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 #else /* not HAVE_STRING_H */
38 #endif /* not HAVE_STRING_H */
39 #include <sys/types.h>
44 # include <sys/mman.h>
55 #ifdef HAVE_SYS_UTIME_H
56 # include <sys/utime.h>
60 # include <libc.h> /* for access() */
64 #ifdef WGET_USE_STDARG
70 /* For TIOCGWINSZ and friends: */
71 #ifdef HAVE_SYS_IOCTL_H
72 # include <sys/ioctl.h>
78 /* Needed for run_with_timeout. */
79 #undef USE_SIGNAL_TIMEOUT
87 #ifndef HAVE_SIGSETJMP
88 /* If sigsetjmp is a macro, configure won't pick it up. */
90 # define HAVE_SIGSETJMP
95 # ifdef HAVE_SIGSETJMP
96 # define USE_SIGNAL_TIMEOUT
99 # define USE_SIGNAL_TIMEOUT
111 /* Utility function: like xstrdup(), but also lowercases S. */
114 xstrdup_lower (const char *s)
116 char *copy = xstrdup (s);
123 /* Copy the string formed by two pointers (one on the beginning, other
124 on the char after the last char) to a new, malloc-ed location.
127 strdupdelim (const char *beg, const char *end)
129 char *res = (char *)xmalloc (end - beg + 1);
130 memcpy (res, beg, end - beg);
131 res[end - beg] = '\0';
135 /* Parse a string containing comma-separated elements, and return a
136 vector of char pointers with the elements. Spaces following the
137 commas are ignored. */
139 sepstring (const char *s)
153 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
154 res[i] = strdupdelim (p, s);
157 /* Skip the blanks following the ','. */
165 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
166 res[i] = strdupdelim (p, s);
171 #ifdef WGET_USE_STDARG
172 # define VA_START(args, arg1) va_start (args, arg1)
174 # define VA_START(args, ignored) va_start (args)
177 /* Like sprintf, but allocates a string of sufficient size with malloc
178 and returns it. GNU libc has a similar function named asprintf,
179 which requires the pointer to the string to be passed. */
182 aprintf (const char *fmt, ...)
184 /* This function is implemented using vsnprintf, which we provide
185 for the systems that don't have it. Therefore, it should be 100%
189 char *str = xmalloc (size);
196 /* See log_vprintf_internal for explanation why it's OK to rely
197 on the return value of vsnprintf. */
199 VA_START (args, fmt);
200 n = vsnprintf (str, size, fmt, args);
203 /* If the printing worked, return the string. */
204 if (n > -1 && n < size)
207 /* Else try again with a larger buffer. */
208 if (n > -1) /* C99 */
209 size = n + 1; /* precisely what is needed */
211 size <<= 1; /* twice the old size */
212 str = xrealloc (str, size);
216 /* Concatenate the NULL-terminated list of string arguments into
217 freshly allocated space. */
220 concat_strings (const char *str0, ...)
223 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
226 const char *next_str;
227 int total_length = 0;
230 /* Calculate the length of and allocate the resulting string. */
233 VA_START (args, str0);
234 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
236 int len = strlen (next_str);
237 if (argcount < countof (saved_lengths))
238 saved_lengths[argcount++] = len;
242 p = ret = xmalloc (total_length + 1);
244 /* Copy the strings into the allocated space. */
247 VA_START (args, str0);
248 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
251 if (argcount < countof (saved_lengths))
252 len = saved_lengths[argcount++];
254 len = strlen (next_str);
255 memcpy (p, next_str, len);
264 /* Return pointer to a static char[] buffer in which zero-terminated
265 string-representation of TM (in form hh:mm:ss) is printed.
267 If TM is NULL, the current time will be used. */
270 time_str (time_t *tm)
272 static char output[15];
274 time_t secs = tm ? *tm : time (NULL);
278 /* In case of error, return the empty string. Maybe we should
279 just abort if this happens? */
283 ptm = localtime (&secs);
284 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
288 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
291 datetime_str (time_t *tm)
293 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
295 time_t secs = tm ? *tm : time (NULL);
299 /* In case of error, return the empty string. Maybe we should
300 just abort if this happens? */
304 ptm = localtime (&secs);
305 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
306 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
307 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
311 /* The Windows versions of the following two functions are defined in
316 fork_to_background (void)
319 /* Whether we arrange our own version of opt.lfilename here. */
320 int logfile_changed = 0;
324 /* We must create the file immediately to avoid either a race
325 condition (which arises from using unique_name and failing to
326 use fopen_excl) or lying to the user about the log file name
327 (which arises from using unique_name, printing the name, and
328 using fopen_excl later on.) */
329 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
345 /* parent, no error */
346 printf (_("Continuing in background, pid %d.\n"), (int)pid);
348 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
349 exit (0); /* #### should we use _exit()? */
352 /* child: give up the privileges and keep running. */
354 freopen ("/dev/null", "r", stdin);
355 freopen ("/dev/null", "w", stdout);
356 freopen ("/dev/null", "w", stderr);
358 #endif /* not WINDOWS */
360 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
361 specified with TM. The atime ("access time") is set to the current
365 touch (const char *file, time_t tm)
367 #ifdef HAVE_STRUCT_UTIMBUF
368 struct utimbuf times;
376 times.actime = time (NULL);
377 if (utime (file, ×) == -1)
378 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
381 /* Checks if FILE is a symbolic link, and removes it if it is. Does
382 nothing under MS-Windows. */
384 remove_link (const char *file)
389 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
391 DEBUGP (("Unlinking %s (symlink).\n", file));
394 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
395 file, strerror (errno));
400 /* Does FILENAME exist? This is quite a lousy implementation, since
401 it supplies no error codes -- only a yes-or-no answer. Thus it
402 will return that a file does not exist if, e.g., the directory is
403 unreadable. I don't mind it too much currently, though. The
404 proper way should, of course, be to have a third, error state,
405 other than true/false, but that would introduce uncalled-for
406 additional complexity to the callers. */
408 file_exists_p (const char *filename)
411 return access (filename, F_OK) >= 0;
414 return stat (filename, &buf) >= 0;
418 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
419 Returns 0 on error. */
421 file_non_directory_p (const char *path)
424 /* Use lstat() rather than stat() so that symbolic links pointing to
425 directories can be identified correctly. */
426 if (lstat (path, &buf) != 0)
428 return S_ISDIR (buf.st_mode) ? 0 : 1;
431 /* Return the size of file named by FILENAME, or -1 if it cannot be
432 opened or seeked into. */
434 file_size (const char *filename)
436 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
438 /* We use fseek rather than stat to determine the file size because
439 that way we can also verify that the file is readable without
440 explicitly checking for permissions. Inspired by the POST patch
442 FILE *fp = fopen (filename, "rb");
445 fseeko (fp, 0, SEEK_END);
451 if (stat (filename, &st) < 0)
457 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
458 doesn't exist is found. Return a freshly allocated copy of the
462 unique_name_1 (const char *prefix)
465 int plen = strlen (prefix);
466 char *template = (char *)alloca (plen + 1 + 24);
467 char *template_tail = template + plen;
469 memcpy (template, prefix, plen);
470 *template_tail++ = '.';
473 number_to_string (template_tail, count++);
474 while (file_exists_p (template));
476 return xstrdup (template);
479 /* Return a unique file name, based on FILE.
481 More precisely, if FILE doesn't exist, it is returned unmodified.
482 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
483 file name that doesn't exist is returned.
485 The resulting file is not created, only verified that it didn't
486 exist at the point in time when the function was called.
487 Therefore, where security matters, don't rely that the file created
488 by this function exists until you open it with O_EXCL or
491 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
492 string. Otherwise, it may return FILE if the file doesn't exist
493 (and therefore doesn't need changing). */
496 unique_name (const char *file, int allow_passthrough)
498 /* If the FILE itself doesn't exist, return it without
500 if (!file_exists_p (file))
501 return allow_passthrough ? (char *)file : xstrdup (file);
503 /* Otherwise, find a numeric suffix that results in unused file name
505 return unique_name_1 (file);
508 /* Create a file based on NAME, except without overwriting an existing
509 file with that name. Providing O_EXCL is correctly implemented,
510 this function does not have the race condition associated with
511 opening the file returned by unique_name. */
514 unique_create (const char *name, int binary, char **opened_name)
516 /* unique file name, based on NAME */
517 char *uname = unique_name (name, 0);
519 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
522 uname = unique_name (name, 0);
524 if (opened_name && fp != NULL)
527 *opened_name = uname;
539 /* Open the file for writing, with the addition that the file is
540 opened "exclusively". This means that, if the file already exists,
541 this function will *fail* and errno will be set to EEXIST. If
542 BINARY is set, the file will be opened in binary mode, equivalent
545 If opening the file fails for any reason, including the file having
546 previously existed, this function returns NULL and sets errno
550 fopen_excl (const char *fname, int binary)
554 int flags = O_WRONLY | O_CREAT | O_EXCL;
559 fd = open (fname, flags, 0666);
562 return fdopen (fd, binary ? "wb" : "w");
563 #else /* not O_EXCL */
564 /* Manually check whether the file exists. This is prone to race
565 conditions, but systems without O_EXCL haven't deserved
567 if (file_exists_p (fname))
572 return fopen (fname, binary ? "wb" : "w");
573 #endif /* not O_EXCL */
576 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
577 are missing, create them first. In case any mkdir() call fails,
578 return its error status. Returns 0 on successful completion.
580 The behaviour of this function should be identical to the behaviour
581 of `mkdir -p' on systems where mkdir supports the `-p' option. */
583 make_directory (const char *directory)
585 int i, ret, quit = 0;
588 /* Make a copy of dir, to be able to write to it. Otherwise, the
589 function is unsafe if called with a read-only char *argument. */
590 STRDUP_ALLOCA (dir, directory);
592 /* If the first character of dir is '/', skip it (and thus enable
593 creation of absolute-pathname directories. */
594 for (i = (*dir == '/'); 1; ++i)
596 for (; dir[i] && dir[i] != '/'; i++)
601 /* Check whether the directory already exists. Allow creation of
602 of intermediate directories to fail, as the initial path components
603 are not necessarily directories! */
604 if (!file_exists_p (dir))
605 ret = mkdir (dir, 0777);
616 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
617 should be a file name.
619 file_merge("/foo/bar", "baz") => "/foo/baz"
620 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
621 file_merge("foo", "bar") => "bar"
623 In other words, it's a simpler and gentler version of uri_merge_1. */
626 file_merge (const char *base, const char *file)
629 const char *cut = (const char *)strrchr (base, '/');
632 return xstrdup (file);
634 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
635 memcpy (result, base, cut - base);
636 result[cut - base] = '/';
637 strcpy (result + (cut - base) + 1, file);
642 static int in_acclist PARAMS ((const char *const *, const char *, int));
644 /* Determine whether a file is acceptable to be followed, according to
645 lists of patterns to accept/reject. */
647 acceptable (const char *s)
651 while (l && s[l] != '/')
658 return (in_acclist ((const char *const *)opt.accepts, s, 1)
659 && !in_acclist ((const char *const *)opt.rejects, s, 1));
661 return in_acclist ((const char *const *)opt.accepts, s, 1);
663 else if (opt.rejects)
664 return !in_acclist ((const char *const *)opt.rejects, s, 1);
668 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
669 `/something', frontcmp() will return 1 only if S2 begins with
670 `/something'. Otherwise, 0 is returned. */
672 frontcmp (const char *s1, const char *s2)
674 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
678 /* Iterate through STRLIST, and return the first element that matches
679 S, through wildcards or front comparison (as appropriate). */
681 proclist (char **strlist, const char *s, enum accd flags)
684 for (x = strlist; *x; x++)
686 /* Remove leading '/' if ALLABS */
687 char *p = *x + ((flags & ALLABS) && (**x == '/'));
688 if (has_wildcards_p (p))
690 if (fnmatch (p, s, FNM_PATHNAME) == 0)
702 /* Returns whether DIRECTORY is acceptable for download, wrt the
703 include/exclude lists.
705 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
706 and absolute paths may be freely intermixed. */
708 accdir (const char *directory, enum accd flags)
710 /* Remove starting '/'. */
711 if (flags & ALLABS && *directory == '/')
715 if (!proclist (opt.includes, directory, flags))
720 if (proclist (opt.excludes, directory, flags))
726 /* Return non-zero if STRING ends with TAIL. For instance:
728 match_tail ("abc", "bc", 0) -> 1
729 match_tail ("abc", "ab", 0) -> 0
730 match_tail ("abc", "abc", 0) -> 1
732 If FOLD_CASE_P is non-zero, the comparison will be
736 match_tail (const char *string, const char *tail, int fold_case_p)
740 /* We want this to be fast, so we code two loops, one with
741 case-folding, one without. */
745 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
746 if (string[i] != tail[j])
751 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
752 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
756 /* If the tail was exhausted, the match was succesful. */
763 /* Checks whether string S matches each element of ACCEPTS. A list
764 element are matched either with fnmatch() or match_tail(),
765 according to whether the element contains wildcards or not.
767 If the BACKWARD is 0, don't do backward comparison -- just compare
770 in_acclist (const char *const *accepts, const char *s, int backward)
772 for (; *accepts; accepts++)
774 if (has_wildcards_p (*accepts))
776 /* fnmatch returns 0 if the pattern *does* match the
778 if (fnmatch (*accepts, s, 0) == 0)
785 if (match_tail (s, *accepts, 0))
790 if (!strcmp (s, *accepts))
798 /* Return the location of STR's suffix (file extension). Examples:
799 suffix ("foo.bar") -> "bar"
800 suffix ("foo.bar.baz") -> "baz"
801 suffix ("/foo/bar") -> NULL
802 suffix ("/foo.bar/baz") -> NULL */
804 suffix (const char *str)
808 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
812 return (char *)str + i;
817 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
821 has_wildcards_p (const char *s)
824 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
829 /* Return non-zero if FNAME ends with a typical HTML suffix. The
830 following (case-insensitive) suffixes are presumed to be HTML files:
834 ?html (`?' matches one character)
836 #### CAVEAT. This is not necessarily a good indication that FNAME
837 refers to a file that contains HTML! */
839 has_html_suffix_p (const char *fname)
843 if ((suf = suffix (fname)) == NULL)
845 if (!strcasecmp (suf, "html"))
847 if (!strcasecmp (suf, "htm"))
849 if (suf[0] && !strcasecmp (suf + 1, "html"))
854 /* Read a line from FP and return the pointer to freshly allocated
855 storage. The storage space is obtained through malloc() and should
856 be freed with free() when it is no longer needed.
858 The length of the line is not limited, except by available memory.
859 The newline character at the end of line is retained. The line is
860 terminated with a zero character.
862 After end-of-file is encountered without anything being read, NULL
863 is returned. NULL is also returned on error. To distinguish
864 between these two cases, use the stdio function ferror(). */
867 read_whole_line (FILE *fp)
871 char *line = (char *)xmalloc (bufsize);
873 while (fgets (line + length, bufsize - length, fp))
875 length += strlen (line + length);
877 /* Possible for example when reading from a binary file where
878 a line begins with \0. */
881 if (line[length - 1] == '\n')
884 /* fgets() guarantees to read the whole line, or to use up the
885 space we've given it. We can double the buffer
888 line = xrealloc (line, bufsize);
890 if (length == 0 || ferror (fp))
895 if (length + 1 < bufsize)
896 /* Relieve the memory from our exponential greediness. We say
897 `length + 1' because the terminating \0 is not included in
898 LENGTH. We don't need to zero-terminate the string ourselves,
899 though, because fgets() does that. */
900 line = xrealloc (line, length + 1);
904 /* Read FILE into memory. A pointer to `struct file_memory' are
905 returned; use struct element `content' to access file contents, and
906 the element `length' to know the file length. `content' is *not*
907 zero-terminated, and you should *not* read or write beyond the [0,
908 length) range of characters.
910 After you are done with the file contents, call read_file_free to
913 Depending on the operating system and the type of file that is
914 being read, read_file() either mmap's the file into memory, or
915 reads the file into the core using read().
917 If file is named "-", fileno(stdin) is used for reading instead.
918 If you want to read from a real file named "-", use "./-" instead. */
921 read_file (const char *file)
924 struct file_memory *fm;
926 int inhibit_close = 0;
928 /* Some magic in the finest tradition of Perl and its kin: if FILE
929 is "-", just use stdin. */
934 /* Note that we don't inhibit mmap() in this case. If stdin is
935 redirected from a regular file, mmap() will still work. */
938 fd = open (file, O_RDONLY);
941 fm = xnew (struct file_memory);
946 if (fstat (fd, &buf) < 0)
948 fm->length = buf.st_size;
949 /* NOTE: As far as I know, the callers of this function never
950 modify the file text. Relying on this would enable us to
951 specify PROT_READ and MAP_SHARED for a marginal gain in
952 efficiency, but at some cost to generality. */
953 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
955 if (fm->content == (char *)MAP_FAILED)
965 /* The most common reason why mmap() fails is that FD does not point
966 to a plain file. However, it's also possible that mmap() doesn't
967 work for a particular type of file. Therefore, whenever mmap()
968 fails, we just fall back to the regular method. */
969 #endif /* HAVE_MMAP */
972 size = 512; /* number of bytes fm->contents can
973 hold at any given time. */
974 fm->content = xmalloc (size);
978 if (fm->length > size / 2)
980 /* #### I'm not sure whether the whole exponential-growth
981 thing makes sense with kernel read. On Linux at least,
982 read() refuses to read more than 4K from a file at a
983 single chunk anyway. But other Unixes might optimize it
984 better, and it doesn't *hurt* anything, so I'm leaving
987 /* Normally, we grow SIZE exponentially to make the number
988 of calls to read() and realloc() logarithmic in relation
989 to file size. However, read() can read an amount of data
990 smaller than requested, and it would be unreasonable to
991 double SIZE every time *something* was read. Therefore,
992 we double SIZE only when the length exceeds half of the
993 entire allocated size. */
995 fm->content = xrealloc (fm->content, size);
997 nread = read (fd, fm->content + fm->length, size - fm->length);
999 /* Successful read. */
1000 fm->length += nread;
1010 if (size > fm->length && fm->length != 0)
1011 /* Due to exponential growth of fm->content, the allocated region
1012 might be much larger than what is actually needed. */
1013 fm->content = xrealloc (fm->content, fm->length);
1020 xfree (fm->content);
1025 /* Release the resources held by FM. Specifically, this calls
1026 munmap() or xfree() on fm->content, depending whether mmap or
1027 malloc/read were used to read in the file. It also frees the
1028 memory needed to hold the FM structure itself. */
1031 read_file_free (struct file_memory *fm)
1036 munmap (fm->content, fm->length);
1041 xfree (fm->content);
1046 /* Free the pointers in a NULL-terminated vector of pointers, then
1047 free the pointer itself. */
1049 free_vec (char **vec)
1060 /* Append vector V2 to vector V1. The function frees V2 and
1061 reallocates V1 (thus you may not use the contents of neither
1062 pointer after the call). If V1 is NULL, V2 is returned. */
1064 merge_vecs (char **v1, char **v2)
1074 /* To avoid j == 0 */
1079 for (i = 0; v1[i]; i++);
1081 for (j = 0; v2[j]; j++);
1082 /* Reallocate v1. */
1083 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1084 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1089 /* Sometimes it's useful to create "sets" of strings, i.e. special
1090 hash tables where you want to store strings as keys and merely
1091 query for their existence. Here is a set of utility routines that
1092 makes that transparent. */
1095 string_set_add (struct hash_table *ht, const char *s)
1097 /* First check whether the set element already exists. If it does,
1098 do nothing so that we don't have to free() the old element and
1099 then strdup() a new one. */
1100 if (hash_table_contains (ht, s))
1103 /* We use "1" as value. It provides us a useful and clear arbitrary
1104 value, and it consumes no memory -- the pointers to the same
1105 string "1" will be shared by all the key-value pairs in all `set'
1107 hash_table_put (ht, xstrdup (s), "1");
1110 /* Synonym for hash_table_contains... */
1113 string_set_contains (struct hash_table *ht, const char *s)
1115 return hash_table_contains (ht, s);
1119 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1121 char ***arrayptr = (char ***) arg;
1122 *(*arrayptr)++ = (char *) key;
1126 /* Convert the specified string set to array. ARRAY should be large
1127 enough to hold hash_table_count(ht) char pointers. */
1129 void string_set_to_array (struct hash_table *ht, char **array)
1131 hash_table_map (ht, string_set_to_array_mapper, &array);
1135 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1142 string_set_free (struct hash_table *ht)
1144 hash_table_map (ht, string_set_free_mapper, NULL);
1145 hash_table_destroy (ht);
1149 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1156 /* Another utility function: call free() on all keys and values of HT. */
1159 free_keys_and_values (struct hash_table *ht)
1161 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1165 /* Add thousand separators to a number already in string form. Used
1166 by with_thousand_seps and with_thousand_seps_large. */
1169 add_thousand_seps (const char *repr)
1171 static char outbuf[48];
1176 /* Reset the pointers. */
1180 /* Ignore the sign for the purpose of adding thousand
1187 /* How many digits before the first separator? */
1188 mod = strlen (inptr) % 3;
1190 for (i = 0; i < mod; i++)
1191 *outptr++ = inptr[i];
1192 /* Now insert the rest of them, putting separator before every
1194 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1196 if (i % 3 == 0 && i1 != 0)
1198 *outptr++ = inptr[i1];
1200 /* Zero-terminate the string. */
1205 /* Return a static pointer to the number printed with thousand
1206 separators inserted at the right places. */
1209 with_thousand_seps (wgint l)
1212 /* Print the number into the buffer. */
1213 number_to_string (inbuf, l);
1214 return add_thousand_seps (inbuf);
1217 /* Write a string representation of LARGE_INT NUMBER into the provided
1220 It would be dangerous to use sprintf, because the code wouldn't
1221 work on a machine with gcc-provided long long support, but without
1222 libc support for "%lld". However, such old systems platforms
1223 typically lack snprintf and will end up using our version, which
1224 does support "%lld" whereever long longs are available. */
1227 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1229 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1232 /* The same as with_thousand_seps, but works on LARGE_INT. */
1235 with_thousand_seps_large (LARGE_INT l)
1238 large_int_to_string (inbuf, sizeof (inbuf), l);
1239 return add_thousand_seps (inbuf);
1242 /* N, a byte quantity, is converted to a human-readable abberviated
1243 form a la sizes printed by `ls -lh'. The result is written to a
1244 static buffer, a pointer to which is returned.
1246 Unlike `with_thousand_seps', this approximates to the nearest unit.
1247 Quoting GNU libit: "Most people visually process strings of 3-4
1248 digits effectively, but longer strings of digits are more prone to
1249 misinterpretation. Hence, converting to an abbreviated form
1250 usually improves readability."
1252 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1253 original computer science meaning of "powers of 1024". Powers of
1254 1000 would be useless since Wget already displays sizes with
1255 thousand separators. We don't use the "*bibyte" names invented in
1256 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1257 discusses this in some detail. */
1260 human_readable (wgint n)
1262 /* These suffixes are compatible with those of GNU `ls -lh'. */
1263 static char powers[] =
1265 'K', /* kilobyte, 2^10 bytes */
1266 'M', /* megabyte, 2^20 bytes */
1267 'G', /* gigabyte, 2^30 bytes */
1268 'T', /* terabyte, 2^40 bytes */
1269 'P', /* petabyte, 2^50 bytes */
1270 'E', /* exabyte, 2^60 bytes */
1275 /* If the quantity is smaller than 1K, just print it. */
1278 snprintf (buf, sizeof (buf), "%d", (int) n);
1282 /* Loop over powers, dividing N with 1024 in each iteration. This
1283 works unchanged for all sizes of wgint, while still avoiding
1284 non-portable `long double' arithmetic. */
1285 for (i = 0; i < countof (powers); i++)
1287 /* At each iteration N is greater than the *subsequent* power.
1288 That way N/1024.0 produces a decimal number in the units of
1290 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1292 /* Must cast to long first because MS VC can't directly cast
1293 __int64 to double. (This is safe because N is known to
1295 double val = (double) (long) n / 1024.0;
1296 /* Print values smaller than 10 with one decimal digits, and
1297 others without any decimals. */
1298 snprintf (buf, sizeof (buf), "%.*f%c",
1299 val < 10 ? 1 : 0, val, powers[i]);
1304 return NULL; /* unreached */
1307 /* Count the digits in the provided number. Used to allocate space
1308 when printing numbers. */
1311 numdigit (wgint number)
1315 ++cnt; /* accomodate '-' */
1316 while ((number /= 10) != 0)
1321 #define PR(mask) *p++ = n / (mask) + '0'
1323 /* DIGITS_<D> is used to print a D-digit number and should be called
1324 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1325 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1326 Recursively this continues until DIGITS_1 is invoked. */
1328 #define DIGITS_1(mask) PR (mask)
1329 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1330 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1331 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1332 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1333 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1334 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1335 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1336 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1337 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1339 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1341 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1342 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1343 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1344 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1345 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1346 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1347 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1348 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1349 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1351 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1352 cases and to portably support strange sizes of wgint. Ideally this
1353 would just use "%j" and intmax_t, but many systems don't support
1354 it, so it's used only if nothing else works. */
1355 #if SIZEOF_LONG >= SIZEOF_WGINT
1356 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1358 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1359 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1362 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1364 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1369 /* Shorthand for casting to wgint. */
1372 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1373 `sprintf(buffer, "%lld", (long long) number)', only typically much
1374 faster and portable to machines without long long.
1376 The speedup may make a difference in programs that frequently
1377 convert numbers to strings. Some implementations of sprintf,
1378 particularly the one in GNU libc, have been known to be extremely
1379 slow when converting integers to strings.
1381 Return the pointer to the location where the terminating zero was
1382 printed. (Equivalent to calling buffer+strlen(buffer) after the
1385 BUFFER should be big enough to accept as many bytes as you expect
1386 the number to take up. On machines with 64-bit longs the maximum
1387 needed size is 24 bytes. That includes the digits needed for the
1388 largest 64-bit number, the `-' sign in case it's negative, and the
1389 terminating '\0'. */
1392 number_to_string (char *buffer, wgint number)
1397 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1398 /* We are running in a strange or misconfigured environment. Let
1399 sprintf cope with it. */
1400 SPRINTF_WGINT (buffer, n);
1401 p += strlen (buffer);
1402 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1408 /* -n would overflow. Have sprintf deal with this. */
1409 SPRINTF_WGINT (buffer, n);
1410 p += strlen (buffer);
1418 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1419 way printing any N is fully open-coded without a loop or jump.
1420 (Also see description of DIGITS_*.) */
1422 if (n < 10) DIGITS_1 (1);
1423 else if (n < 100) DIGITS_2 (10);
1424 else if (n < 1000) DIGITS_3 (100);
1425 else if (n < 10000) DIGITS_4 (1000);
1426 else if (n < 100000) DIGITS_5 (10000);
1427 else if (n < 1000000) DIGITS_6 (100000);
1428 else if (n < 10000000) DIGITS_7 (1000000);
1429 else if (n < 100000000) DIGITS_8 (10000000);
1430 else if (n < 1000000000) DIGITS_9 (100000000);
1431 #if SIZEOF_WGINT == 4
1432 /* wgint is 32 bits wide: no number has more than 10 digits. */
1433 else DIGITS_10 (1000000000);
1435 /* wgint is 64 bits wide: handle numbers with more than 9 decimal
1436 digits. Constants are constructed by compile-time multiplication
1437 to avoid dealing with different notations for 64-bit constants
1438 (nnnL, nnnLL, and nnnI64, depending on the compiler). */
1439 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1440 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1441 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1442 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1443 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1444 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1445 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1446 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1447 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1448 else DIGITS_19 (1000000000*(W)1000000000);
1452 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1481 /* Print NUMBER to a statically allocated string and return a pointer
1482 to the printed representation.
1484 This function is intended to be used in conjunction with printf.
1485 It is hard to portably print wgint values:
1486 a) you cannot use printf("%ld", number) because wgint can be long
1487 long on 32-bit machines with LFS.
1488 b) you cannot use printf("%lld", number) because NUMBER could be
1489 long on 32-bit machines without LFS, or on 64-bit machines,
1490 which do not require LFS. Also, Windows doesn't support %lld.
1491 c) you cannot use printf("%j", (int_max_t) number) because not all
1492 versions of printf support "%j", the most notable being the one
1494 d) you cannot #define WGINT_FMT to the appropriate format and use
1495 printf(WGINT_FMT, number) because that would break translations
1496 for user-visible messages, such as printf("Downloaded: %d
1499 What you should use instead is printf("%s", number_to_static_string
1502 CAVEAT: since the function returns pointers to static data, you
1503 must be careful to copy its result before calling it again.
1504 However, to make it more useful with printf, the function maintains
1505 an internal ring of static buffers to return. That way things like
1506 printf("%s %s", number_to_static_string (num1),
1507 number_to_static_string (num2)) work as expected. Three buffers
1508 are currently used, which means that "%s %s %s" will work, but "%s
1509 %s %s %s" won't. If you need to print more than three wgints,
1510 bump the RING_SIZE (or rethink your message.) */
1513 number_to_static_string (wgint number)
1515 static char ring[RING_SIZE][24];
1517 char *buf = ring[ringpos];
1518 number_to_string (buf, number);
1519 ringpos = (ringpos + 1) % RING_SIZE;
1523 /* Determine the width of the terminal we're running on. If that's
1524 not possible, return 0. */
1527 determine_screen_width (void)
1529 /* If there's a way to get the terminal size using POSIX
1530 tcgetattr(), somebody please tell me. */
1535 if (opt.lfilename != NULL)
1538 fd = fileno (stderr);
1539 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1540 return 0; /* most likely ENOTTY */
1543 #else /* not TIOCGWINSZ */
1545 CONSOLE_SCREEN_BUFFER_INFO csbi;
1546 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1548 return csbi.dwSize.X;
1549 # else /* neither WINDOWS nor TIOCGWINSZ */
1551 #endif /* neither WINDOWS nor TIOCGWINSZ */
1552 #endif /* not TIOCGWINSZ */
1555 /* Return a random number between 0 and MAX-1, inclusive.
1557 If MAX is greater than the value of RAND_MAX+1 on the system, the
1558 returned value will be in the range [0, RAND_MAX]. This may be
1559 fixed in a future release.
1561 The random number generator is seeded automatically the first time
1564 This uses rand() for portability. It has been suggested that
1565 random() offers better randomness, but this is not required for
1566 Wget, so I chose to go for simplicity and use rand
1569 DO NOT use this for cryptographic purposes. It is only meant to be
1570 used in situations where quality of the random numbers returned
1571 doesn't really matter. */
1574 random_number (int max)
1582 srand (time (NULL));
1587 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1588 and enforce that assumption by masking other bits. */
1590 # define RAND_MAX 32767
1594 /* This is equivalent to rand() % max, but uses the high-order bits
1595 for better randomness on architecture where rand() is implemented
1596 using a simple congruential generator. */
1598 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1599 return (int)bounded;
1602 /* Return a random uniformly distributed floating point number in the
1603 [0, 1) range. The precision of returned numbers is 9 digits.
1605 Modify this to use erand48() where available! */
1610 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1611 sure it's greater than 1000. */
1612 int rnd1 = random_number (1000);
1613 int rnd2 = random_number (1000);
1614 int rnd3 = random_number (1000);
1615 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1618 /* Implementation of run_with_timeout, a generic timeout-forcing
1619 routine for systems with Unix-like signal handling. */
1621 #ifdef USE_SIGNAL_TIMEOUT
1622 # ifdef HAVE_SIGSETJMP
1623 # define SETJMP(env) sigsetjmp (env, 1)
1625 static sigjmp_buf run_with_timeout_env;
1628 abort_run_with_timeout (int sig)
1630 assert (sig == SIGALRM);
1631 siglongjmp (run_with_timeout_env, -1);
1633 # else /* not HAVE_SIGSETJMP */
1634 # define SETJMP(env) setjmp (env)
1636 static jmp_buf run_with_timeout_env;
1639 abort_run_with_timeout (int sig)
1641 assert (sig == SIGALRM);
1642 /* We don't have siglongjmp to preserve the set of blocked signals;
1643 if we longjumped out of the handler at this point, SIGALRM would
1644 remain blocked. We must unblock it manually. */
1645 int mask = siggetmask ();
1646 mask &= ~sigmask (SIGALRM);
1649 /* Now it's safe to longjump. */
1650 longjmp (run_with_timeout_env, -1);
1652 # endif /* not HAVE_SIGSETJMP */
1654 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1655 setitimer where available, alarm otherwise.
1657 TIMEOUT should be non-zero. If the timeout value is so small that
1658 it would be rounded to zero, it is rounded to the least legal value
1659 instead (1us for setitimer, 1s for alarm). That ensures that
1660 SIGALRM will be delivered in all cases. */
1663 alarm_set (double timeout)
1666 /* Use the modern itimer interface. */
1667 struct itimerval itv;
1669 itv.it_value.tv_sec = (long) timeout;
1670 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1671 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1672 /* Ensure that we wait for at least the minimum interval.
1673 Specifying zero would mean "wait forever". */
1674 itv.it_value.tv_usec = 1;
1675 setitimer (ITIMER_REAL, &itv, NULL);
1676 #else /* not ITIMER_REAL */
1677 /* Use the old alarm() interface. */
1678 int secs = (int) timeout;
1680 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1681 because alarm(0) means "never deliver the alarm", i.e. "wait
1682 forever", which is not what someone who specifies a 0.5s
1683 timeout would expect. */
1686 #endif /* not ITIMER_REAL */
1689 /* Cancel the alarm set with alarm_set. */
1695 struct itimerval disable;
1697 setitimer (ITIMER_REAL, &disable, NULL);
1698 #else /* not ITIMER_REAL */
1700 #endif /* not ITIMER_REAL */
1703 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1704 seconds. Returns non-zero if the function was interrupted with a
1705 timeout, zero otherwise.
1707 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1708 using setitimer() or alarm(). The timeout is enforced by
1709 longjumping out of the SIGALRM handler. This has several
1710 advantages compared to the traditional approach of relying on
1711 signals causing system calls to exit with EINTR:
1713 * The callback function is *forcibly* interrupted after the
1714 timeout expires, (almost) regardless of what it was doing and
1715 whether it was in a syscall. For example, a calculation that
1716 takes a long time is interrupted as reliably as an IO
1719 * It works with both SYSV and BSD signals because it doesn't
1720 depend on the default setting of SA_RESTART.
1722 * It doesn't require special handler setup beyond a simple call
1723 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1726 The only downside is that, if FUN allocates internal resources that
1727 are normally freed prior to exit from the functions, they will be
1728 lost in case of timeout. */
1731 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1741 signal (SIGALRM, abort_run_with_timeout);
1742 if (SETJMP (run_with_timeout_env) != 0)
1744 /* Longjumped out of FUN with a timeout. */
1745 signal (SIGALRM, SIG_DFL);
1748 alarm_set (timeout);
1751 /* Preserve errno in case alarm() or signal() modifies it. */
1752 saved_errno = errno;
1754 signal (SIGALRM, SIG_DFL);
1755 errno = saved_errno;
1760 #else /* not USE_SIGNAL_TIMEOUT */
1763 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1764 define it under Windows, because Windows has its own version of
1765 run_with_timeout that uses threads. */
1768 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1773 #endif /* not WINDOWS */
1774 #endif /* not USE_SIGNAL_TIMEOUT */
1778 /* Sleep the specified amount of seconds. On machines without
1779 nanosleep(), this may sleep shorter if interrupted by signals. */
1782 xsleep (double seconds)
1784 #ifdef HAVE_NANOSLEEP
1785 /* nanosleep is the preferred interface because it offers high
1786 accuracy and, more importantly, because it allows us to reliably
1787 restart receiving a signal such as SIGWINCH. (There was an
1788 actual Debian bug report about --limit-rate malfunctioning while
1789 the terminal was being resized.) */
1790 struct timespec sleep, remaining;
1791 sleep.tv_sec = (long) seconds;
1792 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1793 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1794 /* If nanosleep has been interrupted by a signal, adjust the
1795 sleeping period and return to sleep. */
1797 #else /* not HAVE_NANOSLEEP */
1799 /* If usleep is available, use it in preference to select. */
1802 /* On some systems, usleep cannot handle values larger than
1803 1,000,000. If the period is larger than that, use sleep
1804 first, then add usleep for subsecond accuracy. */
1806 seconds -= (long) seconds;
1808 usleep (seconds * 1000000);
1809 #else /* not HAVE_USLEEP */
1811 /* Note that, although Windows supports select, this sleeping
1812 strategy doesn't work there because Winsock's select doesn't
1813 implement timeout when it is passed NULL pointers for all fd
1814 sets. (But it does work under Cygwin, which implements its own
1816 struct timeval sleep;
1817 sleep.tv_sec = (long) seconds;
1818 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1819 select (0, NULL, NULL, NULL, &sleep);
1820 /* If select returns -1 and errno is EINTR, it means we were
1821 interrupted by a signal. But without knowing how long we've
1822 actually slept, we can't return to sleep. Using gettimeofday to
1823 track sleeps is slow and unreliable due to clock skew. */
1824 #else /* not HAVE_SELECT */
1826 #endif /* not HAVE_SELECT */
1827 #endif /* not HAVE_USLEEP */
1828 #endif /* not HAVE_NANOSLEEP */
1831 #endif /* not WINDOWS */
1833 /* Encode the string STR of length LENGTH to base64 format and place it
1834 to B64STORE. The output will be \0-terminated, and must point to a
1835 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1836 returns the length of the resulting base64 data, not counting the
1839 This implementation will not emit newlines after 76 characters of
1843 base64_encode (const char *str, int length, char *b64store)
1845 /* Conversion table. */
1846 static char tbl[64] = {
1847 'A','B','C','D','E','F','G','H',
1848 'I','J','K','L','M','N','O','P',
1849 'Q','R','S','T','U','V','W','X',
1850 'Y','Z','a','b','c','d','e','f',
1851 'g','h','i','j','k','l','m','n',
1852 'o','p','q','r','s','t','u','v',
1853 'w','x','y','z','0','1','2','3',
1854 '4','5','6','7','8','9','+','/'
1857 const unsigned char *s = (const unsigned char *) str;
1860 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1861 for (i = 0; i < length; i += 3)
1863 *p++ = tbl[s[0] >> 2];
1864 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1865 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1866 *p++ = tbl[s[2] & 0x3f];
1870 /* Pad the result if necessary... */
1871 if (i == length + 1)
1873 else if (i == length + 2)
1874 *(p - 1) = *(p - 2) = '=';
1876 /* ...and zero-terminate it. */
1879 return p - b64store;
1882 #define IS_ASCII(c) (((c) & 0x80) == 0)
1883 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1885 /* Get next character from the string, except that non-base64
1886 characters are ignored, as mandated by rfc2045. */
1887 #define NEXT_BASE64_CHAR(c, p) do { \
1889 } while (c != '\0' && !IS_BASE64 (c))
1891 /* Decode data from BASE64 (assumed to be encoded as base64) into
1892 memory pointed to by TO. TO should be large enough to accomodate
1893 the decoded data, which is guaranteed to be less than
1896 Since TO is assumed to contain binary data, it is not
1897 NUL-terminated. The function returns the length of the data
1898 written to TO. -1 is returned in case of error caused by malformed
1902 base64_decode (const char *base64, char *to)
1904 /* Table of base64 values for first 128 characters. Note that this
1905 assumes ASCII (but so does Wget in other places). */
1906 static short base64_char_to_value[128] =
1908 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1909 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1910 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1911 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1912 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1913 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1914 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1915 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1916 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1917 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1918 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1919 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1920 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1923 const char *p = base64;
1929 unsigned long value;
1931 /* Process first byte of a quadruplet. */
1932 NEXT_BASE64_CHAR (c, p);
1936 return -1; /* illegal '=' while decoding base64 */
1937 value = base64_char_to_value[c] << 18;
1939 /* Process scond byte of a quadruplet. */
1940 NEXT_BASE64_CHAR (c, p);
1942 return -1; /* premature EOF while decoding base64 */
1944 return -1; /* illegal `=' while decoding base64 */
1945 value |= base64_char_to_value[c] << 12;
1948 /* Process third byte of a quadruplet. */
1949 NEXT_BASE64_CHAR (c, p);
1951 return -1; /* premature EOF while decoding base64 */
1955 NEXT_BASE64_CHAR (c, p);
1957 return -1; /* premature EOF while decoding base64 */
1959 return -1; /* padding `=' expected but not found */
1963 value |= base64_char_to_value[c] << 6;
1964 *q++ = 0xff & value >> 8;
1966 /* Process fourth byte of a quadruplet. */
1967 NEXT_BASE64_CHAR (c, p);
1969 return -1; /* premature EOF while decoding base64 */
1973 value |= base64_char_to_value[c];
1974 *q++ = 0xff & value;
1982 #undef NEXT_BASE64_CHAR
1984 /* Simple merge sort for use by stable_sort. Implementation courtesy
1985 Zeljko Vrba with additional debugging by Nenad Barbutov. */
1988 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1989 int (*cmpfun) PARAMS ((const void *, const void *)))
1991 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1995 size_t mid = (to + from) / 2;
1996 mergesort_internal (base, temp, size, from, mid, cmpfun);
1997 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2000 for (k = from; (i <= mid) && (j <= to); k++)
2001 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2002 memcpy (ELT (temp, k), ELT (base, i++), size);
2004 memcpy (ELT (temp, k), ELT (base, j++), size);
2006 memcpy (ELT (temp, k++), ELT (base, i++), size);
2008 memcpy (ELT (temp, k++), ELT (base, j++), size);
2009 for (k = from; k <= to; k++)
2010 memcpy (ELT (base, k), ELT (temp, k), size);
2015 /* Stable sort with interface exactly like standard library's qsort.
2016 Uses mergesort internally, allocating temporary storage with
2020 stable_sort (void *base, size_t nmemb, size_t size,
2021 int (*cmpfun) PARAMS ((const void *, const void *)))
2025 void *temp = alloca (nmemb * size * sizeof (void *));
2026 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);