1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010 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 */
96 memfatal (const char *context, long attempted_size)
98 /* Make sure we don't try to store part of the log line, and thus
100 log_set_save_context (false);
102 /* We have different log outputs in different situations:
103 1) output without bytes information
104 2) output with bytes information */
105 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
107 logprintf (LOG_ALWAYS,
108 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
113 logprintf (LOG_ALWAYS,
114 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
115 exec_name, context, attempted_size);
121 /* Character property table for (re-)escaping VMS ODS5 extended file
122 names. Note that this table ignores Unicode.
124 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
126 ODS5 Invalid characters:
127 C0 control codes (0x00 to 0x1F inclusive)
131 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
132 Double quotation marks (")
135 Left angle bracket (<)
136 Right angle bracket (>)
140 Characters escaped by "^":
141 SP ! " # % & ' ( ) + , . : ; =
144 Either "^_" or "^ " is accepted as a space. Period (.) is a special
145 case. Note that un-escaped < and > can also confuse a directory
148 Characters put out as ^xx:
150 80-9F (C1 control characters)
151 A0 (nonbreaking space)
152 FF (Latin small letter y diaeresis)
155 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
157 Property table values:
167 unsigned char char_prop[ 256] = {
169 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
170 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
173 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
175 /* SP ! " # $ % & ' ( ) * + , - . / */
176 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
178 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
179 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
181 /* @ A B C D E F G H I J K L M N O */
182 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
184 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
185 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
187 /* ` a b c d e f g h i j k l m n o */
188 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
190 /* p q r s t u v w x y z { | } ~ DEL */
191 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
193 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
194 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
195 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
196 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
197 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
198 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
199 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
200 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
203 /* Utility function: like xstrdup(), but also lowercases S. */
206 xstrdup_lower (const char *s)
208 char *copy = xstrdup (s);
215 /* Copy the string formed by two pointers (one on the beginning, other
216 on the char after the last char) to a new, malloc-ed location.
219 strdupdelim (const char *beg, const char *end)
221 char *res = xmalloc (end - beg + 1);
222 memcpy (res, beg, end - beg);
223 res[end - beg] = '\0';
227 /* Parse a string containing comma-separated elements, and return a
228 vector of char pointers with the elements. Spaces following the
229 commas are ignored. */
231 sepstring (const char *s)
245 res = xrealloc (res, (i + 2) * sizeof (char *));
246 res[i] = strdupdelim (p, s);
249 /* Skip the blanks following the ','. */
250 while (c_isspace (*s))
257 res = xrealloc (res, (i + 2) * sizeof (char *));
258 res[i] = strdupdelim (p, s);
263 /* Like sprintf, but prints into a string of sufficient size freshly
264 allocated with malloc, which is returned. If unable to print due
265 to invalid format, returns NULL. Inability to allocate needed
266 memory results in abort, as with xmalloc. This is in spirit
267 similar to the GNU/BSD extension asprintf, but somewhat easier to
270 Internally the function either calls vasprintf or loops around
271 vsnprintf until the correct size is found. Since Wget also ships a
272 fallback implementation of vsnprintf, this should be portable. */
274 /* Constant is using for limits memory allocation for text buffer.
275 Applicable in situation when: vasprintf is not available in the system
276 and vsnprintf return -1 when long line is truncated (in old versions of
277 glibc and in other system where C99 doesn`t support) */
279 #define FMT_MAX_LENGTH 1048576
282 aprintf (const char *fmt, ...)
284 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
289 va_start (args, fmt);
290 ret = vasprintf (&str, fmt, args);
292 if (ret < 0 && errno == ENOMEM)
293 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
294 with xmalloc/xrealloc */
298 #else /* not HAVE_VASPRINTF */
300 /* vasprintf is unavailable. snprintf into a small buffer and
301 resize it as necessary. */
303 char *str = xmalloc (size);
305 /* #### This code will infloop and eventually abort in xrealloc if
306 passed a FMT that causes snprintf to consistently return -1. */
313 va_start (args, fmt);
314 n = vsnprintf (str, size, fmt, args);
317 /* If the printing worked, return the string. */
318 if (n > -1 && n < size)
321 /* Else try again with a larger buffer. */
322 if (n > -1) /* C99 */
323 size = n + 1; /* precisely what is needed */
324 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
325 { /* maybe we have some wrong
327 logprintf (LOG_ALWAYS,
328 _("%s: aprintf: text buffer is too big (%ld bytes), "
330 exec_name, size); /* printout a log message */
331 abort (); /* and abort... */
335 /* else, we continue to grow our
336 * buffer: Twice the old size. */
339 str = xrealloc (str, size);
341 #endif /* not HAVE_VASPRINTF */
344 /* Concatenate the NULL-terminated list of string arguments into
345 freshly allocated space. */
348 concat_strings (const char *str0, ...)
351 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
354 const char *next_str;
355 int total_length = 0;
358 /* Calculate the length of and allocate the resulting string. */
361 va_start (args, str0);
362 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
364 int len = strlen (next_str);
365 if (argcount < countof (saved_lengths))
366 saved_lengths[argcount++] = len;
370 p = ret = xmalloc (total_length + 1);
372 /* Copy the strings into the allocated space. */
375 va_start (args, str0);
376 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
379 if (argcount < countof (saved_lengths))
380 len = saved_lengths[argcount++];
382 len = strlen (next_str);
383 memcpy (p, next_str, len);
392 /* Format the provided time according to the specified format. The
393 format is a string with format elements supported by strftime. */
396 fmttime (time_t t, const char *fmt)
398 static char output[32];
399 struct tm *tm = localtime(&t);
402 if (!strftime(output, sizeof(output), fmt, tm))
407 /* Return pointer to a static char[] buffer in which zero-terminated
408 string-representation of TM (in form hh:mm:ss) is printed.
410 If TM is NULL, the current time will be used. */
415 return fmttime(t, "%H:%M:%S");
418 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
421 datetime_str (time_t t)
423 return fmttime(t, "%Y-%m-%d %H:%M:%S");
426 /* The Windows versions of the following two functions are defined in
427 mswindows.c. On MSDOS this function should never be called. */
432 fork_to_background (void)
437 #else /* def __VMS */
439 #if !defined(WINDOWS) && !defined(MSDOS)
441 fork_to_background (void)
444 /* Whether we arrange our own version of opt.lfilename here. */
445 bool logfile_changed = false;
447 if (!opt.lfilename && (!opt.quiet || opt.server_response))
449 /* We must create the file immediately to avoid either a race
450 condition (which arises from using unique_name and failing to
451 use fopen_excl) or lying to the user about the log file name
452 (which arises from using unique_name, printing the name, and
453 using fopen_excl later on.) */
454 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
457 logfile_changed = true;
470 /* parent, no error */
471 printf (_("Continuing in background, pid %d.\n"), (int) pid);
473 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
474 exit (0); /* #### should we use _exit()? */
477 /* child: give up the privileges and keep running. */
479 freopen ("/dev/null", "r", stdin);
480 freopen ("/dev/null", "w", stdout);
481 freopen ("/dev/null", "w", stderr);
483 #endif /* !WINDOWS && !MSDOS */
485 #endif /* def __VMS [else] */
488 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
489 specified with TM. The atime ("access time") is set to the current
493 touch (const char *file, time_t tm)
495 #ifdef HAVE_STRUCT_UTIMBUF
496 struct utimbuf times;
504 times.actime = time (NULL);
505 if (utime (file, ×) == -1)
506 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
509 /* Checks if FILE is a symbolic link, and removes it if it is. Does
510 nothing under MS-Windows. */
512 remove_link (const char *file)
517 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
519 DEBUGP (("Unlinking %s (symlink).\n", file));
522 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
523 quote (file), strerror (errno));
528 /* Does FILENAME exist? This is quite a lousy implementation, since
529 it supplies no error codes -- only a yes-or-no answer. Thus it
530 will return that a file does not exist if, e.g., the directory is
531 unreadable. I don't mind it too much currently, though. The
532 proper way should, of course, be to have a third, error state,
533 other than true/false, but that would introduce uncalled-for
534 additional complexity to the callers. */
536 file_exists_p (const char *filename)
539 return access (filename, F_OK) >= 0;
542 return stat (filename, &buf) >= 0;
546 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
547 Returns 0 on error. */
549 file_non_directory_p (const char *path)
552 /* Use lstat() rather than stat() so that symbolic links pointing to
553 directories can be identified correctly. */
554 if (lstat (path, &buf) != 0)
556 return S_ISDIR (buf.st_mode) ? false : true;
559 /* Return the size of file named by FILENAME, or -1 if it cannot be
560 opened or seeked into. */
562 file_size (const char *filename)
564 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
566 /* We use fseek rather than stat to determine the file size because
567 that way we can also verify that the file is readable without
568 explicitly checking for permissions. Inspired by the POST patch
570 FILE *fp = fopen (filename, "rb");
573 fseeko (fp, 0, SEEK_END);
579 if (stat (filename, &st) < 0)
586 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
587 original name. With the VMS file systems' versioning, everything
588 should be fine, and appending ".NN" just causes trouble.
593 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
594 doesn't exist is found. Return a freshly allocated copy of the
598 unique_name_1 (const char *prefix)
601 int plen = strlen (prefix);
602 char *template = (char *)alloca (plen + 1 + 24);
603 char *template_tail = template + plen;
605 memcpy (template, prefix, plen);
606 *template_tail++ = UNIQ_SEP;
609 number_to_string (template_tail, count++);
610 while (file_exists_p (template));
612 return xstrdup (template);
615 /* Return a unique file name, based on FILE.
617 More precisely, if FILE doesn't exist, it is returned unmodified.
618 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
619 file name that doesn't exist is returned.
621 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
623 The resulting file is not created, only verified that it didn't
624 exist at the point in time when the function was called.
625 Therefore, where security matters, don't rely that the file created
626 by this function exists until you open it with O_EXCL or
629 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
630 string. Otherwise, it may return FILE if the file doesn't exist
631 (and therefore doesn't need changing). */
634 unique_name (const char *file, bool allow_passthrough)
636 /* If the FILE itself doesn't exist, return it without
638 if (!file_exists_p (file))
639 return allow_passthrough ? (char *)file : xstrdup (file);
641 /* Otherwise, find a numeric suffix that results in unused file name
643 return unique_name_1 (file);
646 #else /* def UNIQ_SEP */
648 /* Dummy unique_name() for VMS. Return the original name as easily as
652 unique_name (const char *file, bool allow_passthrough)
654 /* Return the FILE itself, without modification, irregardful. */
655 return allow_passthrough ? (char *)file : xstrdup (file);
658 #endif /* def UNIQ_SEP [else] */
660 /* Create a file based on NAME, except without overwriting an existing
661 file with that name. Providing O_EXCL is correctly implemented,
662 this function does not have the race condition associated with
663 opening the file returned by unique_name. */
666 unique_create (const char *name, bool binary, char **opened_name)
668 /* unique file name, based on NAME */
669 char *uname = unique_name (name, false);
671 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
674 uname = unique_name (name, false);
676 if (opened_name && fp != NULL)
679 *opened_name = uname;
691 /* Open the file for writing, with the addition that the file is
692 opened "exclusively". This means that, if the file already exists,
693 this function will *fail* and errno will be set to EEXIST. If
694 BINARY is set, the file will be opened in binary mode, equivalent
697 If opening the file fails for any reason, including the file having
698 previously existed, this function returns NULL and sets errno
702 fopen_excl (const char *fname, int binary)
708 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
709 It also has file versions which obviate all the O_EXCL effort.
710 O_TRUNC (something of a misnomer) requests a new version.
713 /* Common open() optional arguments:
714 sequential access only, access callback function.
716 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
719 int flags = O_WRONLY | O_CREAT | O_TRUNC;
724 fd = open( fname, /* File name. */
726 0777, /* Mode for default protection. */
727 "ctx=bin,stm", /* Binary, stream access. */
728 "rfm=stmlf", /* Stream_LF. */
729 OPEN_OPT_ARGS); /* Access callback. */
734 fd = open( fname, /* File name. */
736 0777, /* Mode for default protection. */
737 "ctx=bin,stm", /* Binary, stream access. */
738 "rfm=fix", /* Fixed-length, */
739 "mrs=512", /* 512-byte records. */
740 OPEN_OPT_ARGS); /* Access callback. */
745 fd = open( fname, /* File name. */
747 0777, /* Mode for default protection.
749 "rfm=stmlf", /* Stream_LF. */
750 OPEN_OPT_ARGS); /* Access callback. */
752 # else /* def __VMS */
753 int flags = O_WRONLY | O_CREAT | O_EXCL;
758 fd = open (fname, flags, 0666);
759 # endif /* def __VMS [else] */
763 return fdopen (fd, binary ? "wb" : "w");
764 #else /* not O_EXCL */
765 /* Manually check whether the file exists. This is prone to race
766 conditions, but systems without O_EXCL haven't deserved
768 if (file_exists_p (fname))
773 return fopen (fname, binary ? "wb" : "w");
774 #endif /* not O_EXCL */
777 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
778 are missing, create them first. In case any mkdir() call fails,
779 return its error status. Returns 0 on successful completion.
781 The behaviour of this function should be identical to the behaviour
782 of `mkdir -p' on systems where mkdir supports the `-p' option. */
784 make_directory (const char *directory)
786 int i, ret, quit = 0;
789 /* Make a copy of dir, to be able to write to it. Otherwise, the
790 function is unsafe if called with a read-only char *argument. */
791 STRDUP_ALLOCA (dir, directory);
793 /* If the first character of dir is '/', skip it (and thus enable
794 creation of absolute-pathname directories. */
795 for (i = (*dir == '/'); 1; ++i)
797 for (; dir[i] && dir[i] != '/'; i++)
802 /* Check whether the directory already exists. Allow creation of
803 of intermediate directories to fail, as the initial path components
804 are not necessarily directories! */
805 if (!file_exists_p (dir))
806 ret = mkdir (dir, 0777);
817 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
818 should be a file name.
820 file_merge("/foo/bar", "baz") => "/foo/baz"
821 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
822 file_merge("foo", "bar") => "bar"
824 In other words, it's a simpler and gentler version of uri_merge. */
827 file_merge (const char *base, const char *file)
830 const char *cut = (const char *)strrchr (base, '/');
833 return xstrdup (file);
835 result = xmalloc (cut - base + 1 + strlen (file) + 1);
836 memcpy (result, base, cut - base);
837 result[cut - base] = '/';
838 strcpy (result + (cut - base) + 1, file);
843 /* Like fnmatch, but performs a case-insensitive match. */
846 fnmatch_nocase (const char *pattern, const char *string, int flags)
849 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
850 also present on *BSD platforms, and possibly elsewhere. */
851 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
853 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
854 char *patcopy = (char *) alloca (strlen (pattern) + 1);
855 char *strcopy = (char *) alloca (strlen (string) + 1);
857 for (p = patcopy; *pattern; pattern++, p++)
858 *p = c_tolower (*pattern);
860 for (p = strcopy; *string; string++, p++)
861 *p = c_tolower (*string);
863 return fnmatch (patcopy, strcopy, flags);
867 static bool in_acclist (const char *const *, const char *, bool);
869 /* Determine whether a file is acceptable to be followed, according to
870 lists of patterns to accept/reject. */
872 acceptable (const char *s)
876 while (l && s[l] != '/')
883 return (in_acclist ((const char *const *)opt.accepts, s, true)
884 && !in_acclist ((const char *const *)opt.rejects, s, true));
886 return in_acclist ((const char *const *)opt.accepts, s, true);
888 else if (opt.rejects)
889 return !in_acclist ((const char *const *)opt.rejects, s, true);
893 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
894 will return true if and only if D2 begins with `/something/' or is exactly
897 subdir_p (const char *d1, const char *d2)
901 if (!opt.ignore_case)
902 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
905 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
908 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
911 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
912 first element that matches DIR, through wildcards or front comparison (as
915 dir_matches_p (char **dirlist, const char *dir)
918 int (*matcher) (const char *, const char *, int)
919 = opt.ignore_case ? fnmatch_nocase : fnmatch;
921 for (x = dirlist; *x; x++)
923 /* Remove leading '/' */
924 char *p = *x + (**x == '/');
925 if (has_wildcards_p (p))
927 if (matcher (p, dir, FNM_PATHNAME) == 0)
932 if (subdir_p (p, dir))
937 return *x ? true : false;
940 /* Returns whether DIRECTORY is acceptable for download, wrt the
941 include/exclude lists.
943 The leading `/' is ignored in paths; relative and absolute paths
944 may be freely intermixed. */
947 accdir (const char *directory)
949 /* Remove starting '/'. */
950 if (*directory == '/')
954 if (!dir_matches_p (opt.includes, directory))
959 if (dir_matches_p (opt.excludes, directory))
965 /* Return true if STRING ends with TAIL. For instance:
967 match_tail ("abc", "bc", false) -> 1
968 match_tail ("abc", "ab", false) -> 0
969 match_tail ("abc", "abc", false) -> 1
971 If FOLD_CASE is true, the comparison will be case-insensitive. */
974 match_tail (const char *string, const char *tail, bool fold_case)
978 /* We want this to be fast, so we code two loops, one with
979 case-folding, one without. */
983 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
984 if (string[i] != tail[j])
989 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
990 if (c_tolower (string[i]) != c_tolower (tail[j]))
994 /* If the tail was exhausted, the match was succesful. */
1001 /* Checks whether string S matches each element of ACCEPTS. A list
1002 element are matched either with fnmatch() or match_tail(),
1003 according to whether the element contains wildcards or not.
1005 If the BACKWARD is false, don't do backward comparison -- just compare
1008 in_acclist (const char *const *accepts, const char *s, bool backward)
1010 for (; *accepts; accepts++)
1012 if (has_wildcards_p (*accepts))
1014 int res = opt.ignore_case
1015 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1016 /* fnmatch returns 0 if the pattern *does* match the string. */
1024 if (match_tail (s, *accepts, opt.ignore_case))
1029 int cmp = opt.ignore_case
1030 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1039 /* Return the location of STR's suffix (file extension). Examples:
1040 suffix ("foo.bar") -> "bar"
1041 suffix ("foo.bar.baz") -> "baz"
1042 suffix ("/foo/bar") -> NULL
1043 suffix ("/foo.bar/baz") -> NULL */
1045 suffix (const char *str)
1049 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1052 if (str[i++] == '.')
1053 return (char *)str + i;
1058 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1062 has_wildcards_p (const char *s)
1065 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1070 /* Return true if FNAME ends with a typical HTML suffix. The
1071 following (case-insensitive) suffixes are presumed to be HTML
1076 ?html (`?' matches one character)
1078 #### CAVEAT. This is not necessarily a good indication that FNAME
1079 refers to a file that contains HTML! */
1081 has_html_suffix_p (const char *fname)
1085 if ((suf = suffix (fname)) == NULL)
1087 if (!strcasecmp (suf, "html"))
1089 if (!strcasecmp (suf, "htm"))
1091 if (suf[0] && !strcasecmp (suf + 1, "html"))
1096 /* Read a line from FP and return the pointer to freshly allocated
1097 storage. The storage space is obtained through malloc() and should
1098 be freed with free() when it is no longer needed.
1100 The length of the line is not limited, except by available memory.
1101 The newline character at the end of line is retained. The line is
1102 terminated with a zero character.
1104 After end-of-file is encountered without anything being read, NULL
1105 is returned. NULL is also returned on error. To distinguish
1106 between these two cases, use the stdio function ferror(). */
1109 read_whole_line (FILE *fp)
1113 char *line = xmalloc (bufsize);
1115 while (fgets (line + length, bufsize - length, fp))
1117 length += strlen (line + length);
1119 /* Possible for example when reading from a binary file where
1120 a line begins with \0. */
1123 if (line[length - 1] == '\n')
1126 /* fgets() guarantees to read the whole line, or to use up the
1127 space we've given it. We can double the buffer
1130 line = xrealloc (line, bufsize);
1132 if (length == 0 || ferror (fp))
1137 if (length + 1 < bufsize)
1138 /* Relieve the memory from our exponential greediness. We say
1139 `length + 1' because the terminating \0 is not included in
1140 LENGTH. We don't need to zero-terminate the string ourselves,
1141 though, because fgets() does that. */
1142 line = xrealloc (line, length + 1);
1146 /* Read FILE into memory. A pointer to `struct file_memory' are
1147 returned; use struct element `content' to access file contents, and
1148 the element `length' to know the file length. `content' is *not*
1149 zero-terminated, and you should *not* read or write beyond the [0,
1150 length) range of characters.
1152 After you are done with the file contents, call wget_read_file_free to
1155 Depending on the operating system and the type of file that is
1156 being read, wget_read_file() either mmap's the file into memory, or
1157 reads the file into the core using read().
1159 If file is named "-", fileno(stdin) is used for reading instead.
1160 If you want to read from a real file named "-", use "./-" instead. */
1162 struct file_memory *
1163 wget_read_file (const char *file)
1166 struct file_memory *fm;
1168 bool inhibit_close = false;
1170 /* Some magic in the finest tradition of Perl and its kin: if FILE
1171 is "-", just use stdin. */
1174 fd = fileno (stdin);
1175 inhibit_close = true;
1176 /* Note that we don't inhibit mmap() in this case. If stdin is
1177 redirected from a regular file, mmap() will still work. */
1180 fd = open (file, O_RDONLY);
1183 fm = xnew (struct file_memory);
1188 if (fstat (fd, &buf) < 0)
1190 fm->length = buf.st_size;
1191 /* NOTE: As far as I know, the callers of this function never
1192 modify the file text. Relying on this would enable us to
1193 specify PROT_READ and MAP_SHARED for a marginal gain in
1194 efficiency, but at some cost to generality. */
1195 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1196 MAP_PRIVATE, fd, 0);
1197 if (fm->content == (char *)MAP_FAILED)
1207 /* The most common reason why mmap() fails is that FD does not point
1208 to a plain file. However, it's also possible that mmap() doesn't
1209 work for a particular type of file. Therefore, whenever mmap()
1210 fails, we just fall back to the regular method. */
1211 #endif /* HAVE_MMAP */
1214 size = 512; /* number of bytes fm->contents can
1215 hold at any given time. */
1216 fm->content = xmalloc (size);
1220 if (fm->length > size / 2)
1222 /* #### I'm not sure whether the whole exponential-growth
1223 thing makes sense with kernel read. On Linux at least,
1224 read() refuses to read more than 4K from a file at a
1225 single chunk anyway. But other Unixes might optimize it
1226 better, and it doesn't *hurt* anything, so I'm leaving
1229 /* Normally, we grow SIZE exponentially to make the number
1230 of calls to read() and realloc() logarithmic in relation
1231 to file size. However, read() can read an amount of data
1232 smaller than requested, and it would be unreasonable to
1233 double SIZE every time *something* was read. Therefore,
1234 we double SIZE only when the length exceeds half of the
1235 entire allocated size. */
1237 fm->content = xrealloc (fm->content, size);
1239 nread = read (fd, fm->content + fm->length, size - fm->length);
1241 /* Successful read. */
1242 fm->length += nread;
1252 if (size > fm->length && fm->length != 0)
1253 /* Due to exponential growth of fm->content, the allocated region
1254 might be much larger than what is actually needed. */
1255 fm->content = xrealloc (fm->content, fm->length);
1262 xfree (fm->content);
1267 /* Release the resources held by FM. Specifically, this calls
1268 munmap() or xfree() on fm->content, depending whether mmap or
1269 malloc/read were used to read in the file. It also frees the
1270 memory needed to hold the FM structure itself. */
1273 wget_read_file_free (struct file_memory *fm)
1278 munmap (fm->content, fm->length);
1283 xfree (fm->content);
1288 /* Free the pointers in a NULL-terminated vector of pointers, then
1289 free the pointer itself. */
1291 free_vec (char **vec)
1302 /* Append vector V2 to vector V1. The function frees V2 and
1303 reallocates V1 (thus you may not use the contents of neither
1304 pointer after the call). If V1 is NULL, V2 is returned. */
1306 merge_vecs (char **v1, char **v2)
1316 /* To avoid j == 0 */
1321 for (i = 0; v1[i]; i++)
1324 for (j = 0; v2[j]; j++)
1326 /* Reallocate v1. */
1327 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1328 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1333 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1334 is allocated as needed. Return the new value of the vector. */
1337 vec_append (char **vec, const char *str)
1339 int cnt; /* count of vector elements, including
1340 the one we're about to append */
1343 for (cnt = 0; vec[cnt]; cnt++)
1349 /* Reallocate the array to fit the new element and the NULL. */
1350 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1351 /* Append a copy of STR to the vector. */
1352 vec[cnt - 1] = xstrdup (str);
1357 /* Sometimes it's useful to create "sets" of strings, i.e. special
1358 hash tables where you want to store strings as keys and merely
1359 query for their existence. Here is a set of utility routines that
1360 makes that transparent. */
1363 string_set_add (struct hash_table *ht, const char *s)
1365 /* First check whether the set element already exists. If it does,
1366 do nothing so that we don't have to free() the old element and
1367 then strdup() a new one. */
1368 if (hash_table_contains (ht, s))
1371 /* We use "1" as value. It provides us a useful and clear arbitrary
1372 value, and it consumes no memory -- the pointers to the same
1373 string "1" will be shared by all the key-value pairs in all `set'
1375 hash_table_put (ht, xstrdup (s), "1");
1378 /* Synonym for hash_table_contains... */
1381 string_set_contains (struct hash_table *ht, const char *s)
1383 return hash_table_contains (ht, s);
1386 /* Convert the specified string set to array. ARRAY should be large
1387 enough to hold hash_table_count(ht) char pointers. */
1389 void string_set_to_array (struct hash_table *ht, char **array)
1391 hash_table_iterator iter;
1392 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1393 *array++ = iter.key;
1396 /* Free the string set. This frees both the storage allocated for
1397 keys and the actual hash table. (hash_table_destroy would only
1398 destroy the hash table.) */
1401 string_set_free (struct hash_table *ht)
1403 hash_table_iterator iter;
1404 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1406 hash_table_destroy (ht);
1409 /* Utility function: simply call xfree() on all keys and values of HT. */
1412 free_keys_and_values (struct hash_table *ht)
1414 hash_table_iterator iter;
1415 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1422 /* Get digit grouping data for thousand separors by calling
1423 localeconv(). The data includes separator string and grouping info
1424 and is cached after the first call to the function.
1426 In locales that don't set a thousand separator (such as the "C"
1427 locale), this forces it to be ",". We are now only showing
1428 thousand separators in one place, so this shouldn't be a problem in
1432 get_grouping_data (const char **sep, const char **grouping)
1434 static const char *cached_sep;
1435 static const char *cached_grouping;
1436 static bool initialized;
1439 /* Get the grouping info from the locale. */
1440 struct lconv *lconv = localeconv ();
1441 cached_sep = lconv->thousands_sep;
1442 cached_grouping = lconv->grouping;
1443 #if ! USE_NLS_PROGRESS_BAR
1444 /* We can't count column widths, so ensure that the separator
1445 * is single-byte only (let check below determine what byte). */
1446 if (strlen(cached_sep) > 1)
1451 /* Many locales (such as "C" or "hr_HR") don't specify
1452 grouping, which we still want to use it for legibility.
1453 In those locales set the sep char to ',', unless that
1454 character is used for decimal point, in which case set it
1456 if (*lconv->decimal_point != ',')
1460 cached_grouping = "\x03";
1465 *grouping = cached_grouping;
1468 /* Return a printed representation of N with thousand separators.
1469 This should respect locale settings, with the exception of the "C"
1470 locale which mandates no separator, but we use one anyway.
1472 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1473 the separators because it's too non-portable, and it's hard to test
1474 for this feature at configure time. Besides, it wouldn't display
1475 separators in the "C" locale, still used by many Unix users. */
1478 with_thousand_seps (wgint n)
1480 static char outbuf[48];
1481 char *p = outbuf + sizeof outbuf;
1483 /* Info received from locale */
1484 const char *grouping, *sep;
1487 /* State information */
1488 int i = 0, groupsize;
1489 const char *atgroup;
1491 bool negative = n < 0;
1493 /* Initialize grouping data. */
1494 get_grouping_data (&sep, &grouping);
1495 seplen = strlen (sep);
1497 groupsize = *atgroup++;
1499 /* This would overflow on WGINT_MIN, but printing negative numbers
1500 is not an important goal of this fuinction. */
1504 /* Write the number into the buffer, backwards, inserting the
1505 separators as necessary. */
1509 *--p = n % 10 + '0';
1513 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1514 if (++i == groupsize)
1519 memcpy (p -= seplen, sep, seplen);
1522 groupsize = *atgroup++;
1531 /* N, a byte quantity, is converted to a human-readable abberviated
1532 form a la sizes printed by `ls -lh'. The result is written to a
1533 static buffer, a pointer to which is returned.
1535 Unlike `with_thousand_seps', this approximates to the nearest unit.
1536 Quoting GNU libit: "Most people visually process strings of 3-4
1537 digits effectively, but longer strings of digits are more prone to
1538 misinterpretation. Hence, converting to an abbreviated form
1539 usually improves readability."
1541 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1542 original computer-related meaning of "powers of 1024". We don't
1543 use the "*bibyte" names invented in 1998, and seldom used in
1544 practice. Wikipedia's entry on "binary prefix" discusses this in
1548 human_readable (HR_NUMTYPE n)
1550 /* These suffixes are compatible with those of GNU `ls -lh'. */
1551 static char powers[] =
1553 'K', /* kilobyte, 2^10 bytes */
1554 'M', /* megabyte, 2^20 bytes */
1555 'G', /* gigabyte, 2^30 bytes */
1556 'T', /* terabyte, 2^40 bytes */
1557 'P', /* petabyte, 2^50 bytes */
1558 'E', /* exabyte, 2^60 bytes */
1563 /* If the quantity is smaller than 1K, just print it. */
1566 snprintf (buf, sizeof (buf), "%d", (int) n);
1570 /* Loop over powers, dividing N with 1024 in each iteration. This
1571 works unchanged for all sizes of wgint, while still avoiding
1572 non-portable `long double' arithmetic. */
1573 for (i = 0; i < countof (powers); i++)
1575 /* At each iteration N is greater than the *subsequent* power.
1576 That way N/1024.0 produces a decimal number in the units of
1578 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1580 double val = n / 1024.0;
1581 /* Print values smaller than 10 with one decimal digits, and
1582 others without any decimals. */
1583 snprintf (buf, sizeof (buf), "%.*f%c",
1584 val < 10 ? 1 : 0, val, powers[i]);
1589 return NULL; /* unreached */
1592 /* Count the digits in the provided number. Used to allocate space
1593 when printing numbers. */
1596 numdigit (wgint number)
1600 ++cnt; /* accomodate '-' */
1601 while ((number /= 10) != 0)
1606 #define PR(mask) *p++ = n / (mask) + '0'
1608 /* DIGITS_<D> is used to print a D-digit number and should be called
1609 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1610 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1611 Recursively this continues until DIGITS_1 is invoked. */
1613 #define DIGITS_1(mask) PR (mask)
1614 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1615 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1616 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1617 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1618 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1619 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1620 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1621 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1622 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1624 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1626 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1627 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1628 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1629 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1630 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1631 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1632 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1633 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1634 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1636 /* Shorthand for casting to wgint. */
1639 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1640 `sprintf(buffer, "%lld", (long long) number)', only typically much
1641 faster and portable to machines without long long.
1643 The speedup may make a difference in programs that frequently
1644 convert numbers to strings. Some implementations of sprintf,
1645 particularly the one in some versions of GNU libc, have been known
1646 to be quite slow when converting integers to strings.
1648 Return the pointer to the location where the terminating zero was
1649 printed. (Equivalent to calling buffer+strlen(buffer) after the
1652 BUFFER should be large enough to accept as many bytes as you expect
1653 the number to take up. On machines with 64-bit wgints the maximum
1654 needed size is 24 bytes. That includes the digits needed for the
1655 largest 64-bit number, the `-' sign in case it's negative, and the
1656 terminating '\0'. */
1659 number_to_string (char *buffer, wgint number)
1664 int last_digit_char = 0;
1666 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1667 /* We are running in a very strange environment. Leave the correct
1668 printing to sprintf. */
1669 p += sprintf (buf, "%j", (intmax_t) (n));
1670 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1676 /* n = -n would overflow because -n would evaluate to a
1677 wgint value larger than WGINT_MAX. Need to make n
1678 smaller and handle the last digit separately. */
1679 int last_digit = n % 10;
1680 /* The sign of n%10 is implementation-defined. */
1682 last_digit_char = '0' - last_digit;
1684 last_digit_char = '0' + last_digit;
1685 /* After n is made smaller, -n will not overflow. */
1693 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1694 way printing any N is fully open-coded without a loop or jump.
1695 (Also see description of DIGITS_*.) */
1697 if (n < 10) DIGITS_1 (1);
1698 else if (n < 100) DIGITS_2 (10);
1699 else if (n < 1000) DIGITS_3 (100);
1700 else if (n < 10000) DIGITS_4 (1000);
1701 else if (n < 100000) DIGITS_5 (10000);
1702 else if (n < 1000000) DIGITS_6 (100000);
1703 else if (n < 10000000) DIGITS_7 (1000000);
1704 else if (n < 100000000) DIGITS_8 (10000000);
1705 else if (n < 1000000000) DIGITS_9 (100000000);
1706 #if SIZEOF_WGINT == 4
1707 /* wgint is 32 bits wide: no number has more than 10 digits. */
1708 else DIGITS_10 (1000000000);
1710 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1711 Constants are constructed by compile-time multiplication to avoid
1712 dealing with different notations for 64-bit constants
1713 (nL/nLL/nI64, depending on the compiler and architecture). */
1714 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1715 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1716 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1717 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1718 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1719 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1720 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1721 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1722 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1723 else DIGITS_19 (1000000000*(W)1000000000);
1726 if (last_digit_char)
1727 *p++ = last_digit_char;
1730 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1737 #undef SPRINTF_WGINT
1760 /* Print NUMBER to a statically allocated string and return a pointer
1761 to the printed representation.
1763 This function is intended to be used in conjunction with printf.
1764 It is hard to portably print wgint values:
1765 a) you cannot use printf("%ld", number) because wgint can be long
1766 long on 32-bit machines with LFS.
1767 b) you cannot use printf("%lld", number) because NUMBER could be
1768 long on 32-bit machines without LFS, or on 64-bit machines,
1769 which do not require LFS. Also, Windows doesn't support %lld.
1770 c) you cannot use printf("%j", (int_max_t) number) because not all
1771 versions of printf support "%j", the most notable being the one
1773 d) you cannot #define WGINT_FMT to the appropriate format and use
1774 printf(WGINT_FMT, number) because that would break translations
1775 for user-visible messages, such as printf("Downloaded: %d
1778 What you should use instead is printf("%s", number_to_static_string
1781 CAVEAT: since the function returns pointers to static data, you
1782 must be careful to copy its result before calling it again.
1783 However, to make it more useful with printf, the function maintains
1784 an internal ring of static buffers to return. That way things like
1785 printf("%s %s", number_to_static_string (num1),
1786 number_to_static_string (num2)) work as expected. Three buffers
1787 are currently used, which means that "%s %s %s" will work, but "%s
1788 %s %s %s" won't. If you need to print more than three wgints,
1789 bump the RING_SIZE (or rethink your message.) */
1792 number_to_static_string (wgint number)
1794 static char ring[RING_SIZE][24];
1796 char *buf = ring[ringpos];
1797 number_to_string (buf, number);
1798 ringpos = (ringpos + 1) % RING_SIZE;
1802 /* Determine the width of the terminal we're running on. If that's
1803 not possible, return 0. */
1806 determine_screen_width (void)
1808 /* If there's a way to get the terminal size using POSIX
1809 tcgetattr(), somebody please tell me. */
1814 if (opt.lfilename != NULL)
1817 fd = fileno (stderr);
1818 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1819 return 0; /* most likely ENOTTY */
1822 #elif defined(WINDOWS)
1823 CONSOLE_SCREEN_BUFFER_INFO csbi;
1824 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1826 return csbi.dwSize.X;
1827 #else /* neither TIOCGWINSZ nor WINDOWS */
1829 #endif /* neither TIOCGWINSZ nor WINDOWS */
1832 /* Whether the rnd system (either rand or [dl]rand48) has been
1834 static int rnd_seeded;
1836 /* Return a random number between 0 and MAX-1, inclusive.
1838 If the system does not support lrand48 and MAX is greater than the
1839 value of RAND_MAX+1 on the system, the returned value will be in
1840 the range [0, RAND_MAX]. This may be fixed in a future release.
1841 The random number generator is seeded automatically the first time
1844 This uses lrand48 where available, rand elsewhere. DO NOT use it
1845 for cryptography. It is only meant to be used in situations where
1846 quality of the random numbers returned doesn't really matter. */
1849 random_number (int max)
1854 srand48 ((long) time (NULL) ^ (long) getpid ());
1857 return lrand48 () % max;
1858 #else /* not HAVE_DRAND48 */
1864 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1869 /* Like rand() % max, but uses the high-order bits for better
1870 randomness on architectures where rand() is implemented using a
1871 simple congruential generator. */
1873 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1874 return (int) bounded;
1876 #endif /* not HAVE_DRAND48 */
1879 /* Return a random uniformly distributed floating point number in the
1880 [0, 1) range. Uses drand48 where available, and a really lame
1881 kludge elsewhere. */
1889 srand48 ((long) time (NULL) ^ (long) getpid ());
1893 #else /* not HAVE_DRAND48 */
1894 return ( random_number (10000) / 10000.0
1895 + random_number (10000) / (10000.0 * 10000.0)
1896 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1897 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1898 #endif /* not HAVE_DRAND48 */
1901 /* Implementation of run_with_timeout, a generic timeout-forcing
1902 routine for systems with Unix-like signal handling. */
1904 #ifdef USE_SIGNAL_TIMEOUT
1905 # ifdef HAVE_SIGSETJMP
1906 # define SETJMP(env) sigsetjmp (env, 1)
1908 static sigjmp_buf run_with_timeout_env;
1911 abort_run_with_timeout (int sig)
1913 assert (sig == SIGALRM);
1914 siglongjmp (run_with_timeout_env, -1);
1916 # else /* not HAVE_SIGSETJMP */
1917 # define SETJMP(env) setjmp (env)
1919 static jmp_buf run_with_timeout_env;
1922 abort_run_with_timeout (int sig)
1924 assert (sig == SIGALRM);
1925 /* We don't have siglongjmp to preserve the set of blocked signals;
1926 if we longjumped out of the handler at this point, SIGALRM would
1927 remain blocked. We must unblock it manually. */
1928 int mask = siggetmask ();
1929 mask &= ~sigmask (SIGALRM);
1932 /* Now it's safe to longjump. */
1933 longjmp (run_with_timeout_env, -1);
1935 # endif /* not HAVE_SIGSETJMP */
1937 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1938 setitimer where available, alarm otherwise.
1940 TIMEOUT should be non-zero. If the timeout value is so small that
1941 it would be rounded to zero, it is rounded to the least legal value
1942 instead (1us for setitimer, 1s for alarm). That ensures that
1943 SIGALRM will be delivered in all cases. */
1946 alarm_set (double timeout)
1949 /* Use the modern itimer interface. */
1950 struct itimerval itv;
1952 itv.it_value.tv_sec = (long) timeout;
1953 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1954 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1955 /* Ensure that we wait for at least the minimum interval.
1956 Specifying zero would mean "wait forever". */
1957 itv.it_value.tv_usec = 1;
1958 setitimer (ITIMER_REAL, &itv, NULL);
1959 #else /* not ITIMER_REAL */
1960 /* Use the old alarm() interface. */
1961 int secs = (int) timeout;
1963 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1964 because alarm(0) means "never deliver the alarm", i.e. "wait
1965 forever", which is not what someone who specifies a 0.5s
1966 timeout would expect. */
1969 #endif /* not ITIMER_REAL */
1972 /* Cancel the alarm set with alarm_set. */
1978 struct itimerval disable;
1980 setitimer (ITIMER_REAL, &disable, NULL);
1981 #else /* not ITIMER_REAL */
1983 #endif /* not ITIMER_REAL */
1986 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1987 seconds. Returns true if the function was interrupted with a
1988 timeout, false otherwise.
1990 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1991 using setitimer() or alarm(). The timeout is enforced by
1992 longjumping out of the SIGALRM handler. This has several
1993 advantages compared to the traditional approach of relying on
1994 signals causing system calls to exit with EINTR:
1996 * The callback function is *forcibly* interrupted after the
1997 timeout expires, (almost) regardless of what it was doing and
1998 whether it was in a syscall. For example, a calculation that
1999 takes a long time is interrupted as reliably as an IO
2002 * It works with both SYSV and BSD signals because it doesn't
2003 depend on the default setting of SA_RESTART.
2005 * It doesn't require special handler setup beyond a simple call
2006 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2009 The only downside is that, if FUN allocates internal resources that
2010 are normally freed prior to exit from the functions, they will be
2011 lost in case of timeout. */
2014 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2024 signal (SIGALRM, abort_run_with_timeout);
2025 if (SETJMP (run_with_timeout_env) != 0)
2027 /* Longjumped out of FUN with a timeout. */
2028 signal (SIGALRM, SIG_DFL);
2031 alarm_set (timeout);
2034 /* Preserve errno in case alarm() or signal() modifies it. */
2035 saved_errno = errno;
2037 signal (SIGALRM, SIG_DFL);
2038 errno = saved_errno;
2043 #else /* not USE_SIGNAL_TIMEOUT */
2046 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2047 define it under Windows, because Windows has its own version of
2048 run_with_timeout that uses threads. */
2051 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2056 #endif /* not WINDOWS */
2057 #endif /* not USE_SIGNAL_TIMEOUT */
2061 /* Sleep the specified amount of seconds. On machines without
2062 nanosleep(), this may sleep shorter if interrupted by signals. */
2065 xsleep (double seconds)
2067 #ifdef HAVE_NANOSLEEP
2068 /* nanosleep is the preferred interface because it offers high
2069 accuracy and, more importantly, because it allows us to reliably
2070 restart receiving a signal such as SIGWINCH. (There was an
2071 actual Debian bug report about --limit-rate malfunctioning while
2072 the terminal was being resized.) */
2073 struct timespec sleep, remaining;
2074 sleep.tv_sec = (long) seconds;
2075 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2076 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2077 /* If nanosleep has been interrupted by a signal, adjust the
2078 sleeping period and return to sleep. */
2080 #elif defined(HAVE_USLEEP)
2081 /* If usleep is available, use it in preference to select. */
2084 /* On some systems, usleep cannot handle values larger than
2085 1,000,000. If the period is larger than that, use sleep
2086 first, then add usleep for subsecond accuracy. */
2088 seconds -= (long) seconds;
2090 usleep (seconds * 1000000);
2091 #else /* fall back select */
2092 /* Note that, although Windows supports select, it can't be used to
2093 implement sleeping because Winsock's select doesn't implement
2094 timeout when it is passed NULL pointers for all fd sets. (But it
2095 does under Cygwin, which implements Unix-compatible select.) */
2096 struct timeval sleep;
2097 sleep.tv_sec = (long) seconds;
2098 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2099 select (0, NULL, NULL, NULL, &sleep);
2100 /* If select returns -1 and errno is EINTR, it means we were
2101 interrupted by a signal. But without knowing how long we've
2102 actually slept, we can't return to sleep. Using gettimeofday to
2103 track sleeps is slow and unreliable due to clock skew. */
2107 #endif /* not WINDOWS */
2109 /* Encode the octets in DATA of length LENGTH to base64 format,
2110 storing the result to DEST. The output will be zero-terminated,
2111 and must point to a writable buffer of at least
2112 1+BASE64_LENGTH(length) bytes. The function returns the length of
2113 the resulting base64 data, not counting the terminating zero.
2115 This implementation does not emit newlines after 76 characters of
2119 base64_encode (const void *data, int length, char *dest)
2121 /* Conversion table. */
2122 static const char tbl[64] = {
2123 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2124 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2125 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2126 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2128 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2129 don't work for data with MSB set. */
2130 const unsigned char *s = data;
2131 /* Theoretical ANSI violation when length < 3. */
2132 const unsigned char *end = (const unsigned char *) data + length - 2;
2135 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2136 for (; s < end; s += 3)
2138 *p++ = tbl[s[0] >> 2];
2139 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2140 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2141 *p++ = tbl[s[2] & 0x3f];
2144 /* Pad the result if necessary... */
2148 *p++ = tbl[s[0] >> 2];
2149 *p++ = tbl[(s[0] & 3) << 4];
2154 *p++ = tbl[s[0] >> 2];
2155 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2156 *p++ = tbl[((s[1] & 0xf) << 2)];
2160 /* ...and zero-terminate it. */
2166 /* Store in C the next non-whitespace character from the string, or \0
2167 when end of string is reached. */
2168 #define NEXT_CHAR(c, p) do { \
2169 c = (unsigned char) *p++; \
2170 } while (c_isspace (c))
2172 #define IS_ASCII(c) (((c) & 0x80) == 0)
2174 /* Decode data from BASE64 (a null-terminated string) into memory
2175 pointed to by DEST. DEST is assumed to be large enough to
2176 accomodate the decoded data, which is guaranteed to be no more than
2179 Since DEST is assumed to contain binary data, it is not
2180 NUL-terminated. The function returns the length of the data
2181 written to TO. -1 is returned in case of error caused by malformed
2184 This function originates from Free Recode. */
2187 base64_decode (const char *base64, void *dest)
2189 /* Table of base64 values for first 128 characters. Note that this
2190 assumes ASCII (but so does Wget in other places). */
2191 static const signed char base64_char_to_value[128] =
2193 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2194 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2195 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2196 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2197 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2198 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2199 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2200 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2201 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2202 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2203 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2204 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2205 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2207 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2208 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2210 const char *p = base64;
2216 unsigned long value;
2218 /* Process first byte of a quadruplet. */
2222 if (c == '=' || !IS_BASE64 (c))
2223 return -1; /* illegal char while decoding base64 */
2224 value = BASE64_CHAR_TO_VALUE (c) << 18;
2226 /* Process second byte of a quadruplet. */
2229 return -1; /* premature EOF while decoding base64 */
2230 if (c == '=' || !IS_BASE64 (c))
2231 return -1; /* illegal char while decoding base64 */
2232 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2235 /* Process third byte of a quadruplet. */
2238 return -1; /* premature EOF while decoding base64 */
2240 return -1; /* illegal char while decoding base64 */
2246 return -1; /* premature EOF while decoding base64 */
2248 return -1; /* padding `=' expected but not found */
2252 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2253 *q++ = 0xff & value >> 8;
2255 /* Process fourth byte of a quadruplet. */
2258 return -1; /* premature EOF while decoding base64 */
2262 return -1; /* illegal char while decoding base64 */
2264 value |= BASE64_CHAR_TO_VALUE (c);
2265 *q++ = 0xff & value;
2268 #undef BASE64_CHAR_TO_VALUE
2270 return q - (char *) dest;
2276 /* Simple merge sort for use by stable_sort. Implementation courtesy
2277 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2280 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2281 int (*cmpfun) (const void *, const void *))
2283 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2287 size_t mid = (to + from) / 2;
2288 mergesort_internal (base, temp, size, from, mid, cmpfun);
2289 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2292 for (k = from; (i <= mid) && (j <= to); k++)
2293 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2294 memcpy (ELT (temp, k), ELT (base, i++), size);
2296 memcpy (ELT (temp, k), ELT (base, j++), size);
2298 memcpy (ELT (temp, k++), ELT (base, i++), size);
2300 memcpy (ELT (temp, k++), ELT (base, j++), size);
2301 for (k = from; k <= to; k++)
2302 memcpy (ELT (base, k), ELT (temp, k), size);
2307 /* Stable sort with interface exactly like standard library's qsort.
2308 Uses mergesort internally, allocating temporary storage with
2312 stable_sort (void *base, size_t nmemb, size_t size,
2313 int (*cmpfun) (const void *, const void *))
2317 void *temp = alloca (nmemb * size * sizeof (void *));
2318 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2322 /* Print a decimal number. If it is equal to or larger than ten, the
2323 number is rounded. Otherwise it is printed with one significant
2324 digit without trailing zeros and with no more than three fractional
2325 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2326 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2328 This is useful for displaying durations because it provides
2329 order-of-magnitude information without unnecessary clutter --
2330 long-running downloads are shown without the fractional part, and
2331 short ones still retain one significant digit. */
2334 print_decimal (double number)
2336 static char buf[32];
2337 double n = number >= 0 ? number : -number;
2340 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2341 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2342 snprintf (buf, sizeof buf, "%.0f", number);
2344 snprintf (buf, sizeof buf, "%.1f", number);
2345 else if (n >= 0.001)
2346 snprintf (buf, sizeof buf, "%.1g", number);
2347 else if (n >= 0.0005)
2348 /* round [0.0005, 0.001) to 0.001 */
2349 snprintf (buf, sizeof buf, "%.3f", number);
2351 /* print numbers close to 0 as 0, not 0.000 */
2368 { "/somedir", "/somedir", true },
2369 { "/somedir", "/somedir/d2", true },
2370 { "/somedir/d1", "/somedir", false },
2373 for (i = 0; i < countof(test_array); ++i)
2375 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2377 mu_assert ("test_subdir_p: wrong result",
2378 res == test_array[i].result);
2385 test_dir_matches_p()
2393 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2394 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2395 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2396 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2397 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2398 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2399 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2400 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2401 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2402 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2403 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2404 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2405 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2406 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2407 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2410 for (i = 0; i < countof(test_array); ++i)
2412 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2414 mu_assert ("test_dir_matches_p: wrong result",
2415 res == test_array[i].result);
2421 #endif /* TESTING */