1 /* Various utility functions.
2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #ifdef HAVE_SYS_TIME_H
37 # include <sys/time.h>
43 # include <sys/mman.h>
51 #ifdef HAVE_SYS_UTIME_H
52 # include <sys/utime.h>
56 # include <libc.h> /* for access() */
65 /* For TIOCGWINSZ and friends: */
66 #ifdef HAVE_SYS_IOCTL_H
67 # include <sys/ioctl.h>
73 /* Needed for run_with_timeout. */
79 #ifndef HAVE_SIGSETJMP
80 /* If sigsetjmp is a macro, configure won't pick it up. */
82 # define HAVE_SIGSETJMP
86 #undef USE_SIGNAL_TIMEOUT
87 #if defined(HAVE_SIGSETJMP) || defined(HAVE_SIGBLOCK)
88 # define USE_SIGNAL_TIMEOUT
95 /* Utility function: like xstrdup(), but also lowercases S. */
98 xstrdup_lower (const char *s)
100 char *copy = xstrdup (s);
107 /* Copy the string formed by two pointers (one on the beginning, other
108 on the char after the last char) to a new, malloc-ed location.
111 strdupdelim (const char *beg, const char *end)
113 char *res = xmalloc (end - beg + 1);
114 memcpy (res, beg, end - beg);
115 res[end - beg] = '\0';
119 /* Parse a string containing comma-separated elements, and return a
120 vector of char pointers with the elements. Spaces following the
121 commas are ignored. */
123 sepstring (const char *s)
137 res = xrealloc (res, (i + 2) * sizeof (char *));
138 res[i] = strdupdelim (p, s);
141 /* Skip the blanks following the ','. */
149 res = xrealloc (res, (i + 2) * sizeof (char *));
150 res[i] = strdupdelim (p, s);
155 /* Like sprintf, but allocates a string of sufficient size with malloc
156 and returns it. GNU libc has a similar function named asprintf,
157 which requires the pointer to the string to be passed. */
160 aprintf (const char *fmt, ...)
162 /* This function is implemented using vsnprintf, which we provide
163 for the systems that don't have it. Therefore, it should be 100%
167 char *str = xmalloc (size);
174 /* See log_vprintf_internal for explanation why it's OK to rely
175 on the return value of vsnprintf. */
177 va_start (args, fmt);
178 n = vsnprintf (str, size, fmt, args);
181 /* If the printing worked, return the string. */
182 if (n > -1 && n < size)
185 /* Else try again with a larger buffer. */
186 if (n > -1) /* C99 */
187 size = n + 1; /* precisely what is needed */
189 size <<= 1; /* twice the old size */
190 str = xrealloc (str, size);
194 /* Concatenate the NULL-terminated list of string arguments into
195 freshly allocated space. */
198 concat_strings (const char *str0, ...)
201 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
204 const char *next_str;
205 int total_length = 0;
208 /* Calculate the length of and allocate the resulting string. */
211 va_start (args, str0);
212 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
214 int len = strlen (next_str);
215 if (argcount < countof (saved_lengths))
216 saved_lengths[argcount++] = len;
220 p = ret = xmalloc (total_length + 1);
222 /* Copy the strings into the allocated space. */
225 va_start (args, str0);
226 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
229 if (argcount < countof (saved_lengths))
230 len = saved_lengths[argcount++];
232 len = strlen (next_str);
233 memcpy (p, next_str, len);
242 /* Return pointer to a static char[] buffer in which zero-terminated
243 string-representation of TM (in form hh:mm:ss) is printed.
245 If TM is NULL, the current time will be used. */
248 time_str (time_t *tm)
250 static char output[15];
252 time_t secs = tm ? *tm : time (NULL);
256 /* In case of error, return the empty string. Maybe we should
257 just abort if this happens? */
261 ptm = localtime (&secs);
262 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
266 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
269 datetime_str (time_t *tm)
271 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
273 time_t secs = tm ? *tm : time (NULL);
277 /* In case of error, return the empty string. Maybe we should
278 just abort if this happens? */
282 ptm = localtime (&secs);
283 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
284 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
285 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
289 /* The Windows versions of the following two functions are defined in
294 fork_to_background (void)
297 /* Whether we arrange our own version of opt.lfilename here. */
298 bool logfile_changed = false;
302 /* We must create the file immediately to avoid either a race
303 condition (which arises from using unique_name and failing to
304 use fopen_excl) or lying to the user about the log file name
305 (which arises from using unique_name, printing the name, and
306 using fopen_excl later on.) */
307 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
310 logfile_changed = true;
323 /* parent, no error */
324 printf (_("Continuing in background, pid %d.\n"), (int) pid);
326 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
327 exit (0); /* #### should we use _exit()? */
330 /* child: give up the privileges and keep running. */
332 freopen ("/dev/null", "r", stdin);
333 freopen ("/dev/null", "w", stdout);
334 freopen ("/dev/null", "w", stderr);
336 #endif /* not WINDOWS */
338 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
339 specified with TM. The atime ("access time") is set to the current
343 touch (const char *file, time_t tm)
345 #ifdef HAVE_STRUCT_UTIMBUF
346 struct utimbuf times;
354 times.actime = time (NULL);
355 if (utime (file, ×) == -1)
356 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
359 /* Checks if FILE is a symbolic link, and removes it if it is. Does
360 nothing under MS-Windows. */
362 remove_link (const char *file)
367 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
369 DEBUGP (("Unlinking %s (symlink).\n", file));
372 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
373 file, strerror (errno));
378 /* Does FILENAME exist? This is quite a lousy implementation, since
379 it supplies no error codes -- only a yes-or-no answer. Thus it
380 will return that a file does not exist if, e.g., the directory is
381 unreadable. I don't mind it too much currently, though. The
382 proper way should, of course, be to have a third, error state,
383 other than true/false, but that would introduce uncalled-for
384 additional complexity to the callers. */
386 file_exists_p (const char *filename)
389 return access (filename, F_OK) >= 0;
392 return stat (filename, &buf) >= 0;
396 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
397 Returns 0 on error. */
399 file_non_directory_p (const char *path)
402 /* Use lstat() rather than stat() so that symbolic links pointing to
403 directories can be identified correctly. */
404 if (lstat (path, &buf) != 0)
406 return S_ISDIR (buf.st_mode) ? false : true;
409 /* Return the size of file named by FILENAME, or -1 if it cannot be
410 opened or seeked into. */
412 file_size (const char *filename)
414 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
416 /* We use fseek rather than stat to determine the file size because
417 that way we can also verify that the file is readable without
418 explicitly checking for permissions. Inspired by the POST patch
420 FILE *fp = fopen (filename, "rb");
423 fseeko (fp, 0, SEEK_END);
429 if (stat (filename, &st) < 0)
435 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
436 doesn't exist is found. Return a freshly allocated copy of the
440 unique_name_1 (const char *prefix)
443 int plen = strlen (prefix);
444 char *template = (char *)alloca (plen + 1 + 24);
445 char *template_tail = template + plen;
447 memcpy (template, prefix, plen);
448 *template_tail++ = '.';
451 number_to_string (template_tail, count++);
452 while (file_exists_p (template));
454 return xstrdup (template);
457 /* Return a unique file name, based on FILE.
459 More precisely, if FILE doesn't exist, it is returned unmodified.
460 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
461 file name that doesn't exist is returned.
463 The resulting file is not created, only verified that it didn't
464 exist at the point in time when the function was called.
465 Therefore, where security matters, don't rely that the file created
466 by this function exists until you open it with O_EXCL or
469 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
470 string. Otherwise, it may return FILE if the file doesn't exist
471 (and therefore doesn't need changing). */
474 unique_name (const char *file, bool allow_passthrough)
476 /* If the FILE itself doesn't exist, return it without
478 if (!file_exists_p (file))
479 return allow_passthrough ? (char *)file : xstrdup (file);
481 /* Otherwise, find a numeric suffix that results in unused file name
483 return unique_name_1 (file);
486 /* Create a file based on NAME, except without overwriting an existing
487 file with that name. Providing O_EXCL is correctly implemented,
488 this function does not have the race condition associated with
489 opening the file returned by unique_name. */
492 unique_create (const char *name, bool binary, char **opened_name)
494 /* unique file name, based on NAME */
495 char *uname = unique_name (name, false);
497 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
500 uname = unique_name (name, false);
502 if (opened_name && fp != NULL)
505 *opened_name = uname;
517 /* Open the file for writing, with the addition that the file is
518 opened "exclusively". This means that, if the file already exists,
519 this function will *fail* and errno will be set to EEXIST. If
520 BINARY is set, the file will be opened in binary mode, equivalent
523 If opening the file fails for any reason, including the file having
524 previously existed, this function returns NULL and sets errno
528 fopen_excl (const char *fname, bool binary)
532 int flags = O_WRONLY | O_CREAT | O_EXCL;
537 fd = open (fname, flags, 0666);
540 return fdopen (fd, binary ? "wb" : "w");
541 #else /* not O_EXCL */
542 /* Manually check whether the file exists. This is prone to race
543 conditions, but systems without O_EXCL haven't deserved
545 if (file_exists_p (fname))
550 return fopen (fname, binary ? "wb" : "w");
551 #endif /* not O_EXCL */
554 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
555 are missing, create them first. In case any mkdir() call fails,
556 return its error status. Returns 0 on successful completion.
558 The behaviour of this function should be identical to the behaviour
559 of `mkdir -p' on systems where mkdir supports the `-p' option. */
561 make_directory (const char *directory)
563 int i, ret, quit = 0;
566 /* Make a copy of dir, to be able to write to it. Otherwise, the
567 function is unsafe if called with a read-only char *argument. */
568 STRDUP_ALLOCA (dir, directory);
570 /* If the first character of dir is '/', skip it (and thus enable
571 creation of absolute-pathname directories. */
572 for (i = (*dir == '/'); 1; ++i)
574 for (; dir[i] && dir[i] != '/'; i++)
579 /* Check whether the directory already exists. Allow creation of
580 of intermediate directories to fail, as the initial path components
581 are not necessarily directories! */
582 if (!file_exists_p (dir))
583 ret = mkdir (dir, 0777);
594 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
595 should be a file name.
597 file_merge("/foo/bar", "baz") => "/foo/baz"
598 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
599 file_merge("foo", "bar") => "bar"
601 In other words, it's a simpler and gentler version of uri_merge_1. */
604 file_merge (const char *base, const char *file)
607 const char *cut = (const char *)strrchr (base, '/');
610 return xstrdup (file);
612 result = xmalloc (cut - base + 1 + strlen (file) + 1);
613 memcpy (result, base, cut - base);
614 result[cut - base] = '/';
615 strcpy (result + (cut - base) + 1, file);
620 static bool in_acclist (const char *const *, const char *, bool);
622 /* Determine whether a file is acceptable to be followed, according to
623 lists of patterns to accept/reject. */
625 acceptable (const char *s)
629 while (l && s[l] != '/')
636 return (in_acclist ((const char *const *)opt.accepts, s, true)
637 && !in_acclist ((const char *const *)opt.rejects, s, true));
639 return in_acclist ((const char *const *)opt.accepts, s, true);
641 else if (opt.rejects)
642 return !in_acclist ((const char *const *)opt.rejects, s, true);
646 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
647 `/something', frontcmp() will return 1 only if S2 begins with
648 `/something'. Otherwise, 0 is returned. */
650 frontcmp (const char *s1, const char *s2)
652 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
656 /* Iterate through STRLIST, and return the first element that matches
657 S, through wildcards or front comparison (as appropriate). */
659 proclist (char **strlist, const char *s, enum accd flags)
662 for (x = strlist; *x; x++)
664 /* Remove leading '/' if ALLABS */
665 char *p = *x + ((flags & ALLABS) && (**x == '/'));
666 if (has_wildcards_p (p))
668 if (fnmatch (p, s, FNM_PATHNAME) == 0)
680 /* Returns whether DIRECTORY is acceptable for download, wrt the
681 include/exclude lists.
683 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
684 and absolute paths may be freely intermixed. */
686 accdir (const char *directory, enum accd flags)
688 /* Remove starting '/'. */
689 if (flags & ALLABS && *directory == '/')
693 if (!proclist (opt.includes, directory, flags))
698 if (proclist (opt.excludes, directory, flags))
704 /* Return true if STRING ends with TAIL. For instance:
706 match_tail ("abc", "bc", false) -> 1
707 match_tail ("abc", "ab", false) -> 0
708 match_tail ("abc", "abc", false) -> 1
710 If FOLD_CASE is true, the comparison will be case-insensitive. */
713 match_tail (const char *string, const char *tail, bool fold_case)
717 /* We want this to be fast, so we code two loops, one with
718 case-folding, one without. */
722 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
723 if (string[i] != tail[j])
728 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
729 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
733 /* If the tail was exhausted, the match was succesful. */
740 /* Checks whether string S matches each element of ACCEPTS. A list
741 element are matched either with fnmatch() or match_tail(),
742 according to whether the element contains wildcards or not.
744 If the BACKWARD is false, don't do backward comparison -- just compare
747 in_acclist (const char *const *accepts, const char *s, bool backward)
749 for (; *accepts; accepts++)
751 if (has_wildcards_p (*accepts))
753 /* fnmatch returns 0 if the pattern *does* match the
755 if (fnmatch (*accepts, s, 0) == 0)
762 if (match_tail (s, *accepts, 0))
767 if (!strcmp (s, *accepts))
775 /* Return the location of STR's suffix (file extension). Examples:
776 suffix ("foo.bar") -> "bar"
777 suffix ("foo.bar.baz") -> "baz"
778 suffix ("/foo/bar") -> NULL
779 suffix ("/foo.bar/baz") -> NULL */
781 suffix (const char *str)
785 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
789 return (char *)str + i;
794 /* Return true if S contains globbing wildcards (`*', `?', `[' or
798 has_wildcards_p (const char *s)
801 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
806 /* Return true if FNAME ends with a typical HTML suffix. The
807 following (case-insensitive) suffixes are presumed to be HTML
812 ?html (`?' matches one character)
814 #### CAVEAT. This is not necessarily a good indication that FNAME
815 refers to a file that contains HTML! */
817 has_html_suffix_p (const char *fname)
821 if ((suf = suffix (fname)) == NULL)
823 if (!strcasecmp (suf, "html"))
825 if (!strcasecmp (suf, "htm"))
827 if (suf[0] && !strcasecmp (suf + 1, "html"))
832 /* Read a line from FP and return the pointer to freshly allocated
833 storage. The storage space is obtained through malloc() and should
834 be freed with free() when it is no longer needed.
836 The length of the line is not limited, except by available memory.
837 The newline character at the end of line is retained. The line is
838 terminated with a zero character.
840 After end-of-file is encountered without anything being read, NULL
841 is returned. NULL is also returned on error. To distinguish
842 between these two cases, use the stdio function ferror(). */
845 read_whole_line (FILE *fp)
849 char *line = xmalloc (bufsize);
851 while (fgets (line + length, bufsize - length, fp))
853 length += strlen (line + length);
855 /* Possible for example when reading from a binary file where
856 a line begins with \0. */
859 if (line[length - 1] == '\n')
862 /* fgets() guarantees to read the whole line, or to use up the
863 space we've given it. We can double the buffer
866 line = xrealloc (line, bufsize);
868 if (length == 0 || ferror (fp))
873 if (length + 1 < bufsize)
874 /* Relieve the memory from our exponential greediness. We say
875 `length + 1' because the terminating \0 is not included in
876 LENGTH. We don't need to zero-terminate the string ourselves,
877 though, because fgets() does that. */
878 line = xrealloc (line, length + 1);
882 /* Read FILE into memory. A pointer to `struct file_memory' are
883 returned; use struct element `content' to access file contents, and
884 the element `length' to know the file length. `content' is *not*
885 zero-terminated, and you should *not* read or write beyond the [0,
886 length) range of characters.
888 After you are done with the file contents, call read_file_free to
891 Depending on the operating system and the type of file that is
892 being read, read_file() either mmap's the file into memory, or
893 reads the file into the core using read().
895 If file is named "-", fileno(stdin) is used for reading instead.
896 If you want to read from a real file named "-", use "./-" instead. */
899 read_file (const char *file)
902 struct file_memory *fm;
904 bool inhibit_close = false;
906 /* Some magic in the finest tradition of Perl and its kin: if FILE
907 is "-", just use stdin. */
911 inhibit_close = true;
912 /* Note that we don't inhibit mmap() in this case. If stdin is
913 redirected from a regular file, mmap() will still work. */
916 fd = open (file, O_RDONLY);
919 fm = xnew (struct file_memory);
924 if (fstat (fd, &buf) < 0)
926 fm->length = buf.st_size;
927 /* NOTE: As far as I know, the callers of this function never
928 modify the file text. Relying on this would enable us to
929 specify PROT_READ and MAP_SHARED for a marginal gain in
930 efficiency, but at some cost to generality. */
931 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
933 if (fm->content == (char *)MAP_FAILED)
943 /* The most common reason why mmap() fails is that FD does not point
944 to a plain file. However, it's also possible that mmap() doesn't
945 work for a particular type of file. Therefore, whenever mmap()
946 fails, we just fall back to the regular method. */
947 #endif /* HAVE_MMAP */
950 size = 512; /* number of bytes fm->contents can
951 hold at any given time. */
952 fm->content = xmalloc (size);
956 if (fm->length > size / 2)
958 /* #### I'm not sure whether the whole exponential-growth
959 thing makes sense with kernel read. On Linux at least,
960 read() refuses to read more than 4K from a file at a
961 single chunk anyway. But other Unixes might optimize it
962 better, and it doesn't *hurt* anything, so I'm leaving
965 /* Normally, we grow SIZE exponentially to make the number
966 of calls to read() and realloc() logarithmic in relation
967 to file size. However, read() can read an amount of data
968 smaller than requested, and it would be unreasonable to
969 double SIZE every time *something* was read. Therefore,
970 we double SIZE only when the length exceeds half of the
971 entire allocated size. */
973 fm->content = xrealloc (fm->content, size);
975 nread = read (fd, fm->content + fm->length, size - fm->length);
977 /* Successful read. */
988 if (size > fm->length && fm->length != 0)
989 /* Due to exponential growth of fm->content, the allocated region
990 might be much larger than what is actually needed. */
991 fm->content = xrealloc (fm->content, fm->length);
1003 /* Release the resources held by FM. Specifically, this calls
1004 munmap() or xfree() on fm->content, depending whether mmap or
1005 malloc/read were used to read in the file. It also frees the
1006 memory needed to hold the FM structure itself. */
1009 read_file_free (struct file_memory *fm)
1014 munmap (fm->content, fm->length);
1019 xfree (fm->content);
1024 /* Free the pointers in a NULL-terminated vector of pointers, then
1025 free the pointer itself. */
1027 free_vec (char **vec)
1038 /* Append vector V2 to vector V1. The function frees V2 and
1039 reallocates V1 (thus you may not use the contents of neither
1040 pointer after the call). If V1 is NULL, V2 is returned. */
1042 merge_vecs (char **v1, char **v2)
1052 /* To avoid j == 0 */
1057 for (i = 0; v1[i]; i++);
1059 for (j = 0; v2[j]; j++);
1060 /* Reallocate v1. */
1061 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1062 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1067 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1068 is allocated as needed. Return the new value of the vector. */
1071 vec_append (char **vec, const char *str)
1073 int cnt; /* count of vector elements, including
1074 the one we're about to append */
1077 for (cnt = 0; vec[cnt]; cnt++)
1083 /* Reallocate the array to fit the new element and the NULL. */
1084 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1085 /* Append a copy of STR to the vector. */
1086 vec[cnt - 1] = xstrdup (str);
1091 /* Sometimes it's useful to create "sets" of strings, i.e. special
1092 hash tables where you want to store strings as keys and merely
1093 query for their existence. Here is a set of utility routines that
1094 makes that transparent. */
1097 string_set_add (struct hash_table *ht, const char *s)
1099 /* First check whether the set element already exists. If it does,
1100 do nothing so that we don't have to free() the old element and
1101 then strdup() a new one. */
1102 if (hash_table_contains (ht, s))
1105 /* We use "1" as value. It provides us a useful and clear arbitrary
1106 value, and it consumes no memory -- the pointers to the same
1107 string "1" will be shared by all the key-value pairs in all `set'
1109 hash_table_put (ht, xstrdup (s), "1");
1112 /* Synonym for hash_table_contains... */
1115 string_set_contains (struct hash_table *ht, const char *s)
1117 return hash_table_contains (ht, s);
1121 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1123 char ***arrayptr = (char ***) arg;
1124 *(*arrayptr)++ = (char *) key;
1128 /* Convert the specified string set to array. ARRAY should be large
1129 enough to hold hash_table_count(ht) char pointers. */
1131 void string_set_to_array (struct hash_table *ht, char **array)
1133 hash_table_map (ht, string_set_to_array_mapper, &array);
1137 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1144 string_set_free (struct hash_table *ht)
1146 hash_table_map (ht, string_set_free_mapper, NULL);
1147 hash_table_destroy (ht);
1151 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1158 /* Another utility function: call free() on all keys and values of HT. */
1161 free_keys_and_values (struct hash_table *ht)
1163 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1168 get_grouping_data (const char **sep, const char **grouping)
1170 static const char *cached_sep;
1171 static const char *cached_grouping;
1172 static bool initialized;
1176 /* Get the grouping info from the locale. */
1177 struct lconv *lconv;
1178 const char *oldlocale = setlocale (LC_NUMERIC, "");
1179 lconv = localeconv ();
1180 cached_sep = xstrdup (lconv->thousands_sep);
1181 cached_grouping = xstrdup (lconv->grouping);
1182 /* Restore the locale to previous settings. */
1183 setlocale (LC_NUMERIC, oldlocale);
1186 /* Force separator for locales that specify no separators
1187 ("C", "hr", and probably many more.) */
1188 cached_sep = ",", cached_grouping = "\x03";
1192 *grouping = cached_grouping;
1195 /* Return a printed representation of N with thousand separators.
1196 This should respect locale settings, with the exception of the "C"
1197 locale which mandates no separator, but we use one anyway.
1199 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1200 the separators because it's too non-portable, and it's hard to test
1201 for this feature at configure time. Besides, it wouldn't work in
1202 the "C" locale, which many Unix users still work in. */
1205 with_thousand_seps (wgint n)
1207 static char outbuf[48];
1208 char *p = outbuf + sizeof outbuf;
1210 /* Info received from locale */
1211 const char *grouping, *sep;
1214 /* State information */
1215 int i = 0, groupsize;
1216 const char *atgroup;
1218 bool negative = n < 0;
1220 /* Initialize grouping data. */
1221 get_grouping_data (&sep, &grouping);
1222 seplen = strlen (sep);
1224 groupsize = *atgroup++;
1226 /* This will overflow on WGINT_MIN, but we're not using this to
1227 print negative numbers anyway. */
1231 /* Write the number into the buffer, backwards, inserting the
1232 separators as necessary. */
1236 *--p = n % 10 + '0';
1240 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1241 if (++i == groupsize)
1246 memcpy (p -= seplen, sep, seplen);
1249 groupsize = *atgroup++;
1258 /* N, a byte quantity, is converted to a human-readable abberviated
1259 form a la sizes printed by `ls -lh'. The result is written to a
1260 static buffer, a pointer to which is returned.
1262 Unlike `with_thousand_seps', this approximates to the nearest unit.
1263 Quoting GNU libit: "Most people visually process strings of 3-4
1264 digits effectively, but longer strings of digits are more prone to
1265 misinterpretation. Hence, converting to an abbreviated form
1266 usually improves readability."
1268 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1269 original computer-related meaning of "powers of 1024". Powers of
1270 1000 would be useless since Wget already displays sizes with
1271 thousand separators. We don't use the "*bibyte" names invented in
1272 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1273 discusses this in some detail. */
1276 human_readable (HR_NUMTYPE n)
1278 /* These suffixes are compatible with those of GNU `ls -lh'. */
1279 static char powers[] =
1281 'K', /* kilobyte, 2^10 bytes */
1282 'M', /* megabyte, 2^20 bytes */
1283 'G', /* gigabyte, 2^30 bytes */
1284 'T', /* terabyte, 2^40 bytes */
1285 'P', /* petabyte, 2^50 bytes */
1286 'E', /* exabyte, 2^60 bytes */
1291 /* If the quantity is smaller than 1K, just print it. */
1294 snprintf (buf, sizeof (buf), "%d", (int) n);
1298 /* Loop over powers, dividing N with 1024 in each iteration. This
1299 works unchanged for all sizes of wgint, while still avoiding
1300 non-portable `long double' arithmetic. */
1301 for (i = 0; i < countof (powers); i++)
1303 /* At each iteration N is greater than the *subsequent* power.
1304 That way N/1024.0 produces a decimal number in the units of
1306 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1308 /* Must cast to long first because MS VC can't directly cast
1309 __int64 to double. (This is safe because N is known to
1310 be < 1024^2, so always fits into long.) */
1311 double val = (double) (long) n / 1024.0;
1312 /* Print values smaller than 10 with one decimal digits, and
1313 others without any decimals. */
1314 snprintf (buf, sizeof (buf), "%.*f%c",
1315 val < 10 ? 1 : 0, val, powers[i]);
1320 return NULL; /* unreached */
1323 /* Count the digits in the provided number. Used to allocate space
1324 when printing numbers. */
1327 numdigit (wgint number)
1331 ++cnt; /* accomodate '-' */
1332 while ((number /= 10) != 0)
1337 #define PR(mask) *p++ = n / (mask) + '0'
1339 /* DIGITS_<D> is used to print a D-digit number and should be called
1340 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1341 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1342 Recursively this continues until DIGITS_1 is invoked. */
1344 #define DIGITS_1(mask) PR (mask)
1345 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1346 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1347 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1348 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1349 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1350 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1351 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1352 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1353 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1355 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1357 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1358 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1359 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1360 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1361 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1362 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1363 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1364 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1365 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1367 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1368 cases and to portably support strange sizes of wgint. Ideally this
1369 would just use "%j" and intmax_t, but many systems don't support
1370 it, so it's used only if nothing else works. */
1371 #if SIZEOF_LONG >= SIZEOF_WGINT
1372 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1373 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1374 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1375 #elif defined(WINDOWS)
1376 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64d", (__int64) (n))
1378 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1381 /* Shorthand for casting to wgint. */
1384 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1385 `sprintf(buffer, "%lld", (long long) number)', only typically much
1386 faster and portable to machines without long long.
1388 The speedup may make a difference in programs that frequently
1389 convert numbers to strings. Some implementations of sprintf,
1390 particularly the one in GNU libc, have been known to be extremely
1391 slow when converting integers to strings.
1393 Return the pointer to the location where the terminating zero was
1394 printed. (Equivalent to calling buffer+strlen(buffer) after the
1397 BUFFER should be big enough to accept as many bytes as you expect
1398 the number to take up. On machines with 64-bit longs the maximum
1399 needed size is 24 bytes. That includes the digits needed for the
1400 largest 64-bit number, the `-' sign in case it's negative, and the
1401 terminating '\0'. */
1404 number_to_string (char *buffer, wgint number)
1409 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1410 /* We are running in a strange or misconfigured environment. Let
1411 sprintf cope with it. */
1412 SPRINTF_WGINT (buffer, n);
1413 p += strlen (buffer);
1414 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1420 /* -n would overflow. Have sprintf deal with this. */
1421 SPRINTF_WGINT (buffer, n);
1422 p += strlen (buffer);
1430 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1431 way printing any N is fully open-coded without a loop or jump.
1432 (Also see description of DIGITS_*.) */
1434 if (n < 10) DIGITS_1 (1);
1435 else if (n < 100) DIGITS_2 (10);
1436 else if (n < 1000) DIGITS_3 (100);
1437 else if (n < 10000) DIGITS_4 (1000);
1438 else if (n < 100000) DIGITS_5 (10000);
1439 else if (n < 1000000) DIGITS_6 (100000);
1440 else if (n < 10000000) DIGITS_7 (1000000);
1441 else if (n < 100000000) DIGITS_8 (10000000);
1442 else if (n < 1000000000) DIGITS_9 (100000000);
1443 #if SIZEOF_WGINT == 4
1444 /* wgint is 32 bits wide: no number has more than 10 digits. */
1445 else DIGITS_10 (1000000000);
1447 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1448 Constants are constructed by compile-time multiplication to avoid
1449 dealing with different notations for 64-bit constants
1450 (nL/nLL/nI64, depending on the compiler and architecture). */
1451 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1452 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1453 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1454 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1455 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1456 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1457 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1458 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1459 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1460 else DIGITS_19 (1000000000*(W)1000000000);
1464 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1493 /* Print NUMBER to a statically allocated string and return a pointer
1494 to the printed representation.
1496 This function is intended to be used in conjunction with printf.
1497 It is hard to portably print wgint values:
1498 a) you cannot use printf("%ld", number) because wgint can be long
1499 long on 32-bit machines with LFS.
1500 b) you cannot use printf("%lld", number) because NUMBER could be
1501 long on 32-bit machines without LFS, or on 64-bit machines,
1502 which do not require LFS. Also, Windows doesn't support %lld.
1503 c) you cannot use printf("%j", (int_max_t) number) because not all
1504 versions of printf support "%j", the most notable being the one
1506 d) you cannot #define WGINT_FMT to the appropriate format and use
1507 printf(WGINT_FMT, number) because that would break translations
1508 for user-visible messages, such as printf("Downloaded: %d
1511 What you should use instead is printf("%s", number_to_static_string
1514 CAVEAT: since the function returns pointers to static data, you
1515 must be careful to copy its result before calling it again.
1516 However, to make it more useful with printf, the function maintains
1517 an internal ring of static buffers to return. That way things like
1518 printf("%s %s", number_to_static_string (num1),
1519 number_to_static_string (num2)) work as expected. Three buffers
1520 are currently used, which means that "%s %s %s" will work, but "%s
1521 %s %s %s" won't. If you need to print more than three wgints,
1522 bump the RING_SIZE (or rethink your message.) */
1525 number_to_static_string (wgint number)
1527 static char ring[RING_SIZE][24];
1529 char *buf = ring[ringpos];
1530 number_to_string (buf, number);
1531 ringpos = (ringpos + 1) % RING_SIZE;
1535 /* Determine the width of the terminal we're running on. If that's
1536 not possible, return 0. */
1539 determine_screen_width (void)
1541 /* If there's a way to get the terminal size using POSIX
1542 tcgetattr(), somebody please tell me. */
1547 if (opt.lfilename != NULL)
1550 fd = fileno (stderr);
1551 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1552 return 0; /* most likely ENOTTY */
1555 #elif defined(WINDOWS)
1556 CONSOLE_SCREEN_BUFFER_INFO csbi;
1557 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1559 return csbi.dwSize.X;
1560 #else /* neither TIOCGWINSZ nor WINDOWS */
1562 #endif /* neither TIOCGWINSZ nor WINDOWS */
1565 /* Return a random number between 0 and MAX-1, inclusive.
1567 If MAX is greater than the value of RAND_MAX+1 on the system, the
1568 returned value will be in the range [0, RAND_MAX]. This may be
1569 fixed in a future release.
1571 The random number generator is seeded automatically the first time
1574 This uses rand() for portability. It has been suggested that
1575 random() offers better randomness, but this is not required for
1576 Wget, so I chose to go for simplicity and use rand
1579 DO NOT use this for cryptographic purposes. It is only meant to be
1580 used in situations where quality of the random numbers returned
1581 doesn't really matter. */
1584 random_number (int max)
1592 srand (time (NULL));
1597 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1598 and enforce that assumption by masking other bits. */
1600 # define RAND_MAX 32767
1604 /* This is equivalent to rand() % max, but uses the high-order bits
1605 for better randomness on architecture where rand() is implemented
1606 using a simple congruential generator. */
1608 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1609 return (int)bounded;
1612 /* Return a random uniformly distributed floating point number in the
1613 [0, 1) range. The precision of returned numbers is 9 digits.
1615 Modify this to use erand48() where available! */
1620 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1621 sure it's greater than 1000. */
1622 int rnd1 = random_number (1000);
1623 int rnd2 = random_number (1000);
1624 int rnd3 = random_number (1000);
1625 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1628 /* Implementation of run_with_timeout, a generic timeout-forcing
1629 routine for systems with Unix-like signal handling. */
1631 #ifdef USE_SIGNAL_TIMEOUT
1632 # ifdef HAVE_SIGSETJMP
1633 # define SETJMP(env) sigsetjmp (env, 1)
1635 static sigjmp_buf run_with_timeout_env;
1638 abort_run_with_timeout (int sig)
1640 assert (sig == SIGALRM);
1641 siglongjmp (run_with_timeout_env, -1);
1643 # else /* not HAVE_SIGSETJMP */
1644 # define SETJMP(env) setjmp (env)
1646 static jmp_buf run_with_timeout_env;
1649 abort_run_with_timeout (int sig)
1651 assert (sig == SIGALRM);
1652 /* We don't have siglongjmp to preserve the set of blocked signals;
1653 if we longjumped out of the handler at this point, SIGALRM would
1654 remain blocked. We must unblock it manually. */
1655 int mask = siggetmask ();
1656 mask &= ~sigmask (SIGALRM);
1659 /* Now it's safe to longjump. */
1660 longjmp (run_with_timeout_env, -1);
1662 # endif /* not HAVE_SIGSETJMP */
1664 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1665 setitimer where available, alarm otherwise.
1667 TIMEOUT should be non-zero. If the timeout value is so small that
1668 it would be rounded to zero, it is rounded to the least legal value
1669 instead (1us for setitimer, 1s for alarm). That ensures that
1670 SIGALRM will be delivered in all cases. */
1673 alarm_set (double timeout)
1676 /* Use the modern itimer interface. */
1677 struct itimerval itv;
1679 itv.it_value.tv_sec = (long) timeout;
1680 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1681 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1682 /* Ensure that we wait for at least the minimum interval.
1683 Specifying zero would mean "wait forever". */
1684 itv.it_value.tv_usec = 1;
1685 setitimer (ITIMER_REAL, &itv, NULL);
1686 #else /* not ITIMER_REAL */
1687 /* Use the old alarm() interface. */
1688 int secs = (int) timeout;
1690 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1691 because alarm(0) means "never deliver the alarm", i.e. "wait
1692 forever", which is not what someone who specifies a 0.5s
1693 timeout would expect. */
1696 #endif /* not ITIMER_REAL */
1699 /* Cancel the alarm set with alarm_set. */
1705 struct itimerval disable;
1707 setitimer (ITIMER_REAL, &disable, NULL);
1708 #else /* not ITIMER_REAL */
1710 #endif /* not ITIMER_REAL */
1713 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1714 seconds. Returns true if the function was interrupted with a
1715 timeout, false otherwise.
1717 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1718 using setitimer() or alarm(). The timeout is enforced by
1719 longjumping out of the SIGALRM handler. This has several
1720 advantages compared to the traditional approach of relying on
1721 signals causing system calls to exit with EINTR:
1723 * The callback function is *forcibly* interrupted after the
1724 timeout expires, (almost) regardless of what it was doing and
1725 whether it was in a syscall. For example, a calculation that
1726 takes a long time is interrupted as reliably as an IO
1729 * It works with both SYSV and BSD signals because it doesn't
1730 depend on the default setting of SA_RESTART.
1732 * It doesn't require special handler setup beyond a simple call
1733 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1736 The only downside is that, if FUN allocates internal resources that
1737 are normally freed prior to exit from the functions, they will be
1738 lost in case of timeout. */
1741 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1751 signal (SIGALRM, abort_run_with_timeout);
1752 if (SETJMP (run_with_timeout_env) != 0)
1754 /* Longjumped out of FUN with a timeout. */
1755 signal (SIGALRM, SIG_DFL);
1758 alarm_set (timeout);
1761 /* Preserve errno in case alarm() or signal() modifies it. */
1762 saved_errno = errno;
1764 signal (SIGALRM, SIG_DFL);
1765 errno = saved_errno;
1770 #else /* not USE_SIGNAL_TIMEOUT */
1773 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1774 define it under Windows, because Windows has its own version of
1775 run_with_timeout that uses threads. */
1778 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1783 #endif /* not WINDOWS */
1784 #endif /* not USE_SIGNAL_TIMEOUT */
1788 /* Sleep the specified amount of seconds. On machines without
1789 nanosleep(), this may sleep shorter if interrupted by signals. */
1792 xsleep (double seconds)
1794 #ifdef HAVE_NANOSLEEP
1795 /* nanosleep is the preferred interface because it offers high
1796 accuracy and, more importantly, because it allows us to reliably
1797 restart receiving a signal such as SIGWINCH. (There was an
1798 actual Debian bug report about --limit-rate malfunctioning while
1799 the terminal was being resized.) */
1800 struct timespec sleep, remaining;
1801 sleep.tv_sec = (long) seconds;
1802 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1803 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1804 /* If nanosleep has been interrupted by a signal, adjust the
1805 sleeping period and return to sleep. */
1807 #elif defined(HAVE_USLEEP)
1808 /* If usleep is available, use it in preference to select. */
1811 /* On some systems, usleep cannot handle values larger than
1812 1,000,000. If the period is larger than that, use sleep
1813 first, then add usleep for subsecond accuracy. */
1815 seconds -= (long) seconds;
1817 usleep (seconds * 1000000);
1818 #else /* fall back select */
1819 /* Note that, although Windows supports select, it can't be used to
1820 implement sleeping because Winsock's select doesn't implement
1821 timeout when it is passed NULL pointers for all fd sets. (But it
1822 does under Cygwin, which implements Unix-compatible select.) */
1823 struct timeval sleep;
1824 sleep.tv_sec = (long) seconds;
1825 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1826 select (0, NULL, NULL, NULL, &sleep);
1827 /* If select returns -1 and errno is EINTR, it means we were
1828 interrupted by a signal. But without knowing how long we've
1829 actually slept, we can't return to sleep. Using gettimeofday to
1830 track sleeps is slow and unreliable due to clock skew. */
1834 #endif /* not WINDOWS */
1836 /* Encode the string STR of length LENGTH to base64 format and place it
1837 to B64STORE. The output will be \0-terminated, and must point to a
1838 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1839 returns the length of the resulting base64 data, not counting the
1842 This implementation will not emit newlines after 76 characters of
1846 base64_encode (const char *str, int length, char *b64store)
1848 /* Conversion table. */
1849 static char tbl[64] = {
1850 'A','B','C','D','E','F','G','H',
1851 'I','J','K','L','M','N','O','P',
1852 'Q','R','S','T','U','V','W','X',
1853 'Y','Z','a','b','c','d','e','f',
1854 'g','h','i','j','k','l','m','n',
1855 'o','p','q','r','s','t','u','v',
1856 'w','x','y','z','0','1','2','3',
1857 '4','5','6','7','8','9','+','/'
1860 const unsigned char *s = (const unsigned char *) str;
1863 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1864 for (i = 0; i < length; i += 3)
1866 *p++ = tbl[s[0] >> 2];
1867 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1868 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1869 *p++ = tbl[s[2] & 0x3f];
1873 /* Pad the result if necessary... */
1874 if (i == length + 1)
1876 else if (i == length + 2)
1877 *(p - 1) = *(p - 2) = '=';
1879 /* ...and zero-terminate it. */
1882 return p - b64store;
1885 #define IS_ASCII(c) (((c) & 0x80) == 0)
1886 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1888 /* Get next character from the string, except that non-base64
1889 characters are ignored, as mandated by rfc2045. */
1890 #define NEXT_BASE64_CHAR(c, p) do { \
1892 } while (c != '\0' && !IS_BASE64 (c))
1894 /* Decode data from BASE64 (assumed to be encoded as base64) into
1895 memory pointed to by TO. TO should be large enough to accomodate
1896 the decoded data, which is guaranteed to be less than
1899 Since TO is assumed to contain binary data, it is not
1900 NUL-terminated. The function returns the length of the data
1901 written to TO. -1 is returned in case of error caused by malformed
1905 base64_decode (const char *base64, char *to)
1907 /* Table of base64 values for first 128 characters. Note that this
1908 assumes ASCII (but so does Wget in other places). */
1909 static short base64_char_to_value[128] =
1911 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1912 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1913 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1914 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1915 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1916 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1917 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1918 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1919 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1920 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1921 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1922 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1923 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1926 const char *p = base64;
1932 unsigned long value;
1934 /* Process first byte of a quadruplet. */
1935 NEXT_BASE64_CHAR (c, p);
1939 return -1; /* illegal '=' while decoding base64 */
1940 value = base64_char_to_value[c] << 18;
1942 /* Process scond byte of a quadruplet. */
1943 NEXT_BASE64_CHAR (c, p);
1945 return -1; /* premature EOF while decoding base64 */
1947 return -1; /* illegal `=' while decoding base64 */
1948 value |= base64_char_to_value[c] << 12;
1951 /* Process third byte of a quadruplet. */
1952 NEXT_BASE64_CHAR (c, p);
1954 return -1; /* premature EOF while decoding base64 */
1958 NEXT_BASE64_CHAR (c, p);
1960 return -1; /* premature EOF while decoding base64 */
1962 return -1; /* padding `=' expected but not found */
1966 value |= base64_char_to_value[c] << 6;
1967 *q++ = 0xff & value >> 8;
1969 /* Process fourth byte of a quadruplet. */
1970 NEXT_BASE64_CHAR (c, p);
1972 return -1; /* premature EOF while decoding base64 */
1976 value |= base64_char_to_value[c];
1977 *q++ = 0xff & value;
1985 #undef NEXT_BASE64_CHAR
1987 /* Simple merge sort for use by stable_sort. Implementation courtesy
1988 Zeljko Vrba with additional debugging by Nenad Barbutov. */
1991 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1992 int (*cmpfun) (const void *, const void *))
1994 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1998 size_t mid = (to + from) / 2;
1999 mergesort_internal (base, temp, size, from, mid, cmpfun);
2000 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2003 for (k = from; (i <= mid) && (j <= to); k++)
2004 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2005 memcpy (ELT (temp, k), ELT (base, i++), size);
2007 memcpy (ELT (temp, k), ELT (base, j++), size);
2009 memcpy (ELT (temp, k++), ELT (base, i++), size);
2011 memcpy (ELT (temp, k++), ELT (base, j++), size);
2012 for (k = from; k <= to; k++)
2013 memcpy (ELT (base, k), ELT (temp, k), size);
2018 /* Stable sort with interface exactly like standard library's qsort.
2019 Uses mergesort internally, allocating temporary storage with
2023 stable_sort (void *base, size_t nmemb, size_t size,
2024 int (*cmpfun) (const void *, const void *))
2028 void *temp = alloca (nmemb * size * sizeof (void *));
2029 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);