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 atime and mtime equal to the time
361 specified with TM. */
363 touch (const char *file, time_t tm)
365 #ifdef HAVE_STRUCT_UTIMBUF
366 struct utimbuf times;
367 times.actime = times.modtime = tm;
370 times[0] = times[1] = tm;
373 if (utime (file, ×) == -1)
374 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
377 /* Checks if FILE is a symbolic link, and removes it if it is. Does
378 nothing under MS-Windows. */
380 remove_link (const char *file)
385 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
387 DEBUGP (("Unlinking %s (symlink).\n", file));
390 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
391 file, strerror (errno));
396 /* Does FILENAME exist? This is quite a lousy implementation, since
397 it supplies no error codes -- only a yes-or-no answer. Thus it
398 will return that a file does not exist if, e.g., the directory is
399 unreadable. I don't mind it too much currently, though. The
400 proper way should, of course, be to have a third, error state,
401 other than true/false, but that would introduce uncalled-for
402 additional complexity to the callers. */
404 file_exists_p (const char *filename)
407 return access (filename, F_OK) >= 0;
410 return stat (filename, &buf) >= 0;
414 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
415 Returns 0 on error. */
417 file_non_directory_p (const char *path)
420 /* Use lstat() rather than stat() so that symbolic links pointing to
421 directories can be identified correctly. */
422 if (lstat (path, &buf) != 0)
424 return S_ISDIR (buf.st_mode) ? 0 : 1;
427 /* Return the size of file named by FILENAME, or -1 if it cannot be
428 opened or seeked into. */
430 file_size (const char *filename)
432 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
434 /* We use fseek rather than stat to determine the file size because
435 that way we can also verify that the file is readable without
436 explicitly checking for permissions. Inspired by the POST patch
438 FILE *fp = fopen (filename, "rb");
441 fseeko (fp, 0, SEEK_END);
447 if (stat (filename, &st) < 0)
453 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
454 doesn't exist is found. Return a freshly allocated copy of the
458 unique_name_1 (const char *prefix)
461 int plen = strlen (prefix);
462 char *template = (char *)alloca (plen + 1 + 24);
463 char *template_tail = template + plen;
465 memcpy (template, prefix, plen);
466 *template_tail++ = '.';
469 number_to_string (template_tail, count++);
470 while (file_exists_p (template));
472 return xstrdup (template);
475 /* Return a unique file name, based on FILE.
477 More precisely, if FILE doesn't exist, it is returned unmodified.
478 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
479 file name that doesn't exist is returned.
481 The resulting file is not created, only verified that it didn't
482 exist at the point in time when the function was called.
483 Therefore, where security matters, don't rely that the file created
484 by this function exists until you open it with O_EXCL or
487 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
488 string. Otherwise, it may return FILE if the file doesn't exist
489 (and therefore doesn't need changing). */
492 unique_name (const char *file, int allow_passthrough)
494 /* If the FILE itself doesn't exist, return it without
496 if (!file_exists_p (file))
497 return allow_passthrough ? (char *)file : xstrdup (file);
499 /* Otherwise, find a numeric suffix that results in unused file name
501 return unique_name_1 (file);
504 /* Create a file based on NAME, except without overwriting an existing
505 file with that name. Providing O_EXCL is correctly implemented,
506 this function does not have the race condition associated with
507 opening the file returned by unique_name. */
510 unique_create (const char *name, int binary, char **opened_name)
512 /* unique file name, based on NAME */
513 char *uname = unique_name (name, 0);
515 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
518 uname = unique_name (name, 0);
520 if (opened_name && fp != NULL)
523 *opened_name = uname;
535 /* Open the file for writing, with the addition that the file is
536 opened "exclusively". This means that, if the file already exists,
537 this function will *fail* and errno will be set to EEXIST. If
538 BINARY is set, the file will be opened in binary mode, equivalent
541 If opening the file fails for any reason, including the file having
542 previously existed, this function returns NULL and sets errno
546 fopen_excl (const char *fname, int binary)
550 int flags = O_WRONLY | O_CREAT | O_EXCL;
555 fd = open (fname, flags, 0666);
558 return fdopen (fd, binary ? "wb" : "w");
559 #else /* not O_EXCL */
560 /* Manually check whether the file exists. This is prone to race
561 conditions, but systems without O_EXCL haven't deserved
563 if (file_exists_p (fname))
568 return fopen (fname, binary ? "wb" : "w");
569 #endif /* not O_EXCL */
572 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
573 are missing, create them first. In case any mkdir() call fails,
574 return its error status. Returns 0 on successful completion.
576 The behaviour of this function should be identical to the behaviour
577 of `mkdir -p' on systems where mkdir supports the `-p' option. */
579 make_directory (const char *directory)
581 int i, ret, quit = 0;
584 /* Make a copy of dir, to be able to write to it. Otherwise, the
585 function is unsafe if called with a read-only char *argument. */
586 STRDUP_ALLOCA (dir, directory);
588 /* If the first character of dir is '/', skip it (and thus enable
589 creation of absolute-pathname directories. */
590 for (i = (*dir == '/'); 1; ++i)
592 for (; dir[i] && dir[i] != '/'; i++)
597 /* Check whether the directory already exists. Allow creation of
598 of intermediate directories to fail, as the initial path components
599 are not necessarily directories! */
600 if (!file_exists_p (dir))
601 ret = mkdir (dir, 0777);
612 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
613 should be a file name.
615 file_merge("/foo/bar", "baz") => "/foo/baz"
616 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
617 file_merge("foo", "bar") => "bar"
619 In other words, it's a simpler and gentler version of uri_merge_1. */
622 file_merge (const char *base, const char *file)
625 const char *cut = (const char *)strrchr (base, '/');
628 return xstrdup (file);
630 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
631 memcpy (result, base, cut - base);
632 result[cut - base] = '/';
633 strcpy (result + (cut - base) + 1, file);
638 static int in_acclist PARAMS ((const char *const *, const char *, int));
640 /* Determine whether a file is acceptable to be followed, according to
641 lists of patterns to accept/reject. */
643 acceptable (const char *s)
647 while (l && s[l] != '/')
654 return (in_acclist ((const char *const *)opt.accepts, s, 1)
655 && !in_acclist ((const char *const *)opt.rejects, s, 1));
657 return in_acclist ((const char *const *)opt.accepts, s, 1);
659 else if (opt.rejects)
660 return !in_acclist ((const char *const *)opt.rejects, s, 1);
664 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
665 `/something', frontcmp() will return 1 only if S2 begins with
666 `/something'. Otherwise, 0 is returned. */
668 frontcmp (const char *s1, const char *s2)
670 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
674 /* Iterate through STRLIST, and return the first element that matches
675 S, through wildcards or front comparison (as appropriate). */
677 proclist (char **strlist, const char *s, enum accd flags)
681 for (x = strlist; *x; x++)
682 if (has_wildcards_p (*x))
684 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
689 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
696 /* Returns whether DIRECTORY is acceptable for download, wrt the
697 include/exclude lists.
699 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
700 and absolute paths may be freely intermixed. */
702 accdir (const char *directory, enum accd flags)
704 /* Remove starting '/'. */
705 if (flags & ALLABS && *directory == '/')
709 if (!proclist (opt.includes, directory, flags))
714 if (proclist (opt.excludes, directory, flags))
720 /* Return non-zero if STRING ends with TAIL. For instance:
722 match_tail ("abc", "bc", 0) -> 1
723 match_tail ("abc", "ab", 0) -> 0
724 match_tail ("abc", "abc", 0) -> 1
726 If FOLD_CASE_P is non-zero, the comparison will be
730 match_tail (const char *string, const char *tail, int fold_case_p)
734 /* We want this to be fast, so we code two loops, one with
735 case-folding, one without. */
739 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
740 if (string[i] != tail[j])
745 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
746 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
750 /* If the tail was exhausted, the match was succesful. */
757 /* Checks whether string S matches each element of ACCEPTS. A list
758 element are matched either with fnmatch() or match_tail(),
759 according to whether the element contains wildcards or not.
761 If the BACKWARD is 0, don't do backward comparison -- just compare
764 in_acclist (const char *const *accepts, const char *s, int backward)
766 for (; *accepts; accepts++)
768 if (has_wildcards_p (*accepts))
770 /* fnmatch returns 0 if the pattern *does* match the
772 if (fnmatch (*accepts, s, 0) == 0)
779 if (match_tail (s, *accepts, 0))
784 if (!strcmp (s, *accepts))
792 /* Return the location of STR's suffix (file extension). Examples:
793 suffix ("foo.bar") -> "bar"
794 suffix ("foo.bar.baz") -> "baz"
795 suffix ("/foo/bar") -> NULL
796 suffix ("/foo.bar/baz") -> NULL */
798 suffix (const char *str)
802 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
806 return (char *)str + i;
811 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
815 has_wildcards_p (const char *s)
818 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
823 /* Return non-zero if FNAME ends with a typical HTML suffix. The
824 following (case-insensitive) suffixes are presumed to be HTML files:
828 ?html (`?' matches one character)
830 #### CAVEAT. This is not necessarily a good indication that FNAME
831 refers to a file that contains HTML! */
833 has_html_suffix_p (const char *fname)
837 if ((suf = suffix (fname)) == NULL)
839 if (!strcasecmp (suf, "html"))
841 if (!strcasecmp (suf, "htm"))
843 if (suf[0] && !strcasecmp (suf + 1, "html"))
848 /* Read a line from FP and return the pointer to freshly allocated
849 storage. The storage space is obtained through malloc() and should
850 be freed with free() when it is no longer needed.
852 The length of the line is not limited, except by available memory.
853 The newline character at the end of line is retained. The line is
854 terminated with a zero character.
856 After end-of-file is encountered without anything being read, NULL
857 is returned. NULL is also returned on error. To distinguish
858 between these two cases, use the stdio function ferror(). */
861 read_whole_line (FILE *fp)
865 char *line = (char *)xmalloc (bufsize);
867 while (fgets (line + length, bufsize - length, fp))
869 length += strlen (line + length);
871 /* Possible for example when reading from a binary file where
872 a line begins with \0. */
875 if (line[length - 1] == '\n')
878 /* fgets() guarantees to read the whole line, or to use up the
879 space we've given it. We can double the buffer
882 line = xrealloc (line, bufsize);
884 if (length == 0 || ferror (fp))
889 if (length + 1 < bufsize)
890 /* Relieve the memory from our exponential greediness. We say
891 `length + 1' because the terminating \0 is not included in
892 LENGTH. We don't need to zero-terminate the string ourselves,
893 though, because fgets() does that. */
894 line = xrealloc (line, length + 1);
898 /* Read FILE into memory. A pointer to `struct file_memory' are
899 returned; use struct element `content' to access file contents, and
900 the element `length' to know the file length. `content' is *not*
901 zero-terminated, and you should *not* read or write beyond the [0,
902 length) range of characters.
904 After you are done with the file contents, call read_file_free to
907 Depending on the operating system and the type of file that is
908 being read, read_file() either mmap's the file into memory, or
909 reads the file into the core using read().
911 If file is named "-", fileno(stdin) is used for reading instead.
912 If you want to read from a real file named "-", use "./-" instead. */
915 read_file (const char *file)
918 struct file_memory *fm;
920 int inhibit_close = 0;
922 /* Some magic in the finest tradition of Perl and its kin: if FILE
923 is "-", just use stdin. */
928 /* Note that we don't inhibit mmap() in this case. If stdin is
929 redirected from a regular file, mmap() will still work. */
932 fd = open (file, O_RDONLY);
935 fm = xnew (struct file_memory);
940 if (fstat (fd, &buf) < 0)
942 fm->length = buf.st_size;
943 /* NOTE: As far as I know, the callers of this function never
944 modify the file text. Relying on this would enable us to
945 specify PROT_READ and MAP_SHARED for a marginal gain in
946 efficiency, but at some cost to generality. */
947 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
949 if (fm->content == (char *)MAP_FAILED)
959 /* The most common reason why mmap() fails is that FD does not point
960 to a plain file. However, it's also possible that mmap() doesn't
961 work for a particular type of file. Therefore, whenever mmap()
962 fails, we just fall back to the regular method. */
963 #endif /* HAVE_MMAP */
966 size = 512; /* number of bytes fm->contents can
967 hold at any given time. */
968 fm->content = xmalloc (size);
972 if (fm->length > size / 2)
974 /* #### I'm not sure whether the whole exponential-growth
975 thing makes sense with kernel read. On Linux at least,
976 read() refuses to read more than 4K from a file at a
977 single chunk anyway. But other Unixes might optimize it
978 better, and it doesn't *hurt* anything, so I'm leaving
981 /* Normally, we grow SIZE exponentially to make the number
982 of calls to read() and realloc() logarithmic in relation
983 to file size. However, read() can read an amount of data
984 smaller than requested, and it would be unreasonable to
985 double SIZE every time *something* was read. Therefore,
986 we double SIZE only when the length exceeds half of the
987 entire allocated size. */
989 fm->content = xrealloc (fm->content, size);
991 nread = read (fd, fm->content + fm->length, size - fm->length);
993 /* Successful read. */
1004 if (size > fm->length && fm->length != 0)
1005 /* Due to exponential growth of fm->content, the allocated region
1006 might be much larger than what is actually needed. */
1007 fm->content = xrealloc (fm->content, fm->length);
1014 xfree (fm->content);
1019 /* Release the resources held by FM. Specifically, this calls
1020 munmap() or xfree() on fm->content, depending whether mmap or
1021 malloc/read were used to read in the file. It also frees the
1022 memory needed to hold the FM structure itself. */
1025 read_file_free (struct file_memory *fm)
1030 munmap (fm->content, fm->length);
1035 xfree (fm->content);
1040 /* Free the pointers in a NULL-terminated vector of pointers, then
1041 free the pointer itself. */
1043 free_vec (char **vec)
1054 /* Append vector V2 to vector V1. The function frees V2 and
1055 reallocates V1 (thus you may not use the contents of neither
1056 pointer after the call). If V1 is NULL, V2 is returned. */
1058 merge_vecs (char **v1, char **v2)
1068 /* To avoid j == 0 */
1073 for (i = 0; v1[i]; i++);
1075 for (j = 0; v2[j]; j++);
1076 /* Reallocate v1. */
1077 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1078 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1083 /* Sometimes it's useful to create "sets" of strings, i.e. special
1084 hash tables where you want to store strings as keys and merely
1085 query for their existence. Here is a set of utility routines that
1086 makes that transparent. */
1089 string_set_add (struct hash_table *ht, const char *s)
1091 /* First check whether the set element already exists. If it does,
1092 do nothing so that we don't have to free() the old element and
1093 then strdup() a new one. */
1094 if (hash_table_contains (ht, s))
1097 /* We use "1" as value. It provides us a useful and clear arbitrary
1098 value, and it consumes no memory -- the pointers to the same
1099 string "1" will be shared by all the key-value pairs in all `set'
1101 hash_table_put (ht, xstrdup (s), "1");
1104 /* Synonym for hash_table_contains... */
1107 string_set_contains (struct hash_table *ht, const char *s)
1109 return hash_table_contains (ht, s);
1113 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1115 char ***arrayptr = (char ***) arg;
1116 *(*arrayptr)++ = (char *) key;
1120 /* Convert the specified string set to array. ARRAY should be large
1121 enough to hold hash_table_count(ht) char pointers. */
1123 void string_set_to_array (struct hash_table *ht, char **array)
1125 hash_table_map (ht, string_set_to_array_mapper, &array);
1129 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1136 string_set_free (struct hash_table *ht)
1138 hash_table_map (ht, string_set_free_mapper, NULL);
1139 hash_table_destroy (ht);
1143 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1150 /* Another utility function: call free() on all keys and values of HT. */
1153 free_keys_and_values (struct hash_table *ht)
1155 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1159 /* Add thousand separators to a number already in string form. Used
1160 by with_thousand_seps and with_thousand_seps_large. */
1163 add_thousand_seps (const char *repr)
1165 static char outbuf[48];
1170 /* Reset the pointers. */
1174 /* Ignore the sign for the purpose of adding thousand
1181 /* How many digits before the first separator? */
1182 mod = strlen (inptr) % 3;
1184 for (i = 0; i < mod; i++)
1185 *outptr++ = inptr[i];
1186 /* Now insert the rest of them, putting separator before every
1188 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1190 if (i % 3 == 0 && i1 != 0)
1192 *outptr++ = inptr[i1];
1194 /* Zero-terminate the string. */
1199 /* Return a static pointer to the number printed with thousand
1200 separators inserted at the right places. */
1203 with_thousand_seps (wgint l)
1206 /* Print the number into the buffer. */
1207 number_to_string (inbuf, l);
1208 return add_thousand_seps (inbuf);
1211 /* Write a string representation of LARGE_INT NUMBER into the provided
1214 It would be dangerous to use sprintf, because the code wouldn't
1215 work on a machine with gcc-provided long long support, but without
1216 libc support for "%lld". However, such old systems platforms
1217 typically lack snprintf and will end up using our version, which
1218 does support "%lld" whereever long longs are available. */
1221 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1223 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1226 /* The same as with_thousand_seps, but works on LARGE_INT. */
1229 with_thousand_seps_large (LARGE_INT l)
1232 large_int_to_string (inbuf, sizeof (inbuf), l);
1233 return add_thousand_seps (inbuf);
1236 /* N, a byte quantity, is converted to a human-readable abberviated
1237 form a la sizes printed by `ls -lh'. The result is written to a
1238 static buffer, a pointer to which is returned.
1240 Unlike `with_thousand_seps', this approximates to the nearest unit.
1241 Quoting GNU libit: "Most people visually process strings of 3-4
1242 digits effectively, but longer strings of digits are more prone to
1243 misinterpretation. Hence, converting to an abbreviated form
1244 usually improves readability."
1246 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1247 original computer science meaning of "powers of 1024". Powers of
1248 1000 would be useless since Wget already displays sizes with
1249 thousand separators. We don't use the "*bibyte" names invented in
1250 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1251 discusses this in some detail. */
1254 human_readable (wgint n)
1256 /* These suffixes are compatible with those of GNU `ls -lh'. */
1257 static char powers[] =
1259 'K', /* kilobyte, 2^10 bytes */
1260 'M', /* megabyte, 2^20 bytes */
1261 'G', /* gigabyte, 2^30 bytes */
1262 'T', /* terabyte, 2^40 bytes */
1263 'P', /* petabyte, 2^50 bytes */
1264 'E', /* exabyte, 2^60 bytes */
1269 /* If the quantity is smaller than 1K, just print it. */
1272 snprintf (buf, sizeof (buf), "%d", (int) n);
1276 /* Loop over powers, dividing N with 1024 in each iteration. This
1277 works unchanged for all sizes of wgint, while still avoiding
1278 non-portable `long double' arithmetic. */
1279 for (i = 0; i < countof (powers); i++)
1281 /* At each iteration N is greater than the *subsequent* power.
1282 That way N/1024.0 produces a decimal number in the units of
1284 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1286 /* Must cast to long first because MS VC can't directly cast
1287 __int64 to double. (This is safe because N is known to
1289 double val = (double) (long) n / 1024.0;
1290 /* Print values smaller than 10 with one decimal digits, and
1291 others without any decimals. */
1292 snprintf (buf, sizeof (buf), "%.*f%c",
1293 val < 10 ? 1 : 0, val, powers[i]);
1298 return NULL; /* unreached */
1301 /* Count the digits in the provided number. Used to allocate space
1302 when printing numbers. */
1305 numdigit (wgint number)
1309 ++cnt; /* accomodate '-' */
1310 while ((number /= 10) != 0)
1315 #define PR(mask) *p++ = n / (mask) + '0'
1317 /* DIGITS_<D> is used to print a D-digit number and should be called
1318 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1319 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1320 Recursively this continues until DIGITS_1 is invoked. */
1322 #define DIGITS_1(mask) PR (mask)
1323 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1324 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1325 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1326 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1327 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1328 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1329 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1330 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1331 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1333 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1335 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1336 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1337 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1338 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1339 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1340 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1341 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1342 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1343 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1345 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1346 cases and to portably support strange sizes of wgint. Ideally this
1347 would just use "%j" and intmax_t, but many systems don't support
1348 it, so it's used only if nothing else works. */
1349 #if SIZEOF_LONG >= SIZEOF_WGINT
1350 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1352 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1353 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1356 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1358 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1363 /* Shorthand for casting to wgint. */
1366 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1367 `sprintf(buffer, "%lld", (long long) number)', only typically much
1368 faster and portable to machines without long long.
1370 The speedup may make a difference in programs that frequently
1371 convert numbers to strings. Some implementations of sprintf,
1372 particularly the one in GNU libc, have been known to be extremely
1373 slow when converting integers to strings.
1375 Return the pointer to the location where the terminating zero was
1376 printed. (Equivalent to calling buffer+strlen(buffer) after the
1379 BUFFER should be big enough to accept as many bytes as you expect
1380 the number to take up. On machines with 64-bit longs the maximum
1381 needed size is 24 bytes. That includes the digits needed for the
1382 largest 64-bit number, the `-' sign in case it's negative, and the
1383 terminating '\0'. */
1386 number_to_string (char *buffer, wgint number)
1391 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1392 /* We are running in a strange or misconfigured environment. Let
1393 sprintf cope with it. */
1394 SPRINTF_WGINT (buffer, n);
1395 p += strlen (buffer);
1396 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1402 /* -n would overflow. Have sprintf deal with this. */
1403 SPRINTF_WGINT (buffer, n);
1404 p += strlen (buffer);
1412 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1413 way printing any N is fully open-coded without a loop or jump.
1414 (Also see description of DIGITS_*.) */
1416 if (n < 10) DIGITS_1 (1);
1417 else if (n < 100) DIGITS_2 (10);
1418 else if (n < 1000) DIGITS_3 (100);
1419 else if (n < 10000) DIGITS_4 (1000);
1420 else if (n < 100000) DIGITS_5 (10000);
1421 else if (n < 1000000) DIGITS_6 (100000);
1422 else if (n < 10000000) DIGITS_7 (1000000);
1423 else if (n < 100000000) DIGITS_8 (10000000);
1424 else if (n < 1000000000) DIGITS_9 (100000000);
1425 #if SIZEOF_WGINT == 4
1426 /* wgint is 32 bits wide: no number has more than 10 digits. */
1427 else DIGITS_10 (1000000000);
1429 /* wgint is 64 bits wide: handle numbers with more than 9 decimal
1430 digits. Constants are constructed by compile-time multiplication
1431 to avoid dealing with different notations for 64-bit constants
1432 (nnnL, nnnLL, and nnnI64, depending on the compiler). */
1433 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1434 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1435 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1436 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1437 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1438 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1439 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1440 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1441 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1442 else DIGITS_19 (1000000000*(W)1000000000);
1446 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1475 /* Print NUMBER to a statically allocated string and return a pointer
1476 to the printed representation.
1478 This function is intended to be used in conjunction with printf.
1479 It is hard to portably print wgint values:
1480 a) you cannot use printf("%ld", number) because wgint can be long
1481 long on 32-bit machines with LFS.
1482 b) you cannot use printf("%lld", number) because NUMBER could be
1483 long on 32-bit machines without LFS, or on 64-bit machines,
1484 which do not require LFS. Also, Windows doesn't support %lld.
1485 c) you cannot use printf("%j", (int_max_t) number) because not all
1486 versions of printf support "%j", the most notable being the one
1488 d) you cannot #define WGINT_FMT to the appropriate format and use
1489 printf(WGINT_FMT, number) because that would break translations
1490 for user-visible messages, such as printf("Downloaded: %d
1493 What you should use instead is printf("%s", number_to_static_string
1496 CAVEAT: since the function returns pointers to static data, you
1497 must be careful to copy its result before calling it again.
1498 However, to make it more useful with printf, the function maintains
1499 an internal ring of static buffers to return. That way things like
1500 printf("%s %s", number_to_static_string (num1),
1501 number_to_static_string (num2)) work as expected. Three buffers
1502 are currently used, which means that "%s %s %s" will work, but "%s
1503 %s %s %s" won't. If you need to print more than three wgints,
1504 bump the RING_SIZE (or rethink your message.) */
1507 number_to_static_string (wgint number)
1509 static char ring[RING_SIZE][24];
1511 char *buf = ring[ringpos];
1512 number_to_string (buf, number);
1513 ringpos = (ringpos + 1) % RING_SIZE;
1517 /* Determine the width of the terminal we're running on. If that's
1518 not possible, return 0. */
1521 determine_screen_width (void)
1523 /* If there's a way to get the terminal size using POSIX
1524 tcgetattr(), somebody please tell me. */
1529 if (opt.lfilename != NULL)
1532 fd = fileno (stderr);
1533 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1534 return 0; /* most likely ENOTTY */
1537 #else /* not TIOCGWINSZ */
1539 CONSOLE_SCREEN_BUFFER_INFO csbi;
1540 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1542 return csbi.dwSize.X;
1543 # else /* neither WINDOWS nor TIOCGWINSZ */
1545 #endif /* neither WINDOWS nor TIOCGWINSZ */
1546 #endif /* not TIOCGWINSZ */
1549 /* Return a random number between 0 and MAX-1, inclusive.
1551 If MAX is greater than the value of RAND_MAX+1 on the system, the
1552 returned value will be in the range [0, RAND_MAX]. This may be
1553 fixed in a future release.
1555 The random number generator is seeded automatically the first time
1558 This uses rand() for portability. It has been suggested that
1559 random() offers better randomness, but this is not required for
1560 Wget, so I chose to go for simplicity and use rand
1563 DO NOT use this for cryptographic purposes. It is only meant to be
1564 used in situations where quality of the random numbers returned
1565 doesn't really matter. */
1568 random_number (int max)
1576 srand (time (NULL));
1581 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1582 and enforce that assumption by masking other bits. */
1584 # define RAND_MAX 32767
1588 /* This is equivalent to rand() % max, but uses the high-order bits
1589 for better randomness on architecture where rand() is implemented
1590 using a simple congruential generator. */
1592 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1593 return (int)bounded;
1596 /* Return a random uniformly distributed floating point number in the
1597 [0, 1) range. The precision of returned numbers is 9 digits.
1599 Modify this to use erand48() where available! */
1604 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1605 sure it's greater than 1000. */
1606 int rnd1 = random_number (1000);
1607 int rnd2 = random_number (1000);
1608 int rnd3 = random_number (1000);
1609 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1612 /* Implementation of run_with_timeout, a generic timeout-forcing
1613 routine for systems with Unix-like signal handling. */
1615 #ifdef USE_SIGNAL_TIMEOUT
1616 # ifdef HAVE_SIGSETJMP
1617 # define SETJMP(env) sigsetjmp (env, 1)
1619 static sigjmp_buf run_with_timeout_env;
1622 abort_run_with_timeout (int sig)
1624 assert (sig == SIGALRM);
1625 siglongjmp (run_with_timeout_env, -1);
1627 # else /* not HAVE_SIGSETJMP */
1628 # define SETJMP(env) setjmp (env)
1630 static jmp_buf run_with_timeout_env;
1633 abort_run_with_timeout (int sig)
1635 assert (sig == SIGALRM);
1636 /* We don't have siglongjmp to preserve the set of blocked signals;
1637 if we longjumped out of the handler at this point, SIGALRM would
1638 remain blocked. We must unblock it manually. */
1639 int mask = siggetmask ();
1640 mask &= ~sigmask (SIGALRM);
1643 /* Now it's safe to longjump. */
1644 longjmp (run_with_timeout_env, -1);
1646 # endif /* not HAVE_SIGSETJMP */
1648 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1649 setitimer where available, alarm otherwise.
1651 TIMEOUT should be non-zero. If the timeout value is so small that
1652 it would be rounded to zero, it is rounded to the least legal value
1653 instead (1us for setitimer, 1s for alarm). That ensures that
1654 SIGALRM will be delivered in all cases. */
1657 alarm_set (double timeout)
1660 /* Use the modern itimer interface. */
1661 struct itimerval itv;
1663 itv.it_value.tv_sec = (long) timeout;
1664 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1665 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1666 /* Ensure that we wait for at least the minimum interval.
1667 Specifying zero would mean "wait forever". */
1668 itv.it_value.tv_usec = 1;
1669 setitimer (ITIMER_REAL, &itv, NULL);
1670 #else /* not ITIMER_REAL */
1671 /* Use the old alarm() interface. */
1672 int secs = (int) timeout;
1674 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1675 because alarm(0) means "never deliver the alarm", i.e. "wait
1676 forever", which is not what someone who specifies a 0.5s
1677 timeout would expect. */
1680 #endif /* not ITIMER_REAL */
1683 /* Cancel the alarm set with alarm_set. */
1689 struct itimerval disable;
1691 setitimer (ITIMER_REAL, &disable, NULL);
1692 #else /* not ITIMER_REAL */
1694 #endif /* not ITIMER_REAL */
1697 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1698 seconds. Returns non-zero if the function was interrupted with a
1699 timeout, zero otherwise.
1701 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1702 using setitimer() or alarm(). The timeout is enforced by
1703 longjumping out of the SIGALRM handler. This has several
1704 advantages compared to the traditional approach of relying on
1705 signals causing system calls to exit with EINTR:
1707 * The callback function is *forcibly* interrupted after the
1708 timeout expires, (almost) regardless of what it was doing and
1709 whether it was in a syscall. For example, a calculation that
1710 takes a long time is interrupted as reliably as an IO
1713 * It works with both SYSV and BSD signals because it doesn't
1714 depend on the default setting of SA_RESTART.
1716 * It doesn't require special handler setup beyond a simple call
1717 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1720 The only downside is that, if FUN allocates internal resources that
1721 are normally freed prior to exit from the functions, they will be
1722 lost in case of timeout. */
1725 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1735 signal (SIGALRM, abort_run_with_timeout);
1736 if (SETJMP (run_with_timeout_env) != 0)
1738 /* Longjumped out of FUN with a timeout. */
1739 signal (SIGALRM, SIG_DFL);
1742 alarm_set (timeout);
1745 /* Preserve errno in case alarm() or signal() modifies it. */
1746 saved_errno = errno;
1748 signal (SIGALRM, SIG_DFL);
1749 errno = saved_errno;
1754 #else /* not USE_SIGNAL_TIMEOUT */
1757 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1758 define it under Windows, because Windows has its own version of
1759 run_with_timeout that uses threads. */
1762 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1767 #endif /* not WINDOWS */
1768 #endif /* not USE_SIGNAL_TIMEOUT */
1772 /* Sleep the specified amount of seconds. On machines without
1773 nanosleep(), this may sleep shorter if interrupted by signals. */
1776 xsleep (double seconds)
1778 #ifdef HAVE_NANOSLEEP
1779 /* nanosleep is the preferred interface because it offers high
1780 accuracy and, more importantly, because it allows us to reliably
1781 restart receiving a signal such as SIGWINCH. (There was an
1782 actual Debian bug report about --limit-rate malfunctioning while
1783 the terminal was being resized.) */
1784 struct timespec sleep, remaining;
1785 sleep.tv_sec = (long) seconds;
1786 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1787 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1788 /* If nanosleep has been interrupted by a signal, adjust the
1789 sleeping period and return to sleep. */
1791 #else /* not HAVE_NANOSLEEP */
1793 /* If usleep is available, use it in preference to select. */
1796 /* On some systems, usleep cannot handle values larger than
1797 1,000,000. If the period is larger than that, use sleep
1798 first, then add usleep for subsecond accuracy. */
1800 seconds -= (long) seconds;
1802 usleep (seconds * 1000000);
1803 #else /* not HAVE_USLEEP */
1805 /* Note that, although Windows supports select, this sleeping
1806 strategy doesn't work there because Winsock's select doesn't
1807 implement timeout when it is passed NULL pointers for all fd
1808 sets. (But it does work under Cygwin, which implements its own
1810 struct timeval sleep;
1811 sleep.tv_sec = (long) seconds;
1812 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1813 select (0, NULL, NULL, NULL, &sleep);
1814 /* If select returns -1 and errno is EINTR, it means we were
1815 interrupted by a signal. But without knowing how long we've
1816 actually slept, we can't return to sleep. Using gettimeofday to
1817 track sleeps is slow and unreliable due to clock skew. */
1818 #else /* not HAVE_SELECT */
1820 #endif /* not HAVE_SELECT */
1821 #endif /* not HAVE_USLEEP */
1822 #endif /* not HAVE_NANOSLEEP */
1825 #endif /* not WINDOWS */
1827 /* Encode the string STR of length LENGTH to base64 format and place it
1828 to B64STORE. The output will be \0-terminated, and must point to a
1829 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1830 returns the length of the resulting base64 data, not counting the
1833 This implementation will not emit newlines after 76 characters of
1837 base64_encode (const char *str, int length, char *b64store)
1839 /* Conversion table. */
1840 static char tbl[64] = {
1841 'A','B','C','D','E','F','G','H',
1842 'I','J','K','L','M','N','O','P',
1843 'Q','R','S','T','U','V','W','X',
1844 'Y','Z','a','b','c','d','e','f',
1845 'g','h','i','j','k','l','m','n',
1846 'o','p','q','r','s','t','u','v',
1847 'w','x','y','z','0','1','2','3',
1848 '4','5','6','7','8','9','+','/'
1851 const unsigned char *s = (const unsigned char *) str;
1854 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1855 for (i = 0; i < length; i += 3)
1857 *p++ = tbl[s[0] >> 2];
1858 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1859 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1860 *p++ = tbl[s[2] & 0x3f];
1864 /* Pad the result if necessary... */
1865 if (i == length + 1)
1867 else if (i == length + 2)
1868 *(p - 1) = *(p - 2) = '=';
1870 /* ...and zero-terminate it. */
1873 return p - b64store;
1876 #define IS_ASCII(c) (((c) & 0x80) == 0)
1877 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1879 /* Get next character from the string, except that non-base64
1880 characters are ignored, as mandated by rfc2045. */
1881 #define NEXT_BASE64_CHAR(c, p) do { \
1883 } while (c != '\0' && !IS_BASE64 (c))
1885 /* Decode data from BASE64 (assumed to be encoded as base64) into
1886 memory pointed to by TO. TO should be large enough to accomodate
1887 the decoded data, which is guaranteed to be less than
1890 Since TO is assumed to contain binary data, it is not
1891 NUL-terminated. The function returns the length of the data
1892 written to TO. -1 is returned in case of error caused by malformed
1896 base64_decode (const char *base64, char *to)
1898 /* Table of base64 values for first 128 characters. Note that this
1899 assumes ASCII (but so does Wget in other places). */
1900 static short base64_char_to_value[128] =
1902 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1903 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1904 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1905 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1906 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1907 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1908 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1909 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1910 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1911 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1912 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1913 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1914 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1917 const char *p = base64;
1923 unsigned long value;
1925 /* Process first byte of a quadruplet. */
1926 NEXT_BASE64_CHAR (c, p);
1930 return -1; /* illegal '=' while decoding base64 */
1931 value = base64_char_to_value[c] << 18;
1933 /* Process scond byte of a quadruplet. */
1934 NEXT_BASE64_CHAR (c, p);
1936 return -1; /* premature EOF while decoding base64 */
1938 return -1; /* illegal `=' while decoding base64 */
1939 value |= base64_char_to_value[c] << 12;
1942 /* Process third byte of a quadruplet. */
1943 NEXT_BASE64_CHAR (c, p);
1945 return -1; /* premature EOF while decoding base64 */
1949 NEXT_BASE64_CHAR (c, p);
1951 return -1; /* premature EOF while decoding base64 */
1953 return -1; /* padding `=' expected but not found */
1957 value |= base64_char_to_value[c] << 6;
1958 *q++ = 0xff & value >> 8;
1960 /* Process fourth byte of a quadruplet. */
1961 NEXT_BASE64_CHAR (c, p);
1963 return -1; /* premature EOF while decoding base64 */
1967 value |= base64_char_to_value[c];
1968 *q++ = 0xff & value;
1976 #undef NEXT_BASE64_CHAR
1978 /* Simple merge sort for use by stable_sort. Implementation courtesy
1982 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1983 int (*cmpfun) PARAMS ((const void *, const void *)))
1985 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1989 size_t mid = (to + from) / 2;
1990 mergesort_internal (base, temp, size, from, mid, cmpfun);
1991 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
1994 for (k = from; (i <= mid) && (j <= to); k++)
1995 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
1996 memcpy (ELT (temp, k), ELT (base, i++), size);
1998 memcpy (ELT (temp, k), ELT (base, j++), size);
2000 memcpy (ELT (temp, k++), ELT (base, i++), size);
2002 memcpy (ELT (temp, k++), ELT (base, j++), size);
2003 for (k = from; k <= to; k++)
2004 memcpy (ELT (base, k), ELT (temp, k), size);
2009 /* Stable sort with interface exactly like standard library's qsort.
2010 Uses mergesort internally, allocating temporary storage with
2014 stable_sort (void *base, size_t nmemb, size_t size,
2015 int (*cmpfun) PARAMS ((const void *, const void *)))
2019 void *temp = alloca (nmemb * size * sizeof (void *));
2020 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);