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() */
56 /* For TIOCGWINSZ and friends: */
57 #ifdef HAVE_SYS_IOCTL_H
58 # include <sys/ioctl.h>
64 /* Needed for Unix version of run_with_timeout. */
68 #ifndef HAVE_SIGSETJMP
69 /* If sigsetjmp is a macro, configure won't pick it up. */
71 # define HAVE_SIGSETJMP
75 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
76 # define USE_SIGNAL_TIMEOUT
84 #endif /* def __VMS */
91 memfatal (const char *context, long attempted_size)
93 /* Make sure we don't try to store part of the log line, and thus
95 log_set_save_context (false);
97 /* We have different log outputs in different situations:
98 1) output without bytes information
99 2) output with bytes information */
100 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
102 logprintf (LOG_ALWAYS,
103 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
108 logprintf (LOG_ALWAYS,
109 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
110 exec_name, context, attempted_size);
116 /* Character property table for (re-)escaping VMS ODS5 extended file
117 names. Note that this table ignores Unicode.
119 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
121 ODS5 Invalid characters:
122 C0 control codes (0x00 to 0x1F inclusive)
126 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
127 Double quotation marks (")
130 Left angle bracket (<)
131 Right angle bracket (>)
135 Characters escaped by "^":
136 SP ! " # % & ' ( ) + , . : ; =
139 Either "^_" or "^ " is accepted as a space. Period (.) is a special
140 case. Note that un-escaped < and > can also confuse a directory
143 Characters put out as ^xx:
145 80-9F (C1 control characters)
146 A0 (nonbreaking space)
147 FF (Latin small letter y diaeresis)
150 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
152 Property table values:
162 unsigned char char_prop[ 256] = {
164 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
165 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
167 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
168 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
170 /* SP ! " # $ % & ' ( ) * + , - . / */
171 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
173 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
174 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
176 /* @ A B C D E F G H I J K L M N O */
177 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
179 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
180 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
182 /* ` a b c d e f g h i j k l m n o */
183 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
185 /* p q r s t u v w x y z { | } ~ DEL */
186 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
188 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
189 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
190 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
191 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
192 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
193 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
194 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
195 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
198 /* Utility function: like xstrdup(), but also lowercases S. */
201 xstrdup_lower (const char *s)
203 char *copy = xstrdup (s);
210 /* Copy the string formed by two pointers (one on the beginning, other
211 on the char after the last char) to a new, malloc-ed location.
214 strdupdelim (const char *beg, const char *end)
216 char *res = xmalloc (end - beg + 1);
217 memcpy (res, beg, end - beg);
218 res[end - beg] = '\0';
222 /* Parse a string containing comma-separated elements, and return a
223 vector of char pointers with the elements. Spaces following the
224 commas are ignored. */
226 sepstring (const char *s)
240 res = xrealloc (res, (i + 2) * sizeof (char *));
241 res[i] = strdupdelim (p, s);
244 /* Skip the blanks following the ','. */
245 while (c_isspace (*s))
252 res = xrealloc (res, (i + 2) * sizeof (char *));
253 res[i] = strdupdelim (p, s);
258 /* Like sprintf, but prints into a string of sufficient size freshly
259 allocated with malloc, which is returned. If unable to print due
260 to invalid format, returns NULL. Inability to allocate needed
261 memory results in abort, as with xmalloc. This is in spirit
262 similar to the GNU/BSD extension asprintf, but somewhat easier to
265 Internally the function either calls vasprintf or loops around
266 vsnprintf until the correct size is found. Since Wget also ships a
267 fallback implementation of vsnprintf, this should be portable. */
269 /* Constant is using for limits memory allocation for text buffer.
270 Applicable in situation when: vasprintf is not available in the system
271 and vsnprintf return -1 when long line is truncated (in old versions of
272 glibc and in other system where C99 doesn`t support) */
274 #define FMT_MAX_LENGTH 1048576
277 aprintf (const char *fmt, ...)
279 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
284 va_start (args, fmt);
285 ret = vasprintf (&str, fmt, args);
287 if (ret < 0 && errno == ENOMEM)
288 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
289 with xmalloc/xrealloc */
293 #else /* not HAVE_VASPRINTF */
295 /* vasprintf is unavailable. snprintf into a small buffer and
296 resize it as necessary. */
298 char *str = xmalloc (size);
300 /* #### This code will infloop and eventually abort in xrealloc if
301 passed a FMT that causes snprintf to consistently return -1. */
308 va_start (args, fmt);
309 n = vsnprintf (str, size, fmt, args);
312 /* If the printing worked, return the string. */
313 if (n > -1 && n < size)
316 /* Else try again with a larger buffer. */
317 if (n > -1) /* C99 */
318 size = n + 1; /* precisely what is needed */
319 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
320 { /* maybe we have some wrong
322 logprintf (LOG_ALWAYS,
323 _("%s: aprintf: text buffer is too big (%ld bytes), "
325 exec_name, size); /* printout a log message */
326 abort (); /* and abort... */
330 /* else, we continue to grow our
331 * buffer: Twice the old size. */
334 str = xrealloc (str, size);
336 #endif /* not HAVE_VASPRINTF */
339 /* Concatenate the NULL-terminated list of string arguments into
340 freshly allocated space. */
343 concat_strings (const char *str0, ...)
346 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
349 const char *next_str;
350 int total_length = 0;
353 /* Calculate the length of and allocate the resulting string. */
356 va_start (args, str0);
357 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
359 int len = strlen (next_str);
360 if (argcount < countof (saved_lengths))
361 saved_lengths[argcount++] = len;
365 p = ret = xmalloc (total_length + 1);
367 /* Copy the strings into the allocated space. */
370 va_start (args, str0);
371 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
374 if (argcount < countof (saved_lengths))
375 len = saved_lengths[argcount++];
377 len = strlen (next_str);
378 memcpy (p, next_str, len);
387 /* Format the provided time according to the specified format. The
388 format is a string with format elements supported by strftime. */
391 fmttime (time_t t, const char *fmt)
393 static char output[32];
394 struct tm *tm = localtime(&t);
397 if (!strftime(output, sizeof(output), fmt, tm))
402 /* Return pointer to a static char[] buffer in which zero-terminated
403 string-representation of TM (in form hh:mm:ss) is printed.
405 If TM is NULL, the current time will be used. */
410 return fmttime(t, "%H:%M:%S");
413 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
416 datetime_str (time_t t)
418 return fmttime(t, "%Y-%m-%d %H:%M:%S");
421 /* The Windows versions of the following two functions are defined in
422 mswindows.c. On MSDOS this function should never be called. */
427 fork_to_background (void)
432 #else /* def __VMS */
434 #if !defined(WINDOWS) && !defined(MSDOS)
436 fork_to_background (void)
439 /* Whether we arrange our own version of opt.lfilename here. */
440 bool logfile_changed = false;
442 if (!opt.lfilename && (!opt.quiet || opt.server_response))
444 /* We must create the file immediately to avoid either a race
445 condition (which arises from using unique_name and failing to
446 use fopen_excl) or lying to the user about the log file name
447 (which arises from using unique_name, printing the name, and
448 using fopen_excl later on.) */
449 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
452 logfile_changed = true;
465 /* parent, no error */
466 printf (_("Continuing in background, pid %d.\n"), (int) pid);
468 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
469 exit (0); /* #### should we use _exit()? */
472 /* child: give up the privileges and keep running. */
474 freopen ("/dev/null", "r", stdin);
475 freopen ("/dev/null", "w", stdout);
476 freopen ("/dev/null", "w", stderr);
478 #endif /* !WINDOWS && !MSDOS */
480 #endif /* def __VMS [else] */
483 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
484 specified with TM. The atime ("access time") is set to the current
488 touch (const char *file, time_t tm)
490 struct timespec timespecs[2];
493 fd = open (file, O_WRONLY);
496 logprintf (LOG_NOTQUIET, "open(%s): %s\n", file, strerror (errno));
500 timespecs[0].tv_sec = time (NULL);
501 timespecs[0].tv_nsec = 0L;
502 timespecs[1].tv_sec = tm;
503 timespecs[1].tv_nsec = 0L;
505 if (futimens (fd, timespecs) == -1)
506 logprintf (LOG_NOTQUIET, "futimens(%s): %s\n", file, strerror (errno));
511 /* Checks if FILE is a symbolic link, and removes it if it is. Does
512 nothing under MS-Windows. */
514 remove_link (const char *file)
519 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
521 DEBUGP (("Unlinking %s (symlink).\n", file));
524 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
525 quote (file), strerror (errno));
530 /* Does FILENAME exist? This is quite a lousy implementation, since
531 it supplies no error codes -- only a yes-or-no answer. Thus it
532 will return that a file does not exist if, e.g., the directory is
533 unreadable. I don't mind it too much currently, though. The
534 proper way should, of course, be to have a third, error state,
535 other than true/false, but that would introduce uncalled-for
536 additional complexity to the callers. */
538 file_exists_p (const char *filename)
541 return access (filename, F_OK) >= 0;
544 return stat (filename, &buf) >= 0;
548 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
549 Returns 0 on error. */
551 file_non_directory_p (const char *path)
554 /* Use lstat() rather than stat() so that symbolic links pointing to
555 directories can be identified correctly. */
556 if (lstat (path, &buf) != 0)
558 return S_ISDIR (buf.st_mode) ? false : true;
561 /* Return the size of file named by FILENAME, or -1 if it cannot be
562 opened or seeked into. */
564 file_size (const char *filename)
566 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
568 /* We use fseek rather than stat to determine the file size because
569 that way we can also verify that the file is readable without
570 explicitly checking for permissions. Inspired by the POST patch
572 FILE *fp = fopen (filename, "rb");
575 fseeko (fp, 0, SEEK_END);
581 if (stat (filename, &st) < 0)
588 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
589 original name. With the VMS file systems' versioning, everything
590 should be fine, and appending ".NN" just causes trouble.
595 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
596 doesn't exist is found. Return a freshly allocated copy of the
600 unique_name_1 (const char *prefix)
603 int plen = strlen (prefix);
604 char *template = (char *)alloca (plen + 1 + 24);
605 char *template_tail = template + plen;
607 memcpy (template, prefix, plen);
608 *template_tail++ = UNIQ_SEP;
611 number_to_string (template_tail, count++);
612 while (file_exists_p (template));
614 return xstrdup (template);
617 /* Return a unique file name, based on FILE.
619 More precisely, if FILE doesn't exist, it is returned unmodified.
620 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
621 file name that doesn't exist is returned.
623 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
625 The resulting file is not created, only verified that it didn't
626 exist at the point in time when the function was called.
627 Therefore, where security matters, don't rely that the file created
628 by this function exists until you open it with O_EXCL or
631 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
632 string. Otherwise, it may return FILE if the file doesn't exist
633 (and therefore doesn't need changing). */
636 unique_name (const char *file, bool allow_passthrough)
638 /* If the FILE itself doesn't exist, return it without
640 if (!file_exists_p (file))
641 return allow_passthrough ? (char *)file : xstrdup (file);
643 /* Otherwise, find a numeric suffix that results in unused file name
645 return unique_name_1 (file);
648 #else /* def UNIQ_SEP */
650 /* Dummy unique_name() for VMS. Return the original name as easily as
654 unique_name (const char *file, bool allow_passthrough)
656 /* Return the FILE itself, without modification, irregardful. */
657 return allow_passthrough ? (char *)file : xstrdup (file);
660 #endif /* def UNIQ_SEP [else] */
662 /* Create a file based on NAME, except without overwriting an existing
663 file with that name. Providing O_EXCL is correctly implemented,
664 this function does not have the race condition associated with
665 opening the file returned by unique_name. */
668 unique_create (const char *name, bool binary, char **opened_name)
670 /* unique file name, based on NAME */
671 char *uname = unique_name (name, false);
673 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
676 uname = unique_name (name, false);
678 if (opened_name && fp != NULL)
681 *opened_name = uname;
693 /* Open the file for writing, with the addition that the file is
694 opened "exclusively". This means that, if the file already exists,
695 this function will *fail* and errno will be set to EEXIST. If
696 BINARY is set, the file will be opened in binary mode, equivalent
699 If opening the file fails for any reason, including the file having
700 previously existed, this function returns NULL and sets errno
704 fopen_excl (const char *fname, int binary)
710 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
711 It also has file versions which obviate all the O_EXCL effort.
712 O_TRUNC (something of a misnomer) requests a new version.
715 /* Common open() optional arguments:
716 sequential access only, access callback function.
718 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
721 int flags = O_WRONLY | O_CREAT | O_TRUNC;
726 fd = open( fname, /* File name. */
728 0777, /* Mode for default protection. */
729 "ctx=bin,stm", /* Binary, stream access. */
730 "rfm=stmlf", /* Stream_LF. */
731 OPEN_OPT_ARGS); /* Access callback. */
736 fd = open( fname, /* File name. */
738 0777, /* Mode for default protection. */
739 "ctx=bin,stm", /* Binary, stream access. */
740 "rfm=fix", /* Fixed-length, */
741 "mrs=512", /* 512-byte records. */
742 OPEN_OPT_ARGS); /* Access callback. */
747 fd = open( fname, /* File name. */
749 0777, /* Mode for default protection.
751 "rfm=stmlf", /* Stream_LF. */
752 OPEN_OPT_ARGS); /* Access callback. */
754 # else /* def __VMS */
755 int flags = O_WRONLY | O_CREAT | O_EXCL;
760 fd = open (fname, flags, 0666);
761 # endif /* def __VMS [else] */
765 return fdopen (fd, binary ? "wb" : "w");
766 #else /* not O_EXCL */
767 /* Manually check whether the file exists. This is prone to race
768 conditions, but systems without O_EXCL haven't deserved
770 if (file_exists_p (fname))
775 return fopen (fname, binary ? "wb" : "w");
776 #endif /* not O_EXCL */
779 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
780 are missing, create them first. In case any mkdir() call fails,
781 return its error status. Returns 0 on successful completion.
783 The behaviour of this function should be identical to the behaviour
784 of `mkdir -p' on systems where mkdir supports the `-p' option. */
786 make_directory (const char *directory)
788 int i, ret, quit = 0;
791 /* Make a copy of dir, to be able to write to it. Otherwise, the
792 function is unsafe if called with a read-only char *argument. */
793 STRDUP_ALLOCA (dir, directory);
795 /* If the first character of dir is '/', skip it (and thus enable
796 creation of absolute-pathname directories. */
797 for (i = (*dir == '/'); 1; ++i)
799 for (; dir[i] && dir[i] != '/'; i++)
804 /* Check whether the directory already exists. Allow creation of
805 of intermediate directories to fail, as the initial path components
806 are not necessarily directories! */
807 if (!file_exists_p (dir))
808 ret = mkdir (dir, 0777);
819 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
820 should be a file name.
822 file_merge("/foo/bar", "baz") => "/foo/baz"
823 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
824 file_merge("foo", "bar") => "bar"
826 In other words, it's a simpler and gentler version of uri_merge. */
829 file_merge (const char *base, const char *file)
832 const char *cut = (const char *)strrchr (base, '/');
835 return xstrdup (file);
837 result = xmalloc (cut - base + 1 + strlen (file) + 1);
838 memcpy (result, base, cut - base);
839 result[cut - base] = '/';
840 strcpy (result + (cut - base) + 1, file);
845 /* Like fnmatch, but performs a case-insensitive match. */
848 fnmatch_nocase (const char *pattern, const char *string, int flags)
851 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
852 also present on *BSD platforms, and possibly elsewhere. */
853 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
855 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
856 char *patcopy = (char *) alloca (strlen (pattern) + 1);
857 char *strcopy = (char *) alloca (strlen (string) + 1);
859 for (p = patcopy; *pattern; pattern++, p++)
860 *p = c_tolower (*pattern);
862 for (p = strcopy; *string; string++, p++)
863 *p = c_tolower (*string);
865 return fnmatch (patcopy, strcopy, flags);
869 static bool in_acclist (const char *const *, const char *, bool);
871 /* Determine whether a file is acceptable to be followed, according to
872 lists of patterns to accept/reject. */
874 acceptable (const char *s)
878 while (l && s[l] != '/')
885 return (in_acclist ((const char *const *)opt.accepts, s, true)
886 && !in_acclist ((const char *const *)opt.rejects, s, true));
888 return in_acclist ((const char *const *)opt.accepts, s, true);
890 else if (opt.rejects)
891 return !in_acclist ((const char *const *)opt.rejects, s, true);
895 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
896 will return true if and only if D2 begins with `/something/' or is exactly
899 subdir_p (const char *d1, const char *d2)
903 if (!opt.ignore_case)
904 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
907 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
910 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
913 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
914 first element that matches DIR, through wildcards or front comparison (as
917 dir_matches_p (char **dirlist, const char *dir)
920 int (*matcher) (const char *, const char *, int)
921 = opt.ignore_case ? fnmatch_nocase : fnmatch;
923 for (x = dirlist; *x; x++)
925 /* Remove leading '/' */
926 char *p = *x + (**x == '/');
927 if (has_wildcards_p (p))
929 if (matcher (p, dir, FNM_PATHNAME) == 0)
934 if (subdir_p (p, dir))
939 return *x ? true : false;
942 /* Returns whether DIRECTORY is acceptable for download, wrt the
943 include/exclude lists.
945 The leading `/' is ignored in paths; relative and absolute paths
946 may be freely intermixed. */
949 accdir (const char *directory)
951 /* Remove starting '/'. */
952 if (*directory == '/')
956 if (!dir_matches_p (opt.includes, directory))
961 if (dir_matches_p (opt.excludes, directory))
967 /* Return true if STRING ends with TAIL. For instance:
969 match_tail ("abc", "bc", false) -> 1
970 match_tail ("abc", "ab", false) -> 0
971 match_tail ("abc", "abc", false) -> 1
973 If FOLD_CASE is true, the comparison will be case-insensitive. */
976 match_tail (const char *string, const char *tail, bool fold_case)
980 /* We want this to be fast, so we code two loops, one with
981 case-folding, one without. */
985 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
986 if (string[i] != tail[j])
991 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
992 if (c_tolower (string[i]) != c_tolower (tail[j]))
996 /* If the tail was exhausted, the match was succesful. */
1003 /* Checks whether string S matches each element of ACCEPTS. A list
1004 element are matched either with fnmatch() or match_tail(),
1005 according to whether the element contains wildcards or not.
1007 If the BACKWARD is false, don't do backward comparison -- just compare
1010 in_acclist (const char *const *accepts, const char *s, bool backward)
1012 for (; *accepts; accepts++)
1014 if (has_wildcards_p (*accepts))
1016 int res = opt.ignore_case
1017 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1018 /* fnmatch returns 0 if the pattern *does* match the string. */
1026 if (match_tail (s, *accepts, opt.ignore_case))
1031 int cmp = opt.ignore_case
1032 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1041 /* Return the location of STR's suffix (file extension). Examples:
1042 suffix ("foo.bar") -> "bar"
1043 suffix ("foo.bar.baz") -> "baz"
1044 suffix ("/foo/bar") -> NULL
1045 suffix ("/foo.bar/baz") -> NULL */
1047 suffix (const char *str)
1051 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1054 if (str[i++] == '.')
1055 return (char *)str + i;
1060 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1064 has_wildcards_p (const char *s)
1067 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1072 /* Return true if FNAME ends with a typical HTML suffix. The
1073 following (case-insensitive) suffixes are presumed to be HTML
1078 ?html (`?' matches one character)
1080 #### CAVEAT. This is not necessarily a good indication that FNAME
1081 refers to a file that contains HTML! */
1083 has_html_suffix_p (const char *fname)
1087 if ((suf = suffix (fname)) == NULL)
1089 if (!strcasecmp (suf, "html"))
1091 if (!strcasecmp (suf, "htm"))
1093 if (suf[0] && !strcasecmp (suf + 1, "html"))
1098 /* Read a line from FP and return the pointer to freshly allocated
1099 storage. The storage space is obtained through malloc() and should
1100 be freed with free() when it is no longer needed.
1102 The length of the line is not limited, except by available memory.
1103 The newline character at the end of line is retained. The line is
1104 terminated with a zero character.
1106 After end-of-file is encountered without anything being read, NULL
1107 is returned. NULL is also returned on error. To distinguish
1108 between these two cases, use the stdio function ferror(). */
1111 read_whole_line (FILE *fp)
1115 char *line = xmalloc (bufsize);
1117 while (fgets (line + length, bufsize - length, fp))
1119 length += strlen (line + length);
1121 /* Possible for example when reading from a binary file where
1122 a line begins with \0. */
1125 if (line[length - 1] == '\n')
1128 /* fgets() guarantees to read the whole line, or to use up the
1129 space we've given it. We can double the buffer
1132 line = xrealloc (line, bufsize);
1134 if (length == 0 || ferror (fp))
1139 if (length + 1 < bufsize)
1140 /* Relieve the memory from our exponential greediness. We say
1141 `length + 1' because the terminating \0 is not included in
1142 LENGTH. We don't need to zero-terminate the string ourselves,
1143 though, because fgets() does that. */
1144 line = xrealloc (line, length + 1);
1148 /* Read FILE into memory. A pointer to `struct file_memory' are
1149 returned; use struct element `content' to access file contents, and
1150 the element `length' to know the file length. `content' is *not*
1151 zero-terminated, and you should *not* read or write beyond the [0,
1152 length) range of characters.
1154 After you are done with the file contents, call wget_read_file_free to
1157 Depending on the operating system and the type of file that is
1158 being read, wget_read_file() either mmap's the file into memory, or
1159 reads the file into the core using read().
1161 If file is named "-", fileno(stdin) is used for reading instead.
1162 If you want to read from a real file named "-", use "./-" instead. */
1164 struct file_memory *
1165 wget_read_file (const char *file)
1168 struct file_memory *fm;
1170 bool inhibit_close = false;
1172 /* Some magic in the finest tradition of Perl and its kin: if FILE
1173 is "-", just use stdin. */
1176 fd = fileno (stdin);
1177 inhibit_close = true;
1178 /* Note that we don't inhibit mmap() in this case. If stdin is
1179 redirected from a regular file, mmap() will still work. */
1182 fd = open (file, O_RDONLY);
1185 fm = xnew (struct file_memory);
1190 if (fstat (fd, &buf) < 0)
1192 fm->length = buf.st_size;
1193 /* NOTE: As far as I know, the callers of this function never
1194 modify the file text. Relying on this would enable us to
1195 specify PROT_READ and MAP_SHARED for a marginal gain in
1196 efficiency, but at some cost to generality. */
1197 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1198 MAP_PRIVATE, fd, 0);
1199 if (fm->content == (char *)MAP_FAILED)
1209 /* The most common reason why mmap() fails is that FD does not point
1210 to a plain file. However, it's also possible that mmap() doesn't
1211 work for a particular type of file. Therefore, whenever mmap()
1212 fails, we just fall back to the regular method. */
1213 #endif /* HAVE_MMAP */
1216 size = 512; /* number of bytes fm->contents can
1217 hold at any given time. */
1218 fm->content = xmalloc (size);
1222 if (fm->length > size / 2)
1224 /* #### I'm not sure whether the whole exponential-growth
1225 thing makes sense with kernel read. On Linux at least,
1226 read() refuses to read more than 4K from a file at a
1227 single chunk anyway. But other Unixes might optimize it
1228 better, and it doesn't *hurt* anything, so I'm leaving
1231 /* Normally, we grow SIZE exponentially to make the number
1232 of calls to read() and realloc() logarithmic in relation
1233 to file size. However, read() can read an amount of data
1234 smaller than requested, and it would be unreasonable to
1235 double SIZE every time *something* was read. Therefore,
1236 we double SIZE only when the length exceeds half of the
1237 entire allocated size. */
1239 fm->content = xrealloc (fm->content, size);
1241 nread = read (fd, fm->content + fm->length, size - fm->length);
1243 /* Successful read. */
1244 fm->length += nread;
1254 if (size > fm->length && fm->length != 0)
1255 /* Due to exponential growth of fm->content, the allocated region
1256 might be much larger than what is actually needed. */
1257 fm->content = xrealloc (fm->content, fm->length);
1264 xfree (fm->content);
1269 /* Release the resources held by FM. Specifically, this calls
1270 munmap() or xfree() on fm->content, depending whether mmap or
1271 malloc/read were used to read in the file. It also frees the
1272 memory needed to hold the FM structure itself. */
1275 wget_read_file_free (struct file_memory *fm)
1280 munmap (fm->content, fm->length);
1285 xfree (fm->content);
1290 /* Free the pointers in a NULL-terminated vector of pointers, then
1291 free the pointer itself. */
1293 free_vec (char **vec)
1304 /* Append vector V2 to vector V1. The function frees V2 and
1305 reallocates V1 (thus you may not use the contents of neither
1306 pointer after the call). If V1 is NULL, V2 is returned. */
1308 merge_vecs (char **v1, char **v2)
1318 /* To avoid j == 0 */
1323 for (i = 0; v1[i]; i++)
1326 for (j = 0; v2[j]; j++)
1328 /* Reallocate v1. */
1329 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1330 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1335 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1336 is allocated as needed. Return the new value of the vector. */
1339 vec_append (char **vec, const char *str)
1341 int cnt; /* count of vector elements, including
1342 the one we're about to append */
1345 for (cnt = 0; vec[cnt]; cnt++)
1351 /* Reallocate the array to fit the new element and the NULL. */
1352 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1353 /* Append a copy of STR to the vector. */
1354 vec[cnt - 1] = xstrdup (str);
1359 /* Sometimes it's useful to create "sets" of strings, i.e. special
1360 hash tables where you want to store strings as keys and merely
1361 query for their existence. Here is a set of utility routines that
1362 makes that transparent. */
1365 string_set_add (struct hash_table *ht, const char *s)
1367 /* First check whether the set element already exists. If it does,
1368 do nothing so that we don't have to free() the old element and
1369 then strdup() a new one. */
1370 if (hash_table_contains (ht, s))
1373 /* We use "1" as value. It provides us a useful and clear arbitrary
1374 value, and it consumes no memory -- the pointers to the same
1375 string "1" will be shared by all the key-value pairs in all `set'
1377 hash_table_put (ht, xstrdup (s), "1");
1380 /* Synonym for hash_table_contains... */
1383 string_set_contains (struct hash_table *ht, const char *s)
1385 return hash_table_contains (ht, s);
1388 /* Convert the specified string set to array. ARRAY should be large
1389 enough to hold hash_table_count(ht) char pointers. */
1391 void string_set_to_array (struct hash_table *ht, char **array)
1393 hash_table_iterator iter;
1394 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1395 *array++ = iter.key;
1398 /* Free the string set. This frees both the storage allocated for
1399 keys and the actual hash table. (hash_table_destroy would only
1400 destroy the hash table.) */
1403 string_set_free (struct hash_table *ht)
1405 hash_table_iterator iter;
1406 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1408 hash_table_destroy (ht);
1411 /* Utility function: simply call xfree() on all keys and values of HT. */
1414 free_keys_and_values (struct hash_table *ht)
1416 hash_table_iterator iter;
1417 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1424 /* Get digit grouping data for thousand separors by calling
1425 localeconv(). The data includes separator string and grouping info
1426 and is cached after the first call to the function.
1428 In locales that don't set a thousand separator (such as the "C"
1429 locale), this forces it to be ",". We are now only showing
1430 thousand separators in one place, so this shouldn't be a problem in
1434 get_grouping_data (const char **sep, const char **grouping)
1436 static const char *cached_sep;
1437 static const char *cached_grouping;
1438 static bool initialized;
1441 /* Get the grouping info from the locale. */
1442 struct lconv *lconv = localeconv ();
1443 cached_sep = lconv->thousands_sep;
1444 cached_grouping = lconv->grouping;
1445 #if ! USE_NLS_PROGRESS_BAR
1446 /* We can't count column widths, so ensure that the separator
1447 * is single-byte only (let check below determine what byte). */
1448 if (strlen(cached_sep) > 1)
1453 /* Many locales (such as "C" or "hr_HR") don't specify
1454 grouping, which we still want to use it for legibility.
1455 In those locales set the sep char to ',', unless that
1456 character is used for decimal point, in which case set it
1458 if (*lconv->decimal_point != ',')
1462 cached_grouping = "\x03";
1467 *grouping = cached_grouping;
1470 /* Return a printed representation of N with thousand separators.
1471 This should respect locale settings, with the exception of the "C"
1472 locale which mandates no separator, but we use one anyway.
1474 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1475 the separators because it's too non-portable, and it's hard to test
1476 for this feature at configure time. Besides, it wouldn't display
1477 separators in the "C" locale, still used by many Unix users. */
1480 with_thousand_seps (wgint n)
1482 static char outbuf[48];
1483 char *p = outbuf + sizeof outbuf;
1485 /* Info received from locale */
1486 const char *grouping, *sep;
1489 /* State information */
1490 int i = 0, groupsize;
1491 const char *atgroup;
1493 bool negative = n < 0;
1495 /* Initialize grouping data. */
1496 get_grouping_data (&sep, &grouping);
1497 seplen = strlen (sep);
1499 groupsize = *atgroup++;
1501 /* This would overflow on WGINT_MIN, but printing negative numbers
1502 is not an important goal of this fuinction. */
1506 /* Write the number into the buffer, backwards, inserting the
1507 separators as necessary. */
1511 *--p = n % 10 + '0';
1515 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1516 if (++i == groupsize)
1521 memcpy (p -= seplen, sep, seplen);
1524 groupsize = *atgroup++;
1533 /* N, a byte quantity, is converted to a human-readable abberviated
1534 form a la sizes printed by `ls -lh'. The result is written to a
1535 static buffer, a pointer to which is returned.
1537 Unlike `with_thousand_seps', this approximates to the nearest unit.
1538 Quoting GNU libit: "Most people visually process strings of 3-4
1539 digits effectively, but longer strings of digits are more prone to
1540 misinterpretation. Hence, converting to an abbreviated form
1541 usually improves readability."
1543 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1544 original computer-related meaning of "powers of 1024". We don't
1545 use the "*bibyte" names invented in 1998, and seldom used in
1546 practice. Wikipedia's entry on "binary prefix" discusses this in
1550 human_readable (HR_NUMTYPE n)
1552 /* These suffixes are compatible with those of GNU `ls -lh'. */
1553 static char powers[] =
1555 'K', /* kilobyte, 2^10 bytes */
1556 'M', /* megabyte, 2^20 bytes */
1557 'G', /* gigabyte, 2^30 bytes */
1558 'T', /* terabyte, 2^40 bytes */
1559 'P', /* petabyte, 2^50 bytes */
1560 'E', /* exabyte, 2^60 bytes */
1565 /* If the quantity is smaller than 1K, just print it. */
1568 snprintf (buf, sizeof (buf), "%d", (int) n);
1572 /* Loop over powers, dividing N with 1024 in each iteration. This
1573 works unchanged for all sizes of wgint, while still avoiding
1574 non-portable `long double' arithmetic. */
1575 for (i = 0; i < countof (powers); i++)
1577 /* At each iteration N is greater than the *subsequent* power.
1578 That way N/1024.0 produces a decimal number in the units of
1580 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1582 double val = n / 1024.0;
1583 /* Print values smaller than 10 with one decimal digits, and
1584 others without any decimals. */
1585 snprintf (buf, sizeof (buf), "%.*f%c",
1586 val < 10 ? 1 : 0, val, powers[i]);
1591 return NULL; /* unreached */
1594 /* Count the digits in the provided number. Used to allocate space
1595 when printing numbers. */
1598 numdigit (wgint number)
1602 ++cnt; /* accomodate '-' */
1603 while ((number /= 10) != 0)
1608 #define PR(mask) *p++ = n / (mask) + '0'
1610 /* DIGITS_<D> is used to print a D-digit number and should be called
1611 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1612 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1613 Recursively this continues until DIGITS_1 is invoked. */
1615 #define DIGITS_1(mask) PR (mask)
1616 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1617 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1618 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1619 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1620 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1621 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1622 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1623 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1624 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1626 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1628 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1629 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1630 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1631 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1632 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1633 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1634 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1635 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1636 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1638 /* Shorthand for casting to wgint. */
1641 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1642 `sprintf(buffer, "%lld", (long long) number)', only typically much
1643 faster and portable to machines without long long.
1645 The speedup may make a difference in programs that frequently
1646 convert numbers to strings. Some implementations of sprintf,
1647 particularly the one in some versions of GNU libc, have been known
1648 to be quite slow when converting integers to strings.
1650 Return the pointer to the location where the terminating zero was
1651 printed. (Equivalent to calling buffer+strlen(buffer) after the
1654 BUFFER should be large enough to accept as many bytes as you expect
1655 the number to take up. On machines with 64-bit wgints the maximum
1656 needed size is 24 bytes. That includes the digits needed for the
1657 largest 64-bit number, the `-' sign in case it's negative, and the
1658 terminating '\0'. */
1661 number_to_string (char *buffer, wgint number)
1666 int last_digit_char = 0;
1668 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1669 /* We are running in a very strange environment. Leave the correct
1670 printing to sprintf. */
1671 p += sprintf (buf, "%j", (intmax_t) (n));
1672 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1678 /* n = -n would overflow because -n would evaluate to a
1679 wgint value larger than WGINT_MAX. Need to make n
1680 smaller and handle the last digit separately. */
1681 int last_digit = n % 10;
1682 /* The sign of n%10 is implementation-defined. */
1684 last_digit_char = '0' - last_digit;
1686 last_digit_char = '0' + last_digit;
1687 /* After n is made smaller, -n will not overflow. */
1695 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1696 way printing any N is fully open-coded without a loop or jump.
1697 (Also see description of DIGITS_*.) */
1699 if (n < 10) DIGITS_1 (1);
1700 else if (n < 100) DIGITS_2 (10);
1701 else if (n < 1000) DIGITS_3 (100);
1702 else if (n < 10000) DIGITS_4 (1000);
1703 else if (n < 100000) DIGITS_5 (10000);
1704 else if (n < 1000000) DIGITS_6 (100000);
1705 else if (n < 10000000) DIGITS_7 (1000000);
1706 else if (n < 100000000) DIGITS_8 (10000000);
1707 else if (n < 1000000000) DIGITS_9 (100000000);
1708 #if SIZEOF_WGINT == 4
1709 /* wgint is 32 bits wide: no number has more than 10 digits. */
1710 else DIGITS_10 (1000000000);
1712 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1713 Constants are constructed by compile-time multiplication to avoid
1714 dealing with different notations for 64-bit constants
1715 (nL/nLL/nI64, depending on the compiler and architecture). */
1716 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1717 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1718 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1719 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1720 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1721 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1722 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1723 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1724 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1725 else DIGITS_19 (1000000000*(W)1000000000);
1728 if (last_digit_char)
1729 *p++ = last_digit_char;
1732 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1739 #undef SPRINTF_WGINT
1762 /* Print NUMBER to a statically allocated string and return a pointer
1763 to the printed representation.
1765 This function is intended to be used in conjunction with printf.
1766 It is hard to portably print wgint values:
1767 a) you cannot use printf("%ld", number) because wgint can be long
1768 long on 32-bit machines with LFS.
1769 b) you cannot use printf("%lld", number) because NUMBER could be
1770 long on 32-bit machines without LFS, or on 64-bit machines,
1771 which do not require LFS. Also, Windows doesn't support %lld.
1772 c) you cannot use printf("%j", (int_max_t) number) because not all
1773 versions of printf support "%j", the most notable being the one
1775 d) you cannot #define WGINT_FMT to the appropriate format and use
1776 printf(WGINT_FMT, number) because that would break translations
1777 for user-visible messages, such as printf("Downloaded: %d
1780 What you should use instead is printf("%s", number_to_static_string
1783 CAVEAT: since the function returns pointers to static data, you
1784 must be careful to copy its result before calling it again.
1785 However, to make it more useful with printf, the function maintains
1786 an internal ring of static buffers to return. That way things like
1787 printf("%s %s", number_to_static_string (num1),
1788 number_to_static_string (num2)) work as expected. Three buffers
1789 are currently used, which means that "%s %s %s" will work, but "%s
1790 %s %s %s" won't. If you need to print more than three wgints,
1791 bump the RING_SIZE (or rethink your message.) */
1794 number_to_static_string (wgint number)
1796 static char ring[RING_SIZE][24];
1798 char *buf = ring[ringpos];
1799 number_to_string (buf, number);
1800 ringpos = (ringpos + 1) % RING_SIZE;
1804 /* Determine the width of the terminal we're running on. If that's
1805 not possible, return 0. */
1808 determine_screen_width (void)
1810 /* If there's a way to get the terminal size using POSIX
1811 tcgetattr(), somebody please tell me. */
1816 if (opt.lfilename != NULL)
1819 fd = fileno (stderr);
1820 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1821 return 0; /* most likely ENOTTY */
1824 #elif defined(WINDOWS)
1825 CONSOLE_SCREEN_BUFFER_INFO csbi;
1826 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1828 return csbi.dwSize.X;
1829 #else /* neither TIOCGWINSZ nor WINDOWS */
1831 #endif /* neither TIOCGWINSZ nor WINDOWS */
1834 /* Whether the rnd system (either rand or [dl]rand48) has been
1836 static int rnd_seeded;
1838 /* Return a random number between 0 and MAX-1, inclusive.
1840 If the system does not support lrand48 and MAX is greater than the
1841 value of RAND_MAX+1 on the system, the returned value will be in
1842 the range [0, RAND_MAX]. This may be fixed in a future release.
1843 The random number generator is seeded automatically the first time
1846 This uses lrand48 where available, rand elsewhere. DO NOT use it
1847 for cryptography. It is only meant to be used in situations where
1848 quality of the random numbers returned doesn't really matter. */
1851 random_number (int max)
1856 srand48 ((long) time (NULL) ^ (long) getpid ());
1859 return lrand48 () % max;
1860 #else /* not HAVE_DRAND48 */
1866 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1871 /* Like rand() % max, but uses the high-order bits for better
1872 randomness on architectures where rand() is implemented using a
1873 simple congruential generator. */
1875 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1876 return (int) bounded;
1878 #endif /* not HAVE_DRAND48 */
1881 /* Return a random uniformly distributed floating point number in the
1882 [0, 1) range. Uses drand48 where available, and a really lame
1883 kludge elsewhere. */
1891 srand48 ((long) time (NULL) ^ (long) getpid ());
1895 #else /* not HAVE_DRAND48 */
1896 return ( random_number (10000) / 10000.0
1897 + random_number (10000) / (10000.0 * 10000.0)
1898 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1899 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1900 #endif /* not HAVE_DRAND48 */
1903 /* Implementation of run_with_timeout, a generic timeout-forcing
1904 routine for systems with Unix-like signal handling. */
1906 #ifdef USE_SIGNAL_TIMEOUT
1907 # ifdef HAVE_SIGSETJMP
1908 # define SETJMP(env) sigsetjmp (env, 1)
1910 static sigjmp_buf run_with_timeout_env;
1913 abort_run_with_timeout (int sig)
1915 assert (sig == SIGALRM);
1916 siglongjmp (run_with_timeout_env, -1);
1918 # else /* not HAVE_SIGSETJMP */
1919 # define SETJMP(env) setjmp (env)
1921 static jmp_buf run_with_timeout_env;
1924 abort_run_with_timeout (int sig)
1926 assert (sig == SIGALRM);
1927 /* We don't have siglongjmp to preserve the set of blocked signals;
1928 if we longjumped out of the handler at this point, SIGALRM would
1929 remain blocked. We must unblock it manually. */
1930 int mask = siggetmask ();
1931 mask &= ~sigmask (SIGALRM);
1934 /* Now it's safe to longjump. */
1935 longjmp (run_with_timeout_env, -1);
1937 # endif /* not HAVE_SIGSETJMP */
1939 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1940 setitimer where available, alarm otherwise.
1942 TIMEOUT should be non-zero. If the timeout value is so small that
1943 it would be rounded to zero, it is rounded to the least legal value
1944 instead (1us for setitimer, 1s for alarm). That ensures that
1945 SIGALRM will be delivered in all cases. */
1948 alarm_set (double timeout)
1951 /* Use the modern itimer interface. */
1952 struct itimerval itv;
1954 itv.it_value.tv_sec = (long) timeout;
1955 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1956 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1957 /* Ensure that we wait for at least the minimum interval.
1958 Specifying zero would mean "wait forever". */
1959 itv.it_value.tv_usec = 1;
1960 setitimer (ITIMER_REAL, &itv, NULL);
1961 #else /* not ITIMER_REAL */
1962 /* Use the old alarm() interface. */
1963 int secs = (int) timeout;
1965 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1966 because alarm(0) means "never deliver the alarm", i.e. "wait
1967 forever", which is not what someone who specifies a 0.5s
1968 timeout would expect. */
1971 #endif /* not ITIMER_REAL */
1974 /* Cancel the alarm set with alarm_set. */
1980 struct itimerval disable;
1982 setitimer (ITIMER_REAL, &disable, NULL);
1983 #else /* not ITIMER_REAL */
1985 #endif /* not ITIMER_REAL */
1988 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1989 seconds. Returns true if the function was interrupted with a
1990 timeout, false otherwise.
1992 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1993 using setitimer() or alarm(). The timeout is enforced by
1994 longjumping out of the SIGALRM handler. This has several
1995 advantages compared to the traditional approach of relying on
1996 signals causing system calls to exit with EINTR:
1998 * The callback function is *forcibly* interrupted after the
1999 timeout expires, (almost) regardless of what it was doing and
2000 whether it was in a syscall. For example, a calculation that
2001 takes a long time is interrupted as reliably as an IO
2004 * It works with both SYSV and BSD signals because it doesn't
2005 depend on the default setting of SA_RESTART.
2007 * It doesn't require special handler setup beyond a simple call
2008 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2011 The only downside is that, if FUN allocates internal resources that
2012 are normally freed prior to exit from the functions, they will be
2013 lost in case of timeout. */
2016 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2026 signal (SIGALRM, abort_run_with_timeout);
2027 if (SETJMP (run_with_timeout_env) != 0)
2029 /* Longjumped out of FUN with a timeout. */
2030 signal (SIGALRM, SIG_DFL);
2033 alarm_set (timeout);
2036 /* Preserve errno in case alarm() or signal() modifies it. */
2037 saved_errno = errno;
2039 signal (SIGALRM, SIG_DFL);
2040 errno = saved_errno;
2045 #else /* not USE_SIGNAL_TIMEOUT */
2048 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2049 define it under Windows, because Windows has its own version of
2050 run_with_timeout that uses threads. */
2053 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2058 #endif /* not WINDOWS */
2059 #endif /* not USE_SIGNAL_TIMEOUT */
2063 /* Sleep the specified amount of seconds. On machines without
2064 nanosleep(), this may sleep shorter if interrupted by signals. */
2067 xsleep (double seconds)
2069 #ifdef HAVE_NANOSLEEP
2070 /* nanosleep is the preferred interface because it offers high
2071 accuracy and, more importantly, because it allows us to reliably
2072 restart receiving a signal such as SIGWINCH. (There was an
2073 actual Debian bug report about --limit-rate malfunctioning while
2074 the terminal was being resized.) */
2075 struct timespec sleep, remaining;
2076 sleep.tv_sec = (long) seconds;
2077 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2078 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2079 /* If nanosleep has been interrupted by a signal, adjust the
2080 sleeping period and return to sleep. */
2082 #elif defined(HAVE_USLEEP)
2083 /* If usleep is available, use it in preference to select. */
2086 /* On some systems, usleep cannot handle values larger than
2087 1,000,000. If the period is larger than that, use sleep
2088 first, then add usleep for subsecond accuracy. */
2090 seconds -= (long) seconds;
2092 usleep (seconds * 1000000);
2093 #else /* fall back select */
2094 /* Note that, although Windows supports select, it can't be used to
2095 implement sleeping because Winsock's select doesn't implement
2096 timeout when it is passed NULL pointers for all fd sets. (But it
2097 does under Cygwin, which implements Unix-compatible select.) */
2098 struct timeval sleep;
2099 sleep.tv_sec = (long) seconds;
2100 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2101 select (0, NULL, NULL, NULL, &sleep);
2102 /* If select returns -1 and errno is EINTR, it means we were
2103 interrupted by a signal. But without knowing how long we've
2104 actually slept, we can't return to sleep. Using gettimeofday to
2105 track sleeps is slow and unreliable due to clock skew. */
2109 #endif /* not WINDOWS */
2111 /* Encode the octets in DATA of length LENGTH to base64 format,
2112 storing the result to DEST. The output will be zero-terminated,
2113 and must point to a writable buffer of at least
2114 1+BASE64_LENGTH(length) bytes. The function returns the length of
2115 the resulting base64 data, not counting the terminating zero.
2117 This implementation does not emit newlines after 76 characters of
2121 base64_encode (const void *data, int length, char *dest)
2123 /* Conversion table. */
2124 static const char tbl[64] = {
2125 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2126 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2127 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2128 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2130 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2131 don't work for data with MSB set. */
2132 const unsigned char *s = data;
2133 /* Theoretical ANSI violation when length < 3. */
2134 const unsigned char *end = (const unsigned char *) data + length - 2;
2137 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2138 for (; s < end; s += 3)
2140 *p++ = tbl[s[0] >> 2];
2141 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2142 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2143 *p++ = tbl[s[2] & 0x3f];
2146 /* Pad the result if necessary... */
2150 *p++ = tbl[s[0] >> 2];
2151 *p++ = tbl[(s[0] & 3) << 4];
2156 *p++ = tbl[s[0] >> 2];
2157 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2158 *p++ = tbl[((s[1] & 0xf) << 2)];
2162 /* ...and zero-terminate it. */
2168 /* Store in C the next non-whitespace character from the string, or \0
2169 when end of string is reached. */
2170 #define NEXT_CHAR(c, p) do { \
2171 c = (unsigned char) *p++; \
2172 } while (c_isspace (c))
2174 #define IS_ASCII(c) (((c) & 0x80) == 0)
2176 /* Decode data from BASE64 (a null-terminated string) into memory
2177 pointed to by DEST. DEST is assumed to be large enough to
2178 accomodate the decoded data, which is guaranteed to be no more than
2181 Since DEST is assumed to contain binary data, it is not
2182 NUL-terminated. The function returns the length of the data
2183 written to TO. -1 is returned in case of error caused by malformed
2186 This function originates from Free Recode. */
2189 base64_decode (const char *base64, void *dest)
2191 /* Table of base64 values for first 128 characters. Note that this
2192 assumes ASCII (but so does Wget in other places). */
2193 static const signed char base64_char_to_value[128] =
2195 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2196 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2197 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2198 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2199 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2200 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2201 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2202 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2203 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2204 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2205 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2206 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2207 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2209 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2210 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2212 const char *p = base64;
2218 unsigned long value;
2220 /* Process first byte of a quadruplet. */
2224 if (c == '=' || !IS_BASE64 (c))
2225 return -1; /* illegal char while decoding base64 */
2226 value = BASE64_CHAR_TO_VALUE (c) << 18;
2228 /* Process second byte of a quadruplet. */
2231 return -1; /* premature EOF while decoding base64 */
2232 if (c == '=' || !IS_BASE64 (c))
2233 return -1; /* illegal char while decoding base64 */
2234 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2237 /* Process third byte of a quadruplet. */
2240 return -1; /* premature EOF while decoding base64 */
2242 return -1; /* illegal char while decoding base64 */
2248 return -1; /* premature EOF while decoding base64 */
2250 return -1; /* padding `=' expected but not found */
2254 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2255 *q++ = 0xff & value >> 8;
2257 /* Process fourth byte of a quadruplet. */
2260 return -1; /* premature EOF while decoding base64 */
2264 return -1; /* illegal char while decoding base64 */
2266 value |= BASE64_CHAR_TO_VALUE (c);
2267 *q++ = 0xff & value;
2270 #undef BASE64_CHAR_TO_VALUE
2272 return q - (char *) dest;
2278 /* Simple merge sort for use by stable_sort. Implementation courtesy
2279 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2282 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2283 int (*cmpfun) (const void *, const void *))
2285 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2289 size_t mid = (to + from) / 2;
2290 mergesort_internal (base, temp, size, from, mid, cmpfun);
2291 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2294 for (k = from; (i <= mid) && (j <= to); k++)
2295 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2296 memcpy (ELT (temp, k), ELT (base, i++), size);
2298 memcpy (ELT (temp, k), ELT (base, j++), size);
2300 memcpy (ELT (temp, k++), ELT (base, i++), size);
2302 memcpy (ELT (temp, k++), ELT (base, j++), size);
2303 for (k = from; k <= to; k++)
2304 memcpy (ELT (base, k), ELT (temp, k), size);
2309 /* Stable sort with interface exactly like standard library's qsort.
2310 Uses mergesort internally, allocating temporary storage with
2314 stable_sort (void *base, size_t nmemb, size_t size,
2315 int (*cmpfun) (const void *, const void *))
2319 void *temp = alloca (nmemb * size * sizeof (void *));
2320 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2324 /* Print a decimal number. If it is equal to or larger than ten, the
2325 number is rounded. Otherwise it is printed with one significant
2326 digit without trailing zeros and with no more than three fractional
2327 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2328 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2330 This is useful for displaying durations because it provides
2331 order-of-magnitude information without unnecessary clutter --
2332 long-running downloads are shown without the fractional part, and
2333 short ones still retain one significant digit. */
2336 print_decimal (double number)
2338 static char buf[32];
2339 double n = number >= 0 ? number : -number;
2342 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2343 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2344 snprintf (buf, sizeof buf, "%.0f", number);
2346 snprintf (buf, sizeof buf, "%.1f", number);
2347 else if (n >= 0.001)
2348 snprintf (buf, sizeof buf, "%.1g", number);
2349 else if (n >= 0.0005)
2350 /* round [0.0005, 0.001) to 0.001 */
2351 snprintf (buf, sizeof buf, "%.3f", number);
2353 /* print numbers close to 0 as 0, not 0.000 */
2370 { "/somedir", "/somedir", true },
2371 { "/somedir", "/somedir/d2", true },
2372 { "/somedir/d1", "/somedir", false },
2375 for (i = 0; i < countof(test_array); ++i)
2377 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2379 mu_assert ("test_subdir_p: wrong result",
2380 res == test_array[i].result);
2387 test_dir_matches_p()
2395 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2396 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2397 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2398 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2399 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2400 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2401 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2402 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2403 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2404 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2405 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2406 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2407 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2408 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2409 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2412 for (i = 0; i < countof(test_array); ++i)
2414 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2416 mu_assert ("test_dir_matches_p: wrong result",
2417 res == test_array[i].result);
2423 #endif /* TESTING */