1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation,
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
40 # include <sys/mman.h>
43 # include <process.h> /* getpid() */
57 /* For TIOCGWINSZ and friends: */
58 #ifdef HAVE_SYS_IOCTL_H
59 # include <sys/ioctl.h>
65 /* Needed for Unix version of run_with_timeout. */
69 #ifndef HAVE_SIGSETJMP
70 /* If sigsetjmp is a macro, configure won't pick it up. */
72 # define HAVE_SIGSETJMP
76 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
77 # define USE_SIGNAL_TIMEOUT
85 #endif /* def __VMS */
92 memfatal (const char *context, long attempted_size)
94 /* Make sure we don't try to store part of the log line, and thus
96 log_set_save_context (false);
98 /* We have different log outputs in different situations:
99 1) output without bytes information
100 2) output with bytes information */
101 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
103 logprintf (LOG_ALWAYS,
104 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
109 logprintf (LOG_ALWAYS,
110 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
111 exec_name, context, attempted_size);
117 /* Character property table for (re-)escaping VMS ODS5 extended file
118 names. Note that this table ignores Unicode.
120 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
122 ODS5 Invalid characters:
123 C0 control codes (0x00 to 0x1F inclusive)
127 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
128 Double quotation marks (")
131 Left angle bracket (<)
132 Right angle bracket (>)
136 Characters escaped by "^":
137 SP ! " # % & ' ( ) + , . : ; =
140 Either "^_" or "^ " is accepted as a space. Period (.) is a special
141 case. Note that un-escaped < and > can also confuse a directory
144 Characters put out as ^xx:
146 80-9F (C1 control characters)
147 A0 (nonbreaking space)
148 FF (Latin small letter y diaeresis)
151 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
153 Property table values:
163 unsigned char char_prop[ 256] = {
165 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
166 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
168 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
169 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171 /* SP ! " # $ % & ' ( ) * + , - . / */
172 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
174 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
175 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
177 /* @ A B C D E F G H I J K L M N O */
178 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
180 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
181 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
183 /* ` a b c d e f g h i j k l m n o */
184 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
186 /* p q r s t u v w x y z { | } ~ DEL */
187 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
189 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
190 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
191 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
192 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
193 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
194 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
195 0, 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, 8
199 /* Utility function: like xstrdup(), but also lowercases S. */
202 xstrdup_lower (const char *s)
204 char *copy = xstrdup (s);
211 /* Copy the string formed by two pointers (one on the beginning, other
212 on the char after the last char) to a new, malloc-ed location.
215 strdupdelim (const char *beg, const char *end)
217 char *res = xmalloc (end - beg + 1);
218 memcpy (res, beg, end - beg);
219 res[end - beg] = '\0';
223 /* Parse a string containing comma-separated elements, and return a
224 vector of char pointers with the elements. Spaces following the
225 commas are ignored. */
227 sepstring (const char *s)
241 res = xrealloc (res, (i + 2) * sizeof (char *));
242 res[i] = strdupdelim (p, s);
245 /* Skip the blanks following the ','. */
246 while (c_isspace (*s))
253 res = xrealloc (res, (i + 2) * sizeof (char *));
254 res[i] = strdupdelim (p, s);
259 /* Like sprintf, but prints into a string of sufficient size freshly
260 allocated with malloc, which is returned. If unable to print due
261 to invalid format, returns NULL. Inability to allocate needed
262 memory results in abort, as with xmalloc. This is in spirit
263 similar to the GNU/BSD extension asprintf, but somewhat easier to
266 Internally the function either calls vasprintf or loops around
267 vsnprintf until the correct size is found. Since Wget also ships a
268 fallback implementation of vsnprintf, this should be portable. */
270 /* Constant is using for limits memory allocation for text buffer.
271 Applicable in situation when: vasprintf is not available in the system
272 and vsnprintf return -1 when long line is truncated (in old versions of
273 glibc and in other system where C99 doesn`t support) */
275 #define FMT_MAX_LENGTH 1048576
278 aprintf (const char *fmt, ...)
280 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
285 va_start (args, fmt);
286 ret = vasprintf (&str, fmt, args);
288 if (ret < 0 && errno == ENOMEM)
289 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
290 with xmalloc/xrealloc */
294 #else /* not HAVE_VASPRINTF */
296 /* vasprintf is unavailable. snprintf into a small buffer and
297 resize it as necessary. */
299 char *str = xmalloc (size);
301 /* #### This code will infloop and eventually abort in xrealloc if
302 passed a FMT that causes snprintf to consistently return -1. */
309 va_start (args, fmt);
310 n = vsnprintf (str, size, fmt, args);
313 /* If the printing worked, return the string. */
314 if (n > -1 && n < size)
317 /* Else try again with a larger buffer. */
318 if (n > -1) /* C99 */
319 size = n + 1; /* precisely what is needed */
320 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
321 { /* maybe we have some wrong
323 logprintf (LOG_ALWAYS,
324 _("%s: aprintf: text buffer is too big (%ld bytes), "
326 exec_name, size); /* printout a log message */
327 abort (); /* and abort... */
331 /* else, we continue to grow our
332 * buffer: Twice the old size. */
335 str = xrealloc (str, size);
337 #endif /* not HAVE_VASPRINTF */
340 /* Concatenate the NULL-terminated list of string arguments into
341 freshly allocated space. */
344 concat_strings (const char *str0, ...)
347 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
350 const char *next_str;
351 int total_length = 0;
354 /* Calculate the length of and allocate the resulting string. */
357 va_start (args, str0);
358 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
360 int len = strlen (next_str);
361 if (argcount < countof (saved_lengths))
362 saved_lengths[argcount++] = len;
366 p = ret = xmalloc (total_length + 1);
368 /* Copy the strings into the allocated space. */
371 va_start (args, str0);
372 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
375 if (argcount < countof (saved_lengths))
376 len = saved_lengths[argcount++];
378 len = strlen (next_str);
379 memcpy (p, next_str, len);
388 /* Format the provided time according to the specified format. The
389 format is a string with format elements supported by strftime. */
392 fmttime (time_t t, const char *fmt)
394 static char output[32];
395 struct tm *tm = localtime(&t);
398 if (!strftime(output, sizeof(output), fmt, tm))
403 /* Return pointer to a static char[] buffer in which zero-terminated
404 string-representation of TM (in form hh:mm:ss) is printed.
406 If TM is NULL, the current time will be used. */
411 return fmttime(t, "%H:%M:%S");
414 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
417 datetime_str (time_t t)
419 return fmttime(t, "%Y-%m-%d %H:%M:%S");
422 /* The Windows versions of the following two functions are defined in
423 mswindows.c. On MSDOS this function should never be called. */
428 fork_to_background (void)
433 #else /* def __VMS */
435 #if !defined(WINDOWS) && !defined(MSDOS)
437 fork_to_background (void)
440 /* Whether we arrange our own version of opt.lfilename here. */
441 bool logfile_changed = false;
443 if (!opt.lfilename && (!opt.quiet || opt.server_response))
445 /* We must create the file immediately to avoid either a race
446 condition (which arises from using unique_name and failing to
447 use fopen_excl) or lying to the user about the log file name
448 (which arises from using unique_name, printing the name, and
449 using fopen_excl later on.) */
450 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
453 logfile_changed = true;
466 /* parent, no error */
467 printf (_("Continuing in background, pid %d.\n"), (int) pid);
469 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
470 exit (0); /* #### should we use _exit()? */
473 /* child: give up the privileges and keep running. */
475 freopen ("/dev/null", "r", stdin);
476 freopen ("/dev/null", "w", stdout);
477 freopen ("/dev/null", "w", stderr);
479 #endif /* !WINDOWS && !MSDOS */
481 #endif /* def __VMS [else] */
484 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
485 specified with TM. The atime ("access time") is set to the current
489 touch (const char *file, time_t tm)
491 struct timeval timevals[2];
493 timevals[0].tv_sec = time (NULL);
494 timevals[0].tv_usec = 0L;
495 timevals[1].tv_sec = tm;
496 timevals[1].tv_usec = 0L;
498 if (utimes (file, timevals) == -1)
499 logprintf (LOG_NOTQUIET, "utimes(%s): %s\n", file, strerror (errno));
502 /* Checks if FILE is a symbolic link, and removes it if it is. Does
503 nothing under MS-Windows. */
505 remove_link (const char *file)
510 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
512 DEBUGP (("Unlinking %s (symlink).\n", file));
515 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
516 quote (file), strerror (errno));
521 /* Does FILENAME exist? This is quite a lousy implementation, since
522 it supplies no error codes -- only a yes-or-no answer. Thus it
523 will return that a file does not exist if, e.g., the directory is
524 unreadable. I don't mind it too much currently, though. The
525 proper way should, of course, be to have a third, error state,
526 other than true/false, but that would introduce uncalled-for
527 additional complexity to the callers. */
529 file_exists_p (const char *filename)
532 return access (filename, F_OK) >= 0;
535 return stat (filename, &buf) >= 0;
539 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
540 Returns 0 on error. */
542 file_non_directory_p (const char *path)
545 /* Use lstat() rather than stat() so that symbolic links pointing to
546 directories can be identified correctly. */
547 if (lstat (path, &buf) != 0)
549 return S_ISDIR (buf.st_mode) ? false : true;
552 /* Return the size of file named by FILENAME, or -1 if it cannot be
553 opened or seeked into. */
555 file_size (const char *filename)
557 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
559 /* We use fseek rather than stat to determine the file size because
560 that way we can also verify that the file is readable without
561 explicitly checking for permissions. Inspired by the POST patch
563 FILE *fp = fopen (filename, "rb");
566 fseeko (fp, 0, SEEK_END);
572 if (stat (filename, &st) < 0)
579 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
580 original name. With the VMS file systems' versioning, everything
581 should be fine, and appending ".NN" just causes trouble.
586 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
587 doesn't exist is found. Return a freshly allocated copy of the
591 unique_name_1 (const char *prefix)
594 int plen = strlen (prefix);
595 char *template = (char *)alloca (plen + 1 + 24);
596 char *template_tail = template + plen;
598 memcpy (template, prefix, plen);
599 *template_tail++ = UNIQ_SEP;
602 number_to_string (template_tail, count++);
603 while (file_exists_p (template));
605 return xstrdup (template);
608 /* Return a unique file name, based on FILE.
610 More precisely, if FILE doesn't exist, it is returned unmodified.
611 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
612 file name that doesn't exist is returned.
614 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
616 The resulting file is not created, only verified that it didn't
617 exist at the point in time when the function was called.
618 Therefore, where security matters, don't rely that the file created
619 by this function exists until you open it with O_EXCL or
622 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
623 string. Otherwise, it may return FILE if the file doesn't exist
624 (and therefore doesn't need changing). */
627 unique_name (const char *file, bool allow_passthrough)
629 /* If the FILE itself doesn't exist, return it without
631 if (!file_exists_p (file))
632 return allow_passthrough ? (char *)file : xstrdup (file);
634 /* Otherwise, find a numeric suffix that results in unused file name
636 return unique_name_1 (file);
639 #else /* def UNIQ_SEP */
641 /* Dummy unique_name() for VMS. Return the original name as easily as
645 unique_name (const char *file, bool allow_passthrough)
647 /* Return the FILE itself, without modification, irregardful. */
648 return allow_passthrough ? (char *)file : xstrdup (file);
651 #endif /* def UNIQ_SEP [else] */
653 /* Create a file based on NAME, except without overwriting an existing
654 file with that name. Providing O_EXCL is correctly implemented,
655 this function does not have the race condition associated with
656 opening the file returned by unique_name. */
659 unique_create (const char *name, bool binary, char **opened_name)
661 /* unique file name, based on NAME */
662 char *uname = unique_name (name, false);
664 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
667 uname = unique_name (name, false);
669 if (opened_name && fp != NULL)
672 *opened_name = uname;
684 /* Open the file for writing, with the addition that the file is
685 opened "exclusively". This means that, if the file already exists,
686 this function will *fail* and errno will be set to EEXIST. If
687 BINARY is set, the file will be opened in binary mode, equivalent
690 If opening the file fails for any reason, including the file having
691 previously existed, this function returns NULL and sets errno
695 fopen_excl (const char *fname, int binary)
701 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
702 It also has file versions which obviate all the O_EXCL effort.
703 O_TRUNC (something of a misnomer) requests a new version.
706 /* Common open() optional arguments:
707 sequential access only, access callback function.
709 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
712 int flags = O_WRONLY | O_CREAT | O_TRUNC;
717 fd = open( fname, /* File name. */
719 0777, /* Mode for default protection. */
720 "ctx=bin,stm", /* Binary, stream access. */
721 "rfm=stmlf", /* Stream_LF. */
722 OPEN_OPT_ARGS); /* Access callback. */
727 fd = open( fname, /* File name. */
729 0777, /* Mode for default protection. */
730 "ctx=bin,stm", /* Binary, stream access. */
731 "rfm=fix", /* Fixed-length, */
732 "mrs=512", /* 512-byte records. */
733 OPEN_OPT_ARGS); /* Access callback. */
738 fd = open( fname, /* File name. */
740 0777, /* Mode for default protection.
742 "rfm=stmlf", /* Stream_LF. */
743 OPEN_OPT_ARGS); /* Access callback. */
745 # else /* def __VMS */
746 int flags = O_WRONLY | O_CREAT | O_EXCL;
751 fd = open (fname, flags, 0666);
752 # endif /* def __VMS [else] */
756 return fdopen (fd, binary ? "wb" : "w");
757 #else /* not O_EXCL */
758 /* Manually check whether the file exists. This is prone to race
759 conditions, but systems without O_EXCL haven't deserved
761 if (file_exists_p (fname))
766 return fopen (fname, binary ? "wb" : "w");
767 #endif /* not O_EXCL */
770 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
771 are missing, create them first. In case any mkdir() call fails,
772 return its error status. Returns 0 on successful completion.
774 The behaviour of this function should be identical to the behaviour
775 of `mkdir -p' on systems where mkdir supports the `-p' option. */
777 make_directory (const char *directory)
779 int i, ret, quit = 0;
782 /* Make a copy of dir, to be able to write to it. Otherwise, the
783 function is unsafe if called with a read-only char *argument. */
784 STRDUP_ALLOCA (dir, directory);
786 /* If the first character of dir is '/', skip it (and thus enable
787 creation of absolute-pathname directories. */
788 for (i = (*dir == '/'); 1; ++i)
790 for (; dir[i] && dir[i] != '/'; i++)
795 /* Check whether the directory already exists. Allow creation of
796 of intermediate directories to fail, as the initial path components
797 are not necessarily directories! */
798 if (!file_exists_p (dir))
799 ret = mkdir (dir, 0777);
810 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
811 should be a file name.
813 file_merge("/foo/bar", "baz") => "/foo/baz"
814 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
815 file_merge("foo", "bar") => "bar"
817 In other words, it's a simpler and gentler version of uri_merge. */
820 file_merge (const char *base, const char *file)
823 const char *cut = (const char *)strrchr (base, '/');
826 return xstrdup (file);
828 result = xmalloc (cut - base + 1 + strlen (file) + 1);
829 memcpy (result, base, cut - base);
830 result[cut - base] = '/';
831 strcpy (result + (cut - base) + 1, file);
836 /* Like fnmatch, but performs a case-insensitive match. */
839 fnmatch_nocase (const char *pattern, const char *string, int flags)
842 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
843 also present on *BSD platforms, and possibly elsewhere. */
844 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
846 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
847 char *patcopy = (char *) alloca (strlen (pattern) + 1);
848 char *strcopy = (char *) alloca (strlen (string) + 1);
850 for (p = patcopy; *pattern; pattern++, p++)
851 *p = c_tolower (*pattern);
853 for (p = strcopy; *string; string++, p++)
854 *p = c_tolower (*string);
856 return fnmatch (patcopy, strcopy, flags);
860 static bool in_acclist (const char *const *, const char *, bool);
862 /* Determine whether a file is acceptable to be followed, according to
863 lists of patterns to accept/reject. */
865 acceptable (const char *s)
869 while (l && s[l] != '/')
876 return (in_acclist ((const char *const *)opt.accepts, s, true)
877 && !in_acclist ((const char *const *)opt.rejects, s, true));
879 return in_acclist ((const char *const *)opt.accepts, s, true);
881 else if (opt.rejects)
882 return !in_acclist ((const char *const *)opt.rejects, s, true);
886 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
887 will return true if and only if D2 begins with `/something/' or is exactly
890 subdir_p (const char *d1, const char *d2)
894 if (!opt.ignore_case)
895 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
898 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
901 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
904 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
905 first element that matches DIR, through wildcards or front comparison (as
908 dir_matches_p (char **dirlist, const char *dir)
911 int (*matcher) (const char *, const char *, int)
912 = opt.ignore_case ? fnmatch_nocase : fnmatch;
914 for (x = dirlist; *x; x++)
916 /* Remove leading '/' */
917 char *p = *x + (**x == '/');
918 if (has_wildcards_p (p))
920 if (matcher (p, dir, FNM_PATHNAME) == 0)
925 if (subdir_p (p, dir))
930 return *x ? true : false;
933 /* Returns whether DIRECTORY is acceptable for download, wrt the
934 include/exclude lists.
936 The leading `/' is ignored in paths; relative and absolute paths
937 may be freely intermixed. */
940 accdir (const char *directory)
942 /* Remove starting '/'. */
943 if (*directory == '/')
947 if (!dir_matches_p (opt.includes, directory))
952 if (dir_matches_p (opt.excludes, directory))
958 /* Return true if STRING ends with TAIL. For instance:
960 match_tail ("abc", "bc", false) -> 1
961 match_tail ("abc", "ab", false) -> 0
962 match_tail ("abc", "abc", false) -> 1
964 If FOLD_CASE is true, the comparison will be case-insensitive. */
967 match_tail (const char *string, const char *tail, bool fold_case)
971 /* We want this to be fast, so we code two loops, one with
972 case-folding, one without. */
976 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
977 if (string[i] != tail[j])
982 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
983 if (c_tolower (string[i]) != c_tolower (tail[j]))
987 /* If the tail was exhausted, the match was succesful. */
994 /* Checks whether string S matches each element of ACCEPTS. A list
995 element are matched either with fnmatch() or match_tail(),
996 according to whether the element contains wildcards or not.
998 If the BACKWARD is false, don't do backward comparison -- just compare
1001 in_acclist (const char *const *accepts, const char *s, bool backward)
1003 for (; *accepts; accepts++)
1005 if (has_wildcards_p (*accepts))
1007 int res = opt.ignore_case
1008 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1009 /* fnmatch returns 0 if the pattern *does* match the string. */
1017 if (match_tail (s, *accepts, opt.ignore_case))
1022 int cmp = opt.ignore_case
1023 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1032 /* Return the location of STR's suffix (file extension). Examples:
1033 suffix ("foo.bar") -> "bar"
1034 suffix ("foo.bar.baz") -> "baz"
1035 suffix ("/foo/bar") -> NULL
1036 suffix ("/foo.bar/baz") -> NULL */
1038 suffix (const char *str)
1042 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1045 if (str[i++] == '.')
1046 return (char *)str + i;
1051 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1055 has_wildcards_p (const char *s)
1058 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1063 /* Return true if FNAME ends with a typical HTML suffix. The
1064 following (case-insensitive) suffixes are presumed to be HTML
1069 ?html (`?' matches one character)
1071 #### CAVEAT. This is not necessarily a good indication that FNAME
1072 refers to a file that contains HTML! */
1074 has_html_suffix_p (const char *fname)
1078 if ((suf = suffix (fname)) == NULL)
1080 if (!strcasecmp (suf, "html"))
1082 if (!strcasecmp (suf, "htm"))
1084 if (suf[0] && !strcasecmp (suf + 1, "html"))
1089 /* Read a line from FP and return the pointer to freshly allocated
1090 storage. The storage space is obtained through malloc() and should
1091 be freed with free() when it is no longer needed.
1093 The length of the line is not limited, except by available memory.
1094 The newline character at the end of line is retained. The line is
1095 terminated with a zero character.
1097 After end-of-file is encountered without anything being read, NULL
1098 is returned. NULL is also returned on error. To distinguish
1099 between these two cases, use the stdio function ferror(). */
1102 read_whole_line (FILE *fp)
1106 char *line = xmalloc (bufsize);
1108 while (fgets (line + length, bufsize - length, fp))
1110 length += strlen (line + length);
1112 /* Possible for example when reading from a binary file where
1113 a line begins with \0. */
1116 if (line[length - 1] == '\n')
1119 /* fgets() guarantees to read the whole line, or to use up the
1120 space we've given it. We can double the buffer
1123 line = xrealloc (line, bufsize);
1125 if (length == 0 || ferror (fp))
1130 if (length + 1 < bufsize)
1131 /* Relieve the memory from our exponential greediness. We say
1132 `length + 1' because the terminating \0 is not included in
1133 LENGTH. We don't need to zero-terminate the string ourselves,
1134 though, because fgets() does that. */
1135 line = xrealloc (line, length + 1);
1139 /* Read FILE into memory. A pointer to `struct file_memory' are
1140 returned; use struct element `content' to access file contents, and
1141 the element `length' to know the file length. `content' is *not*
1142 zero-terminated, and you should *not* read or write beyond the [0,
1143 length) range of characters.
1145 After you are done with the file contents, call wget_read_file_free to
1148 Depending on the operating system and the type of file that is
1149 being read, wget_read_file() either mmap's the file into memory, or
1150 reads the file into the core using read().
1152 If file is named "-", fileno(stdin) is used for reading instead.
1153 If you want to read from a real file named "-", use "./-" instead. */
1155 struct file_memory *
1156 wget_read_file (const char *file)
1159 struct file_memory *fm;
1161 bool inhibit_close = false;
1163 /* Some magic in the finest tradition of Perl and its kin: if FILE
1164 is "-", just use stdin. */
1167 fd = fileno (stdin);
1168 inhibit_close = true;
1169 /* Note that we don't inhibit mmap() in this case. If stdin is
1170 redirected from a regular file, mmap() will still work. */
1173 fd = open (file, O_RDONLY);
1176 fm = xnew (struct file_memory);
1181 if (fstat (fd, &buf) < 0)
1183 fm->length = buf.st_size;
1184 /* NOTE: As far as I know, the callers of this function never
1185 modify the file text. Relying on this would enable us to
1186 specify PROT_READ and MAP_SHARED for a marginal gain in
1187 efficiency, but at some cost to generality. */
1188 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1189 MAP_PRIVATE, fd, 0);
1190 if (fm->content == (char *)MAP_FAILED)
1200 /* The most common reason why mmap() fails is that FD does not point
1201 to a plain file. However, it's also possible that mmap() doesn't
1202 work for a particular type of file. Therefore, whenever mmap()
1203 fails, we just fall back to the regular method. */
1204 #endif /* HAVE_MMAP */
1207 size = 512; /* number of bytes fm->contents can
1208 hold at any given time. */
1209 fm->content = xmalloc (size);
1213 if (fm->length > size / 2)
1215 /* #### I'm not sure whether the whole exponential-growth
1216 thing makes sense with kernel read. On Linux at least,
1217 read() refuses to read more than 4K from a file at a
1218 single chunk anyway. But other Unixes might optimize it
1219 better, and it doesn't *hurt* anything, so I'm leaving
1222 /* Normally, we grow SIZE exponentially to make the number
1223 of calls to read() and realloc() logarithmic in relation
1224 to file size. However, read() can read an amount of data
1225 smaller than requested, and it would be unreasonable to
1226 double SIZE every time *something* was read. Therefore,
1227 we double SIZE only when the length exceeds half of the
1228 entire allocated size. */
1230 fm->content = xrealloc (fm->content, size);
1232 nread = read (fd, fm->content + fm->length, size - fm->length);
1234 /* Successful read. */
1235 fm->length += nread;
1245 if (size > fm->length && fm->length != 0)
1246 /* Due to exponential growth of fm->content, the allocated region
1247 might be much larger than what is actually needed. */
1248 fm->content = xrealloc (fm->content, fm->length);
1255 xfree (fm->content);
1260 /* Release the resources held by FM. Specifically, this calls
1261 munmap() or xfree() on fm->content, depending whether mmap or
1262 malloc/read were used to read in the file. It also frees the
1263 memory needed to hold the FM structure itself. */
1266 wget_read_file_free (struct file_memory *fm)
1271 munmap (fm->content, fm->length);
1276 xfree (fm->content);
1281 /* Free the pointers in a NULL-terminated vector of pointers, then
1282 free the pointer itself. */
1284 free_vec (char **vec)
1295 /* Append vector V2 to vector V1. The function frees V2 and
1296 reallocates V1 (thus you may not use the contents of neither
1297 pointer after the call). If V1 is NULL, V2 is returned. */
1299 merge_vecs (char **v1, char **v2)
1309 /* To avoid j == 0 */
1314 for (i = 0; v1[i]; i++)
1317 for (j = 0; v2[j]; j++)
1319 /* Reallocate v1. */
1320 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1321 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1326 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1327 is allocated as needed. Return the new value of the vector. */
1330 vec_append (char **vec, const char *str)
1332 int cnt; /* count of vector elements, including
1333 the one we're about to append */
1336 for (cnt = 0; vec[cnt]; cnt++)
1342 /* Reallocate the array to fit the new element and the NULL. */
1343 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1344 /* Append a copy of STR to the vector. */
1345 vec[cnt - 1] = xstrdup (str);
1350 /* Sometimes it's useful to create "sets" of strings, i.e. special
1351 hash tables where you want to store strings as keys and merely
1352 query for their existence. Here is a set of utility routines that
1353 makes that transparent. */
1356 string_set_add (struct hash_table *ht, const char *s)
1358 /* First check whether the set element already exists. If it does,
1359 do nothing so that we don't have to free() the old element and
1360 then strdup() a new one. */
1361 if (hash_table_contains (ht, s))
1364 /* We use "1" as value. It provides us a useful and clear arbitrary
1365 value, and it consumes no memory -- the pointers to the same
1366 string "1" will be shared by all the key-value pairs in all `set'
1368 hash_table_put (ht, xstrdup (s), "1");
1371 /* Synonym for hash_table_contains... */
1374 string_set_contains (struct hash_table *ht, const char *s)
1376 return hash_table_contains (ht, s);
1379 /* Convert the specified string set to array. ARRAY should be large
1380 enough to hold hash_table_count(ht) char pointers. */
1382 void string_set_to_array (struct hash_table *ht, char **array)
1384 hash_table_iterator iter;
1385 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1386 *array++ = iter.key;
1389 /* Free the string set. This frees both the storage allocated for
1390 keys and the actual hash table. (hash_table_destroy would only
1391 destroy the hash table.) */
1394 string_set_free (struct hash_table *ht)
1396 hash_table_iterator iter;
1397 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1399 hash_table_destroy (ht);
1402 /* Utility function: simply call xfree() on all keys and values of HT. */
1405 free_keys_and_values (struct hash_table *ht)
1407 hash_table_iterator iter;
1408 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1415 /* Get digit grouping data for thousand separors by calling
1416 localeconv(). The data includes separator string and grouping info
1417 and is cached after the first call to the function.
1419 In locales that don't set a thousand separator (such as the "C"
1420 locale), this forces it to be ",". We are now only showing
1421 thousand separators in one place, so this shouldn't be a problem in
1425 get_grouping_data (const char **sep, const char **grouping)
1427 static const char *cached_sep;
1428 static const char *cached_grouping;
1429 static bool initialized;
1432 /* Get the grouping info from the locale. */
1433 struct lconv *lconv = localeconv ();
1434 cached_sep = lconv->thousands_sep;
1435 cached_grouping = lconv->grouping;
1436 #if ! USE_NLS_PROGRESS_BAR
1437 /* We can't count column widths, so ensure that the separator
1438 * is single-byte only (let check below determine what byte). */
1439 if (strlen(cached_sep) > 1)
1444 /* Many locales (such as "C" or "hr_HR") don't specify
1445 grouping, which we still want to use it for legibility.
1446 In those locales set the sep char to ',', unless that
1447 character is used for decimal point, in which case set it
1449 if (*lconv->decimal_point != ',')
1453 cached_grouping = "\x03";
1458 *grouping = cached_grouping;
1461 /* Return a printed representation of N with thousand separators.
1462 This should respect locale settings, with the exception of the "C"
1463 locale which mandates no separator, but we use one anyway.
1465 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1466 the separators because it's too non-portable, and it's hard to test
1467 for this feature at configure time. Besides, it wouldn't display
1468 separators in the "C" locale, still used by many Unix users. */
1471 with_thousand_seps (wgint n)
1473 static char outbuf[48];
1474 char *p = outbuf + sizeof outbuf;
1476 /* Info received from locale */
1477 const char *grouping, *sep;
1480 /* State information */
1481 int i = 0, groupsize;
1482 const char *atgroup;
1484 bool negative = n < 0;
1486 /* Initialize grouping data. */
1487 get_grouping_data (&sep, &grouping);
1488 seplen = strlen (sep);
1490 groupsize = *atgroup++;
1492 /* This would overflow on WGINT_MIN, but printing negative numbers
1493 is not an important goal of this fuinction. */
1497 /* Write the number into the buffer, backwards, inserting the
1498 separators as necessary. */
1502 *--p = n % 10 + '0';
1506 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1507 if (++i == groupsize)
1512 memcpy (p -= seplen, sep, seplen);
1515 groupsize = *atgroup++;
1524 /* N, a byte quantity, is converted to a human-readable abberviated
1525 form a la sizes printed by `ls -lh'. The result is written to a
1526 static buffer, a pointer to which is returned.
1528 Unlike `with_thousand_seps', this approximates to the nearest unit.
1529 Quoting GNU libit: "Most people visually process strings of 3-4
1530 digits effectively, but longer strings of digits are more prone to
1531 misinterpretation. Hence, converting to an abbreviated form
1532 usually improves readability."
1534 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1535 original computer-related meaning of "powers of 1024". We don't
1536 use the "*bibyte" names invented in 1998, and seldom used in
1537 practice. Wikipedia's entry on "binary prefix" discusses this in
1541 human_readable (HR_NUMTYPE n)
1543 /* These suffixes are compatible with those of GNU `ls -lh'. */
1544 static char powers[] =
1546 'K', /* kilobyte, 2^10 bytes */
1547 'M', /* megabyte, 2^20 bytes */
1548 'G', /* gigabyte, 2^30 bytes */
1549 'T', /* terabyte, 2^40 bytes */
1550 'P', /* petabyte, 2^50 bytes */
1551 'E', /* exabyte, 2^60 bytes */
1556 /* If the quantity is smaller than 1K, just print it. */
1559 snprintf (buf, sizeof (buf), "%d", (int) n);
1563 /* Loop over powers, dividing N with 1024 in each iteration. This
1564 works unchanged for all sizes of wgint, while still avoiding
1565 non-portable `long double' arithmetic. */
1566 for (i = 0; i < countof (powers); i++)
1568 /* At each iteration N is greater than the *subsequent* power.
1569 That way N/1024.0 produces a decimal number in the units of
1571 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1573 double val = n / 1024.0;
1574 /* Print values smaller than 10 with one decimal digits, and
1575 others without any decimals. */
1576 snprintf (buf, sizeof (buf), "%.*f%c",
1577 val < 10 ? 1 : 0, val, powers[i]);
1582 return NULL; /* unreached */
1585 /* Count the digits in the provided number. Used to allocate space
1586 when printing numbers. */
1589 numdigit (wgint number)
1593 ++cnt; /* accomodate '-' */
1594 while ((number /= 10) != 0)
1599 #define PR(mask) *p++ = n / (mask) + '0'
1601 /* DIGITS_<D> is used to print a D-digit number and should be called
1602 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1603 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1604 Recursively this continues until DIGITS_1 is invoked. */
1606 #define DIGITS_1(mask) PR (mask)
1607 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1608 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1609 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1610 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1611 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1612 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1613 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1614 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1615 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1617 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1619 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1620 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1621 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1622 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1623 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1624 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1625 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1626 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1627 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1629 /* Shorthand for casting to wgint. */
1632 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1633 `sprintf(buffer, "%lld", (long long) number)', only typically much
1634 faster and portable to machines without long long.
1636 The speedup may make a difference in programs that frequently
1637 convert numbers to strings. Some implementations of sprintf,
1638 particularly the one in some versions of GNU libc, have been known
1639 to be quite slow when converting integers to strings.
1641 Return the pointer to the location where the terminating zero was
1642 printed. (Equivalent to calling buffer+strlen(buffer) after the
1645 BUFFER should be large enough to accept as many bytes as you expect
1646 the number to take up. On machines with 64-bit wgints the maximum
1647 needed size is 24 bytes. That includes the digits needed for the
1648 largest 64-bit number, the `-' sign in case it's negative, and the
1649 terminating '\0'. */
1652 number_to_string (char *buffer, wgint number)
1657 int last_digit_char = 0;
1659 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1660 /* We are running in a very strange environment. Leave the correct
1661 printing to sprintf. */
1662 p += sprintf (buf, "%j", (intmax_t) (n));
1663 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1669 /* n = -n would overflow because -n would evaluate to a
1670 wgint value larger than WGINT_MAX. Need to make n
1671 smaller and handle the last digit separately. */
1672 int last_digit = n % 10;
1673 /* The sign of n%10 is implementation-defined. */
1675 last_digit_char = '0' - last_digit;
1677 last_digit_char = '0' + last_digit;
1678 /* After n is made smaller, -n will not overflow. */
1686 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1687 way printing any N is fully open-coded without a loop or jump.
1688 (Also see description of DIGITS_*.) */
1690 if (n < 10) DIGITS_1 (1);
1691 else if (n < 100) DIGITS_2 (10);
1692 else if (n < 1000) DIGITS_3 (100);
1693 else if (n < 10000) DIGITS_4 (1000);
1694 else if (n < 100000) DIGITS_5 (10000);
1695 else if (n < 1000000) DIGITS_6 (100000);
1696 else if (n < 10000000) DIGITS_7 (1000000);
1697 else if (n < 100000000) DIGITS_8 (10000000);
1698 else if (n < 1000000000) DIGITS_9 (100000000);
1699 #if SIZEOF_WGINT == 4
1700 /* wgint is 32 bits wide: no number has more than 10 digits. */
1701 else DIGITS_10 (1000000000);
1703 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1704 Constants are constructed by compile-time multiplication to avoid
1705 dealing with different notations for 64-bit constants
1706 (nL/nLL/nI64, depending on the compiler and architecture). */
1707 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1708 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1709 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1710 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1711 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1712 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1713 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1714 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1715 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1716 else DIGITS_19 (1000000000*(W)1000000000);
1719 if (last_digit_char)
1720 *p++ = last_digit_char;
1723 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1730 #undef SPRINTF_WGINT
1753 /* Print NUMBER to a statically allocated string and return a pointer
1754 to the printed representation.
1756 This function is intended to be used in conjunction with printf.
1757 It is hard to portably print wgint values:
1758 a) you cannot use printf("%ld", number) because wgint can be long
1759 long on 32-bit machines with LFS.
1760 b) you cannot use printf("%lld", number) because NUMBER could be
1761 long on 32-bit machines without LFS, or on 64-bit machines,
1762 which do not require LFS. Also, Windows doesn't support %lld.
1763 c) you cannot use printf("%j", (int_max_t) number) because not all
1764 versions of printf support "%j", the most notable being the one
1766 d) you cannot #define WGINT_FMT to the appropriate format and use
1767 printf(WGINT_FMT, number) because that would break translations
1768 for user-visible messages, such as printf("Downloaded: %d
1771 What you should use instead is printf("%s", number_to_static_string
1774 CAVEAT: since the function returns pointers to static data, you
1775 must be careful to copy its result before calling it again.
1776 However, to make it more useful with printf, the function maintains
1777 an internal ring of static buffers to return. That way things like
1778 printf("%s %s", number_to_static_string (num1),
1779 number_to_static_string (num2)) work as expected. Three buffers
1780 are currently used, which means that "%s %s %s" will work, but "%s
1781 %s %s %s" won't. If you need to print more than three wgints,
1782 bump the RING_SIZE (or rethink your message.) */
1785 number_to_static_string (wgint number)
1787 static char ring[RING_SIZE][24];
1789 char *buf = ring[ringpos];
1790 number_to_string (buf, number);
1791 ringpos = (ringpos + 1) % RING_SIZE;
1795 /* Determine the width of the terminal we're running on. If that's
1796 not possible, return 0. */
1799 determine_screen_width (void)
1801 /* If there's a way to get the terminal size using POSIX
1802 tcgetattr(), somebody please tell me. */
1807 if (opt.lfilename != NULL)
1810 fd = fileno (stderr);
1811 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1812 return 0; /* most likely ENOTTY */
1815 #elif defined(WINDOWS)
1816 CONSOLE_SCREEN_BUFFER_INFO csbi;
1817 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1819 return csbi.dwSize.X;
1820 #else /* neither TIOCGWINSZ nor WINDOWS */
1822 #endif /* neither TIOCGWINSZ nor WINDOWS */
1825 /* Whether the rnd system (either rand or [dl]rand48) has been
1827 static int rnd_seeded;
1829 /* Return a random number between 0 and MAX-1, inclusive.
1831 If the system does not support lrand48 and MAX is greater than the
1832 value of RAND_MAX+1 on the system, the returned value will be in
1833 the range [0, RAND_MAX]. This may be fixed in a future release.
1834 The random number generator is seeded automatically the first time
1837 This uses lrand48 where available, rand elsewhere. DO NOT use it
1838 for cryptography. It is only meant to be used in situations where
1839 quality of the random numbers returned doesn't really matter. */
1842 random_number (int max)
1847 srand48 ((long) time (NULL) ^ (long) getpid ());
1850 return lrand48 () % max;
1851 #else /* not HAVE_DRAND48 */
1857 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1862 /* Like rand() % max, but uses the high-order bits for better
1863 randomness on architectures where rand() is implemented using a
1864 simple congruential generator. */
1866 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1867 return (int) bounded;
1869 #endif /* not HAVE_DRAND48 */
1872 /* Return a random uniformly distributed floating point number in the
1873 [0, 1) range. Uses drand48 where available, and a really lame
1874 kludge elsewhere. */
1882 srand48 ((long) time (NULL) ^ (long) getpid ());
1886 #else /* not HAVE_DRAND48 */
1887 return ( random_number (10000) / 10000.0
1888 + random_number (10000) / (10000.0 * 10000.0)
1889 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1890 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1891 #endif /* not HAVE_DRAND48 */
1894 /* Implementation of run_with_timeout, a generic timeout-forcing
1895 routine for systems with Unix-like signal handling. */
1897 #ifdef USE_SIGNAL_TIMEOUT
1898 # ifdef HAVE_SIGSETJMP
1899 # define SETJMP(env) sigsetjmp (env, 1)
1901 static sigjmp_buf run_with_timeout_env;
1904 abort_run_with_timeout (int sig)
1906 assert (sig == SIGALRM);
1907 siglongjmp (run_with_timeout_env, -1);
1909 # else /* not HAVE_SIGSETJMP */
1910 # define SETJMP(env) setjmp (env)
1912 static jmp_buf run_with_timeout_env;
1915 abort_run_with_timeout (int sig)
1917 assert (sig == SIGALRM);
1918 /* We don't have siglongjmp to preserve the set of blocked signals;
1919 if we longjumped out of the handler at this point, SIGALRM would
1920 remain blocked. We must unblock it manually. */
1921 int mask = siggetmask ();
1922 mask &= ~sigmask (SIGALRM);
1925 /* Now it's safe to longjump. */
1926 longjmp (run_with_timeout_env, -1);
1928 # endif /* not HAVE_SIGSETJMP */
1930 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1931 setitimer where available, alarm otherwise.
1933 TIMEOUT should be non-zero. If the timeout value is so small that
1934 it would be rounded to zero, it is rounded to the least legal value
1935 instead (1us for setitimer, 1s for alarm). That ensures that
1936 SIGALRM will be delivered in all cases. */
1939 alarm_set (double timeout)
1942 /* Use the modern itimer interface. */
1943 struct itimerval itv;
1945 itv.it_value.tv_sec = (long) timeout;
1946 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1947 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1948 /* Ensure that we wait for at least the minimum interval.
1949 Specifying zero would mean "wait forever". */
1950 itv.it_value.tv_usec = 1;
1951 setitimer (ITIMER_REAL, &itv, NULL);
1952 #else /* not ITIMER_REAL */
1953 /* Use the old alarm() interface. */
1954 int secs = (int) timeout;
1956 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1957 because alarm(0) means "never deliver the alarm", i.e. "wait
1958 forever", which is not what someone who specifies a 0.5s
1959 timeout would expect. */
1962 #endif /* not ITIMER_REAL */
1965 /* Cancel the alarm set with alarm_set. */
1971 struct itimerval disable;
1973 setitimer (ITIMER_REAL, &disable, NULL);
1974 #else /* not ITIMER_REAL */
1976 #endif /* not ITIMER_REAL */
1979 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1980 seconds. Returns true if the function was interrupted with a
1981 timeout, false otherwise.
1983 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1984 using setitimer() or alarm(). The timeout is enforced by
1985 longjumping out of the SIGALRM handler. This has several
1986 advantages compared to the traditional approach of relying on
1987 signals causing system calls to exit with EINTR:
1989 * The callback function is *forcibly* interrupted after the
1990 timeout expires, (almost) regardless of what it was doing and
1991 whether it was in a syscall. For example, a calculation that
1992 takes a long time is interrupted as reliably as an IO
1995 * It works with both SYSV and BSD signals because it doesn't
1996 depend on the default setting of SA_RESTART.
1998 * It doesn't require special handler setup beyond a simple call
1999 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2002 The only downside is that, if FUN allocates internal resources that
2003 are normally freed prior to exit from the functions, they will be
2004 lost in case of timeout. */
2007 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2017 signal (SIGALRM, abort_run_with_timeout);
2018 if (SETJMP (run_with_timeout_env) != 0)
2020 /* Longjumped out of FUN with a timeout. */
2021 signal (SIGALRM, SIG_DFL);
2024 alarm_set (timeout);
2027 /* Preserve errno in case alarm() or signal() modifies it. */
2028 saved_errno = errno;
2030 signal (SIGALRM, SIG_DFL);
2031 errno = saved_errno;
2036 #else /* not USE_SIGNAL_TIMEOUT */
2039 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2040 define it under Windows, because Windows has its own version of
2041 run_with_timeout that uses threads. */
2044 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2049 #endif /* not WINDOWS */
2050 #endif /* not USE_SIGNAL_TIMEOUT */
2054 /* Sleep the specified amount of seconds. On machines without
2055 nanosleep(), this may sleep shorter if interrupted by signals. */
2058 xsleep (double seconds)
2060 #ifdef HAVE_NANOSLEEP
2061 /* nanosleep is the preferred interface because it offers high
2062 accuracy and, more importantly, because it allows us to reliably
2063 restart receiving a signal such as SIGWINCH. (There was an
2064 actual Debian bug report about --limit-rate malfunctioning while
2065 the terminal was being resized.) */
2066 struct timespec sleep, remaining;
2067 sleep.tv_sec = (long) seconds;
2068 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2069 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2070 /* If nanosleep has been interrupted by a signal, adjust the
2071 sleeping period and return to sleep. */
2073 #elif defined(HAVE_USLEEP)
2074 /* If usleep is available, use it in preference to select. */
2077 /* On some systems, usleep cannot handle values larger than
2078 1,000,000. If the period is larger than that, use sleep
2079 first, then add usleep for subsecond accuracy. */
2081 seconds -= (long) seconds;
2083 usleep (seconds * 1000000);
2084 #else /* fall back select */
2085 /* Note that, although Windows supports select, it can't be used to
2086 implement sleeping because Winsock's select doesn't implement
2087 timeout when it is passed NULL pointers for all fd sets. (But it
2088 does under Cygwin, which implements Unix-compatible select.) */
2089 struct timeval sleep;
2090 sleep.tv_sec = (long) seconds;
2091 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2092 select (0, NULL, NULL, NULL, &sleep);
2093 /* If select returns -1 and errno is EINTR, it means we were
2094 interrupted by a signal. But without knowing how long we've
2095 actually slept, we can't return to sleep. Using gettimeofday to
2096 track sleeps is slow and unreliable due to clock skew. */
2100 #endif /* not WINDOWS */
2102 /* Encode the octets in DATA of length LENGTH to base64 format,
2103 storing the result to DEST. The output will be zero-terminated,
2104 and must point to a writable buffer of at least
2105 1+BASE64_LENGTH(length) bytes. The function returns the length of
2106 the resulting base64 data, not counting the terminating zero.
2108 This implementation does not emit newlines after 76 characters of
2112 base64_encode (const void *data, int length, char *dest)
2114 /* Conversion table. */
2115 static const char tbl[64] = {
2116 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2117 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2118 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2119 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2121 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2122 don't work for data with MSB set. */
2123 const unsigned char *s = data;
2124 /* Theoretical ANSI violation when length < 3. */
2125 const unsigned char *end = (const unsigned char *) data + length - 2;
2128 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2129 for (; s < end; s += 3)
2131 *p++ = tbl[s[0] >> 2];
2132 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2133 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2134 *p++ = tbl[s[2] & 0x3f];
2137 /* Pad the result if necessary... */
2141 *p++ = tbl[s[0] >> 2];
2142 *p++ = tbl[(s[0] & 3) << 4];
2147 *p++ = tbl[s[0] >> 2];
2148 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2149 *p++ = tbl[((s[1] & 0xf) << 2)];
2153 /* ...and zero-terminate it. */
2159 /* Store in C the next non-whitespace character from the string, or \0
2160 when end of string is reached. */
2161 #define NEXT_CHAR(c, p) do { \
2162 c = (unsigned char) *p++; \
2163 } while (c_isspace (c))
2165 #define IS_ASCII(c) (((c) & 0x80) == 0)
2167 /* Decode data from BASE64 (a null-terminated string) into memory
2168 pointed to by DEST. DEST is assumed to be large enough to
2169 accomodate the decoded data, which is guaranteed to be no more than
2172 Since DEST is assumed to contain binary data, it is not
2173 NUL-terminated. The function returns the length of the data
2174 written to TO. -1 is returned in case of error caused by malformed
2177 This function originates from Free Recode. */
2180 base64_decode (const char *base64, void *dest)
2182 /* Table of base64 values for first 128 characters. Note that this
2183 assumes ASCII (but so does Wget in other places). */
2184 static const signed char base64_char_to_value[128] =
2186 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2187 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2188 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2189 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2190 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2191 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2192 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2193 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2194 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2195 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2196 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2197 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2198 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2200 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2201 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2203 const char *p = base64;
2209 unsigned long value;
2211 /* Process first byte of a quadruplet. */
2215 if (c == '=' || !IS_BASE64 (c))
2216 return -1; /* illegal char while decoding base64 */
2217 value = BASE64_CHAR_TO_VALUE (c) << 18;
2219 /* Process second byte of a quadruplet. */
2222 return -1; /* premature EOF while decoding base64 */
2223 if (c == '=' || !IS_BASE64 (c))
2224 return -1; /* illegal char while decoding base64 */
2225 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2228 /* Process third byte of a quadruplet. */
2231 return -1; /* premature EOF while decoding base64 */
2233 return -1; /* illegal char while decoding base64 */
2239 return -1; /* premature EOF while decoding base64 */
2241 return -1; /* padding `=' expected but not found */
2245 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2246 *q++ = 0xff & value >> 8;
2248 /* Process fourth byte of a quadruplet. */
2251 return -1; /* premature EOF while decoding base64 */
2255 return -1; /* illegal char while decoding base64 */
2257 value |= BASE64_CHAR_TO_VALUE (c);
2258 *q++ = 0xff & value;
2261 #undef BASE64_CHAR_TO_VALUE
2263 return q - (char *) dest;
2269 /* Simple merge sort for use by stable_sort. Implementation courtesy
2270 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2273 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2274 int (*cmpfun) (const void *, const void *))
2276 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2280 size_t mid = (to + from) / 2;
2281 mergesort_internal (base, temp, size, from, mid, cmpfun);
2282 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2285 for (k = from; (i <= mid) && (j <= to); k++)
2286 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2287 memcpy (ELT (temp, k), ELT (base, i++), size);
2289 memcpy (ELT (temp, k), ELT (base, j++), size);
2291 memcpy (ELT (temp, k++), ELT (base, i++), size);
2293 memcpy (ELT (temp, k++), ELT (base, j++), size);
2294 for (k = from; k <= to; k++)
2295 memcpy (ELT (base, k), ELT (temp, k), size);
2300 /* Stable sort with interface exactly like standard library's qsort.
2301 Uses mergesort internally, allocating temporary storage with
2305 stable_sort (void *base, size_t nmemb, size_t size,
2306 int (*cmpfun) (const void *, const void *))
2310 void *temp = alloca (nmemb * size * sizeof (void *));
2311 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2315 /* Print a decimal number. If it is equal to or larger than ten, the
2316 number is rounded. Otherwise it is printed with one significant
2317 digit without trailing zeros and with no more than three fractional
2318 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2319 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2321 This is useful for displaying durations because it provides
2322 order-of-magnitude information without unnecessary clutter --
2323 long-running downloads are shown without the fractional part, and
2324 short ones still retain one significant digit. */
2327 print_decimal (double number)
2329 static char buf[32];
2330 double n = number >= 0 ? number : -number;
2333 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2334 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2335 snprintf (buf, sizeof buf, "%.0f", number);
2337 snprintf (buf, sizeof buf, "%.1f", number);
2338 else if (n >= 0.001)
2339 snprintf (buf, sizeof buf, "%.1g", number);
2340 else if (n >= 0.0005)
2341 /* round [0.0005, 0.001) to 0.001 */
2342 snprintf (buf, sizeof buf, "%.3f", number);
2344 /* print numbers close to 0 as 0, not 0.000 */
2361 { "/somedir", "/somedir", true },
2362 { "/somedir", "/somedir/d2", true },
2363 { "/somedir/d1", "/somedir", false },
2366 for (i = 0; i < countof(test_array); ++i)
2368 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2370 mu_assert ("test_subdir_p: wrong result",
2371 res == test_array[i].result);
2378 test_dir_matches_p()
2386 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2387 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2388 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2389 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2390 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2391 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2392 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2393 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2394 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2395 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2396 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2397 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2398 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2399 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2400 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2403 for (i = 0; i < countof(test_array); ++i)
2405 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2407 mu_assert ("test_dir_matches_p: wrong result",
2408 res == test_array[i].result);
2414 #endif /* TESTING */