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>
42 # include <sys/mman.h>
45 # include <process.h> /* getpid() */
50 #ifdef HAVE_SYS_UTIME_H
51 # include <sys/utime.h>
59 /* For TIOCGWINSZ and friends: */
60 #ifdef HAVE_SYS_IOCTL_H
61 # include <sys/ioctl.h>
67 /* Needed for Unix version of run_with_timeout. */
71 #ifndef HAVE_SIGSETJMP
72 /* If sigsetjmp is a macro, configure won't pick it up. */
74 # define HAVE_SIGSETJMP
78 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
79 # define USE_SIGNAL_TIMEOUT
87 #endif /* def __VMS */
94 memfatal (const char *context, long attempted_size)
96 /* Make sure we don't try to store part of the log line, and thus
98 log_set_save_context (false);
100 /* We have different log outputs in different situations:
101 1) output without bytes information
102 2) output with bytes information */
103 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
105 logprintf (LOG_ALWAYS,
106 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
111 logprintf (LOG_ALWAYS,
112 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
113 exec_name, context, attempted_size);
119 /* Character property table for (re-)escaping VMS ODS5 extended file
120 names. Note that this table ignores Unicode.
122 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
124 ODS5 Invalid characters:
125 C0 control codes (0x00 to 0x1F inclusive)
129 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
130 Double quotation marks (")
133 Left angle bracket (<)
134 Right angle bracket (>)
138 Characters escaped by "^":
139 SP ! " # % & ' ( ) + , . : ; =
142 Either "^_" or "^ " is accepted as a space. Period (.) is a special
143 case. Note that un-escaped < and > can also confuse a directory
146 Characters put out as ^xx:
148 80-9F (C1 control characters)
149 A0 (nonbreaking space)
150 FF (Latin small letter y diaeresis)
153 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
155 Property table values:
165 unsigned char char_prop[ 256] = {
167 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
168 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
170 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
171 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
173 /* SP ! " # $ % & ' ( ) * + , - . / */
174 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
176 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
177 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
179 /* @ A B C D E F G H I J K L M N O */
180 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
182 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
183 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
185 /* ` a b c d e f g h i j k l m n o */
186 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
188 /* p q r s t u v w x y z { | } ~ DEL */
189 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
191 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
192 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
193 8, 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, 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, 8
201 /* Utility function: like xstrdup(), but also lowercases S. */
204 xstrdup_lower (const char *s)
206 char *copy = xstrdup (s);
213 /* Copy the string formed by two pointers (one on the beginning, other
214 on the char after the last char) to a new, malloc-ed location.
217 strdupdelim (const char *beg, const char *end)
219 char *res = xmalloc (end - beg + 1);
220 memcpy (res, beg, end - beg);
221 res[end - beg] = '\0';
225 /* Parse a string containing comma-separated elements, and return a
226 vector of char pointers with the elements. Spaces following the
227 commas are ignored. */
229 sepstring (const char *s)
243 res = xrealloc (res, (i + 2) * sizeof (char *));
244 res[i] = strdupdelim (p, s);
247 /* Skip the blanks following the ','. */
248 while (c_isspace (*s))
255 res = xrealloc (res, (i + 2) * sizeof (char *));
256 res[i] = strdupdelim (p, s);
261 /* Like sprintf, but prints into a string of sufficient size freshly
262 allocated with malloc, which is returned. If unable to print due
263 to invalid format, returns NULL. Inability to allocate needed
264 memory results in abort, as with xmalloc. This is in spirit
265 similar to the GNU/BSD extension asprintf, but somewhat easier to
268 Internally the function either calls vasprintf or loops around
269 vsnprintf until the correct size is found. Since Wget also ships a
270 fallback implementation of vsnprintf, this should be portable. */
272 /* Constant is using for limits memory allocation for text buffer.
273 Applicable in situation when: vasprintf is not available in the system
274 and vsnprintf return -1 when long line is truncated (in old versions of
275 glibc and in other system where C99 doesn`t support) */
277 #define FMT_MAX_LENGTH 1048576
280 aprintf (const char *fmt, ...)
282 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
287 va_start (args, fmt);
288 ret = vasprintf (&str, fmt, args);
290 if (ret < 0 && errno == ENOMEM)
291 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
292 with xmalloc/xrealloc */
296 #else /* not HAVE_VASPRINTF */
298 /* vasprintf is unavailable. snprintf into a small buffer and
299 resize it as necessary. */
301 char *str = xmalloc (size);
303 /* #### This code will infloop and eventually abort in xrealloc if
304 passed a FMT that causes snprintf to consistently return -1. */
311 va_start (args, fmt);
312 n = vsnprintf (str, size, fmt, args);
315 /* If the printing worked, return the string. */
316 if (n > -1 && n < size)
319 /* Else try again with a larger buffer. */
320 if (n > -1) /* C99 */
321 size = n + 1; /* precisely what is needed */
322 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
323 { /* maybe we have some wrong
325 logprintf (LOG_ALWAYS,
326 _("%s: aprintf: text buffer is too big (%ld bytes), "
328 exec_name, size); /* printout a log message */
329 abort (); /* and abort... */
333 /* else, we continue to grow our
334 * buffer: Twice the old size. */
337 str = xrealloc (str, size);
339 #endif /* not HAVE_VASPRINTF */
342 /* Concatenate the NULL-terminated list of string arguments into
343 freshly allocated space. */
346 concat_strings (const char *str0, ...)
349 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
352 const char *next_str;
353 int total_length = 0;
356 /* Calculate the length of and allocate the resulting string. */
359 va_start (args, str0);
360 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
362 int len = strlen (next_str);
363 if (argcount < countof (saved_lengths))
364 saved_lengths[argcount++] = len;
368 p = ret = xmalloc (total_length + 1);
370 /* Copy the strings into the allocated space. */
373 va_start (args, str0);
374 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
377 if (argcount < countof (saved_lengths))
378 len = saved_lengths[argcount++];
380 len = strlen (next_str);
381 memcpy (p, next_str, len);
390 /* Format the provided time according to the specified format. The
391 format is a string with format elements supported by strftime. */
394 fmttime (time_t t, const char *fmt)
396 static char output[32];
397 struct tm *tm = localtime(&t);
400 if (!strftime(output, sizeof(output), fmt, tm))
405 /* Return pointer to a static char[] buffer in which zero-terminated
406 string-representation of TM (in form hh:mm:ss) is printed.
408 If TM is NULL, the current time will be used. */
413 return fmttime(t, "%H:%M:%S");
416 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
419 datetime_str (time_t t)
421 return fmttime(t, "%Y-%m-%d %H:%M:%S");
424 /* The Windows versions of the following two functions are defined in
425 mswindows.c. On MSDOS this function should never be called. */
430 fork_to_background (void)
435 #else /* def __VMS */
437 #if !defined(WINDOWS) && !defined(MSDOS)
439 fork_to_background (void)
442 /* Whether we arrange our own version of opt.lfilename here. */
443 bool logfile_changed = false;
445 if (!opt.lfilename && (!opt.quiet || opt.server_response))
447 /* We must create the file immediately to avoid either a race
448 condition (which arises from using unique_name and failing to
449 use fopen_excl) or lying to the user about the log file name
450 (which arises from using unique_name, printing the name, and
451 using fopen_excl later on.) */
452 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
455 logfile_changed = true;
468 /* parent, no error */
469 printf (_("Continuing in background, pid %d.\n"), (int) pid);
471 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
472 exit (0); /* #### should we use _exit()? */
475 /* child: give up the privileges and keep running. */
477 freopen ("/dev/null", "r", stdin);
478 freopen ("/dev/null", "w", stdout);
479 freopen ("/dev/null", "w", stderr);
481 #endif /* !WINDOWS && !MSDOS */
483 #endif /* def __VMS [else] */
486 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
487 specified with TM. The atime ("access time") is set to the current
491 touch (const char *file, time_t tm)
493 #ifdef HAVE_STRUCT_UTIMBUF
494 struct utimbuf times;
502 times.actime = time (NULL);
503 if (utime (file, ×) == -1)
504 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
507 /* Checks if FILE is a symbolic link, and removes it if it is. Does
508 nothing under MS-Windows. */
510 remove_link (const char *file)
515 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
517 DEBUGP (("Unlinking %s (symlink).\n", file));
520 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
521 quote (file), strerror (errno));
526 /* Does FILENAME exist? This is quite a lousy implementation, since
527 it supplies no error codes -- only a yes-or-no answer. Thus it
528 will return that a file does not exist if, e.g., the directory is
529 unreadable. I don't mind it too much currently, though. The
530 proper way should, of course, be to have a third, error state,
531 other than true/false, but that would introduce uncalled-for
532 additional complexity to the callers. */
534 file_exists_p (const char *filename)
537 return access (filename, F_OK) >= 0;
540 return stat (filename, &buf) >= 0;
544 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
545 Returns 0 on error. */
547 file_non_directory_p (const char *path)
550 /* Use lstat() rather than stat() so that symbolic links pointing to
551 directories can be identified correctly. */
552 if (lstat (path, &buf) != 0)
554 return S_ISDIR (buf.st_mode) ? false : true;
557 /* Return the size of file named by FILENAME, or -1 if it cannot be
558 opened or seeked into. */
560 file_size (const char *filename)
562 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
564 /* We use fseek rather than stat to determine the file size because
565 that way we can also verify that the file is readable without
566 explicitly checking for permissions. Inspired by the POST patch
568 FILE *fp = fopen (filename, "rb");
571 fseeko (fp, 0, SEEK_END);
577 if (stat (filename, &st) < 0)
584 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
585 original name. With the VMS file systems' versioning, everything
586 should be fine, and appending ".NN" just causes trouble.
591 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
592 doesn't exist is found. Return a freshly allocated copy of the
596 unique_name_1 (const char *prefix)
599 int plen = strlen (prefix);
600 char *template = (char *)alloca (plen + 1 + 24);
601 char *template_tail = template + plen;
603 memcpy (template, prefix, plen);
604 *template_tail++ = UNIQ_SEP;
607 number_to_string (template_tail, count++);
608 while (file_exists_p (template));
610 return xstrdup (template);
613 /* Return a unique file name, based on FILE.
615 More precisely, if FILE doesn't exist, it is returned unmodified.
616 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
617 file name that doesn't exist is returned.
619 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
621 The resulting file is not created, only verified that it didn't
622 exist at the point in time when the function was called.
623 Therefore, where security matters, don't rely that the file created
624 by this function exists until you open it with O_EXCL or
627 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
628 string. Otherwise, it may return FILE if the file doesn't exist
629 (and therefore doesn't need changing). */
632 unique_name (const char *file, bool allow_passthrough)
634 /* If the FILE itself doesn't exist, return it without
636 if (!file_exists_p (file))
637 return allow_passthrough ? (char *)file : xstrdup (file);
639 /* Otherwise, find a numeric suffix that results in unused file name
641 return unique_name_1 (file);
644 #else /* def UNIQ_SEP */
646 /* Dummy unique_name() for VMS. Return the original name as easily as
650 unique_name (const char *file, bool allow_passthrough)
652 /* Return the FILE itself, without modification, irregardful. */
653 return allow_passthrough ? (char *)file : xstrdup (file);
656 #endif /* def UNIQ_SEP [else] */
658 /* Create a file based on NAME, except without overwriting an existing
659 file with that name. Providing O_EXCL is correctly implemented,
660 this function does not have the race condition associated with
661 opening the file returned by unique_name. */
664 unique_create (const char *name, bool binary, char **opened_name)
666 /* unique file name, based on NAME */
667 char *uname = unique_name (name, false);
669 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
672 uname = unique_name (name, false);
674 if (opened_name && fp != NULL)
677 *opened_name = uname;
689 /* Open the file for writing, with the addition that the file is
690 opened "exclusively". This means that, if the file already exists,
691 this function will *fail* and errno will be set to EEXIST. If
692 BINARY is set, the file will be opened in binary mode, equivalent
695 If opening the file fails for any reason, including the file having
696 previously existed, this function returns NULL and sets errno
700 fopen_excl (const char *fname, int binary)
706 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
707 It also has file versions which obviate all the O_EXCL effort.
708 O_TRUNC (something of a misnomer) requests a new version.
711 /* Common open() optional arguments:
712 sequential access only, access callback function.
714 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
717 int flags = O_WRONLY | O_CREAT | O_TRUNC;
722 fd = open( fname, /* File name. */
724 0777, /* Mode for default protection. */
725 "ctx=bin,stm", /* Binary, stream access. */
726 "rfm=stmlf", /* Stream_LF. */
727 OPEN_OPT_ARGS); /* Access callback. */
732 fd = open( fname, /* File name. */
734 0777, /* Mode for default protection. */
735 "ctx=bin,stm", /* Binary, stream access. */
736 "rfm=fix", /* Fixed-length, */
737 "mrs=512", /* 512-byte records. */
738 OPEN_OPT_ARGS); /* Access callback. */
743 fd = open( fname, /* File name. */
745 0777, /* Mode for default protection.
747 "rfm=stmlf", /* Stream_LF. */
748 OPEN_OPT_ARGS); /* Access callback. */
750 # else /* def __VMS */
751 int flags = O_WRONLY | O_CREAT | O_EXCL;
756 fd = open (fname, flags, 0666);
757 # endif /* def __VMS [else] */
761 return fdopen (fd, binary ? "wb" : "w");
762 #else /* not O_EXCL */
763 /* Manually check whether the file exists. This is prone to race
764 conditions, but systems without O_EXCL haven't deserved
766 if (file_exists_p (fname))
771 return fopen (fname, binary ? "wb" : "w");
772 #endif /* not O_EXCL */
775 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
776 are missing, create them first. In case any mkdir() call fails,
777 return its error status. Returns 0 on successful completion.
779 The behaviour of this function should be identical to the behaviour
780 of `mkdir -p' on systems where mkdir supports the `-p' option. */
782 make_directory (const char *directory)
784 int i, ret, quit = 0;
787 /* Make a copy of dir, to be able to write to it. Otherwise, the
788 function is unsafe if called with a read-only char *argument. */
789 STRDUP_ALLOCA (dir, directory);
791 /* If the first character of dir is '/', skip it (and thus enable
792 creation of absolute-pathname directories. */
793 for (i = (*dir == '/'); 1; ++i)
795 for (; dir[i] && dir[i] != '/'; i++)
800 /* Check whether the directory already exists. Allow creation of
801 of intermediate directories to fail, as the initial path components
802 are not necessarily directories! */
803 if (!file_exists_p (dir))
804 ret = mkdir (dir, 0777);
815 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
816 should be a file name.
818 file_merge("/foo/bar", "baz") => "/foo/baz"
819 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
820 file_merge("foo", "bar") => "bar"
822 In other words, it's a simpler and gentler version of uri_merge. */
825 file_merge (const char *base, const char *file)
828 const char *cut = (const char *)strrchr (base, '/');
831 return xstrdup (file);
833 result = xmalloc (cut - base + 1 + strlen (file) + 1);
834 memcpy (result, base, cut - base);
835 result[cut - base] = '/';
836 strcpy (result + (cut - base) + 1, file);
841 /* Like fnmatch, but performs a case-insensitive match. */
844 fnmatch_nocase (const char *pattern, const char *string, int flags)
847 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
848 also present on *BSD platforms, and possibly elsewhere. */
849 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
851 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
852 char *patcopy = (char *) alloca (strlen (pattern) + 1);
853 char *strcopy = (char *) alloca (strlen (string) + 1);
855 for (p = patcopy; *pattern; pattern++, p++)
856 *p = c_tolower (*pattern);
858 for (p = strcopy; *string; string++, p++)
859 *p = c_tolower (*string);
861 return fnmatch (patcopy, strcopy, flags);
865 static bool in_acclist (const char *const *, const char *, bool);
867 /* Determine whether a file is acceptable to be followed, according to
868 lists of patterns to accept/reject. */
870 acceptable (const char *s)
874 while (l && s[l] != '/')
881 return (in_acclist ((const char *const *)opt.accepts, s, true)
882 && !in_acclist ((const char *const *)opt.rejects, s, true));
884 return in_acclist ((const char *const *)opt.accepts, s, true);
886 else if (opt.rejects)
887 return !in_acclist ((const char *const *)opt.rejects, s, true);
891 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
892 will return true if and only if D2 begins with `/something/' or is exactly
895 subdir_p (const char *d1, const char *d2)
899 if (!opt.ignore_case)
900 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
903 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
906 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
909 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
910 first element that matches DIR, through wildcards or front comparison (as
913 dir_matches_p (char **dirlist, const char *dir)
916 int (*matcher) (const char *, const char *, int)
917 = opt.ignore_case ? fnmatch_nocase : fnmatch;
919 for (x = dirlist; *x; x++)
921 /* Remove leading '/' */
922 char *p = *x + (**x == '/');
923 if (has_wildcards_p (p))
925 if (matcher (p, dir, FNM_PATHNAME) == 0)
930 if (subdir_p (p, dir))
935 return *x ? true : false;
938 /* Returns whether DIRECTORY is acceptable for download, wrt the
939 include/exclude lists.
941 The leading `/' is ignored in paths; relative and absolute paths
942 may be freely intermixed. */
945 accdir (const char *directory)
947 /* Remove starting '/'. */
948 if (*directory == '/')
952 if (!dir_matches_p (opt.includes, directory))
957 if (dir_matches_p (opt.excludes, directory))
963 /* Return true if STRING ends with TAIL. For instance:
965 match_tail ("abc", "bc", false) -> 1
966 match_tail ("abc", "ab", false) -> 0
967 match_tail ("abc", "abc", false) -> 1
969 If FOLD_CASE is true, the comparison will be case-insensitive. */
972 match_tail (const char *string, const char *tail, bool fold_case)
976 /* We want this to be fast, so we code two loops, one with
977 case-folding, one without. */
981 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
982 if (string[i] != tail[j])
987 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
988 if (c_tolower (string[i]) != c_tolower (tail[j]))
992 /* If the tail was exhausted, the match was succesful. */
999 /* Checks whether string S matches each element of ACCEPTS. A list
1000 element are matched either with fnmatch() or match_tail(),
1001 according to whether the element contains wildcards or not.
1003 If the BACKWARD is false, don't do backward comparison -- just compare
1006 in_acclist (const char *const *accepts, const char *s, bool backward)
1008 for (; *accepts; accepts++)
1010 if (has_wildcards_p (*accepts))
1012 int res = opt.ignore_case
1013 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1014 /* fnmatch returns 0 if the pattern *does* match the string. */
1022 if (match_tail (s, *accepts, opt.ignore_case))
1027 int cmp = opt.ignore_case
1028 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1037 /* Return the location of STR's suffix (file extension). Examples:
1038 suffix ("foo.bar") -> "bar"
1039 suffix ("foo.bar.baz") -> "baz"
1040 suffix ("/foo/bar") -> NULL
1041 suffix ("/foo.bar/baz") -> NULL */
1043 suffix (const char *str)
1047 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1050 if (str[i++] == '.')
1051 return (char *)str + i;
1056 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1060 has_wildcards_p (const char *s)
1063 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1068 /* Return true if FNAME ends with a typical HTML suffix. The
1069 following (case-insensitive) suffixes are presumed to be HTML
1074 ?html (`?' matches one character)
1076 #### CAVEAT. This is not necessarily a good indication that FNAME
1077 refers to a file that contains HTML! */
1079 has_html_suffix_p (const char *fname)
1083 if ((suf = suffix (fname)) == NULL)
1085 if (!strcasecmp (suf, "html"))
1087 if (!strcasecmp (suf, "htm"))
1089 if (suf[0] && !strcasecmp (suf + 1, "html"))
1094 /* Read a line from FP and return the pointer to freshly allocated
1095 storage. The storage space is obtained through malloc() and should
1096 be freed with free() when it is no longer needed.
1098 The length of the line is not limited, except by available memory.
1099 The newline character at the end of line is retained. The line is
1100 terminated with a zero character.
1102 After end-of-file is encountered without anything being read, NULL
1103 is returned. NULL is also returned on error. To distinguish
1104 between these two cases, use the stdio function ferror(). */
1107 read_whole_line (FILE *fp)
1111 char *line = xmalloc (bufsize);
1113 while (fgets (line + length, bufsize - length, fp))
1115 length += strlen (line + length);
1117 /* Possible for example when reading from a binary file where
1118 a line begins with \0. */
1121 if (line[length - 1] == '\n')
1124 /* fgets() guarantees to read the whole line, or to use up the
1125 space we've given it. We can double the buffer
1128 line = xrealloc (line, bufsize);
1130 if (length == 0 || ferror (fp))
1135 if (length + 1 < bufsize)
1136 /* Relieve the memory from our exponential greediness. We say
1137 `length + 1' because the terminating \0 is not included in
1138 LENGTH. We don't need to zero-terminate the string ourselves,
1139 though, because fgets() does that. */
1140 line = xrealloc (line, length + 1);
1144 /* Read FILE into memory. A pointer to `struct file_memory' are
1145 returned; use struct element `content' to access file contents, and
1146 the element `length' to know the file length. `content' is *not*
1147 zero-terminated, and you should *not* read or write beyond the [0,
1148 length) range of characters.
1150 After you are done with the file contents, call wget_read_file_free to
1153 Depending on the operating system and the type of file that is
1154 being read, wget_read_file() either mmap's the file into memory, or
1155 reads the file into the core using read().
1157 If file is named "-", fileno(stdin) is used for reading instead.
1158 If you want to read from a real file named "-", use "./-" instead. */
1160 struct file_memory *
1161 wget_read_file (const char *file)
1164 struct file_memory *fm;
1166 bool inhibit_close = false;
1168 /* Some magic in the finest tradition of Perl and its kin: if FILE
1169 is "-", just use stdin. */
1172 fd = fileno (stdin);
1173 inhibit_close = true;
1174 /* Note that we don't inhibit mmap() in this case. If stdin is
1175 redirected from a regular file, mmap() will still work. */
1178 fd = open (file, O_RDONLY);
1181 fm = xnew (struct file_memory);
1186 if (fstat (fd, &buf) < 0)
1188 fm->length = buf.st_size;
1189 /* NOTE: As far as I know, the callers of this function never
1190 modify the file text. Relying on this would enable us to
1191 specify PROT_READ and MAP_SHARED for a marginal gain in
1192 efficiency, but at some cost to generality. */
1193 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1194 MAP_PRIVATE, fd, 0);
1195 if (fm->content == (char *)MAP_FAILED)
1205 /* The most common reason why mmap() fails is that FD does not point
1206 to a plain file. However, it's also possible that mmap() doesn't
1207 work for a particular type of file. Therefore, whenever mmap()
1208 fails, we just fall back to the regular method. */
1209 #endif /* HAVE_MMAP */
1212 size = 512; /* number of bytes fm->contents can
1213 hold at any given time. */
1214 fm->content = xmalloc (size);
1218 if (fm->length > size / 2)
1220 /* #### I'm not sure whether the whole exponential-growth
1221 thing makes sense with kernel read. On Linux at least,
1222 read() refuses to read more than 4K from a file at a
1223 single chunk anyway. But other Unixes might optimize it
1224 better, and it doesn't *hurt* anything, so I'm leaving
1227 /* Normally, we grow SIZE exponentially to make the number
1228 of calls to read() and realloc() logarithmic in relation
1229 to file size. However, read() can read an amount of data
1230 smaller than requested, and it would be unreasonable to
1231 double SIZE every time *something* was read. Therefore,
1232 we double SIZE only when the length exceeds half of the
1233 entire allocated size. */
1235 fm->content = xrealloc (fm->content, size);
1237 nread = read (fd, fm->content + fm->length, size - fm->length);
1239 /* Successful read. */
1240 fm->length += nread;
1250 if (size > fm->length && fm->length != 0)
1251 /* Due to exponential growth of fm->content, the allocated region
1252 might be much larger than what is actually needed. */
1253 fm->content = xrealloc (fm->content, fm->length);
1260 xfree (fm->content);
1265 /* Release the resources held by FM. Specifically, this calls
1266 munmap() or xfree() on fm->content, depending whether mmap or
1267 malloc/read were used to read in the file. It also frees the
1268 memory needed to hold the FM structure itself. */
1271 wget_read_file_free (struct file_memory *fm)
1276 munmap (fm->content, fm->length);
1281 xfree (fm->content);
1286 /* Free the pointers in a NULL-terminated vector of pointers, then
1287 free the pointer itself. */
1289 free_vec (char **vec)
1300 /* Append vector V2 to vector V1. The function frees V2 and
1301 reallocates V1 (thus you may not use the contents of neither
1302 pointer after the call). If V1 is NULL, V2 is returned. */
1304 merge_vecs (char **v1, char **v2)
1314 /* To avoid j == 0 */
1319 for (i = 0; v1[i]; i++)
1322 for (j = 0; v2[j]; j++)
1324 /* Reallocate v1. */
1325 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1326 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1331 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1332 is allocated as needed. Return the new value of the vector. */
1335 vec_append (char **vec, const char *str)
1337 int cnt; /* count of vector elements, including
1338 the one we're about to append */
1341 for (cnt = 0; vec[cnt]; cnt++)
1347 /* Reallocate the array to fit the new element and the NULL. */
1348 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1349 /* Append a copy of STR to the vector. */
1350 vec[cnt - 1] = xstrdup (str);
1355 /* Sometimes it's useful to create "sets" of strings, i.e. special
1356 hash tables where you want to store strings as keys and merely
1357 query for their existence. Here is a set of utility routines that
1358 makes that transparent. */
1361 string_set_add (struct hash_table *ht, const char *s)
1363 /* First check whether the set element already exists. If it does,
1364 do nothing so that we don't have to free() the old element and
1365 then strdup() a new one. */
1366 if (hash_table_contains (ht, s))
1369 /* We use "1" as value. It provides us a useful and clear arbitrary
1370 value, and it consumes no memory -- the pointers to the same
1371 string "1" will be shared by all the key-value pairs in all `set'
1373 hash_table_put (ht, xstrdup (s), "1");
1376 /* Synonym for hash_table_contains... */
1379 string_set_contains (struct hash_table *ht, const char *s)
1381 return hash_table_contains (ht, s);
1384 /* Convert the specified string set to array. ARRAY should be large
1385 enough to hold hash_table_count(ht) char pointers. */
1387 void string_set_to_array (struct hash_table *ht, char **array)
1389 hash_table_iterator iter;
1390 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1391 *array++ = iter.key;
1394 /* Free the string set. This frees both the storage allocated for
1395 keys and the actual hash table. (hash_table_destroy would only
1396 destroy the hash table.) */
1399 string_set_free (struct hash_table *ht)
1401 hash_table_iterator iter;
1402 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1404 hash_table_destroy (ht);
1407 /* Utility function: simply call xfree() on all keys and values of HT. */
1410 free_keys_and_values (struct hash_table *ht)
1412 hash_table_iterator iter;
1413 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1420 /* Get digit grouping data for thousand separors by calling
1421 localeconv(). The data includes separator string and grouping info
1422 and is cached after the first call to the function.
1424 In locales that don't set a thousand separator (such as the "C"
1425 locale), this forces it to be ",". We are now only showing
1426 thousand separators in one place, so this shouldn't be a problem in
1430 get_grouping_data (const char **sep, const char **grouping)
1432 static const char *cached_sep;
1433 static const char *cached_grouping;
1434 static bool initialized;
1437 /* Get the grouping info from the locale. */
1438 struct lconv *lconv = localeconv ();
1439 cached_sep = lconv->thousands_sep;
1440 cached_grouping = lconv->grouping;
1441 #if ! USE_NLS_PROGRESS_BAR
1442 /* We can't count column widths, so ensure that the separator
1443 * is single-byte only (let check below determine what byte). */
1444 if (strlen(cached_sep) > 1)
1449 /* Many locales (such as "C" or "hr_HR") don't specify
1450 grouping, which we still want to use it for legibility.
1451 In those locales set the sep char to ',', unless that
1452 character is used for decimal point, in which case set it
1454 if (*lconv->decimal_point != ',')
1458 cached_grouping = "\x03";
1463 *grouping = cached_grouping;
1466 /* Return a printed representation of N with thousand separators.
1467 This should respect locale settings, with the exception of the "C"
1468 locale which mandates no separator, but we use one anyway.
1470 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1471 the separators because it's too non-portable, and it's hard to test
1472 for this feature at configure time. Besides, it wouldn't display
1473 separators in the "C" locale, still used by many Unix users. */
1476 with_thousand_seps (wgint n)
1478 static char outbuf[48];
1479 char *p = outbuf + sizeof outbuf;
1481 /* Info received from locale */
1482 const char *grouping, *sep;
1485 /* State information */
1486 int i = 0, groupsize;
1487 const char *atgroup;
1489 bool negative = n < 0;
1491 /* Initialize grouping data. */
1492 get_grouping_data (&sep, &grouping);
1493 seplen = strlen (sep);
1495 groupsize = *atgroup++;
1497 /* This would overflow on WGINT_MIN, but printing negative numbers
1498 is not an important goal of this fuinction. */
1502 /* Write the number into the buffer, backwards, inserting the
1503 separators as necessary. */
1507 *--p = n % 10 + '0';
1511 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1512 if (++i == groupsize)
1517 memcpy (p -= seplen, sep, seplen);
1520 groupsize = *atgroup++;
1529 /* N, a byte quantity, is converted to a human-readable abberviated
1530 form a la sizes printed by `ls -lh'. The result is written to a
1531 static buffer, a pointer to which is returned.
1533 Unlike `with_thousand_seps', this approximates to the nearest unit.
1534 Quoting GNU libit: "Most people visually process strings of 3-4
1535 digits effectively, but longer strings of digits are more prone to
1536 misinterpretation. Hence, converting to an abbreviated form
1537 usually improves readability."
1539 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1540 original computer-related meaning of "powers of 1024". We don't
1541 use the "*bibyte" names invented in 1998, and seldom used in
1542 practice. Wikipedia's entry on "binary prefix" discusses this in
1546 human_readable (HR_NUMTYPE n)
1548 /* These suffixes are compatible with those of GNU `ls -lh'. */
1549 static char powers[] =
1551 'K', /* kilobyte, 2^10 bytes */
1552 'M', /* megabyte, 2^20 bytes */
1553 'G', /* gigabyte, 2^30 bytes */
1554 'T', /* terabyte, 2^40 bytes */
1555 'P', /* petabyte, 2^50 bytes */
1556 'E', /* exabyte, 2^60 bytes */
1561 /* If the quantity is smaller than 1K, just print it. */
1564 snprintf (buf, sizeof (buf), "%d", (int) n);
1568 /* Loop over powers, dividing N with 1024 in each iteration. This
1569 works unchanged for all sizes of wgint, while still avoiding
1570 non-portable `long double' arithmetic. */
1571 for (i = 0; i < countof (powers); i++)
1573 /* At each iteration N is greater than the *subsequent* power.
1574 That way N/1024.0 produces a decimal number in the units of
1576 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1578 double val = n / 1024.0;
1579 /* Print values smaller than 10 with one decimal digits, and
1580 others without any decimals. */
1581 snprintf (buf, sizeof (buf), "%.*f%c",
1582 val < 10 ? 1 : 0, val, powers[i]);
1587 return NULL; /* unreached */
1590 /* Count the digits in the provided number. Used to allocate space
1591 when printing numbers. */
1594 numdigit (wgint number)
1598 ++cnt; /* accomodate '-' */
1599 while ((number /= 10) != 0)
1604 #define PR(mask) *p++ = n / (mask) + '0'
1606 /* DIGITS_<D> is used to print a D-digit number and should be called
1607 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1608 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1609 Recursively this continues until DIGITS_1 is invoked. */
1611 #define DIGITS_1(mask) PR (mask)
1612 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1613 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1614 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1615 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1616 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1617 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1618 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1619 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1620 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1622 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1624 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1625 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1626 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1627 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1628 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1629 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1630 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1631 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1632 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1634 /* Shorthand for casting to wgint. */
1637 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1638 `sprintf(buffer, "%lld", (long long) number)', only typically much
1639 faster and portable to machines without long long.
1641 The speedup may make a difference in programs that frequently
1642 convert numbers to strings. Some implementations of sprintf,
1643 particularly the one in some versions of GNU libc, have been known
1644 to be quite slow when converting integers to strings.
1646 Return the pointer to the location where the terminating zero was
1647 printed. (Equivalent to calling buffer+strlen(buffer) after the
1650 BUFFER should be large enough to accept as many bytes as you expect
1651 the number to take up. On machines with 64-bit wgints the maximum
1652 needed size is 24 bytes. That includes the digits needed for the
1653 largest 64-bit number, the `-' sign in case it's negative, and the
1654 terminating '\0'. */
1657 number_to_string (char *buffer, wgint number)
1662 int last_digit_char = 0;
1664 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1665 /* We are running in a very strange environment. Leave the correct
1666 printing to sprintf. */
1667 p += sprintf (buf, "%j", (intmax_t) (n));
1668 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1674 /* n = -n would overflow because -n would evaluate to a
1675 wgint value larger than WGINT_MAX. Need to make n
1676 smaller and handle the last digit separately. */
1677 int last_digit = n % 10;
1678 /* The sign of n%10 is implementation-defined. */
1680 last_digit_char = '0' - last_digit;
1682 last_digit_char = '0' + last_digit;
1683 /* After n is made smaller, -n will not overflow. */
1691 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1692 way printing any N is fully open-coded without a loop or jump.
1693 (Also see description of DIGITS_*.) */
1695 if (n < 10) DIGITS_1 (1);
1696 else if (n < 100) DIGITS_2 (10);
1697 else if (n < 1000) DIGITS_3 (100);
1698 else if (n < 10000) DIGITS_4 (1000);
1699 else if (n < 100000) DIGITS_5 (10000);
1700 else if (n < 1000000) DIGITS_6 (100000);
1701 else if (n < 10000000) DIGITS_7 (1000000);
1702 else if (n < 100000000) DIGITS_8 (10000000);
1703 else if (n < 1000000000) DIGITS_9 (100000000);
1704 #if SIZEOF_WGINT == 4
1705 /* wgint is 32 bits wide: no number has more than 10 digits. */
1706 else DIGITS_10 (1000000000);
1708 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1709 Constants are constructed by compile-time multiplication to avoid
1710 dealing with different notations for 64-bit constants
1711 (nL/nLL/nI64, depending on the compiler and architecture). */
1712 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1713 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1714 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1715 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1716 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1717 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1718 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1719 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1720 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1721 else DIGITS_19 (1000000000*(W)1000000000);
1724 if (last_digit_char)
1725 *p++ = last_digit_char;
1728 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1735 #undef SPRINTF_WGINT
1758 /* Print NUMBER to a statically allocated string and return a pointer
1759 to the printed representation.
1761 This function is intended to be used in conjunction with printf.
1762 It is hard to portably print wgint values:
1763 a) you cannot use printf("%ld", number) because wgint can be long
1764 long on 32-bit machines with LFS.
1765 b) you cannot use printf("%lld", number) because NUMBER could be
1766 long on 32-bit machines without LFS, or on 64-bit machines,
1767 which do not require LFS. Also, Windows doesn't support %lld.
1768 c) you cannot use printf("%j", (int_max_t) number) because not all
1769 versions of printf support "%j", the most notable being the one
1771 d) you cannot #define WGINT_FMT to the appropriate format and use
1772 printf(WGINT_FMT, number) because that would break translations
1773 for user-visible messages, such as printf("Downloaded: %d
1776 What you should use instead is printf("%s", number_to_static_string
1779 CAVEAT: since the function returns pointers to static data, you
1780 must be careful to copy its result before calling it again.
1781 However, to make it more useful with printf, the function maintains
1782 an internal ring of static buffers to return. That way things like
1783 printf("%s %s", number_to_static_string (num1),
1784 number_to_static_string (num2)) work as expected. Three buffers
1785 are currently used, which means that "%s %s %s" will work, but "%s
1786 %s %s %s" won't. If you need to print more than three wgints,
1787 bump the RING_SIZE (or rethink your message.) */
1790 number_to_static_string (wgint number)
1792 static char ring[RING_SIZE][24];
1794 char *buf = ring[ringpos];
1795 number_to_string (buf, number);
1796 ringpos = (ringpos + 1) % RING_SIZE;
1800 /* Determine the width of the terminal we're running on. If that's
1801 not possible, return 0. */
1804 determine_screen_width (void)
1806 /* If there's a way to get the terminal size using POSIX
1807 tcgetattr(), somebody please tell me. */
1812 if (opt.lfilename != NULL)
1815 fd = fileno (stderr);
1816 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1817 return 0; /* most likely ENOTTY */
1820 #elif defined(WINDOWS)
1821 CONSOLE_SCREEN_BUFFER_INFO csbi;
1822 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1824 return csbi.dwSize.X;
1825 #else /* neither TIOCGWINSZ nor WINDOWS */
1827 #endif /* neither TIOCGWINSZ nor WINDOWS */
1830 /* Whether the rnd system (either rand or [dl]rand48) has been
1832 static int rnd_seeded;
1834 /* Return a random number between 0 and MAX-1, inclusive.
1836 If the system does not support lrand48 and MAX is greater than the
1837 value of RAND_MAX+1 on the system, the returned value will be in
1838 the range [0, RAND_MAX]. This may be fixed in a future release.
1839 The random number generator is seeded automatically the first time
1842 This uses lrand48 where available, rand elsewhere. DO NOT use it
1843 for cryptography. It is only meant to be used in situations where
1844 quality of the random numbers returned doesn't really matter. */
1847 random_number (int max)
1852 srand48 ((long) time (NULL) ^ (long) getpid ());
1855 return lrand48 () % max;
1856 #else /* not HAVE_DRAND48 */
1862 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1867 /* Like rand() % max, but uses the high-order bits for better
1868 randomness on architectures where rand() is implemented using a
1869 simple congruential generator. */
1871 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1872 return (int) bounded;
1874 #endif /* not HAVE_DRAND48 */
1877 /* Return a random uniformly distributed floating point number in the
1878 [0, 1) range. Uses drand48 where available, and a really lame
1879 kludge elsewhere. */
1887 srand48 ((long) time (NULL) ^ (long) getpid ());
1891 #else /* not HAVE_DRAND48 */
1892 return ( random_number (10000) / 10000.0
1893 + random_number (10000) / (10000.0 * 10000.0)
1894 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1895 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1896 #endif /* not HAVE_DRAND48 */
1899 /* Implementation of run_with_timeout, a generic timeout-forcing
1900 routine for systems with Unix-like signal handling. */
1902 #ifdef USE_SIGNAL_TIMEOUT
1903 # ifdef HAVE_SIGSETJMP
1904 # define SETJMP(env) sigsetjmp (env, 1)
1906 static sigjmp_buf run_with_timeout_env;
1909 abort_run_with_timeout (int sig)
1911 assert (sig == SIGALRM);
1912 siglongjmp (run_with_timeout_env, -1);
1914 # else /* not HAVE_SIGSETJMP */
1915 # define SETJMP(env) setjmp (env)
1917 static jmp_buf run_with_timeout_env;
1920 abort_run_with_timeout (int sig)
1922 assert (sig == SIGALRM);
1923 /* We don't have siglongjmp to preserve the set of blocked signals;
1924 if we longjumped out of the handler at this point, SIGALRM would
1925 remain blocked. We must unblock it manually. */
1926 int mask = siggetmask ();
1927 mask &= ~sigmask (SIGALRM);
1930 /* Now it's safe to longjump. */
1931 longjmp (run_with_timeout_env, -1);
1933 # endif /* not HAVE_SIGSETJMP */
1935 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1936 setitimer where available, alarm otherwise.
1938 TIMEOUT should be non-zero. If the timeout value is so small that
1939 it would be rounded to zero, it is rounded to the least legal value
1940 instead (1us for setitimer, 1s for alarm). That ensures that
1941 SIGALRM will be delivered in all cases. */
1944 alarm_set (double timeout)
1947 /* Use the modern itimer interface. */
1948 struct itimerval itv;
1950 itv.it_value.tv_sec = (long) timeout;
1951 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1952 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1953 /* Ensure that we wait for at least the minimum interval.
1954 Specifying zero would mean "wait forever". */
1955 itv.it_value.tv_usec = 1;
1956 setitimer (ITIMER_REAL, &itv, NULL);
1957 #else /* not ITIMER_REAL */
1958 /* Use the old alarm() interface. */
1959 int secs = (int) timeout;
1961 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1962 because alarm(0) means "never deliver the alarm", i.e. "wait
1963 forever", which is not what someone who specifies a 0.5s
1964 timeout would expect. */
1967 #endif /* not ITIMER_REAL */
1970 /* Cancel the alarm set with alarm_set. */
1976 struct itimerval disable;
1978 setitimer (ITIMER_REAL, &disable, NULL);
1979 #else /* not ITIMER_REAL */
1981 #endif /* not ITIMER_REAL */
1984 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1985 seconds. Returns true if the function was interrupted with a
1986 timeout, false otherwise.
1988 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1989 using setitimer() or alarm(). The timeout is enforced by
1990 longjumping out of the SIGALRM handler. This has several
1991 advantages compared to the traditional approach of relying on
1992 signals causing system calls to exit with EINTR:
1994 * The callback function is *forcibly* interrupted after the
1995 timeout expires, (almost) regardless of what it was doing and
1996 whether it was in a syscall. For example, a calculation that
1997 takes a long time is interrupted as reliably as an IO
2000 * It works with both SYSV and BSD signals because it doesn't
2001 depend on the default setting of SA_RESTART.
2003 * It doesn't require special handler setup beyond a simple call
2004 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2007 The only downside is that, if FUN allocates internal resources that
2008 are normally freed prior to exit from the functions, they will be
2009 lost in case of timeout. */
2012 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2022 signal (SIGALRM, abort_run_with_timeout);
2023 if (SETJMP (run_with_timeout_env) != 0)
2025 /* Longjumped out of FUN with a timeout. */
2026 signal (SIGALRM, SIG_DFL);
2029 alarm_set (timeout);
2032 /* Preserve errno in case alarm() or signal() modifies it. */
2033 saved_errno = errno;
2035 signal (SIGALRM, SIG_DFL);
2036 errno = saved_errno;
2041 #else /* not USE_SIGNAL_TIMEOUT */
2044 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2045 define it under Windows, because Windows has its own version of
2046 run_with_timeout that uses threads. */
2049 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2054 #endif /* not WINDOWS */
2055 #endif /* not USE_SIGNAL_TIMEOUT */
2059 /* Sleep the specified amount of seconds. On machines without
2060 nanosleep(), this may sleep shorter if interrupted by signals. */
2063 xsleep (double seconds)
2065 #ifdef HAVE_NANOSLEEP
2066 /* nanosleep is the preferred interface because it offers high
2067 accuracy and, more importantly, because it allows us to reliably
2068 restart receiving a signal such as SIGWINCH. (There was an
2069 actual Debian bug report about --limit-rate malfunctioning while
2070 the terminal was being resized.) */
2071 struct timespec sleep, remaining;
2072 sleep.tv_sec = (long) seconds;
2073 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2074 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2075 /* If nanosleep has been interrupted by a signal, adjust the
2076 sleeping period and return to sleep. */
2078 #elif defined(HAVE_USLEEP)
2079 /* If usleep is available, use it in preference to select. */
2082 /* On some systems, usleep cannot handle values larger than
2083 1,000,000. If the period is larger than that, use sleep
2084 first, then add usleep for subsecond accuracy. */
2086 seconds -= (long) seconds;
2088 usleep (seconds * 1000000);
2089 #else /* fall back select */
2090 /* Note that, although Windows supports select, it can't be used to
2091 implement sleeping because Winsock's select doesn't implement
2092 timeout when it is passed NULL pointers for all fd sets. (But it
2093 does under Cygwin, which implements Unix-compatible select.) */
2094 struct timeval sleep;
2095 sleep.tv_sec = (long) seconds;
2096 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2097 select (0, NULL, NULL, NULL, &sleep);
2098 /* If select returns -1 and errno is EINTR, it means we were
2099 interrupted by a signal. But without knowing how long we've
2100 actually slept, we can't return to sleep. Using gettimeofday to
2101 track sleeps is slow and unreliable due to clock skew. */
2105 #endif /* not WINDOWS */
2107 /* Encode the octets in DATA of length LENGTH to base64 format,
2108 storing the result to DEST. The output will be zero-terminated,
2109 and must point to a writable buffer of at least
2110 1+BASE64_LENGTH(length) bytes. The function returns the length of
2111 the resulting base64 data, not counting the terminating zero.
2113 This implementation does not emit newlines after 76 characters of
2117 base64_encode (const void *data, int length, char *dest)
2119 /* Conversion table. */
2120 static const char tbl[64] = {
2121 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2122 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2123 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2124 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2126 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2127 don't work for data with MSB set. */
2128 const unsigned char *s = data;
2129 /* Theoretical ANSI violation when length < 3. */
2130 const unsigned char *end = (const unsigned char *) data + length - 2;
2133 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2134 for (; s < end; s += 3)
2136 *p++ = tbl[s[0] >> 2];
2137 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2138 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2139 *p++ = tbl[s[2] & 0x3f];
2142 /* Pad the result if necessary... */
2146 *p++ = tbl[s[0] >> 2];
2147 *p++ = tbl[(s[0] & 3) << 4];
2152 *p++ = tbl[s[0] >> 2];
2153 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2154 *p++ = tbl[((s[1] & 0xf) << 2)];
2158 /* ...and zero-terminate it. */
2164 /* Store in C the next non-whitespace character from the string, or \0
2165 when end of string is reached. */
2166 #define NEXT_CHAR(c, p) do { \
2167 c = (unsigned char) *p++; \
2168 } while (c_isspace (c))
2170 #define IS_ASCII(c) (((c) & 0x80) == 0)
2172 /* Decode data from BASE64 (a null-terminated string) into memory
2173 pointed to by DEST. DEST is assumed to be large enough to
2174 accomodate the decoded data, which is guaranteed to be no more than
2177 Since DEST is assumed to contain binary data, it is not
2178 NUL-terminated. The function returns the length of the data
2179 written to TO. -1 is returned in case of error caused by malformed
2182 This function originates from Free Recode. */
2185 base64_decode (const char *base64, void *dest)
2187 /* Table of base64 values for first 128 characters. Note that this
2188 assumes ASCII (but so does Wget in other places). */
2189 static const signed char base64_char_to_value[128] =
2191 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2192 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2193 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2194 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2195 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2196 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2197 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2198 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2199 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2200 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2201 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2202 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2203 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2205 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2206 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2208 const char *p = base64;
2214 unsigned long value;
2216 /* Process first byte of a quadruplet. */
2220 if (c == '=' || !IS_BASE64 (c))
2221 return -1; /* illegal char while decoding base64 */
2222 value = BASE64_CHAR_TO_VALUE (c) << 18;
2224 /* Process second byte of a quadruplet. */
2227 return -1; /* premature EOF while decoding base64 */
2228 if (c == '=' || !IS_BASE64 (c))
2229 return -1; /* illegal char while decoding base64 */
2230 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2233 /* Process third byte of a quadruplet. */
2236 return -1; /* premature EOF while decoding base64 */
2238 return -1; /* illegal char while decoding base64 */
2244 return -1; /* premature EOF while decoding base64 */
2246 return -1; /* padding `=' expected but not found */
2250 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2251 *q++ = 0xff & value >> 8;
2253 /* Process fourth byte of a quadruplet. */
2256 return -1; /* premature EOF while decoding base64 */
2260 return -1; /* illegal char while decoding base64 */
2262 value |= BASE64_CHAR_TO_VALUE (c);
2263 *q++ = 0xff & value;
2266 #undef BASE64_CHAR_TO_VALUE
2268 return q - (char *) dest;
2274 /* Simple merge sort for use by stable_sort. Implementation courtesy
2275 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2278 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2279 int (*cmpfun) (const void *, const void *))
2281 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2285 size_t mid = (to + from) / 2;
2286 mergesort_internal (base, temp, size, from, mid, cmpfun);
2287 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2290 for (k = from; (i <= mid) && (j <= to); k++)
2291 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2292 memcpy (ELT (temp, k), ELT (base, i++), size);
2294 memcpy (ELT (temp, k), ELT (base, j++), size);
2296 memcpy (ELT (temp, k++), ELT (base, i++), size);
2298 memcpy (ELT (temp, k++), ELT (base, j++), size);
2299 for (k = from; k <= to; k++)
2300 memcpy (ELT (base, k), ELT (temp, k), size);
2305 /* Stable sort with interface exactly like standard library's qsort.
2306 Uses mergesort internally, allocating temporary storage with
2310 stable_sort (void *base, size_t nmemb, size_t size,
2311 int (*cmpfun) (const void *, const void *))
2315 void *temp = alloca (nmemb * size * sizeof (void *));
2316 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2320 /* Print a decimal number. If it is equal to or larger than ten, the
2321 number is rounded. Otherwise it is printed with one significant
2322 digit without trailing zeros and with no more than three fractional
2323 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2324 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2326 This is useful for displaying durations because it provides
2327 order-of-magnitude information without unnecessary clutter --
2328 long-running downloads are shown without the fractional part, and
2329 short ones still retain one significant digit. */
2332 print_decimal (double number)
2334 static char buf[32];
2335 double n = number >= 0 ? number : -number;
2338 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2339 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2340 snprintf (buf, sizeof buf, "%.0f", number);
2342 snprintf (buf, sizeof buf, "%.1f", number);
2343 else if (n >= 0.001)
2344 snprintf (buf, sizeof buf, "%.1g", number);
2345 else if (n >= 0.0005)
2346 /* round [0.0005, 0.001) to 0.001 */
2347 snprintf (buf, sizeof buf, "%.3f", number);
2349 /* print numbers close to 0 as 0, not 0.000 */
2366 { "/somedir", "/somedir", true },
2367 { "/somedir", "/somedir/d2", true },
2368 { "/somedir/d1", "/somedir", false },
2371 for (i = 0; i < countof(test_array); ++i)
2373 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2375 mu_assert ("test_subdir_p: wrong result",
2376 res == test_array[i].result);
2383 test_dir_matches_p()
2391 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2392 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2393 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2394 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2395 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2396 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2397 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2398 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2399 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2400 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2401 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2402 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2403 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2404 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2405 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2408 for (i = 0; i < countof(test_array); ++i)
2410 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2412 mu_assert ("test_dir_matches_p: wrong result",
2413 res == test_array[i].result);
2419 #endif /* TESTING */