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)
685 for (x = strlist; *x; x++)
686 if (has_wildcards_p (*x))
688 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
693 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
700 /* Returns whether DIRECTORY is acceptable for download, wrt the
701 include/exclude lists.
703 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
704 and absolute paths may be freely intermixed. */
706 accdir (const char *directory, enum accd flags)
708 /* Remove starting '/'. */
709 if (flags & ALLABS && *directory == '/')
713 if (!proclist (opt.includes, directory, flags))
718 if (proclist (opt.excludes, directory, flags))
724 /* Return non-zero if STRING ends with TAIL. For instance:
726 match_tail ("abc", "bc", 0) -> 1
727 match_tail ("abc", "ab", 0) -> 0
728 match_tail ("abc", "abc", 0) -> 1
730 If FOLD_CASE_P is non-zero, the comparison will be
734 match_tail (const char *string, const char *tail, int fold_case_p)
738 /* We want this to be fast, so we code two loops, one with
739 case-folding, one without. */
743 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
744 if (string[i] != tail[j])
749 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
750 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
754 /* If the tail was exhausted, the match was succesful. */
761 /* Checks whether string S matches each element of ACCEPTS. A list
762 element are matched either with fnmatch() or match_tail(),
763 according to whether the element contains wildcards or not.
765 If the BACKWARD is 0, don't do backward comparison -- just compare
768 in_acclist (const char *const *accepts, const char *s, int backward)
770 for (; *accepts; accepts++)
772 if (has_wildcards_p (*accepts))
774 /* fnmatch returns 0 if the pattern *does* match the
776 if (fnmatch (*accepts, s, 0) == 0)
783 if (match_tail (s, *accepts, 0))
788 if (!strcmp (s, *accepts))
796 /* Return the location of STR's suffix (file extension). Examples:
797 suffix ("foo.bar") -> "bar"
798 suffix ("foo.bar.baz") -> "baz"
799 suffix ("/foo/bar") -> NULL
800 suffix ("/foo.bar/baz") -> NULL */
802 suffix (const char *str)
806 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
810 return (char *)str + i;
815 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
819 has_wildcards_p (const char *s)
822 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
827 /* Return non-zero if FNAME ends with a typical HTML suffix. The
828 following (case-insensitive) suffixes are presumed to be HTML files:
832 ?html (`?' matches one character)
834 #### CAVEAT. This is not necessarily a good indication that FNAME
835 refers to a file that contains HTML! */
837 has_html_suffix_p (const char *fname)
841 if ((suf = suffix (fname)) == NULL)
843 if (!strcasecmp (suf, "html"))
845 if (!strcasecmp (suf, "htm"))
847 if (suf[0] && !strcasecmp (suf + 1, "html"))
852 /* Read a line from FP and return the pointer to freshly allocated
853 storage. The storage space is obtained through malloc() and should
854 be freed with free() when it is no longer needed.
856 The length of the line is not limited, except by available memory.
857 The newline character at the end of line is retained. The line is
858 terminated with a zero character.
860 After end-of-file is encountered without anything being read, NULL
861 is returned. NULL is also returned on error. To distinguish
862 between these two cases, use the stdio function ferror(). */
865 read_whole_line (FILE *fp)
869 char *line = (char *)xmalloc (bufsize);
871 while (fgets (line + length, bufsize - length, fp))
873 length += strlen (line + length);
875 /* Possible for example when reading from a binary file where
876 a line begins with \0. */
879 if (line[length - 1] == '\n')
882 /* fgets() guarantees to read the whole line, or to use up the
883 space we've given it. We can double the buffer
886 line = xrealloc (line, bufsize);
888 if (length == 0 || ferror (fp))
893 if (length + 1 < bufsize)
894 /* Relieve the memory from our exponential greediness. We say
895 `length + 1' because the terminating \0 is not included in
896 LENGTH. We don't need to zero-terminate the string ourselves,
897 though, because fgets() does that. */
898 line = xrealloc (line, length + 1);
902 /* Read FILE into memory. A pointer to `struct file_memory' are
903 returned; use struct element `content' to access file contents, and
904 the element `length' to know the file length. `content' is *not*
905 zero-terminated, and you should *not* read or write beyond the [0,
906 length) range of characters.
908 After you are done with the file contents, call read_file_free to
911 Depending on the operating system and the type of file that is
912 being read, read_file() either mmap's the file into memory, or
913 reads the file into the core using read().
915 If file is named "-", fileno(stdin) is used for reading instead.
916 If you want to read from a real file named "-", use "./-" instead. */
919 read_file (const char *file)
922 struct file_memory *fm;
924 int inhibit_close = 0;
926 /* Some magic in the finest tradition of Perl and its kin: if FILE
927 is "-", just use stdin. */
932 /* Note that we don't inhibit mmap() in this case. If stdin is
933 redirected from a regular file, mmap() will still work. */
936 fd = open (file, O_RDONLY);
939 fm = xnew (struct file_memory);
944 if (fstat (fd, &buf) < 0)
946 fm->length = buf.st_size;
947 /* NOTE: As far as I know, the callers of this function never
948 modify the file text. Relying on this would enable us to
949 specify PROT_READ and MAP_SHARED for a marginal gain in
950 efficiency, but at some cost to generality. */
951 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
953 if (fm->content == (char *)MAP_FAILED)
963 /* The most common reason why mmap() fails is that FD does not point
964 to a plain file. However, it's also possible that mmap() doesn't
965 work for a particular type of file. Therefore, whenever mmap()
966 fails, we just fall back to the regular method. */
967 #endif /* HAVE_MMAP */
970 size = 512; /* number of bytes fm->contents can
971 hold at any given time. */
972 fm->content = xmalloc (size);
976 if (fm->length > size / 2)
978 /* #### I'm not sure whether the whole exponential-growth
979 thing makes sense with kernel read. On Linux at least,
980 read() refuses to read more than 4K from a file at a
981 single chunk anyway. But other Unixes might optimize it
982 better, and it doesn't *hurt* anything, so I'm leaving
985 /* Normally, we grow SIZE exponentially to make the number
986 of calls to read() and realloc() logarithmic in relation
987 to file size. However, read() can read an amount of data
988 smaller than requested, and it would be unreasonable to
989 double SIZE every time *something* was read. Therefore,
990 we double SIZE only when the length exceeds half of the
991 entire allocated size. */
993 fm->content = xrealloc (fm->content, size);
995 nread = read (fd, fm->content + fm->length, size - fm->length);
997 /* Successful read. */
1008 if (size > fm->length && fm->length != 0)
1009 /* Due to exponential growth of fm->content, the allocated region
1010 might be much larger than what is actually needed. */
1011 fm->content = xrealloc (fm->content, fm->length);
1018 xfree (fm->content);
1023 /* Release the resources held by FM. Specifically, this calls
1024 munmap() or xfree() on fm->content, depending whether mmap or
1025 malloc/read were used to read in the file. It also frees the
1026 memory needed to hold the FM structure itself. */
1029 read_file_free (struct file_memory *fm)
1034 munmap (fm->content, fm->length);
1039 xfree (fm->content);
1044 /* Free the pointers in a NULL-terminated vector of pointers, then
1045 free the pointer itself. */
1047 free_vec (char **vec)
1058 /* Append vector V2 to vector V1. The function frees V2 and
1059 reallocates V1 (thus you may not use the contents of neither
1060 pointer after the call). If V1 is NULL, V2 is returned. */
1062 merge_vecs (char **v1, char **v2)
1072 /* To avoid j == 0 */
1077 for (i = 0; v1[i]; i++);
1079 for (j = 0; v2[j]; j++);
1080 /* Reallocate v1. */
1081 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1082 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1087 /* Sometimes it's useful to create "sets" of strings, i.e. special
1088 hash tables where you want to store strings as keys and merely
1089 query for their existence. Here is a set of utility routines that
1090 makes that transparent. */
1093 string_set_add (struct hash_table *ht, const char *s)
1095 /* First check whether the set element already exists. If it does,
1096 do nothing so that we don't have to free() the old element and
1097 then strdup() a new one. */
1098 if (hash_table_contains (ht, s))
1101 /* We use "1" as value. It provides us a useful and clear arbitrary
1102 value, and it consumes no memory -- the pointers to the same
1103 string "1" will be shared by all the key-value pairs in all `set'
1105 hash_table_put (ht, xstrdup (s), "1");
1108 /* Synonym for hash_table_contains... */
1111 string_set_contains (struct hash_table *ht, const char *s)
1113 return hash_table_contains (ht, s);
1117 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1119 char ***arrayptr = (char ***) arg;
1120 *(*arrayptr)++ = (char *) key;
1124 /* Convert the specified string set to array. ARRAY should be large
1125 enough to hold hash_table_count(ht) char pointers. */
1127 void string_set_to_array (struct hash_table *ht, char **array)
1129 hash_table_map (ht, string_set_to_array_mapper, &array);
1133 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1140 string_set_free (struct hash_table *ht)
1142 hash_table_map (ht, string_set_free_mapper, NULL);
1143 hash_table_destroy (ht);
1147 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1154 /* Another utility function: call free() on all keys and values of HT. */
1157 free_keys_and_values (struct hash_table *ht)
1159 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1163 /* Add thousand separators to a number already in string form. Used
1164 by with_thousand_seps and with_thousand_seps_large. */
1167 add_thousand_seps (const char *repr)
1169 static char outbuf[48];
1174 /* Reset the pointers. */
1178 /* Ignore the sign for the purpose of adding thousand
1185 /* How many digits before the first separator? */
1186 mod = strlen (inptr) % 3;
1188 for (i = 0; i < mod; i++)
1189 *outptr++ = inptr[i];
1190 /* Now insert the rest of them, putting separator before every
1192 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1194 if (i % 3 == 0 && i1 != 0)
1196 *outptr++ = inptr[i1];
1198 /* Zero-terminate the string. */
1203 /* Return a static pointer to the number printed with thousand
1204 separators inserted at the right places. */
1207 with_thousand_seps (wgint l)
1210 /* Print the number into the buffer. */
1211 number_to_string (inbuf, l);
1212 return add_thousand_seps (inbuf);
1215 /* Write a string representation of LARGE_INT NUMBER into the provided
1218 It would be dangerous to use sprintf, because the code wouldn't
1219 work on a machine with gcc-provided long long support, but without
1220 libc support for "%lld". However, such old systems platforms
1221 typically lack snprintf and will end up using our version, which
1222 does support "%lld" whereever long longs are available. */
1225 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1227 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1230 /* The same as with_thousand_seps, but works on LARGE_INT. */
1233 with_thousand_seps_large (LARGE_INT l)
1236 large_int_to_string (inbuf, sizeof (inbuf), l);
1237 return add_thousand_seps (inbuf);
1240 /* N, a byte quantity, is converted to a human-readable abberviated
1241 form a la sizes printed by `ls -lh'. The result is written to a
1242 static buffer, a pointer to which is returned.
1244 Unlike `with_thousand_seps', this approximates to the nearest unit.
1245 Quoting GNU libit: "Most people visually process strings of 3-4
1246 digits effectively, but longer strings of digits are more prone to
1247 misinterpretation. Hence, converting to an abbreviated form
1248 usually improves readability."
1250 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1251 original computer science meaning of "powers of 1024". Powers of
1252 1000 would be useless since Wget already displays sizes with
1253 thousand separators. We don't use the "*bibyte" names invented in
1254 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1255 discusses this in some detail. */
1258 human_readable (wgint n)
1260 /* These suffixes are compatible with those of GNU `ls -lh'. */
1261 static char powers[] =
1263 'K', /* kilobyte, 2^10 bytes */
1264 'M', /* megabyte, 2^20 bytes */
1265 'G', /* gigabyte, 2^30 bytes */
1266 'T', /* terabyte, 2^40 bytes */
1267 'P', /* petabyte, 2^50 bytes */
1268 'E', /* exabyte, 2^60 bytes */
1273 /* If the quantity is smaller than 1K, just print it. */
1276 snprintf (buf, sizeof (buf), "%d", (int) n);
1280 /* Loop over powers, dividing N with 1024 in each iteration. This
1281 works unchanged for all sizes of wgint, while still avoiding
1282 non-portable `long double' arithmetic. */
1283 for (i = 0; i < countof (powers); i++)
1285 /* At each iteration N is greater than the *subsequent* power.
1286 That way N/1024.0 produces a decimal number in the units of
1288 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1290 /* Must cast to long first because MS VC can't directly cast
1291 __int64 to double. (This is safe because N is known to
1293 double val = (double) (long) n / 1024.0;
1294 /* Print values smaller than 10 with one decimal digits, and
1295 others without any decimals. */
1296 snprintf (buf, sizeof (buf), "%.*f%c",
1297 val < 10 ? 1 : 0, val, powers[i]);
1302 return NULL; /* unreached */
1305 /* Count the digits in the provided number. Used to allocate space
1306 when printing numbers. */
1309 numdigit (wgint number)
1313 ++cnt; /* accomodate '-' */
1314 while ((number /= 10) != 0)
1319 #define PR(mask) *p++ = n / (mask) + '0'
1321 /* DIGITS_<D> is used to print a D-digit number and should be called
1322 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1323 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1324 Recursively this continues until DIGITS_1 is invoked. */
1326 #define DIGITS_1(mask) PR (mask)
1327 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1328 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1329 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1330 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1331 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1332 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1333 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1334 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1335 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1337 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1339 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1340 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1341 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1342 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1343 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1344 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1345 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1346 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1347 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1349 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1350 cases and to portably support strange sizes of wgint. Ideally this
1351 would just use "%j" and intmax_t, but many systems don't support
1352 it, so it's used only if nothing else works. */
1353 #if SIZEOF_LONG >= SIZEOF_WGINT
1354 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1356 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1357 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1360 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1362 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1367 /* Shorthand for casting to wgint. */
1370 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1371 `sprintf(buffer, "%lld", (long long) number)', only typically much
1372 faster and portable to machines without long long.
1374 The speedup may make a difference in programs that frequently
1375 convert numbers to strings. Some implementations of sprintf,
1376 particularly the one in GNU libc, have been known to be extremely
1377 slow when converting integers to strings.
1379 Return the pointer to the location where the terminating zero was
1380 printed. (Equivalent to calling buffer+strlen(buffer) after the
1383 BUFFER should be big enough to accept as many bytes as you expect
1384 the number to take up. On machines with 64-bit longs the maximum
1385 needed size is 24 bytes. That includes the digits needed for the
1386 largest 64-bit number, the `-' sign in case it's negative, and the
1387 terminating '\0'. */
1390 number_to_string (char *buffer, wgint number)
1395 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1396 /* We are running in a strange or misconfigured environment. Let
1397 sprintf cope with it. */
1398 SPRINTF_WGINT (buffer, n);
1399 p += strlen (buffer);
1400 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1406 /* -n would overflow. Have sprintf deal with this. */
1407 SPRINTF_WGINT (buffer, n);
1408 p += strlen (buffer);
1416 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1417 way printing any N is fully open-coded without a loop or jump.
1418 (Also see description of DIGITS_*.) */
1420 if (n < 10) DIGITS_1 (1);
1421 else if (n < 100) DIGITS_2 (10);
1422 else if (n < 1000) DIGITS_3 (100);
1423 else if (n < 10000) DIGITS_4 (1000);
1424 else if (n < 100000) DIGITS_5 (10000);
1425 else if (n < 1000000) DIGITS_6 (100000);
1426 else if (n < 10000000) DIGITS_7 (1000000);
1427 else if (n < 100000000) DIGITS_8 (10000000);
1428 else if (n < 1000000000) DIGITS_9 (100000000);
1429 #if SIZEOF_WGINT == 4
1430 /* wgint is 32 bits wide: no number has more than 10 digits. */
1431 else DIGITS_10 (1000000000);
1433 /* wgint is 64 bits wide: handle numbers with more than 9 decimal
1434 digits. Constants are constructed by compile-time multiplication
1435 to avoid dealing with different notations for 64-bit constants
1436 (nnnL, nnnLL, and nnnI64, depending on the compiler). */
1437 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1438 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1439 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1440 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1441 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1442 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1443 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1444 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1445 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1446 else DIGITS_19 (1000000000*(W)1000000000);
1450 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1479 /* Print NUMBER to a statically allocated string and return a pointer
1480 to the printed representation.
1482 This function is intended to be used in conjunction with printf.
1483 It is hard to portably print wgint values:
1484 a) you cannot use printf("%ld", number) because wgint can be long
1485 long on 32-bit machines with LFS.
1486 b) you cannot use printf("%lld", number) because NUMBER could be
1487 long on 32-bit machines without LFS, or on 64-bit machines,
1488 which do not require LFS. Also, Windows doesn't support %lld.
1489 c) you cannot use printf("%j", (int_max_t) number) because not all
1490 versions of printf support "%j", the most notable being the one
1492 d) you cannot #define WGINT_FMT to the appropriate format and use
1493 printf(WGINT_FMT, number) because that would break translations
1494 for user-visible messages, such as printf("Downloaded: %d
1497 What you should use instead is printf("%s", number_to_static_string
1500 CAVEAT: since the function returns pointers to static data, you
1501 must be careful to copy its result before calling it again.
1502 However, to make it more useful with printf, the function maintains
1503 an internal ring of static buffers to return. That way things like
1504 printf("%s %s", number_to_static_string (num1),
1505 number_to_static_string (num2)) work as expected. Three buffers
1506 are currently used, which means that "%s %s %s" will work, but "%s
1507 %s %s %s" won't. If you need to print more than three wgints,
1508 bump the RING_SIZE (or rethink your message.) */
1511 number_to_static_string (wgint number)
1513 static char ring[RING_SIZE][24];
1515 char *buf = ring[ringpos];
1516 number_to_string (buf, number);
1517 ringpos = (ringpos + 1) % RING_SIZE;
1521 /* Determine the width of the terminal we're running on. If that's
1522 not possible, return 0. */
1525 determine_screen_width (void)
1527 /* If there's a way to get the terminal size using POSIX
1528 tcgetattr(), somebody please tell me. */
1533 if (opt.lfilename != NULL)
1536 fd = fileno (stderr);
1537 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1538 return 0; /* most likely ENOTTY */
1541 #else /* not TIOCGWINSZ */
1543 CONSOLE_SCREEN_BUFFER_INFO csbi;
1544 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1546 return csbi.dwSize.X;
1547 # else /* neither WINDOWS nor TIOCGWINSZ */
1549 #endif /* neither WINDOWS nor TIOCGWINSZ */
1550 #endif /* not TIOCGWINSZ */
1553 /* Return a random number between 0 and MAX-1, inclusive.
1555 If MAX is greater than the value of RAND_MAX+1 on the system, the
1556 returned value will be in the range [0, RAND_MAX]. This may be
1557 fixed in a future release.
1559 The random number generator is seeded automatically the first time
1562 This uses rand() for portability. It has been suggested that
1563 random() offers better randomness, but this is not required for
1564 Wget, so I chose to go for simplicity and use rand
1567 DO NOT use this for cryptographic purposes. It is only meant to be
1568 used in situations where quality of the random numbers returned
1569 doesn't really matter. */
1572 random_number (int max)
1580 srand (time (NULL));
1585 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1586 and enforce that assumption by masking other bits. */
1588 # define RAND_MAX 32767
1592 /* This is equivalent to rand() % max, but uses the high-order bits
1593 for better randomness on architecture where rand() is implemented
1594 using a simple congruential generator. */
1596 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1597 return (int)bounded;
1600 /* Return a random uniformly distributed floating point number in the
1601 [0, 1) range. The precision of returned numbers is 9 digits.
1603 Modify this to use erand48() where available! */
1608 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1609 sure it's greater than 1000. */
1610 int rnd1 = random_number (1000);
1611 int rnd2 = random_number (1000);
1612 int rnd3 = random_number (1000);
1613 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1616 /* Implementation of run_with_timeout, a generic timeout-forcing
1617 routine for systems with Unix-like signal handling. */
1619 #ifdef USE_SIGNAL_TIMEOUT
1620 # ifdef HAVE_SIGSETJMP
1621 # define SETJMP(env) sigsetjmp (env, 1)
1623 static sigjmp_buf run_with_timeout_env;
1626 abort_run_with_timeout (int sig)
1628 assert (sig == SIGALRM);
1629 siglongjmp (run_with_timeout_env, -1);
1631 # else /* not HAVE_SIGSETJMP */
1632 # define SETJMP(env) setjmp (env)
1634 static jmp_buf run_with_timeout_env;
1637 abort_run_with_timeout (int sig)
1639 assert (sig == SIGALRM);
1640 /* We don't have siglongjmp to preserve the set of blocked signals;
1641 if we longjumped out of the handler at this point, SIGALRM would
1642 remain blocked. We must unblock it manually. */
1643 int mask = siggetmask ();
1644 mask &= ~sigmask (SIGALRM);
1647 /* Now it's safe to longjump. */
1648 longjmp (run_with_timeout_env, -1);
1650 # endif /* not HAVE_SIGSETJMP */
1652 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1653 setitimer where available, alarm otherwise.
1655 TIMEOUT should be non-zero. If the timeout value is so small that
1656 it would be rounded to zero, it is rounded to the least legal value
1657 instead (1us for setitimer, 1s for alarm). That ensures that
1658 SIGALRM will be delivered in all cases. */
1661 alarm_set (double timeout)
1664 /* Use the modern itimer interface. */
1665 struct itimerval itv;
1667 itv.it_value.tv_sec = (long) timeout;
1668 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1669 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1670 /* Ensure that we wait for at least the minimum interval.
1671 Specifying zero would mean "wait forever". */
1672 itv.it_value.tv_usec = 1;
1673 setitimer (ITIMER_REAL, &itv, NULL);
1674 #else /* not ITIMER_REAL */
1675 /* Use the old alarm() interface. */
1676 int secs = (int) timeout;
1678 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1679 because alarm(0) means "never deliver the alarm", i.e. "wait
1680 forever", which is not what someone who specifies a 0.5s
1681 timeout would expect. */
1684 #endif /* not ITIMER_REAL */
1687 /* Cancel the alarm set with alarm_set. */
1693 struct itimerval disable;
1695 setitimer (ITIMER_REAL, &disable, NULL);
1696 #else /* not ITIMER_REAL */
1698 #endif /* not ITIMER_REAL */
1701 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1702 seconds. Returns non-zero if the function was interrupted with a
1703 timeout, zero otherwise.
1705 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1706 using setitimer() or alarm(). The timeout is enforced by
1707 longjumping out of the SIGALRM handler. This has several
1708 advantages compared to the traditional approach of relying on
1709 signals causing system calls to exit with EINTR:
1711 * The callback function is *forcibly* interrupted after the
1712 timeout expires, (almost) regardless of what it was doing and
1713 whether it was in a syscall. For example, a calculation that
1714 takes a long time is interrupted as reliably as an IO
1717 * It works with both SYSV and BSD signals because it doesn't
1718 depend on the default setting of SA_RESTART.
1720 * It doesn't require special handler setup beyond a simple call
1721 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1724 The only downside is that, if FUN allocates internal resources that
1725 are normally freed prior to exit from the functions, they will be
1726 lost in case of timeout. */
1729 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1739 signal (SIGALRM, abort_run_with_timeout);
1740 if (SETJMP (run_with_timeout_env) != 0)
1742 /* Longjumped out of FUN with a timeout. */
1743 signal (SIGALRM, SIG_DFL);
1746 alarm_set (timeout);
1749 /* Preserve errno in case alarm() or signal() modifies it. */
1750 saved_errno = errno;
1752 signal (SIGALRM, SIG_DFL);
1753 errno = saved_errno;
1758 #else /* not USE_SIGNAL_TIMEOUT */
1761 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1762 define it under Windows, because Windows has its own version of
1763 run_with_timeout that uses threads. */
1766 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1771 #endif /* not WINDOWS */
1772 #endif /* not USE_SIGNAL_TIMEOUT */
1776 /* Sleep the specified amount of seconds. On machines without
1777 nanosleep(), this may sleep shorter if interrupted by signals. */
1780 xsleep (double seconds)
1782 #ifdef HAVE_NANOSLEEP
1783 /* nanosleep is the preferred interface because it offers high
1784 accuracy and, more importantly, because it allows us to reliably
1785 restart receiving a signal such as SIGWINCH. (There was an
1786 actual Debian bug report about --limit-rate malfunctioning while
1787 the terminal was being resized.) */
1788 struct timespec sleep, remaining;
1789 sleep.tv_sec = (long) seconds;
1790 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1791 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1792 /* If nanosleep has been interrupted by a signal, adjust the
1793 sleeping period and return to sleep. */
1795 #else /* not HAVE_NANOSLEEP */
1797 /* If usleep is available, use it in preference to select. */
1800 /* On some systems, usleep cannot handle values larger than
1801 1,000,000. If the period is larger than that, use sleep
1802 first, then add usleep for subsecond accuracy. */
1804 seconds -= (long) seconds;
1806 usleep (seconds * 1000000);
1807 #else /* not HAVE_USLEEP */
1809 /* Note that, although Windows supports select, this sleeping
1810 strategy doesn't work there because Winsock's select doesn't
1811 implement timeout when it is passed NULL pointers for all fd
1812 sets. (But it does work under Cygwin, which implements its own
1814 struct timeval sleep;
1815 sleep.tv_sec = (long) seconds;
1816 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1817 select (0, NULL, NULL, NULL, &sleep);
1818 /* If select returns -1 and errno is EINTR, it means we were
1819 interrupted by a signal. But without knowing how long we've
1820 actually slept, we can't return to sleep. Using gettimeofday to
1821 track sleeps is slow and unreliable due to clock skew. */
1822 #else /* not HAVE_SELECT */
1824 #endif /* not HAVE_SELECT */
1825 #endif /* not HAVE_USLEEP */
1826 #endif /* not HAVE_NANOSLEEP */
1829 #endif /* not WINDOWS */
1831 /* Encode the string STR of length LENGTH to base64 format and place it
1832 to B64STORE. The output will be \0-terminated, and must point to a
1833 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1834 returns the length of the resulting base64 data, not counting the
1837 This implementation will not emit newlines after 76 characters of
1841 base64_encode (const char *str, int length, char *b64store)
1843 /* Conversion table. */
1844 static char tbl[64] = {
1845 'A','B','C','D','E','F','G','H',
1846 'I','J','K','L','M','N','O','P',
1847 'Q','R','S','T','U','V','W','X',
1848 'Y','Z','a','b','c','d','e','f',
1849 'g','h','i','j','k','l','m','n',
1850 'o','p','q','r','s','t','u','v',
1851 'w','x','y','z','0','1','2','3',
1852 '4','5','6','7','8','9','+','/'
1855 const unsigned char *s = (const unsigned char *) str;
1858 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1859 for (i = 0; i < length; i += 3)
1861 *p++ = tbl[s[0] >> 2];
1862 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1863 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1864 *p++ = tbl[s[2] & 0x3f];
1868 /* Pad the result if necessary... */
1869 if (i == length + 1)
1871 else if (i == length + 2)
1872 *(p - 1) = *(p - 2) = '=';
1874 /* ...and zero-terminate it. */
1877 return p - b64store;
1880 #define IS_ASCII(c) (((c) & 0x80) == 0)
1881 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1883 /* Get next character from the string, except that non-base64
1884 characters are ignored, as mandated by rfc2045. */
1885 #define NEXT_BASE64_CHAR(c, p) do { \
1887 } while (c != '\0' && !IS_BASE64 (c))
1889 /* Decode data from BASE64 (assumed to be encoded as base64) into
1890 memory pointed to by TO. TO should be large enough to accomodate
1891 the decoded data, which is guaranteed to be less than
1894 Since TO is assumed to contain binary data, it is not
1895 NUL-terminated. The function returns the length of the data
1896 written to TO. -1 is returned in case of error caused by malformed
1900 base64_decode (const char *base64, char *to)
1902 /* Table of base64 values for first 128 characters. Note that this
1903 assumes ASCII (but so does Wget in other places). */
1904 static short base64_char_to_value[128] =
1906 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1907 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1908 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1909 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1910 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1911 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1912 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1913 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1914 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1915 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1916 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1917 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1918 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1921 const char *p = base64;
1927 unsigned long value;
1929 /* Process first byte of a quadruplet. */
1930 NEXT_BASE64_CHAR (c, p);
1934 return -1; /* illegal '=' while decoding base64 */
1935 value = base64_char_to_value[c] << 18;
1937 /* Process scond byte of a quadruplet. */
1938 NEXT_BASE64_CHAR (c, p);
1940 return -1; /* premature EOF while decoding base64 */
1942 return -1; /* illegal `=' while decoding base64 */
1943 value |= base64_char_to_value[c] << 12;
1946 /* Process third byte of a quadruplet. */
1947 NEXT_BASE64_CHAR (c, p);
1949 return -1; /* premature EOF while decoding base64 */
1953 NEXT_BASE64_CHAR (c, p);
1955 return -1; /* premature EOF while decoding base64 */
1957 return -1; /* padding `=' expected but not found */
1961 value |= base64_char_to_value[c] << 6;
1962 *q++ = 0xff & value >> 8;
1964 /* Process fourth byte of a quadruplet. */
1965 NEXT_BASE64_CHAR (c, p);
1967 return -1; /* premature EOF while decoding base64 */
1971 value |= base64_char_to_value[c];
1972 *q++ = 0xff & value;
1980 #undef NEXT_BASE64_CHAR
1982 /* Simple merge sort for use by stable_sort. Implementation courtesy
1986 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1987 int (*cmpfun) PARAMS ((const void *, const void *)))
1989 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1993 size_t mid = (to + from) / 2;
1994 mergesort_internal (base, temp, size, from, mid, cmpfun);
1995 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
1998 for (k = from; (i <= mid) && (j <= to); k++)
1999 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2000 memcpy (ELT (temp, k), ELT (base, i++), size);
2002 memcpy (ELT (temp, k), ELT (base, j++), size);
2004 memcpy (ELT (temp, k++), ELT (base, i++), size);
2006 memcpy (ELT (temp, k++), ELT (base, j++), size);
2007 for (k = from; k <= to; k++)
2008 memcpy (ELT (base, k), ELT (temp, k), size);
2013 /* Stable sort with interface exactly like standard library's qsort.
2014 Uses mergesort internally, allocating temporary storage with
2018 stable_sort (void *base, size_t nmemb, size_t size,
2019 int (*cmpfun) PARAMS ((const void *, const void *)))
2023 void *temp = alloca (nmemb * size * sizeof (void *));
2024 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);