-dictionary characters (not letters, digits, or blanks). */
204 static bool nondictionary[UCHAR_LIM];
206 /* Translation table folding lower case to upper. */
207 static char fold_toupper[UCHAR_LIM];
209 #define MONTHS_PER_YEAR 12
211 /* Table mapping month names to integers.
212 Alphabetic order allows binary search. */
213 static struct month monthtab[] =
229 /* During the merge phase, the number of files to merge at once. */
230 #define NMERGE_DEFAULT 16
232 /* Minimum size for a merge or check buffer. */
233 #define MIN_MERGE_BUFFER_SIZE (2 + sizeof (struct line))
235 /* Minimum sort size; the code might not work with smaller sizes. */
236 #define MIN_SORT_SIZE (nmerge * MIN_MERGE_BUFFER_SIZE)
238 /* The number of bytes needed for a merge or check buffer, which can
239 function relatively efficiently even if it holds only one line. If
240 a longer line is seen, this value is increased. */
241 static size_t merge_buffer_size = MAX (MIN_MERGE_BUFFER_SIZE, 256 * 1024);
243 /* The approximate maximum number of bytes of main memory to use, as
244 specified by the user. Zero if the user has not specified a size. */
245 static size_t sort_size;
247 /* The guessed size for non-regular files. */
248 #define INPUT_FILE_SIZE_GUESS (1024 * 1024)
250 /* Array of directory names in which any temporary files are to be created. */
251 static char const **temp_dirs;
253 /* Number of temporary directory names used. */
254 static size_t temp_dir_count;
256 /* Number of allocated slots in temp_dirs. */
257 static size_t temp_dir_alloc;
259 /* Flag to reverse the order of all comparisons. */
262 /* Flag for stable sort. This turns off the last ditch bytewise
263 comparison of lines, and instead leaves lines in the same order
264 they were read if all keys compare equal. */
267 /* If TAB has this value, blanks separate fields. */
268 enum { TAB_DEFAULT = CHAR_MAX + 1 };
270 /* Tab character separating fields. If TAB_DEFAULT, then fields are
271 separated by the empty string between a non-blank character and a blank
273 static int tab = TAB_DEFAULT;
275 /* Flag to remove consecutive duplicate lines from the output.
276 Only the last of a sequence of equal lines will be output. */
279 /* Nonzero if any of the input files are the standard input. */
280 static bool have_read_stdin;
282 /* List of key field comparisons to be tried. */
283 static struct keyfield *keylist;
285 /* Program used to (de)compress temp files. Must accept -d. */
286 static char const *compress_program;
288 /* Maximum number of files to merge in one go. If more than this
289 number are present, temp files will be used. */
290 static unsigned int nmerge = NMERGE_DEFAULT;
292 static void sortlines_temp (struct line *, size_t, struct line *);
294 /* Report MESSAGE for FILE, then clean up and exit.
295 If FILE is null, it represents standard output. */
297 static void die (char const *, char const *) ATTRIBUTE_NORETURN;
299 die (char const *message, char const *file)
301 error (0, errno, "%s: %s", message, file ? file : _("standard output"));
308 if (status != EXIT_SUCCESS)
309 fprintf (stderr, _("Try `%s --help' for more information.\n"),
314 Usage: %s [OPTION]... [FILE]...\n\
315 or: %s [OPTION]... --files0-from=F\n\
317 program_name, program_name);
319 Write sorted concatenation of all FILE(s) to standard output.\n\
323 Mandatory arguments to long options are mandatory for short options too.\n\
330 -b, --ignore-leading-blanks ignore leading blanks\n\
331 -d, --dictionary-order consider only blanks and alphanumeric characters\n\
332 -f, --ignore-case fold lower case to upper case characters\n\
335 -g, --general-numeric-sort compare according to general numerical value\n\
336 -i, --ignore-nonprinting consider only printable characters\n\
337 -M, --month-sort compare (unknown) < `JAN' < ... < `DEC'\n\
338 -n, --numeric-sort compare according to string numerical value\n\
339 -R, --random-sort sort by random hash of keys\n\
340 -V, --version-sort sort by numeric version (see strverscmp(3C))\n\
341 --random-source=FILE get random bytes from FILE (default /dev/urandom)\n\
342 --sort=WORD sort according to WORD:\n\
343 general-numeric -g, month -M, numeric -n,\n\
344 random -R, version -V\n\
345 -r, --reverse reverse the result of comparisons\n\
353 --batch-size=NMERGE merge at most NMERGE inputs at once;\n\
354 for more use temp files\n\
357 -c, --check, --check=diagnose-first check for sorted input; do not sort\n\
358 -C, --check=quiet, --check=silent like -c, but do not report first bad line\n\
359 --compress-program=PROG compress temporaries with PROG;\n\
360 decompress them with PROG -d\n\
361 --files0-from=F read input from the files specified by\n\
362 NUL-terminated names in file F\n\
365 -k, --key=POS1[,POS2] start a key at POS1, end it at POS2 (origin 1)\n\
366 -m, --merge merge already sorted files; do not sort\n\
369 -o, --output=FILE write result to FILE instead of standard output\n\
370 -s, --stable stabilize sort by disabling last-resort comparison\n\
371 -S, --buffer-size=SIZE use SIZE for main memory buffer\n\
374 -t, --field-separator=SEP use SEP instead of non-blank to blank transition\n\
375 -T, --temporary-directory=DIR use DIR for temporaries, not $TMPDIR or %s;\n\
376 multiple options specify multiple directories\n\
377 -u, --unique with -c, check for strict ordering;\n\
378 without -c, output only the first of an equal run\n\
381 -z, --zero-terminated end lines with 0 byte, not newline\n\
383 fputs (HELP_OPTION_DESCRIPTION, stdout);
384 fputs (VERSION_OPTION_DESCRIPTION, stdout);
387 POS is F[.C][OPTS], where F is the field number and C the character position\n\
388 in the field; both are origin 1. If neither -t nor -b is in effect, characters\n\
389 in a field are counted from the beginning of the preceding whitespace. OPTS is\n\
390 one or more single-letter ordering options, which override global ordering\n\
391 options for that key. If no key is given, use the entire line as the key.\n\
393 SIZE may be followed by the following multiplicative suffixes:\n\
396 % 1% of memory, b 1, K 1024 (default), and so on for M, G, T, P, E, Z, Y.\n\
398 With no FILE, or when FILE is -, read standard input.\n\
401 The locale specified by the environment affects sort order.\n\
402 Set LC_ALL=C to get the traditional sort order that uses\n\
403 native byte values.\n\
405 emit_bug_reporting_address ();
411 /* For long options that have no equivalent short option, use a
412 non-character as a pseudo short option, starting with CHAR_MAX + 1. */
415 CHECK_OPTION = CHAR_MAX + 1,
416 COMPRESS_PROGRAM_OPTION,
419 RANDOM_SOURCE_OPTION,
423 static char const short_options[] = "-bcCdfgik:mMno:rRsS:t:T:uVy:z";
425 static struct option const long_options[] =
427 {"ignore-leading-blanks", no_argument, NULL, 'b'},
428 {"check", optional_argument, NULL, CHECK_OPTION},
429 {"compress-program", required_argument, NULL, COMPRESS_PROGRAM_OPTION},
430 {"dictionary-order", no_argument, NULL, 'd'},
431 {"ignore-case", no_argument, NULL, 'f'},
432 {"files0-from", required_argument, NULL, FILES0_FROM_OPTION},
433 {"general-numeric-sort", no_argument, NULL, 'g'},
434 {"ignore-nonprinting", no_argument, NULL, 'i'},
435 {"key", required_argument, NULL, 'k'},
436 {"merge", no_argument, NULL, 'm'},
437 {"month-sort", no_argument, NULL, 'M'},
438 {"numeric-sort", no_argument, NULL, 'n'},
439 {"version-sort", no_argument, NULL, 'V'},
440 {"random-sort", no_argument, NULL, 'R'},
441 {"random-source", required_argument, NULL, RANDOM_SOURCE_OPTION},
442 {"sort", required_argument, NULL, SORT_OPTION},
443 {"output", required_argument, NULL, 'o'},
444 {"reverse", no_argument, NULL, 'r'},
445 {"stable", no_argument, NULL, 's'},
446 {"batch-size", required_argument, NULL, NMERGE_OPTION},
447 {"buffer-size", required_argument, NULL, 'S'},
448 {"field-separator", required_argument, NULL, 't'},
449 {"temporary-directory", required_argument, NULL, 'T'},
450 {"unique", no_argument, NULL, 'u'},
451 {"zero-terminated", no_argument, NULL, 'z'},
452 {GETOPT_HELP_OPTION_DECL},
453 {GETOPT_VERSION_OPTION_DECL},
457 #define CHECK_TABLE \
459 _ct_("silent", 'C') \
460 _ct_("diagnose-first", 'c')
462 static char const *const check_args[] =
464 #define _ct_(_s, _c) _s,
468 static char const check_types[] =
470 #define _ct_(_s, _c) _c,
476 _st_("general-numeric", 'g') \
478 _st_("numeric", 'n') \
479 _st_("random", 'R') \
482 static char const *const sort_args[] =
484 #define _st_(_s, _c) _s,
488 static char const sort_types[] =
490 #define _st_(_s, _c) _c,
495 /* The set of signals that are caught. */
496 static sigset_t caught_signals;
498 /* Critical section status. */
505 /* Enter a critical section. */
506 static struct cs_status
509 struct cs_status status;
510 status.valid = (sigprocmask (SIG_BLOCK, &caught_signals, &status.sigs) == 0);
514 /* Leave a critical section. */
516 cs_leave (struct cs_status status)
520 /* Ignore failure when restoring the signal mask. */
521 sigprocmask (SIG_SETMASK, &status.sigs, NULL);
525 /* The list of temporary files. */
528 struct tempnode *volatile next;
529 pid_t pid; /* If compressed, the pid of compressor, else zero */
530 char name[1]; /* Actual size is 1 + file name length. */
532 static struct tempnode *volatile temphead;
533 static struct tempnode *volatile *temptail = &temphead;
538 pid_t pid; /* If compressed, the pid of compressor, else zero */
541 /* A table where we store compression process states. We clean up all
542 processes in a timely manner so as not to exhaust system resources,
543 so we store the info on whether the process is still running, or has
545 static Hash_table *proctab;
547 enum { INIT_PROCTAB_SIZE = 47 };
549 enum procstate { ALIVE, ZOMBIE };
551 /* A proctab entry. The COUNT field is there in case we fork a new
552 compression process that has the same PID as an old zombie process
553 that is still in the table (because the process to decompress the
554 temp file it was associated with hasn't started yet). */
558 enum procstate state;
563 proctab_hasher (const void *entry, size_t tabsize)
565 const struct procnode *node = entry;
566 return node->pid % tabsize;
570 proctab_comparator (const void *e1, const void *e2)
572 const struct procnode *n1 = e1, *n2 = e2;
573 return n1->pid == n2->pid;
576 /* The total number of forked processes (compressors and decompressors)
577 that have not been reaped yet. */
578 static size_t nprocs;
580 /* The number of child processes we'll allow before we try to reap some. */
581 enum { MAX_PROCS_BEFORE_REAP = 2 };
583 /* If 0 < PID, wait for the child process with that PID to exit.
584 If PID is -1, clean up a random child process which has finished and
585 return the process ID of that child. If PID is -1 and no processes
586 have quit yet, return 0 without waiting. */
592 pid_t cpid = waitpid (pid, &status, pid < 0 ? WNOHANG : 0);
595 error (SORT_FAILURE, errno, _("waiting for %s [-d]"),
599 if (! WIFEXITED (status) || WEXITSTATUS (status))
600 error (SORT_FAILURE, 0, _("%s [-d] terminated abnormally"),
608 /* Add the PID of a running compression process to proctab, or update
609 the entry COUNT and STATE fields if it's already there. This also
610 creates the table for us the first time it's called. */
613 register_proc (pid_t pid)
615 struct procnode test, *node;
619 proctab = hash_initialize (INIT_PROCTAB_SIZE, NULL,
628 node = hash_lookup (proctab, &test);
636 node = xmalloc (sizeof *node);
640 hash_insert (proctab, node);
644 /* This is called when we reap a random process. We don't know
645 whether we have reaped a compression process or a decompression
646 process until we look in the table. If there's an ALIVE entry for
647 it, then we have reaped a compression process, so change the state
648 to ZOMBIE. Otherwise, it's a decompression processes, so ignore it. */
651 update_proc (pid_t pid)
653 struct procnode test, *node;
656 node = hash_lookup (proctab, &test);
658 node->state = ZOMBIE;
661 /* This is for when we need to wait for a compression process to exit.
662 If it has a ZOMBIE entry in the table then it's already dead and has
663 been reaped. Note that if there's an ALIVE entry for it, it still may
664 already have died and been reaped if a second process was created with
665 the same PID. This is probably exceedingly rare, but to be on the safe
666 side we will have to wait for any compression process with this PID. */
669 wait_proc (pid_t pid)
671 struct procnode test, *node;
674 node = hash_lookup (proctab, &test);
675 if (node->state == ALIVE)
678 node->state = ZOMBIE;
681 hash_delete (proctab, node);
686 /* Keep reaping finished children as long as there are more to reap.
687 This doesn't block waiting for any of them, it only reaps those
688 that are already dead. */
695 while (0 < nprocs && (pid = reap (-1)))
699 /* Clean up any remaining temporary files. */
704 struct tempnode const *node;
706 for (node = temphead; node; node = node->next)
711 /* Cleanup actions to take when exiting. */
718 /* Clean up any remaining temporary files in a critical section so
719 that a signal handler does not try to clean them too. */
720 struct cs_status cs = cs_enter ();
728 /* Create a new temporary file, returning its newly allocated tempnode.
729 Store into *PFD the file descriptor open for writing. */
731 static struct tempnode *
732 create_temp_file (int *pfd)
734 static char const slashbase[] = "/sortXXXXXX";
735 static size_t temp_dir_index;
738 char const *temp_dir = temp_dirs[temp_dir_index];
739 size_t len = strlen (temp_dir);
740 struct tempnode *node =
741 xmalloc (offsetof (struct tempnode, name) + len + sizeof slashbase);
742 char *file = node->name;
745 memcpy (file, temp_dir, len);
746 memcpy (file + len, slashbase, sizeof slashbase);
749 if (++temp_dir_index == temp_dir_count)
752 /* Create the temporary file in a critical section, to avoid races. */
758 temptail = &node->next;
765 error (SORT_FAILURE, errno, _("cannot create temporary file in %s"),
772 /* Return a stream for FILE, opened with mode HOW. A null FILE means
773 standard output; HOW should be "w". When opening for input, "-"
774 means standard input. To avoid confusion, do not return file
775 descriptors STDIN_FILENO, STDOUT_FILENO, or STDERR_FILENO when
776 opening an ordinary FILE. */
779 xfopen (const char *file, const char *how)
785 else if (STREQ (file, "-") && *how == 'r')
787 have_read_stdin = true;
792 fp = fopen (file, how);
794 die (_("open failed"), file);
800 /* Close FP, whose name is FILE, and report any errors. */
803 xfclose (FILE *fp, char const *file)
808 /* Allow reading stdin from tty more than once. */
814 /* Don't close stdout just yet. close_stdout does that. */
815 if (fflush (fp) != 0)
816 die (_("fflush failed"), file);
820 if (fclose (fp) != 0)
821 die (_("close failed"), file);
827 dup2_or_die (int oldfd, int newfd)
829 if (dup2 (oldfd, newfd) < 0)
830 error (SORT_FAILURE, errno, _("dup2 failed"));
833 /* Fork a child process for piping to and do common cleanup. The
834 TRIES parameter tells us how many times to try to fork before
835 giving up. Return the PID of the child or -1 if fork failed. */
838 pipe_fork (int pipefds[2], size_t tries)
840 #if HAVE_WORKING_FORK
841 struct tempnode *saved_temphead;
843 unsigned int wait_retry = 1;
844 pid_t pid IF_LINT (= -1);
847 if (pipe (pipefds) < 0)
852 /* This is so the child process won't delete our temp files
853 if it receives a signal before exec-ing. */
855 saved_temphead = temphead;
861 temphead = saved_temphead;
866 if (0 <= pid || errno != EAGAIN)
883 close (STDIN_FILENO);
884 close (STDOUT_FILENO);
891 #else /* ! HAVE_WORKING_FORK */
896 /* Create a temporary file and start a compression program to filter output
897 to that file. Set *PFP to the file handle and if *PPID is non-NULL,
898 set it to the PID of the newly-created process. */
901 create_temp (FILE **pfp, pid_t *ppid)
904 struct tempnode *node = create_temp_file (&tempfd);
905 char *name = node->name;
907 if (compress_program)
911 node->pid = pipe_fork (pipefds, MAX_FORK_TRIES_COMPRESS);
918 register_proc (node->pid);
920 else if (node->pid == 0)
923 dup2_or_die (tempfd, STDOUT_FILENO);
925 dup2_or_die (pipefds[0], STDIN_FILENO);
928 if (execlp (compress_program, compress_program, (char *) NULL) < 0)
929 error (SORT_FAILURE, errno, _("couldn't execute %s"),
936 *pfp = fdopen (tempfd, "w");
938 die (_("couldn't create temporary file"), name);
946 /* Open a compressed temp file and start a decompression process through
947 which to filter the input. PID must be the valid processes ID of the
948 process used to compress the file. */
951 open_temp (const char *name, pid_t pid)
953 int tempfd, pipefds[2];
959 tempfd = open (name, O_RDONLY);
961 die (_("couldn't open temporary file"), name);
963 child_pid = pipe_fork (pipefds, MAX_FORK_TRIES_DECOMPRESS);
969 else if (child_pid == 0)
972 dup2_or_die (tempfd, STDIN_FILENO);
974 dup2_or_die (pipefds[1], STDOUT_FILENO);
977 if (execlp (compress_program, compress_program, "-d", (char *) NULL) < 0)
978 error (SORT_FAILURE, errno, _("couldn't execute %s -d"),
982 error (SORT_FAILURE, errno, _("couldn't create process for %s -d"),
985 fp = fdopen (pipefds[0], "r");
987 die (_("couldn't create temporary file"), name);
993 write_bytes (const char *buf, size_t n_bytes, FILE *fp, const char *output_file)
995 if (fwrite (buf, 1, n_bytes, fp) != n_bytes)
996 die (_("write failed"), output_file);
999 /* Append DIR to the array of temporary directory names. */
1001 add_temp_dir (char const *dir)
1003 if (temp_dir_count == temp_dir_alloc)
1004 temp_dirs = X2NREALLOC (temp_dirs, &temp_dir_alloc);
1006 temp_dirs[temp_dir_count++] = dir;
1009 /* Remove NAME from the list of temporary files. */
1012 zaptemp (const char *name)
1014 struct tempnode *volatile *pnode;
1015 struct tempnode *node;
1016 struct tempnode *next;
1018 int unlink_errno = 0;
1019 struct cs_status cs;
1021 for (pnode = &temphead; (node = *pnode)->name != name; pnode = &node->next)
1024 /* Unlink the temporary file in a critical section to avoid races. */
1027 unlink_status = unlink (name);
1028 unlink_errno = errno;
1032 if (unlink_status != 0)
1033 error (0, unlink_errno, _("warning: cannot remove: %s"), name);
1039 #if HAVE_NL_LANGINFO
1042 struct_month_cmp (const void *m1, const void *m2)
1044 struct month const *month1 = m1;
1045 struct month const *month2 = m2;
1046 return strcmp (month1->name, month2->name);
1051 /* Initialize the character class tables. */
1058 for (i = 0; i < UCHAR_LIM; ++i)
1060 blanks[i] = !! isblank (i);
1061 nonprinting[i] = ! isprint (i);
1062 nondictionary[i] = ! isalnum (i) && ! isblank (i);
1063 fold_toupper[i] = toupper (i);
1066 #if HAVE_NL_LANGINFO
1067 /* If we're not in the "C" locale, read different names for months. */
1070 for (i = 0; i < MONTHS_PER_YEAR; i++)
1077 s = (char *) nl_langinfo (ABMON_1 + i);
1079 monthtab[i].name = name = xmalloc (s_len + 1);
1080 monthtab[i].val = i + 1;
1082 for (j = 0; j < s_len; j++)
1083 name[j] = fold_toupper[to_uchar (s[j])];
1086 qsort ((void *) monthtab, MONTHS_PER_YEAR,
1087 sizeof *monthtab, struct_month_cmp);
1092 /* Specify how many inputs may be merged at once.
1093 This may be set on the command-line with the
1094 --batch-size option. */
1096 specify_nmerge (int oi, char c, char const *s)
1099 struct rlimit rlimit;
1100 enum strtol_error e = xstrtoumax (s, NULL, 10, &n, NULL);
1102 /* Try to find out how many file descriptors we'll be able
1103 to open. We need at least nmerge + 3 (STDIN_FILENO,
1104 STDOUT_FILENO and STDERR_FILENO). */
1105 unsigned int max_nmerge = ((getrlimit (RLIMIT_NOFILE, &rlimit) == 0
1110 if (e == LONGINT_OK)
1114 e = LONGINT_OVERFLOW;
1119 error (0, 0, _("invalid --%s argument %s"),
1120 long_options[oi].name, quote(s));
1121 error (SORT_FAILURE, 0,
1122 _("minimum --%s argument is %s"),
1123 long_options[oi].name, quote("2"));
1125 else if (max_nmerge < nmerge)
1127 e = LONGINT_OVERFLOW;
1134 if (e == LONGINT_OVERFLOW)
1136 char max_nmerge_buf[INT_BUFSIZE_BOUND (unsigned int)];
1137 error (0, 0, _("--%s argument %s too large"),
1138 long_options[oi].name, quote(s));
1139 error (SORT_FAILURE, 0,
1140 _("maximum --%s argument with current rlimit is %s"),
1141 long_options[oi].name,
1142 uinttostr (max_nmerge, max_nmerge_buf));
1145 xstrtol_fatal (e, oi, c, long_options, s);
1148 /* Specify the amount of main memory to use when sorting. */
1150 specify_sort_size (int oi, char c, char const *s)
1154 enum strtol_error e = xstrtoumax (s, &suffix, 10, &n, "EgGkKmMPtTYZ");
1156 /* The default unit is KiB. */
1157 if (e == LONGINT_OK && ISDIGIT (suffix[-1]))
1159 if (n <= UINTMAX_MAX / 1024)
1162 e = LONGINT_OVERFLOW;
1165 /* A 'b' suffix means bytes; a '%' suffix means percent of memory. */
1166 if (e == LONGINT_INVALID_SUFFIX_CHAR && ISDIGIT (suffix[-1]) && ! suffix[1])
1175 double mem = physmem_total () * n / 100;
1177 /* Use "<", not "<=", to avoid problems with rounding. */
1178 if (mem < UINTMAX_MAX)
1184 e = LONGINT_OVERFLOW;
1189 if (e == LONGINT_OK)
1191 /* If multiple sort sizes are specified, take the maximum, so
1192 that option order does not matter. */
1199 sort_size = MAX (sort_size, MIN_SORT_SIZE);
1203 e = LONGINT_OVERFLOW;
1206 xstrtol_fatal (e, oi, c, long_options, s);
1209 /* Return the default sort size. */
1211 default_sort_size (void)
1213 /* Let MEM be available memory or 1/8 of total memory, whichever
1215 double avail = physmem_available ();
1216 double total = physmem_total ();
1217 double mem = MAX (avail, total / 8);
1218 struct rlimit rlimit;
1220 /* Let SIZE be MEM, but no more than the maximum object size or
1221 system resource limits. Avoid the MIN macro here, as it is not
1222 quite right when only one argument is floating point. Don't
1223 bother to check for values like RLIM_INFINITY since in practice
1224 they are not much less than SIZE_MAX. */
1225 size_t size = SIZE_MAX;
1228 if (getrlimit (RLIMIT_DATA, &rlimit) == 0 && rlimit.rlim_cur < size)
1229 size = rlimit.rlim_cur;
1231 if (getrlimit (RLIMIT_AS, &rlimit) == 0 && rlimit.rlim_cur < size)
1232 size = rlimit.rlim_cur;
1235 /* Leave a large safety margin for the above limits, as failure can
1236 occur when they are exceeded. */
1240 /* Leave a 1/16 margin for RSS to leave room for code, stack, etc.
1241 Exceeding RSS is not fatal, but can be quite slow. */
1242 if (getrlimit (RLIMIT_RSS, &rlimit) == 0 && rlimit.rlim_cur / 16 * 15 < size)
1243 size = rlimit.rlim_cur / 16 * 15;
1246 /* Use no less than the minimum. */
1247 return MAX (size, MIN_SORT_SIZE);
1250 /* Return the sort buffer size to use with the input files identified
1251 by FPS and FILES, which are alternate names of the same files.
1252 NFILES gives the number of input files; NFPS may be less. Assume
1253 that each input line requires LINE_BYTES extra bytes' worth of line
1254 information. Do not exceed the size bound specified by the user
1255 (or a default size bound, if the user does not specify one). */
1258 sort_buffer_size (FILE *const *fps, size_t nfps,
1259 char *const *files, size_t nfiles,
1262 /* A bound on the input size. If zero, the bound hasn't been
1264 static size_t size_bound;
1266 /* In the worst case, each input byte is a newline. */
1267 size_t worst_case_per_input_byte = line_bytes + 1;
1269 /* Keep enough room for one extra input line and an extra byte.
1270 This extra room might be needed when preparing to read EOF. */
1271 size_t size = worst_case_per_input_byte + 1;
1275 for (i = 0; i < nfiles; i++)
1281 if ((i < nfps ? fstat (fileno (fps[i]), &st)
1282 : STREQ (files[i], "-") ? fstat (STDIN_FILENO, &st)
1283 : stat (files[i], &st))
1285 die (_("stat failed"), files[i]);
1287 if (S_ISREG (st.st_mode))
1288 file_size = st.st_size;
1291 /* The file has unknown size. If the user specified a sort
1292 buffer size, use that; otherwise, guess the size. */
1295 file_size = INPUT_FILE_SIZE_GUESS;
1300 size_bound = sort_size;
1302 size_bound = default_sort_size ();
1305 /* Add the amount of memory needed to represent the worst case
1306 where the input consists entirely of newlines followed by a
1307 single non-newline. Check for overflow. */
1308 worst_case = file_size * worst_case_per_input_byte + 1;
1309 if (file_size != worst_case / worst_case_per_input_byte
1310 || size_bound - size <= worst_case)
1318 /* Initialize BUF. Reserve LINE_BYTES bytes for each line; LINE_BYTES
1319 must be at least sizeof (struct line). Allocate ALLOC bytes
1323 initbuf (struct buffer *buf, size_t line_bytes, size_t alloc)
1325 /* Ensure that the line array is properly aligned. If the desired
1326 size cannot be allocated, repeatedly halve it until allocation
1327 succeeds. The smaller allocation may hurt overall performance,
1328 but that's better than failing. */
1331 alloc += sizeof (struct line) - alloc % sizeof (struct line);
1332 buf->buf = malloc (alloc);
1336 if (alloc <= line_bytes + 1)
1340 buf->line_bytes = line_bytes;
1342 buf->used = buf->left = buf->nlines = 0;
1346 /* Return one past the limit of the line array. */
1348 static inline struct line *
1349 buffer_linelim (struct buffer const *buf)
1351 return (struct line *) (buf->buf + buf->alloc);
1354 /* Return a pointer to the first character of the field specified
1358 begfield (const struct line *line, const struct keyfield *key)
1360 char *ptr = line->text, *lim = ptr + line->length - 1;
1361 size_t sword = key->sword;
1362 size_t schar = key->schar;
1363 size_t remaining_bytes;
1365 /* The leading field separator itself is included in a field when -t
1368 if (tab != TAB_DEFAULT)
1369 while (ptr < lim && sword--)
1371 while (ptr < lim && *ptr != tab)
1377 while (ptr < lim && sword--)
1379 while (ptr < lim && blanks[to_uchar (*ptr)])
1381 while (ptr < lim && !blanks[to_uchar (*ptr)])
1385 if (key->skipsblanks)
1386 while (ptr < lim && blanks[to_uchar (*ptr)])
1389 /* Advance PTR by SCHAR (if possible), but no further than LIM. */
1390 remaining_bytes = lim - ptr;
1391 if (schar < remaining_bytes)
1399 /* Return the limit of (a pointer to the first character after) the field
1400 in LINE specified by KEY. */
1403 limfield (const struct line *line, const struct keyfield *key)
1405 char *ptr = line->text, *lim = ptr + line->length - 1;
1406 size_t eword = key->eword, echar = key->echar;
1407 size_t remaining_bytes;
1409 /* Move PTR past EWORD fields or to one past the last byte on LINE,
1410 whichever comes first. If there are more than EWORD fields, leave
1411 PTR pointing at the beginning of the field having zero-based index,
1412 EWORD. If a delimiter character was specified (via -t), then that
1413 `beginning' is the first character following the delimiting TAB.
1414 Otherwise, leave PTR pointing at the first `blank' character after
1415 the preceding field. */
1416 if (tab != TAB_DEFAULT)
1417 while (ptr < lim && eword--)
1419 while (ptr < lim && *ptr != tab)
1421 if (ptr < lim && (eword | echar))
1425 while (ptr < lim && eword--)
1427 while (ptr < lim && blanks[to_uchar (*ptr)])
1429 while (ptr < lim && !blanks[to_uchar (*ptr)])
1433 #ifdef POSIX_UNSPECIFIED
1434 /* The following block of code makes GNU sort incompatible with
1435 standard Unix sort, so it's ifdef'd out for now.
1436 The POSIX spec isn't clear on how to interpret this.
1437 FIXME: request clarification.
1439 From: kwzh@gnu.ai.mit.edu (Karl Heuer)
1440 Date: Thu, 30 May 96 12:20:41 -0400
1441 [Translated to POSIX 1003.1-2001 terminology by Paul Eggert.]
1443 [...]I believe I've found another bug in `sort'.
1448 $ textutils-1.15/src/sort -k1.7,1.7 </tmp/sort.in
1451 $ /bin/sort -k1.7,1.7 </tmp/sort.in
1455 Unix sort produced the answer I expected: sort on the single character
1456 in column 7. GNU sort produced different results, because it disagrees
1457 on the interpretation of the key-end spec "M.N". Unix sort reads this
1458 as "skip M-1 fields, then N-1 characters"; but GNU sort wants it to mean
1459 "skip M-1 fields, then either N-1 characters or the rest of the current
1460 field, whichever comes first". This extra clause applies only to
1461 key-ends, not key-starts.
1464 /* Make LIM point to the end of (one byte past) the current field. */
1465 if (tab != TAB_DEFAULT)
1468 newlim = memchr (ptr, tab, lim - ptr);
1476 while (newlim < lim && blanks[to_uchar (*newlim)])
1478 while (newlim < lim && !blanks[to_uchar (*newlim)])
1484 /* If we're ignoring leading blanks when computing the End
1485 of the field, don't start counting bytes until after skipping
1486 past any leading blanks. */
1487 if (key->skipeblanks)
1488 while (ptr < lim && blanks[to_uchar (*ptr)])
1491 /* Advance PTR by ECHAR (if possible), but no further than LIM. */
1492 remaining_bytes = lim - ptr;
1493 if (echar < remaining_bytes)
1501 /* Fill BUF reading from FP, moving buf->left bytes from the end
1502 of buf->buf to the beginning first. If EOF is reached and the
1503 file wasn't terminated by a newline, supply one. Set up BUF's line
1504 table too. FILE is the name of the file corresponding to FP.
1505 Return true if some input was read. */
1508 fillbuf (struct buffer *buf, FILE *fp, char const *file)
1510 struct keyfield const *key = keylist;
1512 size_t line_bytes = buf->line_bytes;
1513 size_t mergesize = merge_buffer_size - MIN_MERGE_BUFFER_SIZE;
1518 if (buf->used != buf->left)
1520 memmove (buf->buf, buf->buf + buf->used - buf->left, buf->left);
1521 buf->used = buf->left;
1527 char *ptr = buf->buf + buf->used;
1528 struct line *linelim = buffer_linelim (buf);
1529 struct line *line = linelim - buf->nlines;
1530 size_t avail = (char *) linelim - buf->nlines * line_bytes - ptr;
1531 char *line_start = buf->nlines ? line->text + line->length : buf->buf;
1533 while (line_bytes + 1 < avail)
1535 /* Read as many bytes as possible, but do not read so many
1536 bytes that there might not be enough room for the
1537 corresponding line array. The worst case is when the
1538 rest of the input file consists entirely of newlines,
1539 except that the last byte is not a newline. */
1540 size_t readsize = (avail - 1) / (line_bytes + 1);
1541 size_t bytes_read = fread (ptr, 1, readsize, fp);
1542 char *ptrlim = ptr + bytes_read;
1544 avail -= bytes_read;
1546 if (bytes_read != readsize)
1549 die (_("read failed"), file);
1553 if (buf->buf == ptrlim)
1555 if (ptrlim[-1] != eol)
1560 /* Find and record each line in the just-read input. */
1561 while ((p = memchr (ptr, eol, ptrlim - ptr)))
1565 line->text = line_start;
1566 line->length = ptr - line_start;
1567 mergesize = MAX (mergesize, line->length);
1568 avail -= line_bytes;
1572 /* Precompute the position of the first key for
1574 line->keylim = (key->eword == SIZE_MAX
1576 : limfield (line, key));
1578 if (key->sword != SIZE_MAX)
1579 line->keybeg = begfield (line, key);
1582 if (key->skipsblanks)
1583 while (blanks[to_uchar (*line_start)])
1585 line->keybeg = line_start;
1597 buf->used = ptr - buf->buf;
1598 buf->nlines = buffer_linelim (buf) - line;
1599 if (buf->nlines != 0)
1601 buf->left = ptr - line_start;
1602 merge_buffer_size = mergesize + MIN_MERGE_BUFFER_SIZE;
1607 /* The current input line is too long to fit in the buffer.
1608 Double the buffer size and try again, keeping it properly
1610 size_t line_alloc = buf->alloc / sizeof (struct line);
1611 buf->buf = x2nrealloc (buf->buf, &line_alloc, sizeof (struct line));
1612 buf->alloc = line_alloc * sizeof (struct line);
1617 /* Compare strings A and B as numbers without explicitly converting them to
1618 machine numbers. Comparatively slow for short strings, but asymptotically
1622 numcompare (const char *a, const char *b)
1624 while (blanks[to_uchar (*a)])
1626 while (blanks[to_uchar (*b)])
1629 return strnumcmp (a, b, decimal_point, thousands_sep);
1633 general_numcompare (const char *sa, const char *sb)
1635 /* FIXME: add option to warn about failed conversions. */
1636 /* FIXME: maybe add option to try expensive FP conversion
1637 only if A and B can't be compared more cheaply/accurately. */
1641 double a = strtod (sa, &ea);
1642 double b = strtod (sb, &eb);
1644 /* Put conversion errors at the start of the collating sequence. */
1646 return sb == eb ? 0 : -1;
1650 /* Sort numbers in the usual way, where -0 == +0. Put NaNs after
1651 conversion errors but before numbers; sort them by internal
1652 bit-pattern, for lack of a more portable alternative. */
1658 : memcmp ((char *) &a, (char *) &b, sizeof a));
1661 /* Return an integer in 1..12 of the month name MONTH with length LEN.
1662 Return 0 if the name in S is not recognized. */
1665 getmonth (char const *month, size_t len)
1668 size_t hi = MONTHS_PER_YEAR;
1669 char const *monthlim = month + len;
1673 if (month == monthlim)
1675 if (!blanks[to_uchar (*month)])
1682 size_t ix = (lo + hi) / 2;
1683 char const *m = month;
1684 char const *n = monthtab[ix].name;
1689 return monthtab[ix].val;
1690 if (m == monthlim || fold_toupper[to_uchar (*m)] < to_uchar (*n))
1695 else if (fold_toupper[to_uchar (*m)] > to_uchar (*n))
1707 /* A source of random data. */
1708 static struct randread_source *randread_source;
1710 /* Return the Ith randomly-generated state. The caller must invoke
1711 random_state (H) for all H less than I before invoking random_state
1714 static struct md5_ctx
1715 random_state (size_t i)
1717 /* An array of states resulting from the random data, and counts of
1718 its used and allocated members. */
1719 static struct md5_ctx *state;
1721 static size_t allocated;
1723 struct md5_ctx *s = &state[i];
1727 unsigned char buf[MD5_DIGEST_SIZE];
1733 state = X2NREALLOC (state, &allocated);
1737 randread (randread_source, buf, sizeof buf);
1739 md5_process_bytes (buf, sizeof buf, s);
1745 /* Compare the hashes of TEXTA with length LENGTHA to those of TEXTB
1746 with length LENGTHB. Return negative if less, zero if equal,
1747 positive if greater. */
1750 cmp_hashes (char const *texta, size_t lena,
1751 char const *textb, size_t lenb)
1753 /* Try random hashes until a pair of hashes disagree. But if the
1754 first pair of random hashes agree, check whether the keys are
1755 identical and if so report no difference. */
1760 uint32_t dig[2][MD5_DIGEST_SIZE / sizeof (uint32_t)];
1761 struct md5_ctx s[2];
1762 s[0] = s[1] = random_state (i);
1763 md5_process_bytes (texta, lena, &s[0]); md5_finish_ctx (&s[0], dig[0]);
1764 md5_process_bytes (textb, lenb, &s[1]); md5_finish_ctx (&s[1], dig[1]);
1765 diff = memcmp (dig[0], dig[1], sizeof dig[0]);
1768 if (i == 0 && lena == lenb && memcmp (texta, textb, lena) == 0)
1775 /* Compare the keys TEXTA (of length LENA) and TEXTB (of length LENB)
1776 using one or more random hash functions. */
1779 compare_random (char *restrict texta, size_t lena,
1780 char *restrict textb, size_t lenb)
1784 if (! hard_LC_COLLATE)
1785 diff = cmp_hashes (texta, lena, textb, lenb);
1788 /* Transform the text into the basis of comparison, so that byte
1789 strings that would otherwise considered to be equal are
1790 considered equal here even if their bytes differ. */
1793 char stackbuf[4000];
1794 size_t tlena = xmemxfrm (stackbuf, sizeof stackbuf, texta, lena);
1795 bool a_fits = tlena <= sizeof stackbuf;
1796 size_t tlenb = xmemxfrm ((a_fits ? stackbuf + tlena : NULL),
1797 (a_fits ? sizeof stackbuf - tlena : 0),
1800 if (a_fits && tlena + tlenb <= sizeof stackbuf)
1804 /* Adding 1 to the buffer size lets xmemxfrm run a bit
1805 faster by avoiding the need for an extra buffer copy. */
1806 buf = xmalloc (tlena + tlenb + 1);
1807 xmemxfrm (buf, tlena + 1, texta, lena);
1808 xmemxfrm (buf + tlena, tlenb + 1, textb, lenb);
1811 diff = cmp_hashes (buf, tlena, buf + tlena, tlenb);
1813 if (buf != stackbuf)
1820 /* Compare the keys TEXTA (of length LENA) and TEXTB (of length LENB)
1821 using strverscmp. */
1824 compare_version (char *restrict texta, size_t lena,
1825 char *restrict textb, size_t lenb)
1829 /* It is necessary to save the character after the end of the field.
1830 "strverscmp" works with NUL terminated strings. Our blocks of
1831 text are not necessarily terminated with a NUL byte. */
1832 char sv_a = texta[lena];
1833 char sv_b = textb[lenb];
1838 diff = strverscmp (texta, textb);
1846 /* Compare two lines A and B trying every key in sequence until there
1847 are no more keys or a difference is found. */
1850 keycompare (const struct line *a, const struct line *b)
1852 struct keyfield const *key = keylist;
1854 /* For the first iteration only, the key positions have been
1855 precomputed for us. */
1856 char *texta = a->keybeg;
1857 char *textb = b->keybeg;
1858 char *lima = a->keylim;
1859 char *limb = b->keylim;
1865 char const *translate = key->translate;
1866 bool const *ignore = key->ignore;
1868 /* Find the lengths. */
1869 size_t lena = lima <= texta ? 0 : lima - texta;
1870 size_t lenb = limb <= textb ? 0 : limb - textb;
1872 /* Actually compare the fields. */
1875 diff = compare_random (texta, lena, textb, lenb);
1876 else if (key->numeric | key->general_numeric)
1878 char savea = *lima, saveb = *limb;
1880 *lima = *limb = '\0';
1881 diff = ((key->numeric ? numcompare : general_numcompare)
1883 *lima = savea, *limb = saveb;
1885 else if (key->version)
1886 diff = compare_version (texta, lena, textb, lenb);
1887 else if (key->month)
1888 diff = getmonth (texta, lena) - getmonth (textb, lenb);
1889 /* Sorting like this may become slow, so in a simple locale the user
1890 can select a faster sort that is similar to ascii sort. */
1891 else if (hard_LC_COLLATE)
1893 if (ignore || translate)
1896 size_t size = lena + 1 + lenb + 1;
1897 char *copy_a = (size <= sizeof buf ? buf : xmalloc (size));
1898 char *copy_b = copy_a + lena + 1;
1899 size_t new_len_a, new_len_b, i;
1901 /* Ignore and/or translate chars before comparing. */
1902 for (new_len_a = new_len_b = i = 0; i < MAX (lena, lenb); i++)
1906 copy_a[new_len_a] = (translate
1907 ? translate[to_uchar (texta[i])]
1909 if (!ignore || !ignore[to_uchar (texta[i])])
1914 copy_b[new_len_b] = (translate
1915 ? translate[to_uchar (textb[i])]
1917 if (!ignore || !ignore[to_uchar (textb[i])])
1922 diff = xmemcoll (copy_a, new_len_a, copy_b, new_len_b);
1924 if (sizeof buf < size)
1928 diff = - NONZERO (lenb);
1932 diff = xmemcoll (texta, lena, textb, lenb);
1936 #define CMP_WITH_IGNORE(A, B) \
1941 while (texta < lima && ignore[to_uchar (*texta)]) \
1943 while (textb < limb && ignore[to_uchar (*textb)]) \
1945 if (! (texta < lima && textb < limb)) \
1947 diff = to_uchar (A) - to_uchar (B); \
1954 diff = (texta < lima) - (textb < limb); \
1959 CMP_WITH_IGNORE (translate[to_uchar (*texta)],
1960 translate[to_uchar (*textb)]);
1962 CMP_WITH_IGNORE (*texta, *textb);
1965 diff = - NONZERO (lenb);
1972 while (texta < lima && textb < limb)
1974 diff = (to_uchar (translate[to_uchar (*texta++)])
1975 - to_uchar (translate[to_uchar (*textb++)]));
1982 diff = memcmp (texta, textb, MIN (lena, lenb));
1986 diff = lena < lenb ? -1 : lena != lenb;
1996 /* Find the beginning and limit of the next field. */
1997 if (key->eword != SIZE_MAX)
1998 lima = limfield (a, key), limb = limfield (b, key);
2000 lima = a->text + a->length - 1, limb = b->text + b->length - 1;
2002 if (key->sword != SIZE_MAX)
2003 texta = begfield (a, key), textb = begfield (b, key);
2006 texta = a->text, textb = b->text;
2007 if (key->skipsblanks)
2009 while (texta < lima && blanks[to_uchar (*texta)])
2011 while (textb < limb && blanks[to_uchar (*textb)])
2022 return key->reverse ? -diff : diff;
2025 /* Compare two lines A and B, returning negative, zero, or positive
2026 depending on whether A compares less than, equal to, or greater than B. */
2029 compare (const struct line *a, const struct line *b)
2034 /* First try to compare on the specified keys (if any).
2035 The only two cases with no key at all are unadorned sort,
2036 and unadorned sort -r. */
2039 diff = keycompare (a, b);
2040 if (diff | unique | stable)
2044 /* If the keys all compare equal (or no keys were specified)
2045 fall through to the default comparison. */
2046 alen = a->length - 1, blen = b->length - 1;
2049 diff = - NONZERO (blen);
2052 else if (hard_LC_COLLATE)
2053 diff = xmemcoll (a->text, alen, b->text, blen);
2054 else if (! (diff = memcmp (a->text, b->text, MIN (alen, blen))))
2055 diff = alen < blen ? -1 : alen != blen;
2057 return reverse ? -diff : diff;
2060 /* Check that the lines read from FILE_NAME come in order. Return
2061 true if they are in order. If CHECKONLY == 'c', also print a
2062 diagnostic (FILE_NAME, line number, contents of line) to stderr if
2063 they are not in order. */
2066 check (char const *file_name, char checkonly)
2068 FILE *fp = xfopen (file_name, "r");
2069 struct buffer buf; /* Input buffer. */
2070 struct line temp; /* Copy of previous line. */
2072 uintmax_t line_number = 0;
2073 struct keyfield const *key = keylist;
2074 bool nonunique = ! unique;
2075 bool ordered = true;
2077 initbuf (&buf, sizeof (struct line),
2078 MAX (merge_buffer_size, sort_size));
2081 while (fillbuf (&buf, fp, file_name))
2083 struct line const *line = buffer_linelim (&buf);
2084 struct line const *linebase = line - buf.nlines;
2086 /* Make sure the line saved from the old buffer contents is
2087 less than or equal to the first line of the new buffer. */
2088 if (alloc && nonunique <= compare (&temp, line - 1))
2092 if (checkonly == 'c')
2094 struct line const *disorder_line = line - 1;
2095 uintmax_t disorder_line_number =
2096 buffer_linelim (&buf) - disorder_line + line_number;
2097 char hr_buf[INT_BUFSIZE_BOUND (uintmax_t)];
2098 fprintf (stderr, _("%s: %s:%s: disorder: "),
2099 program_name, file_name,
2100 umaxtostr (disorder_line_number, hr_buf));
2101 write_bytes (disorder_line->text, disorder_line->length,
2102 stderr, _("standard error"));
2110 /* Compare each line in the buffer with its successor. */
2111 while (linebase < --line)
2112 if (nonunique <= compare (line, line - 1))
2113 goto found_disorder;
2115 line_number += buf.nlines;
2117 /* Save the last line of the buffer. */
2118 if (alloc < line->length)
2125 alloc = line->length;
2129 while (alloc < line->length);
2131 temp.text = xrealloc (temp.text, alloc);
2133 memcpy (temp.text, line->text, line->length);
2134 temp.length = line->length;
2137 temp.keybeg = temp.text + (line->keybeg - line->text);
2138 temp.keylim = temp.text + (line->keylim - line->text);
2142 xfclose (fp, file_name);
2148 /* Merge lines from FILES onto OFP. NTEMPS is the number of temporary
2149 files (all of which are at the start of the FILES array), and
2150 NFILES is the number of files; 0 <= NTEMPS <= NFILES <= NMERGE.
2151 Close input and output files before returning.
2152 OUTPUT_FILE gives the name of the output file. If it is NULL,
2153 the output file is standard output. If OFP is NULL, the output
2154 file has not been opened yet (or written to, if standard output). */
2157 mergefps (struct sortfile *files, size_t ntemps, size_t nfiles,
2158 FILE *ofp, char const *output_file)
2160 FILE **fps = xnmalloc (nmerge, sizeof *fps);
2161 /* Input streams for each file. */
2162 struct buffer *buffer = xnmalloc (nmerge, sizeof *buffer);
2163 /* Input buffers for each file. */
2164 struct line saved; /* Saved line storage for unique check. */
2165 struct line const *savedline = NULL;
2166 /* &saved if there is a saved line. */
2167 size_t savealloc = 0; /* Size allocated for the saved line. */
2168 struct line const **cur = xnmalloc (nmerge, sizeof *cur);
2169 /* Current line in each line table. */
2170 struct line const **base = xnmalloc (nmerge, sizeof *base);
2171 /* Base of each line table. */
2172 size_t *ord = xnmalloc (nmerge, sizeof *ord);
2173 /* Table representing a permutation of fps,
2174 such that cur[ord[0]] is the smallest line
2175 and will be next output. */
2179 struct keyfield const *key = keylist;
2182 /* Read initial lines from each input file. */
2183 for (i = 0; i < nfiles; )
2185 fps[i] = (files[i].pid
2186 ? open_temp (files[i].name, files[i].pid)
2187 : xfopen (files[i].name, "r"));
2188 initbuf (&buffer[i], sizeof (struct line),
2189 MAX (merge_buffer_size, sort_size / nfiles));
2190 if (fillbuf (&buffer[i], fps[i], files[i].name))
2192 struct line const *linelim = buffer_linelim (&buffer[i]);
2193 cur[i] = linelim - 1;
2194 base[i] = linelim - buffer[i].nlines;
2199 /* fps[i] is empty; eliminate it from future consideration. */
2200 xfclose (fps[i], files[i].name);
2204 zaptemp (files[i].name);
2206 free (buffer[i].buf);
2208 for (j = i; j < nfiles; ++j)
2209 files[j] = files[j + 1];
2214 ofp = xfopen (output_file, "w");
2216 /* Set up the ord table according to comparisons among input lines.
2217 Since this only reorders two items if one is strictly greater than
2218 the other, it is stable. */
2219 for (i = 0; i < nfiles; ++i)
2221 for (i = 1; i < nfiles; ++i)
2222 if (0 < compare (cur[ord[i - 1]], cur[ord[i]]))
2223 t = ord[i - 1], ord[i - 1] = ord[i], ord[i] = t, i = 0;
2225 /* Repeatedly output the smallest line until no input remains. */
2228 struct line const *smallest = cur[ord[0]];
2230 /* If uniquified output is turned on, output only the first of
2231 an identical series of lines. */
2234 if (savedline && compare (savedline, smallest))
2237 write_bytes (saved.text, saved.length, ofp, output_file);
2242 if (savealloc < smallest->length)
2247 savealloc = smallest->length;
2250 while ((savealloc *= 2) < smallest->length);
2252 saved.text = xrealloc (saved.text, savealloc);
2254 saved.length = smallest->length;
2255 memcpy (saved.text, smallest->text, saved.length);
2259 saved.text + (smallest->keybeg - smallest->text);
2261 saved.text + (smallest->keylim - smallest->text);
2266 write_bytes (smallest->text, smallest->length, ofp, output_file);
2268 /* Check if we need to read more lines into core. */
2269 if (base[ord[0]] < smallest)
2270 cur[ord[0]] = smallest - 1;
2273 if (fillbuf (&buffer[ord[0]], fps[ord[0]], files[ord[0]].name))
2275 struct line const *linelim = buffer_linelim (&buffer[ord[0]]);
2276 cur[ord[0]] = linelim - 1;
2277 base[ord[0]] = linelim - buffer[ord[0]].nlines;
2281 /* We reached EOF on fps[ord[0]]. */
2282 for (i = 1; i < nfiles; ++i)
2283 if (ord[i] > ord[0])
2286 xfclose (fps[ord[0]], files[ord[0]].name);
2287 if (ord[0] < ntemps)
2290 zaptemp (files[ord[0]].name);
2292 free (buffer[ord[0]].buf);
2293 for (i = ord[0]; i < nfiles; ++i)
2295 fps[i] = fps[i + 1];
2296 files[i] = files[i + 1];
2297 buffer[i] = buffer[i + 1];
2298 cur[i] = cur[i + 1];
2299 base[i] = base[i + 1];
2301 for (i = 0; i < nfiles; ++i)
2302 ord[i] = ord[i + 1];
2307 /* The new line just read in may be larger than other lines
2308 already in main memory; push it back in the queue until we
2309 encounter a line larger than it. Optimize for the common
2310 case where the new line is smallest. */
2315 size_t ord0 = ord[0];
2316 size_t count_of_smaller_lines;
2320 int cmp = compare (cur[ord0], cur[ord[probe]]);
2321 if (cmp < 0 || (cmp == 0 && ord0 < ord[probe]))
2325 probe = (lo + hi) / 2;
2328 count_of_smaller_lines = lo - 1;
2329 for (j = 0; j < count_of_smaller_lines; j++)
2330 ord[j] = ord[j + 1];
2331 ord[count_of_smaller_lines] = ord0;
2334 /* Free up some resources every once in a while. */
2335 if (MAX_PROCS_BEFORE_REAP < nprocs)
2339 if (unique && savedline)
2341 write_bytes (saved.text, saved.length, ofp, output_file);
2345 xfclose (ofp, output_file);
2353 /* Merge into T the two sorted arrays of lines LO (with NLO members)
2354 and HI (with NHI members). T, LO, and HI point just past their
2355 respective arrays, and the arrays are in reverse order. NLO and
2356 NHI must be positive, and HI - NHI must equal T - (NLO + NHI). */
2359 mergelines (struct line *t,
2360 struct line const *lo, size_t nlo,
2361 struct line const *hi, size_t nhi)
2364 if (compare (lo - 1, hi - 1) <= 0)
2369 /* HI - NHI equalled T - (NLO + NHI) when this function
2370 began. Therefore HI must equal T now, and there is no
2371 need to copy from HI to T. */
2389 /* Sort the array LINES with NLINES members, using TEMP for temporary space.
2390 NLINES must be at least 2.
2391 The input and output arrays are in reverse order, and LINES and
2392 TEMP point just past the end of their respective arrays.
2394 Use a recursive divide-and-conquer algorithm, in the style
2395 suggested by Knuth volume 3 (2nd edition), exercise 5.2.4-23. Use
2396 the optimization suggested by exercise 5.2.4-10; this requires room
2397 for only 1.5*N lines, rather than the usual 2*N lines. Knuth
2398 writes that this memory optimization was originally published by
2399 D. A. Bell, Comp J. 1 (1958), 75. */
2402 sortlines (struct line *lines, size_t nlines, struct line *temp)
2406 if (0 < compare (&lines[-1], &lines[-2]))
2408 struct line tmp = lines[-1];
2409 lines[-1] = lines[-2];
2415 size_t nlo = nlines / 2;
2416 size_t nhi = nlines - nlo;
2417 struct line *lo = lines;
2418 struct line *hi = lines - nlo;
2419 struct line *sorted_lo = temp;
2421 sortlines (hi, nhi, temp);
2423 sortlines_temp (lo, nlo, sorted_lo);
2425 sorted_lo[-1] = lo[-1];
2427 mergelines (lines, sorted_lo, nlo, hi, nhi);
2431 /* Like sortlines (LINES, NLINES, TEMP), except output into TEMP
2432 rather than sorting in place. */
2435 sortlines_temp (struct line *lines, size_t nlines, struct line *temp)
2439 /* Declare `swap' as int, not bool, to work around a bug
2440 <http://lists.gnu.org/archive/html/bug-coreutils/2005-10/msg00086.html>
2441 in the IBM xlc 6.0.0.0 compiler in 64-bit mode. */
2442 int swap = (0 < compare (&lines[-1], &lines[-2]));
2443 temp[-1] = lines[-1 - swap];
2444 temp[-2] = lines[-2 + swap];
2448 size_t nlo = nlines / 2;
2449 size_t nhi = nlines - nlo;
2450 struct line *lo = lines;
2451 struct line *hi = lines - nlo;
2452 struct line *sorted_hi = temp - nlo;
2454 sortlines_temp (hi, nhi, sorted_hi);
2456 sortlines (lo, nlo, temp);
2458 mergelines (temp, lo, nlo, sorted_hi, nhi);
2462 /* Scan through FILES[NTEMPS .. NFILES-1] looking for a file that is
2463 the same as OUTFILE. If found, merge the found instances (and perhaps
2464 some other files) into a temporary file so that it can in turn be
2465 merged into OUTFILE without destroying OUTFILE before it is completely
2466 read. Return the new value of NFILES, which differs from the old if
2467 some merging occurred.
2469 This test ensures that an otherwise-erroneous use like
2470 "sort -m -o FILE ... FILE ..." copies FILE before writing to it.
2471 It's not clear that POSIX requires this nicety.
2472 Detect common error cases, but don't try to catch obscure cases like
2473 "cat ... FILE ... | sort -m -o FILE"
2474 where traditional "sort" doesn't copy the input and where
2475 people should know that they're getting into trouble anyway.
2476 Catching these obscure cases would slow down performance in
2480 avoid_trashing_input (struct sortfile *files, size_t ntemps,
2481 size_t nfiles, char const *outfile)
2484 bool got_outstat = false;
2485 struct stat outstat;
2487 for (i = ntemps; i < nfiles; i++)
2489 bool is_stdin = STREQ (files[i].name, "-");
2493 if (outfile && STREQ (outfile, files[i].name) && !is_stdin)
2500 ? stat (outfile, &outstat)
2501 : fstat (STDOUT_FILENO, &outstat))
2508 ? fstat (STDIN_FILENO, &instat)
2509 : stat (files[i].name, &instat))
2511 && SAME_INODE (instat, outstat));
2518 char *temp = create_temp (&tftp, &pid);
2519 mergefps (&files[i],0, nfiles - i, tftp, temp);
2520 files[i].name = temp;
2529 /* Merge the input FILES. NTEMPS is the number of files at the
2530 start of FILES that are temporary; it is zero at the top level.
2531 NFILES is the total number of files. Put the output in
2532 OUTPUT_FILE; a null OUTPUT_FILE stands for standard output. */
2535 merge (struct sortfile *files, size_t ntemps, size_t nfiles,
2536 char const *output_file)
2538 while (nmerge < nfiles)
2540 /* Number of input files processed so far. */
2543 /* Number of output files generated so far. */
2546 /* nfiles % NMERGE; this counts input files that are left over
2547 after all full-sized merges have been done. */
2550 /* Number of easily-available slots at the next loop iteration. */
2553 /* Do as many NMERGE-size merges as possible. */
2554 for (out = in = 0; out < nfiles / nmerge; out++, in += nmerge)
2558 char *temp = create_temp (&tfp, &pid);
2559 size_t nt = MIN (ntemps, nmerge);
2561 mergefps (&files[in], nt, nmerge, tfp, temp);
2562 files[out].name = temp;
2563 files[out].pid = pid;
2566 remainder = nfiles - in;
2567 cheap_slots = nmerge - out % nmerge;
2569 if (cheap_slots < remainder)
2571 /* So many files remain that they can't all be put into the last
2572 NMERGE-sized output window. Do one more merge. Merge as few
2573 files as possible, to avoid needless I/O. */
2574 size_t nshortmerge = remainder - cheap_slots + 1;
2577 char *temp = create_temp (&tfp, &pid);
2578 size_t nt = MIN (ntemps, nshortmerge);
2580 mergefps (&files[in], nt, nshortmerge, tfp, temp);
2581 files[out].name = temp;
2582 files[out++].pid = pid;
2586 /* Put the remaining input files into the last NMERGE-sized output
2587 window, so they will be merged in the next pass. */
2588 memmove(&files[out], &files[in], (nfiles - in) * sizeof *files);
2593 nfiles = avoid_trashing_input (files, ntemps, nfiles, output_file);
2594 mergefps (files, ntemps, nfiles, NULL, output_file);
2597 /* Sort NFILES FILES onto OUTPUT_FILE. */
2600 sort (char * const *files, size_t nfiles, char const *output_file)
2604 bool output_file_created = false;
2610 char const *temp_output;
2611 char const *file = *files;
2612 FILE *fp = xfopen (file, "r");
2614 size_t bytes_per_line = (2 * sizeof (struct line)
2615 - sizeof (struct line) / 2);
2618 initbuf (&buf, bytes_per_line,
2619 sort_buffer_size (&fp, 1, files, nfiles, bytes_per_line));
2624 while (fillbuf (&buf, fp, file))
2627 struct line *linebase;
2629 if (buf.eof && nfiles
2630 && (bytes_per_line + 1
2631 < (buf.alloc - buf.used - bytes_per_line * buf.nlines)))
2633 /* End of file, but there is more input and buffer room.
2634 Concatenate the next input file; this is faster in
2636 buf.left = buf.used;
2640 line = buffer_linelim (&buf);
2641 linebase = line - buf.nlines;
2643 sortlines (line, buf.nlines, linebase);
2644 if (buf.eof && !nfiles && !ntemps && !buf.left)
2647 tfp = xfopen (output_file, "w");
2648 temp_output = output_file;
2649 output_file_created = true;
2654 temp_output = create_temp (&tfp, NULL);
2660 write_bytes (line->text, line->length, tfp, temp_output);
2662 while (linebase < line && compare (line, line - 1) == 0)
2665 while (linebase < line);
2667 xfclose (tfp, temp_output);
2669 /* Free up some resources every once in a while. */
2670 if (MAX_PROCS_BEFORE_REAP < nprocs)
2673 if (output_file_created)
2682 if (! output_file_created)
2685 struct tempnode *node = temphead;
2686 struct sortfile *tempfiles = xnmalloc (ntemps, sizeof *tempfiles);
2687 for (i = 0; node; i++)
2689 tempfiles[i].name = node->name;
2690 tempfiles[i].pid = node->pid;
2693 merge (tempfiles, ntemps, ntemps, output_file);
2698 /* Insert a malloc'd copy of key KEY_ARG at the end of the key list. */
2701 insertkey (struct keyfield *key_arg)
2703 struct keyfield **p;
2704 struct keyfield *key = xmemdup (key_arg, sizeof *key);
2706 for (p = &keylist; *p; p = &(*p)->next)
2712 /* Report a bad field specification SPEC, with extra info MSGID. */
2714 static void badfieldspec (char const *, char const *)
2717 badfieldspec (char const *spec, char const *msgid)
2719 error (SORT_FAILURE, 0, _("%s: invalid field specification %s"),
2720 _(msgid), quote (spec));
2724 /* Report incompatible options. */
2726 static void incompatible_options (char const *) ATTRIBUTE_NORETURN;
2728 incompatible_options (char const *opts)
2730 error (SORT_FAILURE, 0, _("options `-%s' are incompatible"), opts);
2734 /* Check compatibility of ordering options. */
2737 check_ordering_compatibility (void)
2739 struct keyfield const *key;
2741 for (key = keylist; key; key = key->next)
2742 if ((1 < (key->random + key->numeric + key->general_numeric + key->month
2743 + key->version + !!key->ignore))
2744 || (key->random && key->translate))
2746 /* The following is too big, but guaranteed to be "big enough". */
2747 char opts[sizeof short_options];
2749 if (key->ignore == nondictionary)
2753 if (key->general_numeric)
2755 if (key->ignore == nonprinting)
2766 incompatible_options (opts);
2770 /* Parse the leading integer in STRING and store the resulting value
2771 (which must fit into size_t) into *VAL. Return the address of the
2772 suffix after the integer. If the value is too large, silently
2773 substitute SIZE_MAX. If MSGID is NULL, return NULL after
2774 failure; otherwise, report MSGID and exit on failure. */
2777 parse_field_count (char const *string, size_t *val, char const *msgid)
2782 switch (xstrtoumax (string, &suffix, 10, &n, ""))
2785 case LONGINT_INVALID_SUFFIX_CHAR:
2790 case LONGINT_OVERFLOW:
2791 case LONGINT_OVERFLOW | LONGINT_INVALID_SUFFIX_CHAR:
2795 case LONGINT_INVALID:
2797 error (SORT_FAILURE, 0, _("%s: invalid count at start of %s"),
2798 _(msgid), quote (string));
2805 /* Handle interrupts and hangups. */
2808 sighandler (int sig)
2811 signal (sig, SIG_IGN);
2815 signal (sig, SIG_DFL);
2819 /* Set the ordering options for KEY specified in S.
2820 Return the address of the first character in S that
2821 is not a valid ordering option.
2822 BLANKTYPE is the kind of blanks that 'b' should skip. */
2825 set_ordering (const char *s, struct keyfield *key, enum blanktype blanktype)
2832 if (blanktype == bl_start || blanktype == bl_both)
2833 key->skipsblanks = true;
2834 if (blanktype == bl_end || blanktype == bl_both)
2835 key->skipeblanks = true;
2838 key->ignore = nondictionary;
2841 key->translate = fold_toupper;
2844 key->general_numeric = true;
2847 /* Option order should not matter, so don't let -i override
2848 -d. -d implies -i, but -i does not imply -d. */
2850 key->ignore = nonprinting;
2856 key->numeric = true;
2862 key->reverse = true;
2865 key->version = true;
2875 static struct keyfield *
2876 key_init (struct keyfield *key)
2878 memset (key, 0, sizeof *key);
2879 key->eword = SIZE_MAX;
2884 main (int argc, char **argv)
2886 struct keyfield *key;
2887 struct keyfield key_buf;
2888 struct keyfield gkey;
2892 bool mergeonly = false;
2893 char *random_source = NULL;
2894 bool need_random = false;
2896 bool posixly_correct = (getenv ("POSIXLY_CORRECT") != NULL);
2897 bool obsolete_usage = (posix2_version () < 200112);
2899 char *files_from = NULL;
2901 char const *outfile = NULL;
2903 initialize_main (&argc, &argv);
2904 set_program_name (argv[0]);
2905 setlocale (LC_ALL, "");
2906 bindtextdomain (PACKAGE, LOCALEDIR);
2907 textdomain (PACKAGE);
2909 initialize_exit_failure (SORT_FAILURE);
2911 hard_LC_COLLATE = hard_locale (LC_COLLATE);
2912 #if HAVE_NL_LANGINFO
2913 hard_LC_TIME = hard_locale (LC_TIME);
2916 /* Get locale's representation of the decimal point. */
2918 struct lconv const *locale = localeconv ();
2920 /* If the locale doesn't define a decimal point, or if the decimal
2921 point is multibyte, use the C locale's decimal point. FIXME:
2922 add support for multibyte decimal points. */
2923 decimal_point = to_uchar (locale->decimal_point[0]);
2924 if (! decimal_point || locale->decimal_point[1])
2925 decimal_point = '.';
2927 /* FIXME: add support for multibyte thousands separators. */
2928 thousands_sep = to_uchar (*locale->thousands_sep);
2929 if (! thousands_sep || locale->thousands_sep[1])
2933 have_read_stdin = false;
2938 static int const sig[] =
2940 /* The usual suspects. */
2941 SIGALRM, SIGHUP, SIGINT, SIGPIPE, SIGQUIT, SIGTERM,
2958 enum { nsigs = sizeof sig / sizeof sig[0] };
2961 struct sigaction act;
2963 sigemptyset (&caught_signals);
2964 for (i = 0; i < nsigs; i++)
2966 sigaction (sig[i], NULL, &act);
2967 if (act.sa_handler != SIG_IGN)
2968 sigaddset (&caught_signals, sig[i]);
2971 act.sa_handler = sighandler;
2972 act.sa_mask = caught_signals;
2975 for (i = 0; i < nsigs; i++)
2976 if (sigismember (&caught_signals, sig[i]))
2977 sigaction (sig[i], &act, NULL);
2979 for (i = 0; i < nsigs; i++)
2980 if (signal (sig[i], SIG_IGN) != SIG_IGN)
2982 signal (sig[i], sighandler);
2983 siginterrupt (sig[i], 1);
2988 /* The signal mask is known, so it is safe to invoke exit_cleanup. */
2989 atexit (exit_cleanup);
2991 gkey.sword = gkey.eword = SIZE_MAX;
2993 gkey.translate = NULL;
2994 gkey.numeric = gkey.general_numeric = gkey.random = gkey.version = false;
2995 gkey.month = gkey.reverse = false;
2996 gkey.skipsblanks = gkey.skipeblanks = false;
2998 files = xnmalloc (argc, sizeof *files);
3002 /* Parse an operand as a file after "--" was seen; or if
3003 pedantic and a file was seen, unless the POSIX version
3004 predates 1003.1-2001 and -c was not seen and the operand is
3005 "-o FILE" or "-oFILE". */
3009 || (posixly_correct && nfiles != 0
3010 && ! (obsolete_usage
3013 && argv[optind][0] == '-' && argv[optind][1] == 'o'
3014 && (argv[optind][2] || optind + 1 != argc)))
3015 || ((c = getopt_long (argc, argv, short_options,
3021 files[nfiles++] = argv[optind++];
3027 if (optarg[0] == '+')
3029 bool minus_pos_usage = (optind != argc && argv[optind][0] == '-'
3030 && ISDIGIT (argv[optind][1]));
3031 obsolete_usage |= minus_pos_usage & ~posixly_correct;
3034 /* Treat +POS1 [-POS2] as a key if possible; but silently
3035 treat an operand as a file if it is not a valid +POS1. */
3036 key = key_init (&key_buf);
3037 s = parse_field_count (optarg + 1, &key->sword, NULL);
3039 s = parse_field_count (s + 1, &key->schar, NULL);
3040 if (! (key->sword | key->schar))
3041 key->sword = SIZE_MAX;
3042 if (! s || *set_ordering (s, key, bl_start))
3046 if (minus_pos_usage)
3048 char const *optarg1 = argv[optind++];
3049 s = parse_field_count (optarg1 + 1, &key->eword,
3050 N_("invalid number after `-'"));
3052 s = parse_field_count (s + 1, &key->echar,
3053 N_("invalid number after `.'"));
3054 if (*set_ordering (s, key, bl_end))
3055 badfieldspec (optarg1,
3056 N_("stray character in field spec"));
3063 files[nfiles++] = optarg;
3067 c = XARGMATCH ("--sort", optarg, sort_args, sort_types);
3083 set_ordering (str, &gkey, bl_both);
3089 ? XARGMATCH ("--check", optarg, check_args, check_types)
3094 if (checkonly && checkonly != c)
3095 incompatible_options ("cC");
3099 case COMPRESS_PROGRAM_OPTION:
3100 if (compress_program && !STREQ (compress_program, optarg))
3101 error (SORT_FAILURE, 0, _("multiple compress programs specified"));
3102 compress_program = optarg;
3105 case FILES0_FROM_OPTION:
3106 files_from = optarg;
3110 key = key_init (&key_buf);
3113 s = parse_field_count (optarg, &key->sword,
3114 N_("invalid number at field start"));
3117 /* Provoke with `sort -k0' */
3118 badfieldspec (optarg, N_("field number is zero"));
3122 s = parse_field_count (s + 1, &key->schar,
3123 N_("invalid number after `.'"));
3126 /* Provoke with `sort -k1.0' */
3127 badfieldspec (optarg, N_("character offset is zero"));
3130 if (! (key->sword | key->schar))
3131 key->sword = SIZE_MAX;
3132 s = set_ordering (s, key, bl_start);
3135 key->eword = SIZE_MAX;
3141 s = parse_field_count (s + 1, &key->eword,
3142 N_("invalid number after `,'"));
3145 /* Provoke with `sort -k1,0' */
3146 badfieldspec (optarg, N_("field number is zero"));
3149 s = parse_field_count (s + 1, &key->echar,
3150 N_("invalid number after `.'"));
3153 /* `-k 2,3' is equivalent to `+1 -3'. */
3156 s = set_ordering (s, key, bl_end);
3159 badfieldspec (optarg, N_("stray character in field spec"));
3168 specify_nmerge (oi, c, optarg);
3172 if (outfile && !STREQ (outfile, optarg))
3173 error (SORT_FAILURE, 0, _("multiple output files specified"));
3177 case RANDOM_SOURCE_OPTION:
3178 if (random_source && !STREQ (random_source, optarg))
3179 error (SORT_FAILURE, 0, _("multiple random sources specified"));
3180 random_source = optarg;
3188 specify_sort_size (oi, c, optarg);
3193 char newtab = optarg[0];
3195 error (SORT_FAILURE, 0, _("empty tab"));
3198 if (STREQ (optarg, "\\0"))
3202 /* Provoke with `sort -txx'. Complain about
3203 "multi-character tab" instead of "multibyte tab", so
3204 that the diagnostic's wording does not need to be
3205 changed once multibyte characters are supported. */
3206 error (SORT_FAILURE, 0, _("multi-character tab %s"),
3210 if (tab != TAB_DEFAULT && tab != newtab)
3211 error (SORT_FAILURE, 0, _("incompatible tabs"));
3217 add_temp_dir (optarg);
3225 /* Accept and ignore e.g. -y0 for compatibility with Solaris 2.x
3226 through Solaris 7. It is also accepted by many non-Solaris
3227 "sort" implementations, e.g., AIX 5.2, HP-UX 11i v2, IRIX 6.5.
3228 -y is marked as obsolete starting with Solaris 8 (1999), but is
3229 still accepted as of Solaris 10 prerelease (2004).
3231 Solaris 2.5.1 "sort -y 100" reads the input file "100", but
3232 emulate Solaris 8 and 9 "sort -y 100" which ignores the "100",
3233 and which in general ignores the argument after "-y" if it
3234 consists entirely of digits (it can even be empty). */
3235 if (optarg == argv[optind - 1])
3238 for (p = optarg; ISDIGIT (*p); p++)
3240 optind -= (*p != '\0');
3248 case_GETOPT_HELP_CHAR;
3250 case_GETOPT_VERSION_CHAR (PROGRAM_NAME, AUTHORS);
3253 usage (SORT_FAILURE);
3261 /* When using --files0-from=F, you may not specify any files
3262 on the command-line. */
3265 error (0, 0, _("extra operand %s"), quote (files[0]));
3266 fprintf (stderr, "%s\n",
3267 _("file operands cannot be combined with --files0-from"));
3268 usage (SORT_FAILURE);
3271 if (STREQ (files_from, "-"))
3275 stream = fopen (files_from, "r");
3277 error (SORT_FAILURE, errno, _("cannot open %s for reading"),
3278 quote (files_from));
3281 readtokens0_init (&tok);
3283 if (! readtokens0 (stream, &tok) || fclose (stream) != 0)
3284 error (SORT_FAILURE, 0, _("cannot read file names from %s"),
3285 quote (files_from));
3293 for (i = 0; i < nfiles; i++)
3295 if (STREQ (files[i], "-"))
3296 error (SORT_FAILURE, 0, _("when reading file names from stdin, "
3297 "no file name of %s allowed"),
3299 else if (files[i][0] == '\0')
3301 /* Using the standard `filename:line-number:' prefix here is
3302 not totally appropriate, since NUL is the separator, not NL,
3303 but it might be better than nothing. */
3304 unsigned long int file_number = i + 1;
3305 error (SORT_FAILURE, 0,
3306 _("%s:%lu: invalid zero-length file name"),
3307 quotearg_colon (files_from), file_number);
3312 error (SORT_FAILURE, 0, _("no input from %s"),
3313 quote (files_from));
3316 /* Inheritance of global options to individual keys. */
3317 for (key = keylist; key; key = key->next)
3321 || (key->skipsblanks
3327 | key->general_numeric
3330 key->ignore = gkey.ignore;
3331 key->translate = gkey.translate;
3332 key->skipsblanks = gkey.skipsblanks;
3333 key->skipeblanks = gkey.skipeblanks;
3334 key->month = gkey.month;
3335 key->numeric = gkey.numeric;
3336 key->general_numeric = gkey.general_numeric;
3337 key->random = gkey.random;
3338 key->reverse = gkey.reverse;
3339 key->version = gkey.version;
3342 need_random |= key->random;
3345 if (!keylist && (gkey.ignore
3347 || (gkey.skipsblanks
3351 | gkey.general_numeric
3356 need_random |= gkey.random;
3359 check_ordering_compatibility ();
3361 reverse = gkey.reverse;
3365 randread_source = randread_new (random_source, MD5_DIGEST_SIZE);
3366 if (! randread_source)
3367 die (_("open failed"), random_source);
3370 if (temp_dir_count == 0)
3372 char const *tmp_dir = getenv ("TMPDIR");
3373 add_temp_dir (tmp_dir ? tmp_dir : DEFAULT_TMPDIR);
3378 static char *minus = "-";
3384 /* Need to re-check that we meet the minimum requirement for memory
3385 usage with the final value for NMERGE. */
3387 sort_size = MAX (sort_size, MIN_SORT_SIZE);
3392 error (SORT_FAILURE, 0, _("extra operand %s not allowed with -%c"),
3393 quote (files[1]), checkonly);
3397 static char opts[] = {0, 'o', 0};
3398 opts[0] = checkonly;
3399 incompatible_options (opts);
3402 /* POSIX requires that sort return 1 IFF invoked with -c or -C and the
3403 input is not properly sorted. */
3404 exit (check (files[0], checkonly) ? EXIT_SUCCESS : SORT_OUT_OF_ORDER);
3409 struct sortfile *sortfiles = xcalloc (nfiles, sizeof *sortfiles);
3412 for (i = 0; i < nfiles; ++i)
3413 sortfiles[i].name = files[i];
3415 merge (sortfiles, 0, nfiles, outfile);
3416 IF_LINT (free (sortfiles));
3419 sort (files, nfiles, outfile);
3421 if (have_read_stdin && fclose (stdin) == EOF)
3422 die (_("close failed"), "-");
3424 exit (EXIT_SUCCESS);