1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
37 #ifdef HAVE_SYS_TIME_H
38 # include <sys/time.h>
44 # include <sys/mman.h>
47 # include <process.h> /* getpid() */
52 #ifdef HAVE_SYS_UTIME_H
53 # include <sys/utime.h>
61 /* For TIOCGWINSZ and friends: */
62 #ifdef HAVE_SYS_IOCTL_H
63 # include <sys/ioctl.h>
69 /* Needed for Unix version of run_with_timeout. */
73 #ifndef HAVE_SIGSETJMP
74 /* If sigsetjmp is a macro, configure won't pick it up. */
76 # define HAVE_SIGSETJMP
80 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
81 # define USE_SIGNAL_TIMEOUT
89 #endif /* def __VMS */
95 /* Character property table for (re-)escaping VMS ODS5 extended file
96 names. Note that this table ignores Unicode.
98 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
100 ODS5 Invalid characters:
101 C0 control codes (0x00 to 0x1F inclusive)
105 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
106 Double quotation marks (")
109 Left angle bracket (<)
110 Right angle bracket (>)
114 Characters escaped by "^":
115 SP ! # % & ' ( ) + , . ; = @ [ ] ^ ` { } ~
117 Either "^_" or "^ " is accepted as a space. Period (.) is a special
118 case. Note that un-escaped < and > can also confuse a directory
121 Characters put out as ^xx:
123 80-9F (C1 control characters)
124 A0 (nonbreaking space)
125 FF (Latin small letter y diaeresis)
128 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
130 Property table values:
140 unsigned char char_prop[ 256] = {
142 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
143 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
145 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
146 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
148 /* SP ! " # $ % & ' ( ) * + , - . / */
149 2, 1, 0, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
151 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
152 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 0, 1, 1, 1, 1, 1,
154 /* @ A B C D E F G H I J K L M N O */
155 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
157 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
158 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 0, 1, 1, 16,
160 /* ` a b c d e f g h i j k l m n o */
161 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
163 /* p q r s t u v w x y z { | } ~ DEL */
164 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 0, 1, 17, 8,
166 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
167 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
168 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
169 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
170 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
173 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
176 /* Utility function: like xstrdup(), but also lowercases S. */
179 xstrdup_lower (const char *s)
181 char *copy = xstrdup (s);
188 /* Copy the string formed by two pointers (one on the beginning, other
189 on the char after the last char) to a new, malloc-ed location.
192 strdupdelim (const char *beg, const char *end)
194 char *res = xmalloc (end - beg + 1);
195 memcpy (res, beg, end - beg);
196 res[end - beg] = '\0';
200 /* Parse a string containing comma-separated elements, and return a
201 vector of char pointers with the elements. Spaces following the
202 commas are ignored. */
204 sepstring (const char *s)
218 res = xrealloc (res, (i + 2) * sizeof (char *));
219 res[i] = strdupdelim (p, s);
222 /* Skip the blanks following the ','. */
223 while (c_isspace (*s))
230 res = xrealloc (res, (i + 2) * sizeof (char *));
231 res[i] = strdupdelim (p, s);
236 /* Like sprintf, but prints into a string of sufficient size freshly
237 allocated with malloc, which is returned. If unable to print due
238 to invalid format, returns NULL. Inability to allocate needed
239 memory results in abort, as with xmalloc. This is in spirit
240 similar to the GNU/BSD extension asprintf, but somewhat easier to
243 Internally the function either calls vasprintf or loops around
244 vsnprintf until the correct size is found. Since Wget also ships a
245 fallback implementation of vsnprintf, this should be portable. */
247 /* Constant is using for limits memory allocation for text buffer.
248 Applicable in situation when: vasprintf is not available in the system
249 and vsnprintf return -1 when long line is truncated (in old versions of
250 glibc and in other system where C99 doesn`t support) */
252 #define FMT_MAX_LENGTH 1048576
255 aprintf (const char *fmt, ...)
257 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
262 va_start (args, fmt);
263 ret = vasprintf (&str, fmt, args);
265 if (ret < 0 && errno == ENOMEM)
266 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
267 with xmalloc/xrealloc */
271 #else /* not HAVE_VASPRINTF */
273 /* vasprintf is unavailable. snprintf into a small buffer and
274 resize it as necessary. */
276 char *str = xmalloc (size);
278 /* #### This code will infloop and eventually abort in xrealloc if
279 passed a FMT that causes snprintf to consistently return -1. */
286 va_start (args, fmt);
287 n = vsnprintf (str, size, fmt, args);
290 /* If the printing worked, return the string. */
291 if (n > -1 && n < size)
294 /* Else try again with a larger buffer. */
295 if (n > -1) /* C99 */
296 size = n + 1; /* precisely what is needed */
297 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
298 { /* maybe we have some wrong
300 logprintf (LOG_ALWAYS,
301 _("%s: aprintf: text buffer is too big (%ld bytes), "
303 exec_name, size); /* printout a log message */
304 abort (); /* and abort... */
308 /* else, we continue to grow our
309 * buffer: Twice the old size. */
312 str = xrealloc (str, size);
314 #endif /* not HAVE_VASPRINTF */
317 /* Concatenate the NULL-terminated list of string arguments into
318 freshly allocated space. */
321 concat_strings (const char *str0, ...)
324 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
327 const char *next_str;
328 int total_length = 0;
331 /* Calculate the length of and allocate the resulting string. */
334 va_start (args, str0);
335 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
337 int len = strlen (next_str);
338 if (argcount < countof (saved_lengths))
339 saved_lengths[argcount++] = len;
343 p = ret = xmalloc (total_length + 1);
345 /* Copy the strings into the allocated space. */
348 va_start (args, str0);
349 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
352 if (argcount < countof (saved_lengths))
353 len = saved_lengths[argcount++];
355 len = strlen (next_str);
356 memcpy (p, next_str, len);
365 /* Format the provided time according to the specified format. The
366 format is a string with format elements supported by strftime. */
369 fmttime (time_t t, const char *fmt)
371 static char output[32];
372 struct tm *tm = localtime(&t);
375 if (!strftime(output, sizeof(output), fmt, tm))
380 /* Return pointer to a static char[] buffer in which zero-terminated
381 string-representation of TM (in form hh:mm:ss) is printed.
383 If TM is NULL, the current time will be used. */
388 return fmttime(t, "%H:%M:%S");
391 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
394 datetime_str (time_t t)
396 return fmttime(t, "%Y-%m-%d %H:%M:%S");
399 /* The Windows versions of the following two functions are defined in
400 mswindows.c. On MSDOS this function should never be called. */
405 fork_to_background (void)
410 #else /* def __VMS */
412 #if !defined(WINDOWS) && !defined(MSDOS)
414 fork_to_background (void)
417 /* Whether we arrange our own version of opt.lfilename here. */
418 bool logfile_changed = false;
420 if (!opt.lfilename && (!opt.quiet || opt.server_response))
422 /* We must create the file immediately to avoid either a race
423 condition (which arises from using unique_name and failing to
424 use fopen_excl) or lying to the user about the log file name
425 (which arises from using unique_name, printing the name, and
426 using fopen_excl later on.) */
427 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
430 logfile_changed = true;
443 /* parent, no error */
444 printf (_("Continuing in background, pid %d.\n"), (int) pid);
446 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
447 exit (0); /* #### should we use _exit()? */
450 /* child: give up the privileges and keep running. */
452 freopen ("/dev/null", "r", stdin);
453 freopen ("/dev/null", "w", stdout);
454 freopen ("/dev/null", "w", stderr);
456 #endif /* !WINDOWS && !MSDOS */
458 #endif /* def __VMS [else] */
461 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
462 specified with TM. The atime ("access time") is set to the current
466 touch (const char *file, time_t tm)
468 #ifdef HAVE_STRUCT_UTIMBUF
469 struct utimbuf times;
477 times.actime = time (NULL);
478 if (utime (file, ×) == -1)
479 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
482 /* Checks if FILE is a symbolic link, and removes it if it is. Does
483 nothing under MS-Windows. */
485 remove_link (const char *file)
490 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
492 DEBUGP (("Unlinking %s (symlink).\n", file));
495 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
496 file, strerror (errno));
501 /* Does FILENAME exist? This is quite a lousy implementation, since
502 it supplies no error codes -- only a yes-or-no answer. Thus it
503 will return that a file does not exist if, e.g., the directory is
504 unreadable. I don't mind it too much currently, though. The
505 proper way should, of course, be to have a third, error state,
506 other than true/false, but that would introduce uncalled-for
507 additional complexity to the callers. */
509 file_exists_p (const char *filename)
512 return access (filename, F_OK) >= 0;
515 return stat (filename, &buf) >= 0;
519 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
520 Returns 0 on error. */
522 file_non_directory_p (const char *path)
525 /* Use lstat() rather than stat() so that symbolic links pointing to
526 directories can be identified correctly. */
527 if (lstat (path, &buf) != 0)
529 return S_ISDIR (buf.st_mode) ? false : true;
532 /* Return the size of file named by FILENAME, or -1 if it cannot be
533 opened or seeked into. */
535 file_size (const char *filename)
537 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
539 /* We use fseek rather than stat to determine the file size because
540 that way we can also verify that the file is readable without
541 explicitly checking for permissions. Inspired by the POST patch
543 FILE *fp = fopen (filename, "rb");
546 fseeko (fp, 0, SEEK_END);
552 if (stat (filename, &st) < 0)
559 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
560 original name. With the VMS file systems' versioning, everything
561 should be fine, and appending ".NN" just causes trouble.
566 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
567 doesn't exist is found. Return a freshly allocated copy of the
571 unique_name_1 (const char *prefix)
574 int plen = strlen (prefix);
575 char *template = (char *)alloca (plen + 1 + 24);
576 char *template_tail = template + plen;
578 memcpy (template, prefix, plen);
579 *template_tail++ = UNIQ_SEP;
582 number_to_string (template_tail, count++);
583 while (file_exists_p (template));
585 return xstrdup (template);
588 /* Return a unique file name, based on FILE.
590 More precisely, if FILE doesn't exist, it is returned unmodified.
591 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
592 file name that doesn't exist is returned.
594 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
596 The resulting file is not created, only verified that it didn't
597 exist at the point in time when the function was called.
598 Therefore, where security matters, don't rely that the file created
599 by this function exists until you open it with O_EXCL or
602 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
603 string. Otherwise, it may return FILE if the file doesn't exist
604 (and therefore doesn't need changing). */
607 unique_name (const char *file, bool allow_passthrough)
609 /* If the FILE itself doesn't exist, return it without
611 if (!file_exists_p (file))
612 return allow_passthrough ? (char *)file : xstrdup (file);
614 /* Otherwise, find a numeric suffix that results in unused file name
616 return unique_name_1 (file);
619 #else /* def UNIQ_SEP */
621 /* Dummy unique_name() for VMS. Return the original name as easily as
625 unique_name (const char *file, bool allow_passthrough)
627 /* Return the FILE itself, without modification, irregardful. */
628 return allow_passthrough ? (char *)file : xstrdup (file);
631 #endif /* def UNIQ_SEP [else] */
633 /* Create a file based on NAME, except without overwriting an existing
634 file with that name. Providing O_EXCL is correctly implemented,
635 this function does not have the race condition associated with
636 opening the file returned by unique_name. */
639 unique_create (const char *name, bool binary, char **opened_name)
641 /* unique file name, based on NAME */
642 char *uname = unique_name (name, false);
644 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
647 uname = unique_name (name, false);
649 if (opened_name && fp != NULL)
652 *opened_name = uname;
664 /* Open the file for writing, with the addition that the file is
665 opened "exclusively". This means that, if the file already exists,
666 this function will *fail* and errno will be set to EEXIST. If
667 BINARY is set, the file will be opened in binary mode, equivalent
670 If opening the file fails for any reason, including the file having
671 previously existed, this function returns NULL and sets errno
675 fopen_excl (const char *fname, bool binary)
681 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
682 It also has file versions which obviate all the O_EXCL effort.
683 O_TRUNC (something of a misnomer) requests a new version.
686 /* Common open() optional arguments:
687 sequential access only, access callback function.
689 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
692 int flags = O_WRONLY | O_CREAT | O_TRUNC;
697 fd = open( fname, /* File name. */
699 0777, /* Mode for default protection. */
700 "ctx=bin,stm", /* Binary, stream access. */
701 "rfm=stmlf", /* Stream_LF. */
702 OPEN_OPT_ARGS); /* Access callback. */
707 fd = open( fname, /* File name. */
709 0777, /* Mode for default protection. */
710 "ctx=bin,stm", /* Binary, stream access. */
711 "rfm=fix", /* Fixed-length, */
712 "mrs=512", /* 512-byte records. */
713 OPEN_OPT_ARGS); /* Access callback. */
718 fd = open( fname, /* File name. */
720 0777, /* Mode for default protection.
722 "rfm=stmlf", /* Stream_LF. */
723 OPEN_OPT_ARGS); /* Access callback. */
725 # else /* def __VMS */
726 int flags = O_WRONLY | O_CREAT | O_EXCL;
731 fd = open (fname, flags, 0666);
732 # endif /* def __VMS [else] */
736 return fdopen (fd, binary ? "wb" : "w");
737 #else /* not O_EXCL */
738 /* Manually check whether the file exists. This is prone to race
739 conditions, but systems without O_EXCL haven't deserved
741 if (file_exists_p (fname))
746 return fopen (fname, binary ? "wb" : "w");
747 #endif /* not O_EXCL */
750 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
751 are missing, create them first. In case any mkdir() call fails,
752 return its error status. Returns 0 on successful completion.
754 The behaviour of this function should be identical to the behaviour
755 of `mkdir -p' on systems where mkdir supports the `-p' option. */
757 make_directory (const char *directory)
759 int i, ret, quit = 0;
762 /* Make a copy of dir, to be able to write to it. Otherwise, the
763 function is unsafe if called with a read-only char *argument. */
764 STRDUP_ALLOCA (dir, directory);
766 /* If the first character of dir is '/', skip it (and thus enable
767 creation of absolute-pathname directories. */
768 for (i = (*dir == '/'); 1; ++i)
770 for (; dir[i] && dir[i] != '/'; i++)
775 /* Check whether the directory already exists. Allow creation of
776 of intermediate directories to fail, as the initial path components
777 are not necessarily directories! */
778 if (!file_exists_p (dir))
779 ret = mkdir (dir, 0777);
790 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
791 should be a file name.
793 file_merge("/foo/bar", "baz") => "/foo/baz"
794 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
795 file_merge("foo", "bar") => "bar"
797 In other words, it's a simpler and gentler version of uri_merge. */
800 file_merge (const char *base, const char *file)
803 const char *cut = (const char *)strrchr (base, '/');
806 return xstrdup (file);
808 result = xmalloc (cut - base + 1 + strlen (file) + 1);
809 memcpy (result, base, cut - base);
810 result[cut - base] = '/';
811 strcpy (result + (cut - base) + 1, file);
816 /* Like fnmatch, but performs a case-insensitive match. */
819 fnmatch_nocase (const char *pattern, const char *string, int flags)
822 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
823 also present on *BSD platforms, and possibly elsewhere. */
824 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
826 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
827 char *patcopy = (char *) alloca (strlen (pattern) + 1);
828 char *strcopy = (char *) alloca (strlen (string) + 1);
830 for (p = patcopy; *pattern; pattern++, p++)
831 *p = c_tolower (*pattern);
833 for (p = strcopy; *string; string++, p++)
834 *p = c_tolower (*string);
836 return fnmatch (patcopy, strcopy, flags);
840 static bool in_acclist (const char *const *, const char *, bool);
842 /* Determine whether a file is acceptable to be followed, according to
843 lists of patterns to accept/reject. */
845 acceptable (const char *s)
849 while (l && s[l] != '/')
856 return (in_acclist ((const char *const *)opt.accepts, s, true)
857 && !in_acclist ((const char *const *)opt.rejects, s, true));
859 return in_acclist ((const char *const *)opt.accepts, s, true);
861 else if (opt.rejects)
862 return !in_acclist ((const char *const *)opt.rejects, s, true);
866 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
867 will return true if and only if D2 begins with `/something/' or is exactly
870 subdir_p (const char *d1, const char *d2)
874 if (!opt.ignore_case)
875 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
878 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
881 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
884 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
885 first element that matches DIR, through wildcards or front comparison (as
888 dir_matches_p (char **dirlist, const char *dir)
891 int (*matcher) (const char *, const char *, int)
892 = opt.ignore_case ? fnmatch_nocase : fnmatch;
894 for (x = dirlist; *x; x++)
896 /* Remove leading '/' */
897 char *p = *x + (**x == '/');
898 if (has_wildcards_p (p))
900 if (matcher (p, dir, FNM_PATHNAME) == 0)
905 if (subdir_p (p, dir))
910 return *x ? true : false;
913 /* Returns whether DIRECTORY is acceptable for download, wrt the
914 include/exclude lists.
916 The leading `/' is ignored in paths; relative and absolute paths
917 may be freely intermixed. */
920 accdir (const char *directory)
922 /* Remove starting '/'. */
923 if (*directory == '/')
927 if (!dir_matches_p (opt.includes, directory))
932 if (dir_matches_p (opt.excludes, directory))
938 /* Return true if STRING ends with TAIL. For instance:
940 match_tail ("abc", "bc", false) -> 1
941 match_tail ("abc", "ab", false) -> 0
942 match_tail ("abc", "abc", false) -> 1
944 If FOLD_CASE is true, the comparison will be case-insensitive. */
947 match_tail (const char *string, const char *tail, bool fold_case)
951 /* We want this to be fast, so we code two loops, one with
952 case-folding, one without. */
956 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
957 if (string[i] != tail[j])
962 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
963 if (c_tolower (string[i]) != c_tolower (tail[j]))
967 /* If the tail was exhausted, the match was succesful. */
974 /* Checks whether string S matches each element of ACCEPTS. A list
975 element are matched either with fnmatch() or match_tail(),
976 according to whether the element contains wildcards or not.
978 If the BACKWARD is false, don't do backward comparison -- just compare
981 in_acclist (const char *const *accepts, const char *s, bool backward)
983 for (; *accepts; accepts++)
985 if (has_wildcards_p (*accepts))
987 int res = opt.ignore_case
988 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
989 /* fnmatch returns 0 if the pattern *does* match the string. */
997 if (match_tail (s, *accepts, opt.ignore_case))
1002 int cmp = opt.ignore_case
1003 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1012 /* Return the location of STR's suffix (file extension). Examples:
1013 suffix ("foo.bar") -> "bar"
1014 suffix ("foo.bar.baz") -> "baz"
1015 suffix ("/foo/bar") -> NULL
1016 suffix ("/foo.bar/baz") -> NULL */
1018 suffix (const char *str)
1022 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1025 if (str[i++] == '.')
1026 return (char *)str + i;
1031 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1035 has_wildcards_p (const char *s)
1038 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1043 /* Return true if FNAME ends with a typical HTML suffix. The
1044 following (case-insensitive) suffixes are presumed to be HTML
1049 ?html (`?' matches one character)
1051 #### CAVEAT. This is not necessarily a good indication that FNAME
1052 refers to a file that contains HTML! */
1054 has_html_suffix_p (const char *fname)
1058 if ((suf = suffix (fname)) == NULL)
1060 if (!strcasecmp (suf, "html"))
1062 if (!strcasecmp (suf, "htm"))
1064 if (suf[0] && !strcasecmp (suf + 1, "html"))
1069 /* Read a line from FP and return the pointer to freshly allocated
1070 storage. The storage space is obtained through malloc() and should
1071 be freed with free() when it is no longer needed.
1073 The length of the line is not limited, except by available memory.
1074 The newline character at the end of line is retained. The line is
1075 terminated with a zero character.
1077 After end-of-file is encountered without anything being read, NULL
1078 is returned. NULL is also returned on error. To distinguish
1079 between these two cases, use the stdio function ferror(). */
1082 read_whole_line (FILE *fp)
1086 char *line = xmalloc (bufsize);
1088 while (fgets (line + length, bufsize - length, fp))
1090 length += strlen (line + length);
1092 /* Possible for example when reading from a binary file where
1093 a line begins with \0. */
1096 if (line[length - 1] == '\n')
1099 /* fgets() guarantees to read the whole line, or to use up the
1100 space we've given it. We can double the buffer
1103 line = xrealloc (line, bufsize);
1105 if (length == 0 || ferror (fp))
1110 if (length + 1 < bufsize)
1111 /* Relieve the memory from our exponential greediness. We say
1112 `length + 1' because the terminating \0 is not included in
1113 LENGTH. We don't need to zero-terminate the string ourselves,
1114 though, because fgets() does that. */
1115 line = xrealloc (line, length + 1);
1119 /* Read FILE into memory. A pointer to `struct file_memory' are
1120 returned; use struct element `content' to access file contents, and
1121 the element `length' to know the file length. `content' is *not*
1122 zero-terminated, and you should *not* read or write beyond the [0,
1123 length) range of characters.
1125 After you are done with the file contents, call read_file_free to
1128 Depending on the operating system and the type of file that is
1129 being read, read_file() either mmap's the file into memory, or
1130 reads the file into the core using read().
1132 If file is named "-", fileno(stdin) is used for reading instead.
1133 If you want to read from a real file named "-", use "./-" instead. */
1135 struct file_memory *
1136 read_file (const char *file)
1139 struct file_memory *fm;
1141 bool inhibit_close = false;
1143 /* Some magic in the finest tradition of Perl and its kin: if FILE
1144 is "-", just use stdin. */
1147 fd = fileno (stdin);
1148 inhibit_close = true;
1149 /* Note that we don't inhibit mmap() in this case. If stdin is
1150 redirected from a regular file, mmap() will still work. */
1153 fd = open (file, O_RDONLY);
1156 fm = xnew (struct file_memory);
1161 if (fstat (fd, &buf) < 0)
1163 fm->length = buf.st_size;
1164 /* NOTE: As far as I know, the callers of this function never
1165 modify the file text. Relying on this would enable us to
1166 specify PROT_READ and MAP_SHARED for a marginal gain in
1167 efficiency, but at some cost to generality. */
1168 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1169 MAP_PRIVATE, fd, 0);
1170 if (fm->content == (char *)MAP_FAILED)
1180 /* The most common reason why mmap() fails is that FD does not point
1181 to a plain file. However, it's also possible that mmap() doesn't
1182 work for a particular type of file. Therefore, whenever mmap()
1183 fails, we just fall back to the regular method. */
1184 #endif /* HAVE_MMAP */
1187 size = 512; /* number of bytes fm->contents can
1188 hold at any given time. */
1189 fm->content = xmalloc (size);
1193 if (fm->length > size / 2)
1195 /* #### I'm not sure whether the whole exponential-growth
1196 thing makes sense with kernel read. On Linux at least,
1197 read() refuses to read more than 4K from a file at a
1198 single chunk anyway. But other Unixes might optimize it
1199 better, and it doesn't *hurt* anything, so I'm leaving
1202 /* Normally, we grow SIZE exponentially to make the number
1203 of calls to read() and realloc() logarithmic in relation
1204 to file size. However, read() can read an amount of data
1205 smaller than requested, and it would be unreasonable to
1206 double SIZE every time *something* was read. Therefore,
1207 we double SIZE only when the length exceeds half of the
1208 entire allocated size. */
1210 fm->content = xrealloc (fm->content, size);
1212 nread = read (fd, fm->content + fm->length, size - fm->length);
1214 /* Successful read. */
1215 fm->length += nread;
1225 if (size > fm->length && fm->length != 0)
1226 /* Due to exponential growth of fm->content, the allocated region
1227 might be much larger than what is actually needed. */
1228 fm->content = xrealloc (fm->content, fm->length);
1235 xfree (fm->content);
1240 /* Release the resources held by FM. Specifically, this calls
1241 munmap() or xfree() on fm->content, depending whether mmap or
1242 malloc/read were used to read in the file. It also frees the
1243 memory needed to hold the FM structure itself. */
1246 read_file_free (struct file_memory *fm)
1251 munmap (fm->content, fm->length);
1256 xfree (fm->content);
1261 /* Free the pointers in a NULL-terminated vector of pointers, then
1262 free the pointer itself. */
1264 free_vec (char **vec)
1275 /* Append vector V2 to vector V1. The function frees V2 and
1276 reallocates V1 (thus you may not use the contents of neither
1277 pointer after the call). If V1 is NULL, V2 is returned. */
1279 merge_vecs (char **v1, char **v2)
1289 /* To avoid j == 0 */
1294 for (i = 0; v1[i]; i++)
1297 for (j = 0; v2[j]; j++)
1299 /* Reallocate v1. */
1300 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1301 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1306 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1307 is allocated as needed. Return the new value of the vector. */
1310 vec_append (char **vec, const char *str)
1312 int cnt; /* count of vector elements, including
1313 the one we're about to append */
1316 for (cnt = 0; vec[cnt]; cnt++)
1322 /* Reallocate the array to fit the new element and the NULL. */
1323 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1324 /* Append a copy of STR to the vector. */
1325 vec[cnt - 1] = xstrdup (str);
1330 /* Sometimes it's useful to create "sets" of strings, i.e. special
1331 hash tables where you want to store strings as keys and merely
1332 query for their existence. Here is a set of utility routines that
1333 makes that transparent. */
1336 string_set_add (struct hash_table *ht, const char *s)
1338 /* First check whether the set element already exists. If it does,
1339 do nothing so that we don't have to free() the old element and
1340 then strdup() a new one. */
1341 if (hash_table_contains (ht, s))
1344 /* We use "1" as value. It provides us a useful and clear arbitrary
1345 value, and it consumes no memory -- the pointers to the same
1346 string "1" will be shared by all the key-value pairs in all `set'
1348 hash_table_put (ht, xstrdup (s), "1");
1351 /* Synonym for hash_table_contains... */
1354 string_set_contains (struct hash_table *ht, const char *s)
1356 return hash_table_contains (ht, s);
1359 /* Convert the specified string set to array. ARRAY should be large
1360 enough to hold hash_table_count(ht) char pointers. */
1362 void string_set_to_array (struct hash_table *ht, char **array)
1364 hash_table_iterator iter;
1365 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1366 *array++ = iter.key;
1369 /* Free the string set. This frees both the storage allocated for
1370 keys and the actual hash table. (hash_table_destroy would only
1371 destroy the hash table.) */
1374 string_set_free (struct hash_table *ht)
1376 hash_table_iterator iter;
1377 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1379 hash_table_destroy (ht);
1382 /* Utility function: simply call xfree() on all keys and values of HT. */
1385 free_keys_and_values (struct hash_table *ht)
1387 hash_table_iterator iter;
1388 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1395 /* Get digit grouping data for thousand separors by calling
1396 localeconv(). The data includes separator string and grouping info
1397 and is cached after the first call to the function.
1399 In locales that don't set a thousand separator (such as the "C"
1400 locale), this forces it to be ",". We are now only showing
1401 thousand separators in one place, so this shouldn't be a problem in
1405 get_grouping_data (const char **sep, const char **grouping)
1407 static const char *cached_sep;
1408 static const char *cached_grouping;
1409 static bool initialized;
1412 /* Get the grouping info from the locale. */
1413 struct lconv *lconv = localeconv ();
1414 cached_sep = lconv->thousands_sep;
1415 cached_grouping = lconv->grouping;
1416 #if ! USE_NLS_PROGRESS_BAR
1417 /* We can't count column widths, so ensure that the separator
1418 * is single-byte only (let check below determine what byte). */
1419 if (strlen(cached_sep) > 1)
1424 /* Many locales (such as "C" or "hr_HR") don't specify
1425 grouping, which we still want to use it for legibility.
1426 In those locales set the sep char to ',', unless that
1427 character is used for decimal point, in which case set it
1429 if (*lconv->decimal_point != ',')
1433 cached_grouping = "\x03";
1438 *grouping = cached_grouping;
1441 /* Return a printed representation of N with thousand separators.
1442 This should respect locale settings, with the exception of the "C"
1443 locale which mandates no separator, but we use one anyway.
1445 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1446 the separators because it's too non-portable, and it's hard to test
1447 for this feature at configure time. Besides, it wouldn't display
1448 separators in the "C" locale, still used by many Unix users. */
1451 with_thousand_seps (wgint n)
1453 static char outbuf[48];
1454 char *p = outbuf + sizeof outbuf;
1456 /* Info received from locale */
1457 const char *grouping, *sep;
1460 /* State information */
1461 int i = 0, groupsize;
1462 const char *atgroup;
1464 bool negative = n < 0;
1466 /* Initialize grouping data. */
1467 get_grouping_data (&sep, &grouping);
1468 seplen = strlen (sep);
1470 groupsize = *atgroup++;
1472 /* This would overflow on WGINT_MIN, but printing negative numbers
1473 is not an important goal of this fuinction. */
1477 /* Write the number into the buffer, backwards, inserting the
1478 separators as necessary. */
1482 *--p = n % 10 + '0';
1486 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1487 if (++i == groupsize)
1492 memcpy (p -= seplen, sep, seplen);
1495 groupsize = *atgroup++;
1504 /* N, a byte quantity, is converted to a human-readable abberviated
1505 form a la sizes printed by `ls -lh'. The result is written to a
1506 static buffer, a pointer to which is returned.
1508 Unlike `with_thousand_seps', this approximates to the nearest unit.
1509 Quoting GNU libit: "Most people visually process strings of 3-4
1510 digits effectively, but longer strings of digits are more prone to
1511 misinterpretation. Hence, converting to an abbreviated form
1512 usually improves readability."
1514 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1515 original computer-related meaning of "powers of 1024". We don't
1516 use the "*bibyte" names invented in 1998, and seldom used in
1517 practice. Wikipedia's entry on "binary prefix" discusses this in
1521 human_readable (HR_NUMTYPE n)
1523 /* These suffixes are compatible with those of GNU `ls -lh'. */
1524 static char powers[] =
1526 'K', /* kilobyte, 2^10 bytes */
1527 'M', /* megabyte, 2^20 bytes */
1528 'G', /* gigabyte, 2^30 bytes */
1529 'T', /* terabyte, 2^40 bytes */
1530 'P', /* petabyte, 2^50 bytes */
1531 'E', /* exabyte, 2^60 bytes */
1536 /* If the quantity is smaller than 1K, just print it. */
1539 snprintf (buf, sizeof (buf), "%d", (int) n);
1543 /* Loop over powers, dividing N with 1024 in each iteration. This
1544 works unchanged for all sizes of wgint, while still avoiding
1545 non-portable `long double' arithmetic. */
1546 for (i = 0; i < countof (powers); i++)
1548 /* At each iteration N is greater than the *subsequent* power.
1549 That way N/1024.0 produces a decimal number in the units of
1551 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1553 double val = n / 1024.0;
1554 /* Print values smaller than 10 with one decimal digits, and
1555 others without any decimals. */
1556 snprintf (buf, sizeof (buf), "%.*f%c",
1557 val < 10 ? 1 : 0, val, powers[i]);
1562 return NULL; /* unreached */
1565 /* Count the digits in the provided number. Used to allocate space
1566 when printing numbers. */
1569 numdigit (wgint number)
1573 ++cnt; /* accomodate '-' */
1574 while ((number /= 10) != 0)
1579 #define PR(mask) *p++ = n / (mask) + '0'
1581 /* DIGITS_<D> is used to print a D-digit number and should be called
1582 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1583 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1584 Recursively this continues until DIGITS_1 is invoked. */
1586 #define DIGITS_1(mask) PR (mask)
1587 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1588 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1589 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1590 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1591 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1592 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1593 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1594 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1595 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1597 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1599 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1600 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1601 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1602 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1603 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1604 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1605 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1606 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1607 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1609 /* Shorthand for casting to wgint. */
1612 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1613 `sprintf(buffer, "%lld", (long long) number)', only typically much
1614 faster and portable to machines without long long.
1616 The speedup may make a difference in programs that frequently
1617 convert numbers to strings. Some implementations of sprintf,
1618 particularly the one in some versions of GNU libc, have been known
1619 to be quite slow when converting integers to strings.
1621 Return the pointer to the location where the terminating zero was
1622 printed. (Equivalent to calling buffer+strlen(buffer) after the
1625 BUFFER should be large enough to accept as many bytes as you expect
1626 the number to take up. On machines with 64-bit wgints the maximum
1627 needed size is 24 bytes. That includes the digits needed for the
1628 largest 64-bit number, the `-' sign in case it's negative, and the
1629 terminating '\0'. */
1632 number_to_string (char *buffer, wgint number)
1637 int last_digit_char = 0;
1639 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1640 /* We are running in a very strange environment. Leave the correct
1641 printing to sprintf. */
1642 p += sprintf (buf, "%j", (intmax_t) (n));
1643 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1649 /* n = -n would overflow because -n would evaluate to a
1650 wgint value larger than WGINT_MAX. Need to make n
1651 smaller and handle the last digit separately. */
1652 int last_digit = n % 10;
1653 /* The sign of n%10 is implementation-defined. */
1655 last_digit_char = '0' - last_digit;
1657 last_digit_char = '0' + last_digit;
1658 /* After n is made smaller, -n will not overflow. */
1666 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1667 way printing any N is fully open-coded without a loop or jump.
1668 (Also see description of DIGITS_*.) */
1670 if (n < 10) DIGITS_1 (1);
1671 else if (n < 100) DIGITS_2 (10);
1672 else if (n < 1000) DIGITS_3 (100);
1673 else if (n < 10000) DIGITS_4 (1000);
1674 else if (n < 100000) DIGITS_5 (10000);
1675 else if (n < 1000000) DIGITS_6 (100000);
1676 else if (n < 10000000) DIGITS_7 (1000000);
1677 else if (n < 100000000) DIGITS_8 (10000000);
1678 else if (n < 1000000000) DIGITS_9 (100000000);
1679 #if SIZEOF_WGINT == 4
1680 /* wgint is 32 bits wide: no number has more than 10 digits. */
1681 else DIGITS_10 (1000000000);
1683 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1684 Constants are constructed by compile-time multiplication to avoid
1685 dealing with different notations for 64-bit constants
1686 (nL/nLL/nI64, depending on the compiler and architecture). */
1687 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1688 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1689 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1690 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1691 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1692 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1693 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1694 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1695 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1696 else DIGITS_19 (1000000000*(W)1000000000);
1699 if (last_digit_char)
1700 *p++ = last_digit_char;
1703 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1710 #undef SPRINTF_WGINT
1733 /* Print NUMBER to a statically allocated string and return a pointer
1734 to the printed representation.
1736 This function is intended to be used in conjunction with printf.
1737 It is hard to portably print wgint values:
1738 a) you cannot use printf("%ld", number) because wgint can be long
1739 long on 32-bit machines with LFS.
1740 b) you cannot use printf("%lld", number) because NUMBER could be
1741 long on 32-bit machines without LFS, or on 64-bit machines,
1742 which do not require LFS. Also, Windows doesn't support %lld.
1743 c) you cannot use printf("%j", (int_max_t) number) because not all
1744 versions of printf support "%j", the most notable being the one
1746 d) you cannot #define WGINT_FMT to the appropriate format and use
1747 printf(WGINT_FMT, number) because that would break translations
1748 for user-visible messages, such as printf("Downloaded: %d
1751 What you should use instead is printf("%s", number_to_static_string
1754 CAVEAT: since the function returns pointers to static data, you
1755 must be careful to copy its result before calling it again.
1756 However, to make it more useful with printf, the function maintains
1757 an internal ring of static buffers to return. That way things like
1758 printf("%s %s", number_to_static_string (num1),
1759 number_to_static_string (num2)) work as expected. Three buffers
1760 are currently used, which means that "%s %s %s" will work, but "%s
1761 %s %s %s" won't. If you need to print more than three wgints,
1762 bump the RING_SIZE (or rethink your message.) */
1765 number_to_static_string (wgint number)
1767 static char ring[RING_SIZE][24];
1769 char *buf = ring[ringpos];
1770 number_to_string (buf, number);
1771 ringpos = (ringpos + 1) % RING_SIZE;
1775 /* Determine the width of the terminal we're running on. If that's
1776 not possible, return 0. */
1779 determine_screen_width (void)
1781 /* If there's a way to get the terminal size using POSIX
1782 tcgetattr(), somebody please tell me. */
1787 if (opt.lfilename != NULL)
1790 fd = fileno (stderr);
1791 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1792 return 0; /* most likely ENOTTY */
1795 #elif defined(WINDOWS)
1796 CONSOLE_SCREEN_BUFFER_INFO csbi;
1797 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1799 return csbi.dwSize.X;
1800 #else /* neither TIOCGWINSZ nor WINDOWS */
1802 #endif /* neither TIOCGWINSZ nor WINDOWS */
1805 /* Whether the rnd system (either rand or [dl]rand48) has been
1807 static int rnd_seeded;
1809 /* Return a random number between 0 and MAX-1, inclusive.
1811 If the system does not support lrand48 and MAX is greater than the
1812 value of RAND_MAX+1 on the system, the returned value will be in
1813 the range [0, RAND_MAX]. This may be fixed in a future release.
1814 The random number generator is seeded automatically the first time
1817 This uses lrand48 where available, rand elsewhere. DO NOT use it
1818 for cryptography. It is only meant to be used in situations where
1819 quality of the random numbers returned doesn't really matter. */
1822 random_number (int max)
1827 srand48 ((long) time (NULL) ^ (long) getpid ());
1830 return lrand48 () % max;
1831 #else /* not HAVE_DRAND48 */
1837 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1842 /* Like rand() % max, but uses the high-order bits for better
1843 randomness on architectures where rand() is implemented using a
1844 simple congruential generator. */
1846 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1847 return (int) bounded;
1849 #endif /* not HAVE_DRAND48 */
1852 /* Return a random uniformly distributed floating point number in the
1853 [0, 1) range. Uses drand48 where available, and a really lame
1854 kludge elsewhere. */
1862 srand48 ((long) time (NULL) ^ (long) getpid ());
1866 #else /* not HAVE_DRAND48 */
1867 return ( random_number (10000) / 10000.0
1868 + random_number (10000) / (10000.0 * 10000.0)
1869 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1870 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1871 #endif /* not HAVE_DRAND48 */
1874 /* Implementation of run_with_timeout, a generic timeout-forcing
1875 routine for systems with Unix-like signal handling. */
1877 #ifdef USE_SIGNAL_TIMEOUT
1878 # ifdef HAVE_SIGSETJMP
1879 # define SETJMP(env) sigsetjmp (env, 1)
1881 static sigjmp_buf run_with_timeout_env;
1884 abort_run_with_timeout (int sig)
1886 assert (sig == SIGALRM);
1887 siglongjmp (run_with_timeout_env, -1);
1889 # else /* not HAVE_SIGSETJMP */
1890 # define SETJMP(env) setjmp (env)
1892 static jmp_buf run_with_timeout_env;
1895 abort_run_with_timeout (int sig)
1897 assert (sig == SIGALRM);
1898 /* We don't have siglongjmp to preserve the set of blocked signals;
1899 if we longjumped out of the handler at this point, SIGALRM would
1900 remain blocked. We must unblock it manually. */
1901 int mask = siggetmask ();
1902 mask &= ~sigmask (SIGALRM);
1905 /* Now it's safe to longjump. */
1906 longjmp (run_with_timeout_env, -1);
1908 # endif /* not HAVE_SIGSETJMP */
1910 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1911 setitimer where available, alarm otherwise.
1913 TIMEOUT should be non-zero. If the timeout value is so small that
1914 it would be rounded to zero, it is rounded to the least legal value
1915 instead (1us for setitimer, 1s for alarm). That ensures that
1916 SIGALRM will be delivered in all cases. */
1919 alarm_set (double timeout)
1922 /* Use the modern itimer interface. */
1923 struct itimerval itv;
1925 itv.it_value.tv_sec = (long) timeout;
1926 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1927 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1928 /* Ensure that we wait for at least the minimum interval.
1929 Specifying zero would mean "wait forever". */
1930 itv.it_value.tv_usec = 1;
1931 setitimer (ITIMER_REAL, &itv, NULL);
1932 #else /* not ITIMER_REAL */
1933 /* Use the old alarm() interface. */
1934 int secs = (int) timeout;
1936 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1937 because alarm(0) means "never deliver the alarm", i.e. "wait
1938 forever", which is not what someone who specifies a 0.5s
1939 timeout would expect. */
1942 #endif /* not ITIMER_REAL */
1945 /* Cancel the alarm set with alarm_set. */
1951 struct itimerval disable;
1953 setitimer (ITIMER_REAL, &disable, NULL);
1954 #else /* not ITIMER_REAL */
1956 #endif /* not ITIMER_REAL */
1959 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1960 seconds. Returns true if the function was interrupted with a
1961 timeout, false otherwise.
1963 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1964 using setitimer() or alarm(). The timeout is enforced by
1965 longjumping out of the SIGALRM handler. This has several
1966 advantages compared to the traditional approach of relying on
1967 signals causing system calls to exit with EINTR:
1969 * The callback function is *forcibly* interrupted after the
1970 timeout expires, (almost) regardless of what it was doing and
1971 whether it was in a syscall. For example, a calculation that
1972 takes a long time is interrupted as reliably as an IO
1975 * It works with both SYSV and BSD signals because it doesn't
1976 depend on the default setting of SA_RESTART.
1978 * It doesn't require special handler setup beyond a simple call
1979 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1982 The only downside is that, if FUN allocates internal resources that
1983 are normally freed prior to exit from the functions, they will be
1984 lost in case of timeout. */
1987 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1997 signal (SIGALRM, abort_run_with_timeout);
1998 if (SETJMP (run_with_timeout_env) != 0)
2000 /* Longjumped out of FUN with a timeout. */
2001 signal (SIGALRM, SIG_DFL);
2004 alarm_set (timeout);
2007 /* Preserve errno in case alarm() or signal() modifies it. */
2008 saved_errno = errno;
2010 signal (SIGALRM, SIG_DFL);
2011 errno = saved_errno;
2016 #else /* not USE_SIGNAL_TIMEOUT */
2019 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2020 define it under Windows, because Windows has its own version of
2021 run_with_timeout that uses threads. */
2024 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2029 #endif /* not WINDOWS */
2030 #endif /* not USE_SIGNAL_TIMEOUT */
2034 /* Sleep the specified amount of seconds. On machines without
2035 nanosleep(), this may sleep shorter if interrupted by signals. */
2038 xsleep (double seconds)
2040 #ifdef HAVE_NANOSLEEP
2041 /* nanosleep is the preferred interface because it offers high
2042 accuracy and, more importantly, because it allows us to reliably
2043 restart receiving a signal such as SIGWINCH. (There was an
2044 actual Debian bug report about --limit-rate malfunctioning while
2045 the terminal was being resized.) */
2046 struct timespec sleep, remaining;
2047 sleep.tv_sec = (long) seconds;
2048 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2049 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2050 /* If nanosleep has been interrupted by a signal, adjust the
2051 sleeping period and return to sleep. */
2053 #elif defined(HAVE_USLEEP)
2054 /* If usleep is available, use it in preference to select. */
2057 /* On some systems, usleep cannot handle values larger than
2058 1,000,000. If the period is larger than that, use sleep
2059 first, then add usleep for subsecond accuracy. */
2061 seconds -= (long) seconds;
2063 usleep (seconds * 1000000);
2064 #else /* fall back select */
2065 /* Note that, although Windows supports select, it can't be used to
2066 implement sleeping because Winsock's select doesn't implement
2067 timeout when it is passed NULL pointers for all fd sets. (But it
2068 does under Cygwin, which implements Unix-compatible select.) */
2069 struct timeval sleep;
2070 sleep.tv_sec = (long) seconds;
2071 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2072 select (0, NULL, NULL, NULL, &sleep);
2073 /* If select returns -1 and errno is EINTR, it means we were
2074 interrupted by a signal. But without knowing how long we've
2075 actually slept, we can't return to sleep. Using gettimeofday to
2076 track sleeps is slow and unreliable due to clock skew. */
2080 #endif /* not WINDOWS */
2082 /* Encode the octets in DATA of length LENGTH to base64 format,
2083 storing the result to DEST. The output will be zero-terminated,
2084 and must point to a writable buffer of at least
2085 1+BASE64_LENGTH(length) bytes. The function returns the length of
2086 the resulting base64 data, not counting the terminating zero.
2088 This implementation does not emit newlines after 76 characters of
2092 base64_encode (const void *data, int length, char *dest)
2094 /* Conversion table. */
2095 static const char tbl[64] = {
2096 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2097 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2098 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2099 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2101 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2102 don't work for data with MSB set. */
2103 const unsigned char *s = data;
2104 /* Theoretical ANSI violation when length < 3. */
2105 const unsigned char *end = (const unsigned char *) data + length - 2;
2108 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2109 for (; s < end; s += 3)
2111 *p++ = tbl[s[0] >> 2];
2112 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2113 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2114 *p++ = tbl[s[2] & 0x3f];
2117 /* Pad the result if necessary... */
2121 *p++ = tbl[s[0] >> 2];
2122 *p++ = tbl[(s[0] & 3) << 4];
2127 *p++ = tbl[s[0] >> 2];
2128 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2129 *p++ = tbl[((s[1] & 0xf) << 2)];
2133 /* ...and zero-terminate it. */
2139 /* Store in C the next non-whitespace character from the string, or \0
2140 when end of string is reached. */
2141 #define NEXT_CHAR(c, p) do { \
2142 c = (unsigned char) *p++; \
2143 } while (c_isspace (c))
2145 #define IS_ASCII(c) (((c) & 0x80) == 0)
2147 /* Decode data from BASE64 (a null-terminated string) into memory
2148 pointed to by DEST. DEST is assumed to be large enough to
2149 accomodate the decoded data, which is guaranteed to be no more than
2152 Since DEST is assumed to contain binary data, it is not
2153 NUL-terminated. The function returns the length of the data
2154 written to TO. -1 is returned in case of error caused by malformed
2157 This function originates from Free Recode. */
2160 base64_decode (const char *base64, void *dest)
2162 /* Table of base64 values for first 128 characters. Note that this
2163 assumes ASCII (but so does Wget in other places). */
2164 static const signed char base64_char_to_value[128] =
2166 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2167 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2168 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2169 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2170 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2171 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2172 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2173 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2174 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2175 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2176 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2177 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2178 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2180 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2181 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2183 const char *p = base64;
2189 unsigned long value;
2191 /* Process first byte of a quadruplet. */
2195 if (c == '=' || !IS_BASE64 (c))
2196 return -1; /* illegal char while decoding base64 */
2197 value = BASE64_CHAR_TO_VALUE (c) << 18;
2199 /* Process second byte of a quadruplet. */
2202 return -1; /* premature EOF while decoding base64 */
2203 if (c == '=' || !IS_BASE64 (c))
2204 return -1; /* illegal char while decoding base64 */
2205 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2208 /* Process third byte of a quadruplet. */
2211 return -1; /* premature EOF while decoding base64 */
2213 return -1; /* illegal char while decoding base64 */
2219 return -1; /* premature EOF while decoding base64 */
2221 return -1; /* padding `=' expected but not found */
2225 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2226 *q++ = 0xff & value >> 8;
2228 /* Process fourth byte of a quadruplet. */
2231 return -1; /* premature EOF while decoding base64 */
2235 return -1; /* illegal char while decoding base64 */
2237 value |= BASE64_CHAR_TO_VALUE (c);
2238 *q++ = 0xff & value;
2241 #undef BASE64_CHAR_TO_VALUE
2243 return q - (char *) dest;
2249 /* Simple merge sort for use by stable_sort. Implementation courtesy
2250 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2253 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2254 int (*cmpfun) (const void *, const void *))
2256 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2260 size_t mid = (to + from) / 2;
2261 mergesort_internal (base, temp, size, from, mid, cmpfun);
2262 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2265 for (k = from; (i <= mid) && (j <= to); k++)
2266 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2267 memcpy (ELT (temp, k), ELT (base, i++), size);
2269 memcpy (ELT (temp, k), ELT (base, j++), size);
2271 memcpy (ELT (temp, k++), ELT (base, i++), size);
2273 memcpy (ELT (temp, k++), ELT (base, j++), size);
2274 for (k = from; k <= to; k++)
2275 memcpy (ELT (base, k), ELT (temp, k), size);
2280 /* Stable sort with interface exactly like standard library's qsort.
2281 Uses mergesort internally, allocating temporary storage with
2285 stable_sort (void *base, size_t nmemb, size_t size,
2286 int (*cmpfun) (const void *, const void *))
2290 void *temp = alloca (nmemb * size * sizeof (void *));
2291 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2295 /* Print a decimal number. If it is equal to or larger than ten, the
2296 number is rounded. Otherwise it is printed with one significant
2297 digit without trailing zeros and with no more than three fractional
2298 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2299 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2301 This is useful for displaying durations because it provides
2302 order-of-magnitude information without unnecessary clutter --
2303 long-running downloads are shown without the fractional part, and
2304 short ones still retain one significant digit. */
2307 print_decimal (double number)
2309 static char buf[32];
2310 double n = number >= 0 ? number : -number;
2313 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2314 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2315 snprintf (buf, sizeof buf, "%.0f", number);
2317 snprintf (buf, sizeof buf, "%.1f", number);
2318 else if (n >= 0.001)
2319 snprintf (buf, sizeof buf, "%.1g", number);
2320 else if (n >= 0.0005)
2321 /* round [0.0005, 0.001) to 0.001 */
2322 snprintf (buf, sizeof buf, "%.3f", number);
2324 /* print numbers close to 0 as 0, not 0.000 */
2341 { "/somedir", "/somedir", true },
2342 { "/somedir", "/somedir/d2", true },
2343 { "/somedir/d1", "/somedir", false },
2346 for (i = 0; i < countof(test_array); ++i)
2348 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2350 mu_assert ("test_subdir_p: wrong result",
2351 res == test_array[i].result);
2358 test_dir_matches_p()
2366 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2367 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2368 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2369 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2370 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2371 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2372 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2373 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2374 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2375 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2376 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2377 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2378 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2381 for (i = 0; i < countof(test_array); ++i)
2383 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2385 mu_assert ("test_dir_matches_p: wrong result",
2386 res == test_array[i].result);
2392 #endif /* TESTING */