1 /* tac - concatenate and print files in reverse
2 Copyright (C) 1988-1991, 1995-2006 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 by Jay Lepreau (lepreau@cs.utah.edu).
18 GNU enhancements by David MacKenzie (djm@gnu.ai.mit.edu). */
20 /* Copy each FILE, or the standard input if none are given or when a
21 FILE name of "-" is encountered, to the standard output with the
22 order of the records reversed. The records are separated by
23 instances of a string, or a newline if none is given. By default, the
24 separator string is attached to the end of the record that it
28 -b, --before The separator is attached to the beginning
29 of the record that it precedes in the file.
30 -r, --regex The separator is a regular expression.
31 -s, --separator=separator Use SEPARATOR as the record separator.
33 To reverse a file byte by byte, use (in bash, ksh, or sh):
41 #include <sys/types.h>
49 #include "safe-read.h"
52 /* The official name of this program (e.g., no `g' prefix). */
53 #define PROGRAM_NAME "tac"
55 #define AUTHORS "Jay Lepreau", "David MacKenzie"
57 #if defined __MSDOS__ || defined _WIN32
58 /* Define this to non-zero on systems for which the regular mechanism
59 (of unlinking an open file and expecting to be able to write, seek
60 back to the beginning, then reread it) doesn't work. E.g., on Windows
62 # define DONT_UNLINK_WHILE_OPEN 1
66 #ifndef DEFAULT_TMPDIR
67 # define DEFAULT_TMPDIR "/tmp"
70 /* The number of bytes per atomic read. */
71 #define INITIAL_READSIZE 8192
73 /* The number of bytes per atomic write. */
74 #define WRITESIZE 8192
76 /* The name this program was run with. */
79 /* The string that separates the records of the file. */
80 static char const *separator
;
82 /* True if we have ever read standard input. */
83 static bool have_read_stdin
= false;
85 /* If true, print `separator' along with the record preceding it
86 in the file; otherwise with the record following it. */
87 static bool separator_ends_record
;
89 /* 0 if `separator' is to be matched as a regular expression;
90 otherwise, the length of `separator', used as a sentinel to
92 static size_t sentinel_length
;
94 /* The length of a match with `separator'. If `sentinel_length' is 0,
95 `match_length' is computed every time a match succeeds;
96 otherwise, it is simply the length of `separator'. */
97 static size_t match_length
;
99 /* The input buffer. */
100 static char *G_buffer
;
102 /* The number of bytes to read at once into `buffer'. */
103 static size_t read_size
;
105 /* The size of `buffer'. This is read_size * 2 + sentinel_length + 2.
106 The extra 2 bytes allow `past_end' to have a value beyond the
107 end of `G_buffer' and `match_start' to run off the front of `G_buffer'. */
108 static size_t G_buffer_size
;
110 /* The compiled regular expression representing `separator'. */
111 static struct re_pattern_buffer compiled_separator
;
112 static char compiled_separator_fastmap
[UCHAR_MAX
+ 1];
114 static struct option
const longopts
[] =
116 {"before", no_argument
, NULL
, 'b'},
117 {"regex", no_argument
, NULL
, 'r'},
118 {"separator", required_argument
, NULL
, 's'},
119 {GETOPT_HELP_OPTION_DECL
},
120 {GETOPT_VERSION_OPTION_DECL
},
127 if (status
!= EXIT_SUCCESS
)
128 fprintf (stderr
, _("Try `%s --help' for more information.\n"),
133 Usage: %s [OPTION]... [FILE]...\n\
137 Write each FILE to standard output, last line first.\n\
138 With no FILE, or when FILE is -, read standard input.\n\
142 Mandatory arguments to long options are mandatory for short options too.\n\
145 -b, --before attach the separator before instead of after\n\
146 -r, --regex interpret the separator as a regular expression\n\
147 -s, --separator=STRING use STRING as the separator instead of newline\n\
149 fputs (HELP_OPTION_DESCRIPTION
, stdout
);
150 fputs (VERSION_OPTION_DESCRIPTION
, stdout
);
151 emit_bug_reporting_address ();
156 /* Print the characters from START to PAST_END - 1.
157 If START is NULL, just flush the buffer. */
160 output (const char *start
, const char *past_end
)
162 static char buffer
[WRITESIZE
];
163 static size_t bytes_in_buffer
= 0;
164 size_t bytes_to_add
= past_end
- start
;
165 size_t bytes_available
= WRITESIZE
- bytes_in_buffer
;
169 fwrite (buffer
, 1, bytes_in_buffer
, stdout
);
174 /* Write out as many full buffers as possible. */
175 while (bytes_to_add
>= bytes_available
)
177 memcpy (buffer
+ bytes_in_buffer
, start
, bytes_available
);
178 bytes_to_add
-= bytes_available
;
179 start
+= bytes_available
;
180 fwrite (buffer
, 1, WRITESIZE
, stdout
);
182 bytes_available
= WRITESIZE
;
185 memcpy (buffer
+ bytes_in_buffer
, start
, bytes_to_add
);
186 bytes_in_buffer
+= bytes_to_add
;
189 /* Print in reverse the file open on descriptor FD for reading FILE.
190 Return true if successful. */
193 tac_seekable (int input_fd
, const char *file
)
195 /* Pointer to the location in `G_buffer' where the search for
196 the next separator will begin. */
199 /* Pointer to one past the rightmost character in `G_buffer' that
200 has not been printed yet. */
203 /* Length of the record growing in `G_buffer'. */
204 size_t saved_record_size
;
206 /* Offset in the file of the next read. */
209 /* True if `output' has not been called yet for any file.
210 Only used when the separator is attached to the preceding record. */
211 bool first_time
= true;
212 char first_char
= *separator
; /* Speed optimization, non-regexp. */
213 char const *separator1
= separator
+ 1; /* Speed optimization, non-regexp. */
214 size_t match_length1
= match_length
- 1; /* Speed optimization, non-regexp. */
215 struct re_registers regs
;
217 /* Find the size of the input file. */
218 file_pos
= lseek (input_fd
, (off_t
) 0, SEEK_END
);
220 return true; /* It's an empty file. */
222 /* Arrange for the first read to lop off enough to leave the rest of the
223 file a multiple of `read_size'. Since `read_size' can change, this may
224 not always hold during the program run, but since it usually will, leave
225 it here for i/o efficiency (page/sector boundaries and all that).
226 Note: the efficiency gain has not been verified. */
227 saved_record_size
= file_pos
% read_size
;
228 if (saved_record_size
== 0)
229 saved_record_size
= read_size
;
230 file_pos
-= saved_record_size
;
231 /* `file_pos' now points to the start of the last (probably partial) block
232 in the input file. */
234 if (lseek (input_fd
, file_pos
, SEEK_SET
) < 0)
235 error (0, errno
, _("%s: seek failed"), quotearg_colon (file
));
237 if (safe_read (input_fd
, G_buffer
, saved_record_size
) != saved_record_size
)
239 error (0, errno
, _("%s: read error"), quotearg_colon (file
));
243 match_start
= past_end
= G_buffer
+ saved_record_size
;
244 /* For non-regexp search, move past impossible positions for a match. */
246 match_start
-= match_length1
;
250 /* Search backward from `match_start' - 1 to `G_buffer' for a match
251 with `separator'; for speed, use strncmp if `separator' contains no
253 If the match succeeds, set `match_start' to point to the start of
254 the match and `match_length' to the length of the match.
255 Otherwise, make `match_start' < `G_buffer'. */
256 if (sentinel_length
== 0)
258 size_t i
= match_start
- G_buffer
;
260 regoff_t range
= 1 - ri
;
264 error (EXIT_FAILURE
, 0, _("record too large"));
267 || ((ret
= re_search (&compiled_separator
, G_buffer
,
268 i
, i
- 1, range
, ®s
))
270 match_start
= G_buffer
- 1;
273 error (EXIT_FAILURE
, 0,
274 _("error in regular expression search"));
278 match_start
= G_buffer
+ regs
.start
[0];
279 match_length
= regs
.end
[0] - regs
.start
[0];
284 /* `match_length' is constant for non-regexp boundaries. */
285 while (*--match_start
!= first_char
286 || (match_length1
&& strncmp (match_start
+ 1, separator1
,
291 /* Check whether we backed off the front of `G_buffer' without finding
292 a match for `separator'. */
293 if (match_start
< G_buffer
)
297 /* Hit the beginning of the file; print the remaining record. */
298 output (G_buffer
, past_end
);
302 saved_record_size
= past_end
- G_buffer
;
303 if (saved_record_size
> read_size
)
305 /* `G_buffer_size' is about twice `read_size', so since
306 we want to read in another `read_size' bytes before
307 the data already in `G_buffer', we need to increase
310 size_t offset
= sentinel_length
? sentinel_length
: 1;
311 ptrdiff_t match_start_offset
= match_start
- G_buffer
;
312 ptrdiff_t past_end_offset
= past_end
- G_buffer
;
313 size_t old_G_buffer_size
= G_buffer_size
;
316 G_buffer_size
= read_size
* 2 + sentinel_length
+ 2;
317 if (G_buffer_size
< old_G_buffer_size
)
319 newbuffer
= xrealloc (G_buffer
- offset
, G_buffer_size
);
321 /* Adjust the pointers for the new buffer location. */
322 match_start
= newbuffer
+ match_start_offset
;
323 past_end
= newbuffer
+ past_end_offset
;
324 G_buffer
= newbuffer
;
327 /* Back up to the start of the next bufferfull of the file. */
328 if (file_pos
>= read_size
)
329 file_pos
-= read_size
;
332 read_size
= file_pos
;
335 if (lseek (input_fd
, file_pos
, SEEK_SET
) < 0)
336 error (0, errno
, _("%s: seek failed"), quotearg_colon (file
));
338 /* Shift the pending record data right to make room for the new.
339 The source and destination regions probably overlap. */
340 memmove (G_buffer
+ read_size
, G_buffer
, saved_record_size
);
341 past_end
= G_buffer
+ read_size
+ saved_record_size
;
342 /* For non-regexp searches, avoid unneccessary scanning. */
344 match_start
= G_buffer
+ read_size
;
346 match_start
= past_end
;
348 if (safe_read (input_fd
, G_buffer
, read_size
) != read_size
)
350 error (0, errno
, _("%s: read error"), quotearg_colon (file
));
356 /* Found a match of `separator'. */
357 if (separator_ends_record
)
359 char *match_end
= match_start
+ match_length
;
361 /* If this match of `separator' isn't at the end of the
362 file, print the record. */
363 if (!first_time
|| match_end
!= past_end
)
364 output (match_end
, past_end
);
365 past_end
= match_end
;
370 output (match_start
, past_end
);
371 past_end
= match_start
;
374 /* For non-regex matching, we can back up. */
375 if (sentinel_length
> 0)
376 match_start
-= match_length
- 1;
381 #if DONT_UNLINK_WHILE_OPEN
383 /* FIXME-someday: remove all of this DONT_UNLINK_WHILE_OPEN junk.
384 Using atexit like this is wrong, since it can fail
385 when called e.g. 32 or more times.
386 But this isn't a big deal, since the code is used only on WOE/DOS
387 systems, and few people invoke tac on that many nonseekable files. */
389 static const char *file_to_remove
;
390 static FILE *fp_to_close
;
393 unlink_tempfile (void)
395 fclose (fp_to_close
);
396 unlink (file_to_remove
);
400 record_or_unlink_tempfile (char const *fn
, FILE *fp
)
406 atexit (unlink_tempfile
);
413 record_or_unlink_tempfile (char const *fn
, FILE *fp ATTRIBUTE_UNUSED
)
420 /* Copy from file descriptor INPUT_FD (corresponding to the named FILE) to
421 a temporary file, and set *G_TMP and *G_TEMPFILE to the resulting stream
422 and file name. Return true if successful. */
425 copy_to_temp (FILE **g_tmp
, char **g_tempfile
, int input_fd
, char const *file
)
427 static char *template = NULL
;
428 static char const *tempdir
;
433 if (template == NULL
)
435 char const * const Template
= "%s/tacXXXXXX";
436 tempdir
= getenv ("TMPDIR");
438 tempdir
= DEFAULT_TMPDIR
;
440 /* Subtract 2 for `%s' and add 1 for the trailing NUL byte. */
441 template = xmalloc (strlen (tempdir
) + strlen (Template
) - 2 + 1);
442 sprintf (template, Template
, tempdir
);
445 /* FIXME: there's a small window between a successful mkstemp call
446 and the unlink that's performed by record_or_unlink_tempfile.
447 If we're interrupted in that interval, this code fails to remove
448 the temporary file. On systems that define DONT_UNLINK_WHILE_OPEN,
449 the window is much larger -- it extends to the atexit-called
451 FIXME: clean up upon fatal signal. Don't block them, in case
452 $TMPFILE is a remote file system. */
455 fd
= mkstemp (template);
458 error (0, errno
, _("cannot create temporary file %s"), quote (tempfile
));
462 tmp
= fdopen (fd
, (O_BINARY
? "w+b" : "w+"));
465 error (0, errno
, _("cannot open %s for writing"), quote (tempfile
));
471 record_or_unlink_tempfile (tempfile
, tmp
);
475 size_t bytes_read
= safe_read (input_fd
, G_buffer
, read_size
);
478 if (bytes_read
== SAFE_READ_ERROR
)
480 error (0, errno
, _("%s: read error"), quotearg_colon (file
));
484 if (fwrite (G_buffer
, 1, bytes_read
, tmp
) != bytes_read
)
486 error (0, errno
, _("%s: write error"), quotearg_colon (tempfile
));
491 if (fflush (tmp
) != 0)
493 error (0, errno
, _("%s: write error"), quotearg_colon (tempfile
));
498 *g_tempfile
= tempfile
;
506 /* Copy INPUT_FD to a temporary, then tac that file.
507 Return true if successful. */
510 tac_nonseekable (int input_fd
, const char *file
)
514 return (copy_to_temp (&tmp_stream
, &tmp_file
, input_fd
, file
)
515 && tac_seekable (fileno (tmp_stream
), tmp_file
));
518 /* Print FILE in reverse, copying it to a temporary
519 file first if it is not seekable.
520 Return true if successful. */
523 tac_file (const char *filename
)
528 bool is_stdin
= STREQ (filename
, "-");
532 have_read_stdin
= true;
534 filename
= _("standard input");
535 if (O_BINARY
&& ! isatty (STDIN_FILENO
))
536 freopen (NULL
, "rb", stdin
);
540 fd
= open (filename
, O_RDONLY
| O_BINARY
);
543 error (0, errno
, _("cannot open %s for reading"), quote (filename
));
548 file_size
= lseek (fd
, (off_t
) 0, SEEK_END
);
550 ok
= (file_size
< 0 || isatty (fd
)
551 ? tac_nonseekable (fd
, filename
)
552 : tac_seekable (fd
, filename
));
554 if (!is_stdin
&& close (fd
) != 0)
556 error (0, errno
, _("%s: read error"), quotearg_colon (filename
));
563 main (int argc
, char **argv
)
565 const char *error_message
; /* Return value from re_compile_pattern. */
568 size_t half_buffer_size
;
570 /* Initializer for file_list if no file-arguments
571 were specified on the command line. */
572 static char const *const default_file_list
[] = {"-", NULL
};
573 char const *const *file
;
575 initialize_main (&argc
, &argv
);
576 program_name
= argv
[0];
577 setlocale (LC_ALL
, "");
578 bindtextdomain (PACKAGE
, LOCALEDIR
);
579 textdomain (PACKAGE
);
581 atexit (close_stdout
);
585 separator_ends_record
= true;
587 while ((optc
= getopt_long (argc
, argv
, "brs:", longopts
, NULL
)) != -1)
592 separator_ends_record
= false;
600 error (EXIT_FAILURE
, 0, _("separator cannot be empty"));
602 case_GETOPT_HELP_CHAR
;
603 case_GETOPT_VERSION_CHAR (PROGRAM_NAME
, AUTHORS
);
605 usage (EXIT_FAILURE
);
609 if (sentinel_length
== 0)
611 compiled_separator
.buffer
= NULL
;
612 compiled_separator
.allocated
= 0;
613 compiled_separator
.fastmap
= compiled_separator_fastmap
;
614 compiled_separator
.translate
= NULL
;
615 error_message
= re_compile_pattern (separator
, strlen (separator
),
616 &compiled_separator
);
618 error (EXIT_FAILURE
, 0, "%s", error_message
);
621 match_length
= sentinel_length
= strlen (separator
);
623 read_size
= INITIAL_READSIZE
;
624 while (sentinel_length
>= read_size
/ 2)
626 if (SIZE_MAX
/ 2 < read_size
)
630 half_buffer_size
= read_size
+ sentinel_length
+ 1;
631 G_buffer_size
= 2 * half_buffer_size
;
632 if (! (read_size
< half_buffer_size
&& half_buffer_size
< G_buffer_size
))
634 G_buffer
= xmalloc (G_buffer_size
);
637 strcpy (G_buffer
, separator
);
638 G_buffer
+= sentinel_length
;
645 file
= (optind
< argc
646 ? (char const *const *) &argv
[optind
]
647 : default_file_list
);
649 if (O_BINARY
&& ! isatty (STDOUT_FILENO
))
650 freopen (NULL
, "wb", stdout
);
655 for (i
= 0; file
[i
]; ++i
)
656 ok
&= tac_file (file
[i
]);
659 /* Flush the output buffer. */
660 output ((char *) NULL
, (char *) NULL
);
662 if (have_read_stdin
&& close (STDIN_FILENO
) < 0)
663 error (EXIT_FAILURE
, errno
, "-");
664 exit (ok
? EXIT_SUCCESS
: EXIT_FAILURE
);