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() */
52 # include <sys/types.h>
57 # ifdef HAVE_SYS_UTIME_H
58 # include <sys/utime.h>
64 /* For TIOCGWINSZ and friends: */
65 #ifdef HAVE_SYS_IOCTL_H
66 # include <sys/ioctl.h>
72 /* Needed for Unix version of run_with_timeout. */
76 #ifndef HAVE_SIGSETJMP
77 /* If sigsetjmp is a macro, configure won't pick it up. */
79 # define HAVE_SIGSETJMP
83 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
84 # define USE_SIGNAL_TIMEOUT
92 #endif /* def __VMS */
99 memfatal (const char *context, long attempted_size)
101 /* Make sure we don't try to store part of the log line, and thus
103 log_set_save_context (false);
105 /* We have different log outputs in different situations:
106 1) output without bytes information
107 2) output with bytes information */
108 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
110 logprintf (LOG_ALWAYS,
111 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
116 logprintf (LOG_ALWAYS,
117 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
118 exec_name, context, attempted_size);
124 /* Character property table for (re-)escaping VMS ODS5 extended file
125 names. Note that this table ignores Unicode.
127 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
129 ODS5 Invalid characters:
130 C0 control codes (0x00 to 0x1F inclusive)
134 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
135 Double quotation marks (")
138 Left angle bracket (<)
139 Right angle bracket (>)
143 Characters escaped by "^":
144 SP ! " # % & ' ( ) + , . : ; =
147 Either "^_" or "^ " is accepted as a space. Period (.) is a special
148 case. Note that un-escaped < and > can also confuse a directory
151 Characters put out as ^xx:
153 80-9F (C1 control characters)
154 A0 (nonbreaking space)
155 FF (Latin small letter y diaeresis)
158 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
160 Property table values:
170 unsigned char char_prop[ 256] = {
172 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
173 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
175 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
176 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
178 /* SP ! " # $ % & ' ( ) * + , - . / */
179 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
181 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
182 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
184 /* @ A B C D E F G H I J K L M N O */
185 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
187 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
188 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
190 /* ` a b c d e f g h i j k l m n o */
191 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
193 /* p q r s t u v w x y z { | } ~ DEL */
194 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
196 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
197 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
198 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
199 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
200 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
201 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
202 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
203 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
206 /* Utility function: like xstrdup(), but also lowercases S. */
209 xstrdup_lower (const char *s)
211 char *copy = xstrdup (s);
218 /* Copy the string formed by two pointers (one on the beginning, other
219 on the char after the last char) to a new, malloc-ed location.
222 strdupdelim (const char *beg, const char *end)
224 char *res = xmalloc (end - beg + 1);
225 memcpy (res, beg, end - beg);
226 res[end - beg] = '\0';
230 /* Parse a string containing comma-separated elements, and return a
231 vector of char pointers with the elements. Spaces following the
232 commas are ignored. */
234 sepstring (const char *s)
248 res = xrealloc (res, (i + 2) * sizeof (char *));
249 res[i] = strdupdelim (p, s);
252 /* Skip the blanks following the ','. */
253 while (c_isspace (*s))
260 res = xrealloc (res, (i + 2) * sizeof (char *));
261 res[i] = strdupdelim (p, s);
266 /* Like sprintf, but prints into a string of sufficient size freshly
267 allocated with malloc, which is returned. If unable to print due
268 to invalid format, returns NULL. Inability to allocate needed
269 memory results in abort, as with xmalloc. This is in spirit
270 similar to the GNU/BSD extension asprintf, but somewhat easier to
273 Internally the function either calls vasprintf or loops around
274 vsnprintf until the correct size is found. Since Wget also ships a
275 fallback implementation of vsnprintf, this should be portable. */
277 /* Constant is using for limits memory allocation for text buffer.
278 Applicable in situation when: vasprintf is not available in the system
279 and vsnprintf return -1 when long line is truncated (in old versions of
280 glibc and in other system where C99 doesn`t support) */
282 #define FMT_MAX_LENGTH 1048576
285 aprintf (const char *fmt, ...)
287 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
292 va_start (args, fmt);
293 ret = vasprintf (&str, fmt, args);
295 if (ret < 0 && errno == ENOMEM)
296 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
297 with xmalloc/xrealloc */
301 #else /* not HAVE_VASPRINTF */
303 /* vasprintf is unavailable. snprintf into a small buffer and
304 resize it as necessary. */
306 char *str = xmalloc (size);
308 /* #### This code will infloop and eventually abort in xrealloc if
309 passed a FMT that causes snprintf to consistently return -1. */
316 va_start (args, fmt);
317 n = vsnprintf (str, size, fmt, args);
320 /* If the printing worked, return the string. */
321 if (n > -1 && n < size)
324 /* Else try again with a larger buffer. */
325 if (n > -1) /* C99 */
326 size = n + 1; /* precisely what is needed */
327 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
328 { /* maybe we have some wrong
330 logprintf (LOG_ALWAYS,
331 _("%s: aprintf: text buffer is too big (%ld bytes), "
333 exec_name, size); /* printout a log message */
334 abort (); /* and abort... */
338 /* else, we continue to grow our
339 * buffer: Twice the old size. */
342 str = xrealloc (str, size);
344 #endif /* not HAVE_VASPRINTF */
347 /* Concatenate the NULL-terminated list of string arguments into
348 freshly allocated space. */
351 concat_strings (const char *str0, ...)
354 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
357 const char *next_str;
358 int total_length = 0;
361 /* Calculate the length of and allocate the resulting string. */
364 va_start (args, str0);
365 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
367 int len = strlen (next_str);
368 if (argcount < countof (saved_lengths))
369 saved_lengths[argcount++] = len;
373 p = ret = xmalloc (total_length + 1);
375 /* Copy the strings into the allocated space. */
378 va_start (args, str0);
379 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
382 if (argcount < countof (saved_lengths))
383 len = saved_lengths[argcount++];
385 len = strlen (next_str);
386 memcpy (p, next_str, len);
395 /* Format the provided time according to the specified format. The
396 format is a string with format elements supported by strftime. */
399 fmttime (time_t t, const char *fmt)
401 static char output[32];
402 struct tm *tm = localtime(&t);
405 if (!strftime(output, sizeof(output), fmt, tm))
410 /* Return pointer to a static char[] buffer in which zero-terminated
411 string-representation of TM (in form hh:mm:ss) is printed.
413 If TM is NULL, the current time will be used. */
418 return fmttime(t, "%H:%M:%S");
421 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
424 datetime_str (time_t t)
426 return fmttime(t, "%Y-%m-%d %H:%M:%S");
429 /* The Windows versions of the following two functions are defined in
430 mswindows.c. On MSDOS this function should never be called. */
435 fork_to_background (void)
440 #else /* def __VMS */
442 #if !defined(WINDOWS) && !defined(MSDOS)
444 fork_to_background (void)
447 /* Whether we arrange our own version of opt.lfilename here. */
448 bool logfile_changed = false;
450 if (!opt.lfilename && (!opt.quiet || opt.server_response))
452 /* We must create the file immediately to avoid either a race
453 condition (which arises from using unique_name and failing to
454 use fopen_excl) or lying to the user about the log file name
455 (which arises from using unique_name, printing the name, and
456 using fopen_excl later on.) */
457 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
460 logfile_changed = true;
473 /* parent, no error */
474 printf (_("Continuing in background, pid %d.\n"), (int) pid);
476 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
477 exit (0); /* #### should we use _exit()? */
480 /* child: give up the privileges and keep running. */
482 freopen ("/dev/null", "r", stdin);
483 freopen ("/dev/null", "w", stdout);
484 freopen ("/dev/null", "w", stderr);
486 #endif /* !WINDOWS && !MSDOS */
488 #endif /* def __VMS [else] */
491 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
492 specified with TM. The atime ("access time") is set to the current
496 touch (const char *file, time_t tm)
499 # ifdef HAVE_STRUCT_UTIMBUF
500 struct utimbuf times;
508 times.actime = time (NULL);
509 if (utime (file, ×) == -1)
510 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
512 struct timespec timespecs[2];
515 fd = open (file, O_WRONLY);
518 logprintf (LOG_NOTQUIET, "open(%s): %s\n", file, strerror (errno));
522 timespecs[0].tv_sec = time (NULL);
523 timespecs[0].tv_nsec = 0L;
524 timespecs[1].tv_sec = tm;
525 timespecs[1].tv_nsec = 0L;
527 if (futimens (fd, timespecs) == -1)
528 logprintf (LOG_NOTQUIET, "futimens(%s): %s\n", file, strerror (errno));
534 /* Checks if FILE is a symbolic link, and removes it if it is. Does
535 nothing under MS-Windows. */
537 remove_link (const char *file)
542 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
544 DEBUGP (("Unlinking %s (symlink).\n", file));
547 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
548 quote (file), strerror (errno));
553 /* Does FILENAME exist? This is quite a lousy implementation, since
554 it supplies no error codes -- only a yes-or-no answer. Thus it
555 will return that a file does not exist if, e.g., the directory is
556 unreadable. I don't mind it too much currently, though. The
557 proper way should, of course, be to have a third, error state,
558 other than true/false, but that would introduce uncalled-for
559 additional complexity to the callers. */
561 file_exists_p (const char *filename)
564 return access (filename, F_OK) >= 0;
567 return stat (filename, &buf) >= 0;
571 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
572 Returns 0 on error. */
574 file_non_directory_p (const char *path)
577 /* Use lstat() rather than stat() so that symbolic links pointing to
578 directories can be identified correctly. */
579 if (lstat (path, &buf) != 0)
581 return S_ISDIR (buf.st_mode) ? false : true;
584 /* Return the size of file named by FILENAME, or -1 if it cannot be
585 opened or seeked into. */
587 file_size (const char *filename)
589 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
591 /* We use fseek rather than stat to determine the file size because
592 that way we can also verify that the file is readable without
593 explicitly checking for permissions. Inspired by the POST patch
595 FILE *fp = fopen (filename, "rb");
598 fseeko (fp, 0, SEEK_END);
604 if (stat (filename, &st) < 0)
611 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
612 original name. With the VMS file systems' versioning, everything
613 should be fine, and appending ".NN" just causes trouble.
618 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
619 doesn't exist is found. Return a freshly allocated copy of the
623 unique_name_1 (const char *prefix)
626 int plen = strlen (prefix);
627 char *template = (char *)alloca (plen + 1 + 24);
628 char *template_tail = template + plen;
630 memcpy (template, prefix, plen);
631 *template_tail++ = UNIQ_SEP;
634 number_to_string (template_tail, count++);
635 while (file_exists_p (template));
637 return xstrdup (template);
640 /* Return a unique file name, based on FILE.
642 More precisely, if FILE doesn't exist, it is returned unmodified.
643 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
644 file name that doesn't exist is returned.
646 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
648 The resulting file is not created, only verified that it didn't
649 exist at the point in time when the function was called.
650 Therefore, where security matters, don't rely that the file created
651 by this function exists until you open it with O_EXCL or
654 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
655 string. Otherwise, it may return FILE if the file doesn't exist
656 (and therefore doesn't need changing). */
659 unique_name (const char *file, bool allow_passthrough)
661 /* If the FILE itself doesn't exist, return it without
663 if (!file_exists_p (file))
664 return allow_passthrough ? (char *)file : xstrdup (file);
666 /* Otherwise, find a numeric suffix that results in unused file name
668 return unique_name_1 (file);
671 #else /* def UNIQ_SEP */
673 /* Dummy unique_name() for VMS. Return the original name as easily as
677 unique_name (const char *file, bool allow_passthrough)
679 /* Return the FILE itself, without modification, irregardful. */
680 return allow_passthrough ? (char *)file : xstrdup (file);
683 #endif /* def UNIQ_SEP [else] */
685 /* Create a file based on NAME, except without overwriting an existing
686 file with that name. Providing O_EXCL is correctly implemented,
687 this function does not have the race condition associated with
688 opening the file returned by unique_name. */
691 unique_create (const char *name, bool binary, char **opened_name)
693 /* unique file name, based on NAME */
694 char *uname = unique_name (name, false);
696 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
699 uname = unique_name (name, false);
701 if (opened_name && fp != NULL)
704 *opened_name = uname;
716 /* Open the file for writing, with the addition that the file is
717 opened "exclusively". This means that, if the file already exists,
718 this function will *fail* and errno will be set to EEXIST. If
719 BINARY is set, the file will be opened in binary mode, equivalent
722 If opening the file fails for any reason, including the file having
723 previously existed, this function returns NULL and sets errno
727 fopen_excl (const char *fname, int binary)
733 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
734 It also has file versions which obviate all the O_EXCL effort.
735 O_TRUNC (something of a misnomer) requests a new version.
738 /* Common open() optional arguments:
739 sequential access only, access callback function.
741 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
744 int flags = O_WRONLY | O_CREAT | O_TRUNC;
749 fd = open( fname, /* File name. */
751 0777, /* Mode for default protection. */
752 "ctx=bin,stm", /* Binary, stream access. */
753 "rfm=stmlf", /* Stream_LF. */
754 OPEN_OPT_ARGS); /* Access callback. */
759 fd = open( fname, /* File name. */
761 0777, /* Mode for default protection. */
762 "ctx=bin,stm", /* Binary, stream access. */
763 "rfm=fix", /* Fixed-length, */
764 "mrs=512", /* 512-byte records. */
765 OPEN_OPT_ARGS); /* Access callback. */
770 fd = open( fname, /* File name. */
772 0777, /* Mode for default protection. */
773 "rfm=stmlf", /* Stream_LF. */
774 OPEN_OPT_ARGS); /* Access callback. */
776 # else /* def __VMS */
777 int flags = O_WRONLY | O_CREAT | O_EXCL;
782 fd = open (fname, flags, 0666);
783 # endif /* def __VMS [else] */
787 return fdopen (fd, binary ? "wb" : "w");
788 #else /* not O_EXCL */
789 /* Manually check whether the file exists. This is prone to race
790 conditions, but systems without O_EXCL haven't deserved
792 if (file_exists_p (fname))
797 return fopen (fname, binary ? "wb" : "w");
798 #endif /* not O_EXCL */
801 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
802 are missing, create them first. In case any mkdir() call fails,
803 return its error status. Returns 0 on successful completion.
805 The behaviour of this function should be identical to the behaviour
806 of `mkdir -p' on systems where mkdir supports the `-p' option. */
808 make_directory (const char *directory)
810 int i, ret, quit = 0;
813 /* Make a copy of dir, to be able to write to it. Otherwise, the
814 function is unsafe if called with a read-only char *argument. */
815 STRDUP_ALLOCA (dir, directory);
817 /* If the first character of dir is '/', skip it (and thus enable
818 creation of absolute-pathname directories. */
819 for (i = (*dir == '/'); 1; ++i)
821 for (; dir[i] && dir[i] != '/'; i++)
826 /* Check whether the directory already exists. Allow creation of
827 of intermediate directories to fail, as the initial path components
828 are not necessarily directories! */
829 if (!file_exists_p (dir))
830 ret = mkdir (dir, 0777);
841 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
842 should be a file name.
844 file_merge("/foo/bar", "baz") => "/foo/baz"
845 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
846 file_merge("foo", "bar") => "bar"
848 In other words, it's a simpler and gentler version of uri_merge. */
851 file_merge (const char *base, const char *file)
854 const char *cut = (const char *)strrchr (base, '/');
857 return xstrdup (file);
859 result = xmalloc (cut - base + 1 + strlen (file) + 1);
860 memcpy (result, base, cut - base);
861 result[cut - base] = '/';
862 strcpy (result + (cut - base) + 1, file);
867 /* Like fnmatch, but performs a case-insensitive match. */
870 fnmatch_nocase (const char *pattern, const char *string, int flags)
873 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
874 also present on *BSD platforms, and possibly elsewhere. */
875 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
877 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
878 char *patcopy = (char *) alloca (strlen (pattern) + 1);
879 char *strcopy = (char *) alloca (strlen (string) + 1);
881 for (p = patcopy; *pattern; pattern++, p++)
882 *p = c_tolower (*pattern);
884 for (p = strcopy; *string; string++, p++)
885 *p = c_tolower (*string);
887 return fnmatch (patcopy, strcopy, flags);
891 static bool in_acclist (const char *const *, const char *, bool);
893 /* Determine whether a file is acceptable to be followed, according to
894 lists of patterns to accept/reject. */
896 acceptable (const char *s)
900 if (opt.output_document && strcmp (s, opt.output_document) == 0)
903 while (l && s[l] != '/')
910 return (in_acclist ((const char *const *)opt.accepts, s, true)
911 && !in_acclist ((const char *const *)opt.rejects, s, true));
913 return in_acclist ((const char *const *)opt.accepts, s, true);
915 else if (opt.rejects)
916 return !in_acclist ((const char *const *)opt.rejects, s, true);
920 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
921 will return true if and only if D2 begins with `/something/' or is exactly
924 subdir_p (const char *d1, const char *d2)
928 if (!opt.ignore_case)
929 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
932 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
935 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
938 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
939 first element that matches DIR, through wildcards or front comparison (as
942 dir_matches_p (char **dirlist, const char *dir)
945 int (*matcher) (const char *, const char *, int)
946 = opt.ignore_case ? fnmatch_nocase : fnmatch;
948 for (x = dirlist; *x; x++)
950 /* Remove leading '/' */
951 char *p = *x + (**x == '/');
952 if (has_wildcards_p (p))
954 if (matcher (p, dir, FNM_PATHNAME) == 0)
959 if (subdir_p (p, dir))
964 return *x ? true : false;
967 /* Returns whether DIRECTORY is acceptable for download, wrt the
968 include/exclude lists.
970 The leading `/' is ignored in paths; relative and absolute paths
971 may be freely intermixed. */
974 accdir (const char *directory)
976 /* Remove starting '/'. */
977 if (*directory == '/')
981 if (!dir_matches_p (opt.includes, directory))
986 if (dir_matches_p (opt.excludes, directory))
992 /* Return true if STRING ends with TAIL. For instance:
994 match_tail ("abc", "bc", false) -> 1
995 match_tail ("abc", "ab", false) -> 0
996 match_tail ("abc", "abc", false) -> 1
998 If FOLD_CASE is true, the comparison will be case-insensitive. */
1001 match_tail (const char *string, const char *tail, bool fold_case)
1005 /* We want this to be fast, so we code two loops, one with
1006 case-folding, one without. */
1010 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
1011 if (string[i] != tail[j])
1016 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
1017 if (c_tolower (string[i]) != c_tolower (tail[j]))
1021 /* If the tail was exhausted, the match was succesful. */
1028 /* Checks whether string S matches each element of ACCEPTS. A list
1029 element are matched either with fnmatch() or match_tail(),
1030 according to whether the element contains wildcards or not.
1032 If the BACKWARD is false, don't do backward comparison -- just compare
1035 in_acclist (const char *const *accepts, const char *s, bool backward)
1037 for (; *accepts; accepts++)
1039 if (has_wildcards_p (*accepts))
1041 int res = opt.ignore_case
1042 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1043 /* fnmatch returns 0 if the pattern *does* match the string. */
1051 if (match_tail (s, *accepts, opt.ignore_case))
1056 int cmp = opt.ignore_case
1057 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1066 /* Return the location of STR's suffix (file extension). Examples:
1067 suffix ("foo.bar") -> "bar"
1068 suffix ("foo.bar.baz") -> "baz"
1069 suffix ("/foo/bar") -> NULL
1070 suffix ("/foo.bar/baz") -> NULL */
1072 suffix (const char *str)
1076 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1079 if (str[i++] == '.')
1080 return (char *)str + i;
1085 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1089 has_wildcards_p (const char *s)
1092 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1097 /* Return true if FNAME ends with a typical HTML suffix. The
1098 following (case-insensitive) suffixes are presumed to be HTML
1103 ?html (`?' matches one character)
1105 #### CAVEAT. This is not necessarily a good indication that FNAME
1106 refers to a file that contains HTML! */
1108 has_html_suffix_p (const char *fname)
1112 if ((suf = suffix (fname)) == NULL)
1114 if (!strcasecmp (suf, "html"))
1116 if (!strcasecmp (suf, "htm"))
1118 if (suf[0] && !strcasecmp (suf + 1, "html"))
1123 /* Read a line from FP and return the pointer to freshly allocated
1124 storage. The storage space is obtained through malloc() and should
1125 be freed with free() when it is no longer needed.
1127 The length of the line is not limited, except by available memory.
1128 The newline character at the end of line is retained. The line is
1129 terminated with a zero character.
1131 After end-of-file is encountered without anything being read, NULL
1132 is returned. NULL is also returned on error. To distinguish
1133 between these two cases, use the stdio function ferror(). */
1136 read_whole_line (FILE *fp)
1140 char *line = xmalloc (bufsize);
1142 while (fgets (line + length, bufsize - length, fp))
1144 length += strlen (line + length);
1146 /* Possible for example when reading from a binary file where
1147 a line begins with \0. */
1150 if (line[length - 1] == '\n')
1153 /* fgets() guarantees to read the whole line, or to use up the
1154 space we've given it. We can double the buffer
1157 line = xrealloc (line, bufsize);
1159 if (length == 0 || ferror (fp))
1164 if (length + 1 < bufsize)
1165 /* Relieve the memory from our exponential greediness. We say
1166 `length + 1' because the terminating \0 is not included in
1167 LENGTH. We don't need to zero-terminate the string ourselves,
1168 though, because fgets() does that. */
1169 line = xrealloc (line, length + 1);
1173 /* Read FILE into memory. A pointer to `struct file_memory' are
1174 returned; use struct element `content' to access file contents, and
1175 the element `length' to know the file length. `content' is *not*
1176 zero-terminated, and you should *not* read or write beyond the [0,
1177 length) range of characters.
1179 After you are done with the file contents, call wget_read_file_free to
1182 Depending on the operating system and the type of file that is
1183 being read, wget_read_file() either mmap's the file into memory, or
1184 reads the file into the core using read().
1186 If file is named "-", fileno(stdin) is used for reading instead.
1187 If you want to read from a real file named "-", use "./-" instead. */
1189 struct file_memory *
1190 wget_read_file (const char *file)
1193 struct file_memory *fm;
1195 bool inhibit_close = false;
1197 /* Some magic in the finest tradition of Perl and its kin: if FILE
1198 is "-", just use stdin. */
1201 fd = fileno (stdin);
1202 inhibit_close = true;
1203 /* Note that we don't inhibit mmap() in this case. If stdin is
1204 redirected from a regular file, mmap() will still work. */
1207 fd = open (file, O_RDONLY);
1210 fm = xnew (struct file_memory);
1215 if (fstat (fd, &buf) < 0)
1217 fm->length = buf.st_size;
1218 /* NOTE: As far as I know, the callers of this function never
1219 modify the file text. Relying on this would enable us to
1220 specify PROT_READ and MAP_SHARED for a marginal gain in
1221 efficiency, but at some cost to generality. */
1222 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1223 MAP_PRIVATE, fd, 0);
1224 if (fm->content == (char *)MAP_FAILED)
1234 /* The most common reason why mmap() fails is that FD does not point
1235 to a plain file. However, it's also possible that mmap() doesn't
1236 work for a particular type of file. Therefore, whenever mmap()
1237 fails, we just fall back to the regular method. */
1238 #endif /* HAVE_MMAP */
1241 size = 512; /* number of bytes fm->contents can
1242 hold at any given time. */
1243 fm->content = xmalloc (size);
1247 if (fm->length > size / 2)
1249 /* #### I'm not sure whether the whole exponential-growth
1250 thing makes sense with kernel read. On Linux at least,
1251 read() refuses to read more than 4K from a file at a
1252 single chunk anyway. But other Unixes might optimize it
1253 better, and it doesn't *hurt* anything, so I'm leaving
1256 /* Normally, we grow SIZE exponentially to make the number
1257 of calls to read() and realloc() logarithmic in relation
1258 to file size. However, read() can read an amount of data
1259 smaller than requested, and it would be unreasonable to
1260 double SIZE every time *something* was read. Therefore,
1261 we double SIZE only when the length exceeds half of the
1262 entire allocated size. */
1264 fm->content = xrealloc (fm->content, size);
1266 nread = read (fd, fm->content + fm->length, size - fm->length);
1268 /* Successful read. */
1269 fm->length += nread;
1279 if (size > fm->length && fm->length != 0)
1280 /* Due to exponential growth of fm->content, the allocated region
1281 might be much larger than what is actually needed. */
1282 fm->content = xrealloc (fm->content, fm->length);
1289 xfree (fm->content);
1294 /* Release the resources held by FM. Specifically, this calls
1295 munmap() or xfree() on fm->content, depending whether mmap or
1296 malloc/read were used to read in the file. It also frees the
1297 memory needed to hold the FM structure itself. */
1300 wget_read_file_free (struct file_memory *fm)
1305 munmap (fm->content, fm->length);
1310 xfree (fm->content);
1315 /* Free the pointers in a NULL-terminated vector of pointers, then
1316 free the pointer itself. */
1318 free_vec (char **vec)
1329 /* Append vector V2 to vector V1. The function frees V2 and
1330 reallocates V1 (thus you may not use the contents of neither
1331 pointer after the call). If V1 is NULL, V2 is returned. */
1333 merge_vecs (char **v1, char **v2)
1343 /* To avoid j == 0 */
1348 for (i = 0; v1[i]; i++)
1351 for (j = 0; v2[j]; j++)
1353 /* Reallocate v1. */
1354 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1355 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1360 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1361 is allocated as needed. Return the new value of the vector. */
1364 vec_append (char **vec, const char *str)
1366 int cnt; /* count of vector elements, including
1367 the one we're about to append */
1370 for (cnt = 0; vec[cnt]; cnt++)
1376 /* Reallocate the array to fit the new element and the NULL. */
1377 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1378 /* Append a copy of STR to the vector. */
1379 vec[cnt - 1] = xstrdup (str);
1384 /* Sometimes it's useful to create "sets" of strings, i.e. special
1385 hash tables where you want to store strings as keys and merely
1386 query for their existence. Here is a set of utility routines that
1387 makes that transparent. */
1390 string_set_add (struct hash_table *ht, const char *s)
1392 /* First check whether the set element already exists. If it does,
1393 do nothing so that we don't have to free() the old element and
1394 then strdup() a new one. */
1395 if (hash_table_contains (ht, s))
1398 /* We use "1" as value. It provides us a useful and clear arbitrary
1399 value, and it consumes no memory -- the pointers to the same
1400 string "1" will be shared by all the key-value pairs in all `set'
1402 hash_table_put (ht, xstrdup (s), "1");
1405 /* Synonym for hash_table_contains... */
1408 string_set_contains (struct hash_table *ht, const char *s)
1410 return hash_table_contains (ht, s);
1413 /* Convert the specified string set to array. ARRAY should be large
1414 enough to hold hash_table_count(ht) char pointers. */
1416 void string_set_to_array (struct hash_table *ht, char **array)
1418 hash_table_iterator iter;
1419 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1420 *array++ = iter.key;
1423 /* Free the string set. This frees both the storage allocated for
1424 keys and the actual hash table. (hash_table_destroy would only
1425 destroy the hash table.) */
1428 string_set_free (struct hash_table *ht)
1430 hash_table_iterator iter;
1431 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1433 hash_table_destroy (ht);
1436 /* Utility function: simply call xfree() on all keys and values of HT. */
1439 free_keys_and_values (struct hash_table *ht)
1441 hash_table_iterator iter;
1442 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1449 /* Get digit grouping data for thousand separors by calling
1450 localeconv(). The data includes separator string and grouping info
1451 and is cached after the first call to the function.
1453 In locales that don't set a thousand separator (such as the "C"
1454 locale), this forces it to be ",". We are now only showing
1455 thousand separators in one place, so this shouldn't be a problem in
1459 get_grouping_data (const char **sep, const char **grouping)
1461 static const char *cached_sep;
1462 static const char *cached_grouping;
1463 static bool initialized;
1466 /* Get the grouping info from the locale. */
1467 struct lconv *lconv = localeconv ();
1468 cached_sep = lconv->thousands_sep;
1469 cached_grouping = lconv->grouping;
1470 #if ! USE_NLS_PROGRESS_BAR
1471 /* We can't count column widths, so ensure that the separator
1472 * is single-byte only (let check below determine what byte). */
1473 if (strlen(cached_sep) > 1)
1478 /* Many locales (such as "C" or "hr_HR") don't specify
1479 grouping, which we still want to use it for legibility.
1480 In those locales set the sep char to ',', unless that
1481 character is used for decimal point, in which case set it
1483 if (*lconv->decimal_point != ',')
1487 cached_grouping = "\x03";
1492 *grouping = cached_grouping;
1495 /* Return a printed representation of N with thousand separators.
1496 This should respect locale settings, with the exception of the "C"
1497 locale which mandates no separator, but we use one anyway.
1499 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1500 the separators because it's too non-portable, and it's hard to test
1501 for this feature at configure time. Besides, it wouldn't display
1502 separators in the "C" locale, still used by many Unix users. */
1505 with_thousand_seps (wgint n)
1507 static char outbuf[48];
1508 char *p = outbuf + sizeof outbuf;
1510 /* Info received from locale */
1511 const char *grouping, *sep;
1514 /* State information */
1515 int i = 0, groupsize;
1516 const char *atgroup;
1518 bool negative = n < 0;
1520 /* Initialize grouping data. */
1521 get_grouping_data (&sep, &grouping);
1522 seplen = strlen (sep);
1524 groupsize = *atgroup++;
1526 /* This would overflow on WGINT_MIN, but printing negative numbers
1527 is not an important goal of this fuinction. */
1531 /* Write the number into the buffer, backwards, inserting the
1532 separators as necessary. */
1536 *--p = n % 10 + '0';
1540 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1541 if (++i == groupsize)
1546 memcpy (p -= seplen, sep, seplen);
1549 groupsize = *atgroup++;
1558 /* N, a byte quantity, is converted to a human-readable abberviated
1559 form a la sizes printed by `ls -lh'. The result is written to a
1560 static buffer, a pointer to which is returned.
1562 Unlike `with_thousand_seps', this approximates to the nearest unit.
1563 Quoting GNU libit: "Most people visually process strings of 3-4
1564 digits effectively, but longer strings of digits are more prone to
1565 misinterpretation. Hence, converting to an abbreviated form
1566 usually improves readability."
1568 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1569 original computer-related meaning of "powers of 1024". We don't
1570 use the "*bibyte" names invented in 1998, and seldom used in
1571 practice. Wikipedia's entry on "binary prefix" discusses this in
1575 human_readable (HR_NUMTYPE n)
1577 /* These suffixes are compatible with those of GNU `ls -lh'. */
1578 static char powers[] =
1580 'K', /* kilobyte, 2^10 bytes */
1581 'M', /* megabyte, 2^20 bytes */
1582 'G', /* gigabyte, 2^30 bytes */
1583 'T', /* terabyte, 2^40 bytes */
1584 'P', /* petabyte, 2^50 bytes */
1585 'E', /* exabyte, 2^60 bytes */
1590 /* If the quantity is smaller than 1K, just print it. */
1593 snprintf (buf, sizeof (buf), "%d", (int) n);
1597 /* Loop over powers, dividing N with 1024 in each iteration. This
1598 works unchanged for all sizes of wgint, while still avoiding
1599 non-portable `long double' arithmetic. */
1600 for (i = 0; i < countof (powers); i++)
1602 /* At each iteration N is greater than the *subsequent* power.
1603 That way N/1024.0 produces a decimal number in the units of
1605 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1607 double val = n / 1024.0;
1608 /* Print values smaller than 10 with one decimal digits, and
1609 others without any decimals. */
1610 snprintf (buf, sizeof (buf), "%.*f%c",
1611 val < 10 ? 1 : 0, val, powers[i]);
1616 return NULL; /* unreached */
1619 /* Count the digits in the provided number. Used to allocate space
1620 when printing numbers. */
1623 numdigit (wgint number)
1627 ++cnt; /* accomodate '-' */
1628 while ((number /= 10) != 0)
1633 #define PR(mask) *p++ = n / (mask) + '0'
1635 /* DIGITS_<D> is used to print a D-digit number and should be called
1636 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1637 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1638 Recursively this continues until DIGITS_1 is invoked. */
1640 #define DIGITS_1(mask) PR (mask)
1641 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1642 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1643 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1644 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1645 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1646 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1647 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1648 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1649 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1651 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1653 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1654 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1655 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1656 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1657 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1658 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1659 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1660 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1661 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1663 /* Shorthand for casting to wgint. */
1666 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1667 `sprintf(buffer, "%lld", (long long) number)', only typically much
1668 faster and portable to machines without long long.
1670 The speedup may make a difference in programs that frequently
1671 convert numbers to strings. Some implementations of sprintf,
1672 particularly the one in some versions of GNU libc, have been known
1673 to be quite slow when converting integers to strings.
1675 Return the pointer to the location where the terminating zero was
1676 printed. (Equivalent to calling buffer+strlen(buffer) after the
1679 BUFFER should be large enough to accept as many bytes as you expect
1680 the number to take up. On machines with 64-bit wgints the maximum
1681 needed size is 24 bytes. That includes the digits needed for the
1682 largest 64-bit number, the `-' sign in case it's negative, and the
1683 terminating '\0'. */
1686 number_to_string (char *buffer, wgint number)
1691 int last_digit_char = 0;
1693 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1694 /* We are running in a very strange environment. Leave the correct
1695 printing to sprintf. */
1696 p += sprintf (buf, "%j", (intmax_t) (n));
1697 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1703 /* n = -n would overflow because -n would evaluate to a
1704 wgint value larger than WGINT_MAX. Need to make n
1705 smaller and handle the last digit separately. */
1706 int last_digit = n % 10;
1707 /* The sign of n%10 is implementation-defined. */
1709 last_digit_char = '0' - last_digit;
1711 last_digit_char = '0' + last_digit;
1712 /* After n is made smaller, -n will not overflow. */
1720 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1721 way printing any N is fully open-coded without a loop or jump.
1722 (Also see description of DIGITS_*.) */
1724 if (n < 10) DIGITS_1 (1);
1725 else if (n < 100) DIGITS_2 (10);
1726 else if (n < 1000) DIGITS_3 (100);
1727 else if (n < 10000) DIGITS_4 (1000);
1728 else if (n < 100000) DIGITS_5 (10000);
1729 else if (n < 1000000) DIGITS_6 (100000);
1730 else if (n < 10000000) DIGITS_7 (1000000);
1731 else if (n < 100000000) DIGITS_8 (10000000);
1732 else if (n < 1000000000) DIGITS_9 (100000000);
1733 #if SIZEOF_WGINT == 4
1734 /* wgint is 32 bits wide: no number has more than 10 digits. */
1735 else DIGITS_10 (1000000000);
1737 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1738 Constants are constructed by compile-time multiplication to avoid
1739 dealing with different notations for 64-bit constants
1740 (nL/nLL/nI64, depending on the compiler and architecture). */
1741 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1742 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1743 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1744 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1745 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1746 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1747 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1748 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1749 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1750 else DIGITS_19 (1000000000*(W)1000000000);
1753 if (last_digit_char)
1754 *p++ = last_digit_char;
1757 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1764 #undef SPRINTF_WGINT
1787 /* Print NUMBER to a statically allocated string and return a pointer
1788 to the printed representation.
1790 This function is intended to be used in conjunction with printf.
1791 It is hard to portably print wgint values:
1792 a) you cannot use printf("%ld", number) because wgint can be long
1793 long on 32-bit machines with LFS.
1794 b) you cannot use printf("%lld", number) because NUMBER could be
1795 long on 32-bit machines without LFS, or on 64-bit machines,
1796 which do not require LFS. Also, Windows doesn't support %lld.
1797 c) you cannot use printf("%j", (int_max_t) number) because not all
1798 versions of printf support "%j", the most notable being the one
1800 d) you cannot #define WGINT_FMT to the appropriate format and use
1801 printf(WGINT_FMT, number) because that would break translations
1802 for user-visible messages, such as printf("Downloaded: %d
1805 What you should use instead is printf("%s", number_to_static_string
1808 CAVEAT: since the function returns pointers to static data, you
1809 must be careful to copy its result before calling it again.
1810 However, to make it more useful with printf, the function maintains
1811 an internal ring of static buffers to return. That way things like
1812 printf("%s %s", number_to_static_string (num1),
1813 number_to_static_string (num2)) work as expected. Three buffers
1814 are currently used, which means that "%s %s %s" will work, but "%s
1815 %s %s %s" won't. If you need to print more than three wgints,
1816 bump the RING_SIZE (or rethink your message.) */
1819 number_to_static_string (wgint number)
1821 static char ring[RING_SIZE][24];
1823 char *buf = ring[ringpos];
1824 number_to_string (buf, number);
1825 ringpos = (ringpos + 1) % RING_SIZE;
1829 /* Determine the width of the terminal we're running on. If that's
1830 not possible, return 0. */
1833 determine_screen_width (void)
1835 /* If there's a way to get the terminal size using POSIX
1836 tcgetattr(), somebody please tell me. */
1841 if (opt.lfilename != NULL)
1844 fd = fileno (stderr);
1845 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1846 return 0; /* most likely ENOTTY */
1849 #elif defined(WINDOWS)
1850 CONSOLE_SCREEN_BUFFER_INFO csbi;
1851 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1853 return csbi.dwSize.X;
1854 #else /* neither TIOCGWINSZ nor WINDOWS */
1856 #endif /* neither TIOCGWINSZ nor WINDOWS */
1859 /* Whether the rnd system (either rand or [dl]rand48) has been
1861 static int rnd_seeded;
1863 /* Return a random number between 0 and MAX-1, inclusive.
1865 If the system does not support lrand48 and MAX is greater than the
1866 value of RAND_MAX+1 on the system, the returned value will be in
1867 the range [0, RAND_MAX]. This may be fixed in a future release.
1868 The random number generator is seeded automatically the first time
1871 This uses lrand48 where available, rand elsewhere. DO NOT use it
1872 for cryptography. It is only meant to be used in situations where
1873 quality of the random numbers returned doesn't really matter. */
1876 random_number (int max)
1881 srand48 ((long) time (NULL) ^ (long) getpid ());
1884 return lrand48 () % max;
1885 #else /* not HAVE_DRAND48 */
1891 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1896 /* Like rand() % max, but uses the high-order bits for better
1897 randomness on architectures where rand() is implemented using a
1898 simple congruential generator. */
1900 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1901 return (int) bounded;
1903 #endif /* not HAVE_DRAND48 */
1906 /* Return a random uniformly distributed floating point number in the
1907 [0, 1) range. Uses drand48 where available, and a really lame
1908 kludge elsewhere. */
1916 srand48 ((long) time (NULL) ^ (long) getpid ());
1920 #else /* not HAVE_DRAND48 */
1921 return ( random_number (10000) / 10000.0
1922 + random_number (10000) / (10000.0 * 10000.0)
1923 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1924 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1925 #endif /* not HAVE_DRAND48 */
1928 /* Implementation of run_with_timeout, a generic timeout-forcing
1929 routine for systems with Unix-like signal handling. */
1931 #ifdef USE_SIGNAL_TIMEOUT
1932 # ifdef HAVE_SIGSETJMP
1933 # define SETJMP(env) sigsetjmp (env, 1)
1935 static sigjmp_buf run_with_timeout_env;
1938 abort_run_with_timeout (int sig)
1940 assert (sig == SIGALRM);
1941 siglongjmp (run_with_timeout_env, -1);
1943 # else /* not HAVE_SIGSETJMP */
1944 # define SETJMP(env) setjmp (env)
1946 static jmp_buf run_with_timeout_env;
1949 abort_run_with_timeout (int sig)
1951 assert (sig == SIGALRM);
1952 /* We don't have siglongjmp to preserve the set of blocked signals;
1953 if we longjumped out of the handler at this point, SIGALRM would
1954 remain blocked. We must unblock it manually. */
1957 sigaddset (&set, SIGALRM);
1958 sigprocmask (SIG_BLOCK, &set, NULL);
1960 /* Now it's safe to longjump. */
1961 longjmp (run_with_timeout_env, -1);
1963 # endif /* not HAVE_SIGSETJMP */
1965 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1966 setitimer where available, alarm otherwise.
1968 TIMEOUT should be non-zero. If the timeout value is so small that
1969 it would be rounded to zero, it is rounded to the least legal value
1970 instead (1us for setitimer, 1s for alarm). That ensures that
1971 SIGALRM will be delivered in all cases. */
1974 alarm_set (double timeout)
1977 /* Use the modern itimer interface. */
1978 struct itimerval itv;
1980 itv.it_value.tv_sec = (long) timeout;
1981 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1982 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1983 /* Ensure that we wait for at least the minimum interval.
1984 Specifying zero would mean "wait forever". */
1985 itv.it_value.tv_usec = 1;
1986 setitimer (ITIMER_REAL, &itv, NULL);
1987 #else /* not ITIMER_REAL */
1988 /* Use the old alarm() interface. */
1989 int secs = (int) timeout;
1991 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1992 because alarm(0) means "never deliver the alarm", i.e. "wait
1993 forever", which is not what someone who specifies a 0.5s
1994 timeout would expect. */
1997 #endif /* not ITIMER_REAL */
2000 /* Cancel the alarm set with alarm_set. */
2006 struct itimerval disable;
2008 setitimer (ITIMER_REAL, &disable, NULL);
2009 #else /* not ITIMER_REAL */
2011 #endif /* not ITIMER_REAL */
2014 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2015 seconds. Returns true if the function was interrupted with a
2016 timeout, false otherwise.
2018 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2019 using setitimer() or alarm(). The timeout is enforced by
2020 longjumping out of the SIGALRM handler. This has several
2021 advantages compared to the traditional approach of relying on
2022 signals causing system calls to exit with EINTR:
2024 * The callback function is *forcibly* interrupted after the
2025 timeout expires, (almost) regardless of what it was doing and
2026 whether it was in a syscall. For example, a calculation that
2027 takes a long time is interrupted as reliably as an IO
2030 * It works with both SYSV and BSD signals because it doesn't
2031 depend on the default setting of SA_RESTART.
2033 * It doesn't require special handler setup beyond a simple call
2034 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2037 The only downside is that, if FUN allocates internal resources that
2038 are normally freed prior to exit from the functions, they will be
2039 lost in case of timeout. */
2042 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2052 signal (SIGALRM, abort_run_with_timeout);
2053 if (SETJMP (run_with_timeout_env) != 0)
2055 /* Longjumped out of FUN with a timeout. */
2056 signal (SIGALRM, SIG_DFL);
2059 alarm_set (timeout);
2062 /* Preserve errno in case alarm() or signal() modifies it. */
2063 saved_errno = errno;
2065 signal (SIGALRM, SIG_DFL);
2066 errno = saved_errno;
2071 #else /* not USE_SIGNAL_TIMEOUT */
2074 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2075 define it under Windows, because Windows has its own version of
2076 run_with_timeout that uses threads. */
2079 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2084 #endif /* not WINDOWS */
2085 #endif /* not USE_SIGNAL_TIMEOUT */
2089 /* Sleep the specified amount of seconds. On machines without
2090 nanosleep(), this may sleep shorter if interrupted by signals. */
2093 xsleep (double seconds)
2095 #ifdef HAVE_NANOSLEEP
2096 /* nanosleep is the preferred interface because it offers high
2097 accuracy and, more importantly, because it allows us to reliably
2098 restart receiving a signal such as SIGWINCH. (There was an
2099 actual Debian bug report about --limit-rate malfunctioning while
2100 the terminal was being resized.) */
2101 struct timespec sleep, remaining;
2102 sleep.tv_sec = (long) seconds;
2103 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2104 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2105 /* If nanosleep has been interrupted by a signal, adjust the
2106 sleeping period and return to sleep. */
2108 #elif defined(HAVE_USLEEP)
2109 /* If usleep is available, use it in preference to select. */
2112 /* On some systems, usleep cannot handle values larger than
2113 1,000,000. If the period is larger than that, use sleep
2114 first, then add usleep for subsecond accuracy. */
2116 seconds -= (long) seconds;
2118 usleep (seconds * 1000000);
2119 #else /* fall back select */
2120 /* Note that, although Windows supports select, it can't be used to
2121 implement sleeping because Winsock's select doesn't implement
2122 timeout when it is passed NULL pointers for all fd sets. (But it
2123 does under Cygwin, which implements Unix-compatible select.) */
2124 struct timeval sleep;
2125 sleep.tv_sec = (long) seconds;
2126 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2127 select (0, NULL, NULL, NULL, &sleep);
2128 /* If select returns -1 and errno is EINTR, it means we were
2129 interrupted by a signal. But without knowing how long we've
2130 actually slept, we can't return to sleep. Using gettimeofday to
2131 track sleeps is slow and unreliable due to clock skew. */
2135 #endif /* not WINDOWS */
2137 /* Encode the octets in DATA of length LENGTH to base64 format,
2138 storing the result to DEST. The output will be zero-terminated,
2139 and must point to a writable buffer of at least
2140 1+BASE64_LENGTH(length) bytes. The function returns the length of
2141 the resulting base64 data, not counting the terminating zero.
2143 This implementation does not emit newlines after 76 characters of
2147 base64_encode (const void *data, int length, char *dest)
2149 /* Conversion table. */
2150 static const char tbl[64] = {
2151 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2152 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2153 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2154 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2156 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2157 don't work for data with MSB set. */
2158 const unsigned char *s = data;
2159 /* Theoretical ANSI violation when length < 3. */
2160 const unsigned char *end = (const unsigned char *) data + length - 2;
2163 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2164 for (; s < end; s += 3)
2166 *p++ = tbl[s[0] >> 2];
2167 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2168 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2169 *p++ = tbl[s[2] & 0x3f];
2172 /* Pad the result if necessary... */
2176 *p++ = tbl[s[0] >> 2];
2177 *p++ = tbl[(s[0] & 3) << 4];
2182 *p++ = tbl[s[0] >> 2];
2183 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2184 *p++ = tbl[((s[1] & 0xf) << 2)];
2188 /* ...and zero-terminate it. */
2194 /* Store in C the next non-whitespace character from the string, or \0
2195 when end of string is reached. */
2196 #define NEXT_CHAR(c, p) do { \
2197 c = (unsigned char) *p++; \
2198 } while (c_isspace (c))
2200 #define IS_ASCII(c) (((c) & 0x80) == 0)
2202 /* Decode data from BASE64 (a null-terminated string) into memory
2203 pointed to by DEST. DEST is assumed to be large enough to
2204 accomodate the decoded data, which is guaranteed to be no more than
2207 Since DEST is assumed to contain binary data, it is not
2208 NUL-terminated. The function returns the length of the data
2209 written to TO. -1 is returned in case of error caused by malformed
2212 This function originates from Free Recode. */
2215 base64_decode (const char *base64, void *dest)
2217 /* Table of base64 values for first 128 characters. Note that this
2218 assumes ASCII (but so does Wget in other places). */
2219 static const signed char base64_char_to_value[128] =
2221 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2222 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2223 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2224 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2225 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2226 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2227 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2228 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2229 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2230 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2231 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2232 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2233 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2235 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2236 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2238 const char *p = base64;
2244 unsigned long value;
2246 /* Process first byte of a quadruplet. */
2250 if (c == '=' || !IS_BASE64 (c))
2251 return -1; /* illegal char while decoding base64 */
2252 value = BASE64_CHAR_TO_VALUE (c) << 18;
2254 /* Process second byte of a quadruplet. */
2257 return -1; /* premature EOF while decoding base64 */
2258 if (c == '=' || !IS_BASE64 (c))
2259 return -1; /* illegal char while decoding base64 */
2260 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2263 /* Process third byte of a quadruplet. */
2266 return -1; /* premature EOF while decoding base64 */
2268 return -1; /* illegal char while decoding base64 */
2274 return -1; /* premature EOF while decoding base64 */
2276 return -1; /* padding `=' expected but not found */
2280 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2281 *q++ = 0xff & value >> 8;
2283 /* Process fourth byte of a quadruplet. */
2286 return -1; /* premature EOF while decoding base64 */
2290 return -1; /* illegal char while decoding base64 */
2292 value |= BASE64_CHAR_TO_VALUE (c);
2293 *q++ = 0xff & value;
2296 #undef BASE64_CHAR_TO_VALUE
2298 return q - (char *) dest;
2304 /* Simple merge sort for use by stable_sort. Implementation courtesy
2305 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2308 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2309 int (*cmpfun) (const void *, const void *))
2311 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2315 size_t mid = (to + from) / 2;
2316 mergesort_internal (base, temp, size, from, mid, cmpfun);
2317 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2320 for (k = from; (i <= mid) && (j <= to); k++)
2321 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2322 memcpy (ELT (temp, k), ELT (base, i++), size);
2324 memcpy (ELT (temp, k), ELT (base, j++), size);
2326 memcpy (ELT (temp, k++), ELT (base, i++), size);
2328 memcpy (ELT (temp, k++), ELT (base, j++), size);
2329 for (k = from; k <= to; k++)
2330 memcpy (ELT (base, k), ELT (temp, k), size);
2335 /* Stable sort with interface exactly like standard library's qsort.
2336 Uses mergesort internally, allocating temporary storage with
2340 stable_sort (void *base, size_t nmemb, size_t size,
2341 int (*cmpfun) (const void *, const void *))
2345 void *temp = alloca (nmemb * size * sizeof (void *));
2346 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2350 /* Print a decimal number. If it is equal to or larger than ten, the
2351 number is rounded. Otherwise it is printed with one significant
2352 digit without trailing zeros and with no more than three fractional
2353 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2354 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2356 This is useful for displaying durations because it provides
2357 order-of-magnitude information without unnecessary clutter --
2358 long-running downloads are shown without the fractional part, and
2359 short ones still retain one significant digit. */
2362 print_decimal (double number)
2364 static char buf[32];
2365 double n = number >= 0 ? number : -number;
2368 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2369 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2370 snprintf (buf, sizeof buf, "%.0f", number);
2372 snprintf (buf, sizeof buf, "%.1f", number);
2373 else if (n >= 0.001)
2374 snprintf (buf, sizeof buf, "%.1g", number);
2375 else if (n >= 0.0005)
2376 /* round [0.0005, 0.001) to 0.001 */
2377 snprintf (buf, sizeof buf, "%.3f", number);
2379 /* print numbers close to 0 as 0, not 0.000 */
2396 { "/somedir", "/somedir", true },
2397 { "/somedir", "/somedir/d2", true },
2398 { "/somedir/d1", "/somedir", false },
2401 for (i = 0; i < countof(test_array); ++i)
2403 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2405 mu_assert ("test_subdir_p: wrong result",
2406 res == test_array[i].result);
2413 test_dir_matches_p()
2421 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2422 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2423 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2424 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2425 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2426 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2427 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2428 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2429 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2430 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2431 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2432 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2433 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2434 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2435 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2438 for (i = 0; i < countof(test_array); ++i)
2440 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2442 mu_assert ("test_dir_matches_p: wrong result",
2443 res == test_array[i].result);
2449 #endif /* TESTING */